New Trends in Medical and Service Robotics: MESROB 2021 (Mechanisms and Machine Science, 106) 3030761460, 9783030761462

Citation preview

Mechanisms and Machine Science

Georg Rauter · Giuseppe Carbone · Philippe C. Cattin · Azhar Zam · Doina Pisla · Robert Riener   Editors

New Trends in Medical and Service Robotics MESROB 2021

Mechanisms and Machine Science Volume 106

Series Editor Marco Ceccarelli , Department of Industrial Engineering, University of Rome Tor Vergata, Roma, Italy Advisory Editors Sunil K. Agrawal, Department of Mechanical Engineering, Columbia University, New York, USA Burkhard Corves, RWTH Aachen University, Aachen, Germany Victor Glazunov, Mechanical Engineering Research Institute, Moscow, Russia Alfonso Hernández, University of the Basque Country, Bilbao, Spain Tian Huang, Tianjin University, Tianjin, China Juan Carlos Jauregui Correa, Universidad Autonoma de Queretaro, Queretaro, Mexico Yukio Takeda, Tokyo Institute of Technology, Tokyo, Japan

This book series establishes a well-defined forum for monographs, edited Books, and proceedings on mechanical engineering with particular emphasis on MMS (Mechanism and Machine Science). The final goal is the publication of research that shows the development of mechanical engineering and particularly MMS in all technical aspects, even in very recent assessments. Published works share an approach by which technical details and formulation are discussed, and discuss modern formalisms with the aim to circulate research and technical achievements for use in professional, research, academic, and teaching activities. This technical approach is an essential characteristic of the series. By discussing technical details and formulations in terms of modern formalisms, the possibility is created not only to show technical developments but also to explain achievements for technical teaching and research activity today and for the future. The book series is intended to collect technical views on developments of the broad field of MMS in a unique frame that can be seen in its totality as an Encyclopaedia of MMS but with the additional purpose of archiving and teaching MMS achievements. Therefore, the book series will be of use not only for researchers and teachers in Mechanical Engineering but also for professionals and students for their formation and future work. The series is promoted under the auspices of International Federation for the Promotion of Mechanism and Machine Science (IFToMM). Prospective authors and editors can contact Mr. Pierpaolo Riva (publishing editor, Springer) at: [email protected] Indexed by SCOPUS and Google Scholar.

More information about this series at https://link.springer.com/bookseries/8779

Georg Rauter Giuseppe Carbone Philippe C. Cattin Azhar Zam Doina Pisla Robert Riener •

•

•

•

•

Editors

New Trends in Medical and Service Robotics MESROB 2021

123

Editors Georg Rauter BIROMED-Lab, DBE University of Basel Allschwil, Switzerland Philippe C. Cattin CIAN, DBE University of Basel Allschwil, Switzerland Doina Pisla Technical University of Cluj-Napoca Cluj-Napoca, Romania

Giuseppe Carbone University of Calabria Rende, Italy Azhar Zam BLOG, DBE University of Basel Allschwil, Switzerland Robert Riener Sensory-Motor Systems Lab ETH Zurich Zurich, Switzerland University Hospital Balgrist Medical Faculty University of Zurich Zurich, Switzerland

ISSN 2211-0984 ISSN 2211-0992 (electronic) Mechanisms and Machine Science ISBN 978-3-030-76146-2 ISBN 978-3-030-76147-9 (eBook) https://doi.org/10.1007/978-3-030-76147-9 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Medical and service robots face growing demands on their functionality and performance in a broad range of applications. Therefore, strengthening our community through interdisciplinary work is beneficial for all parties involved: researchers, technology providers, medical healthcare personnel, and most importantly patients. This and last year, methods from laser physics and virtual/augmented reality-based surgical planning have found their way to augment the functionality, possibilities, and safety of medical and service robots. The last years, we had to face difficult circumstances due to the worldwide pandemic situation with COVID-19 that prevented us from realizing MESROB 2020, the 7th International Workshop on New Trends in Medical and Service Robotics, in Basel, Switzerland. Nevertheless, we published a first series of papers in 2020. In 2021, we could finally organize a successful MESROB 2021 conference in virtual format with and for our faithful community. The entire story of MESROB conference events started with the first of its kind in 2012 in Cluj-Napoca, Romania. Following events were: MESROB 2013 at Institute “Mihailo Pupin” in Belgrade, Serbia; MESROB 2014 at EPFL in Lausanne, Switzerland; MESROB2014 at IRCCyN in Nantes, France; MESROB 2016 co-organized by University of Innsbruck and Joanneum Research in Graz, Austria; MESROB 2018 at the School of Engineering of the University of Cassino and South Latium in Cassino, Italy; and MESROB 2020 at University of Basel, Switzerland. This workshop series is also sponsored by IFToMM, the “International Federation for the Promotion of Mechanism and Machine Science”, and is one of the main conferences for the IFToMM Technical Committees on Biomechanical Engineering, Robotics and Mechatronics, and Computational Kinematics. The content of the MESROB 2021 book covers a wide range of aspects and topics such as: 1) rehabilitation robotics, 2) exoskeletons and prostheses, 3) surgical robotics and micromanipulation, and 4) nursing robotics and human performance evaluation. These contributions are provided as a collection of 17 papers that were selected among the 23 submitted contributions on the basis of a blind peer-review process. The MESROB 2021 book completes the collection of papers submitted to v

vi

Preface

MESROB 2020 (37 papers accepted, 49 submitted). So in total, MESROB in Basel successfully incorporates the published work of 54 accepted papers out of 72 submitted. We wish to express our gratitude to the authors, the reviewers, and Scientific Committee for their valuable contribution to ensure the scientific quality of MESROB 2020 and 2021. Finally, we would like to thank our Gold Sponsor “Stäubli AG” and our Silver Sponsors “F. Hoffmann-La Roche Ltd.”, “Stryker GmbH”, and “Advanced Osteotomy Tools AG”. Georg Rauter Giuseppe Carbone Philippe C. Cattin Azhar Zam Doina Pisla Robert Riener

Organization

General Chair Georg Rauter

University of Basel, Switzerland

Conference Chairs Philippe Cattin Giuseppe Carbone Robert Riener Azhar Zam

University of Basel, Switzerland University of Calabria, Italy ETH Zuerich, Switzerland University of Basel, Switzerland

Program Chairs Georg Rauter Giuseppe Carbone Philippe Cattin Azhar Zam Robert Riener

University of Basel, Switzerland University of Calabria, Italy University of Basel, Switzerland University of Basel, Switzerland ETH Zuerich, Switzerland

Program Committee Daniela Tarnita Jean-Pierre Merlet Med Amine Laribi Giuseppe Carbone Michael Hofbaur Philippe Wenger Yeongmi Kim Akio Yamamoto

University of Craiova, Romania Inria, France University of Poitiers, France University of Calabria, Italy Joanneum Research Forschungsgesellschaft mbH, Austria CNRS-LS2N, France MCI, Austria University of Tokyo, Japan

vii

viii

Carlo Ferraresi Niklaus Friederich Teresa Zielinsks Philippe Cattin Marco Ceccarelli Annika Raatz Bernard Bayle Thekla Brunkert Manfred Husty Domen Novak Georg Rauter Azhar Zam Doina Pisla Paolo Fiorini Irini Giannopulu Nicolas Gerig Hannes Bleuler Marco Ceccarelli Robert Riener Ferda Canbaz Mohamed Bouri

Organization

Politecnico di Torino, Italy DBE University of Basel, Switzerland Warsaw University of Technology, Poland University of Basel, Switzerland University of Rome Tor Vergata, Italy Institut für Montagetechnik, Leibniz Universität Hannover, Germany University of Strasbourg, France University of Basel, Switzerland University Innsbruck, Austria University of Wyoming, USA University of Basel, Switzerland University of Basel, Switzerland Technical University of Cluj-Napoca, Romania University of Verona, Italy iCAM, Bond University, Australia University of Basel, Switzerland EPFL, Switzerland University of Rome Tor Vergata, Italy ETH Zuerich, Switzerland University of Basel, Switzerland EPFL, Switzerland

Organization

ix

Gold Sponsor

High-Precision Robots as Medical and Surgical Assistants Helpful hands redefined: Medical robots are on the advance, not only in the production of drugs but also in the operating theater. If you, as a medical equipment manufacturer, require high-precision, quiet, and flexible robots that also meet all the criteria for sterile surgical conditions, look no further than the Stäubli range. Easy to clean, with minimal particle emissions and high precision, they relieve doctors of some of the strain they are under while performing operations that demand their full concentration. Robots are also contributing to the development of innovative surgical techniques. Jean-Marc Collet Stäubli A. G.

x

Organization

Silver Sponsors

Roche and Laboratory Robots in Research Our scientists are among the best in the industry, pursuing new paths to deliver life-changing benefits to patients. We continue to be driven by our long-term aim to deliver medical advances that provide greater benefits to patients and, at the same time, reduce the healthcare burden on society. Roche End-2-End Lab Services Lab Technologies & Robotics department is providing expert solutions for world class science and has more than 40 technology inventions over the past two decades led to patents. We believe that connecting with many stakeholders across the industry (and even beyond) will support our efforts to dramatically speed up the development and implementation of innovative ideas in the field of robotics and automation for the benefit of patients. Tom Kissling

Advanced Osteotomy Tools and CARLO® AOT is reinventing bone surgery by developing novel digital solutions—like CARLO®, the world’s first Laser Osteotome. CARLO® cuts bone with extraordinary precision in any desired geometry using a cold ablation laser combined with robotics, navigation, and smart software. As the laser allows for contactless procedures, it bypasses any complication risks associated with mechanical instruments and enables surgeons to use functional cuts that create interlocking geometries. After more than a decade of R&D, our innovations are protected by more than ten patent families globally and CARLO® is in clinical use helping patients with its unique benefits. We believe that the combination of laser and robotics is the key to the truly digital OR of the future, where a pre-operative plan can be autonomously executed with perfect accuracy every time. Cyrill Bätscher

Organization

xi

Stryker Digitalization and automation represent nowadays in many aspects of our society not only the standard practice but are also expected and, in many instances, even considered as malpractice if not implemented as an integral structural backbone within the life cycle and at the core of services and products. Digitalization of analog processes has enabled their respective automation yielding more reliable, efficient, and efficacious digital enabled processes. In turn, automation has radically changed the status quo of how tasks get done and certainly the associated outcome expectations. Digitalization within the medical domain is starting to undergo an exponential transformational change which will allow for profound changes in the way medicine is practiced today by harnessing not only patient data but also device and process data in a holistic and continuous manner to establish data-driven processes and methods for efficient and efficacious personalized patient care and treatment delivery across the whole care continuum exploiting automation. This scientific series is an important contribution that will decisively promote further growth and advance the status quo in medical and service robotics. Furthermore, this effort will also help enhance the interdisciplinary nature of the complex solutions needed which will certainly go far beyond the amalgamation of the digital and mechatronics domains to incorporate other technical and socio-economical aspects into future services and products. José-Luis Moctezuma de la Barrera

Contents

Rehabilitation Robotics Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thiago Alves, Rogério Sales Gonçalves, and Giuseppe Carbone

3

A Cable-Robot System for Promoting Healthy Postural Stability and Lower-Limb Biomechanics in Gait Rehabilitation . . . . . . . . . . . . . . Carl A. Nelson

12

Designing a Robotized System for Rehabilitation Taking into Account Anthropological Data of Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Artem Voloshkin, Gregory Dubrovin, Anna Nozdracheva, Larisa Rybak, Santhakumar Mohan, and Giuseppe Carbone Design Optimization and Dynamic Control of a 3-d.O.F. Planar Cable-Driven Parallel Robot for Upper Limb Rehabilitation . . . . . . . . . Ferdaws Ennaiem, Hanen El Golli, Abdelbadiâ Chaker, Med Amine Laribi, Juan Sandoval, Sami Bennour, Abdelfattah Mlika, Lotfi Romdhane, and Saïd Zeghloul Novel Design of the ParReEx-Elbow Parallel Robot for the Rehabilitation of Brachial Monoparesis . . . . . . . . . . . . . . . . . . . . . . . . . Bogdan Gherman, Paul Tucan, Calin Vaida, Giuseppe Carbone, and Doina Pisla

19

27

38

Exoskeletons and Prostheses Development of a New Knee Endoprosthesis and Finite Element Analysis of Contact Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Daniela Tarnita, Dan Calafeteanu, Ilie Dunitru, Alin Petcu, Marius Georgescu, and Danut Nicolae Tarnita

49

xiii

xiv

Contents

Observer Based Sliding Mode Control for a Knee Exoskeleton . . . . . . . Yujie Su, Wuxiang Zhang, and Xilung Ding Design and Motion Simulation of a New Exoskeleton Leg Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ionut Geonea, Cristian Copiluși, Sorin Dumitru, and Adrian Sorin Roșca Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mattia Vincenzo Lo Piccolo, Giovanni Gerardo Muscolo, and Carlo Ferraresi

58

70

78

Surgical Robotics and Micro Manipulation Laser-Induced Breakdown Spectroscopy Combined with Artificial Neural Network for Pre-carbonization Detection in Laserosteotomy . . . Ferda Canbaz, Hamed Abbasi, Yakub A. Bayhaqi, Philippe C. Cattin, and Azhar Zam Development and Evaluation of a Force-Sensitive Flexure-Based Microgripper Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cédric Duverney, Mohamed Ali El Bahi, Nicolas Gerig, Philippe C. Cattin, and Georg Rauter

89

97

Universal Mechanical Interface for Surgical Telemanipulation Using Conventional Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Max B. Schäfer, Gerrit R. Friedrich, and Peter P. Pott Volume Rendering-Based Patient Registration for Extended Reality . . . 115 Marek Żelechowski, Balázs Faludi, Georg Rauter, and Philippe C. Cattin Towards Robotic Surgery for Cartilage Replacement: A Review on Cartilage Defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Philipp Krenn, Manuela Eugster, Esther I. Zoller, Niklaus F. Friederich, and Georg Rauter Nursing Robotics and Human Performance Evaluation Investigating the First Robotic Nurses: Humanoid Robot Nightingale and Partners for COVID-19 Preventive Design . . . . . . . . . . . . . . . . . . . 139 Esyin Chew, Pei Lee Lee, Jiaji Yang, and Shuyang Hu Impact of Ear Occlusion on In-Ear Sounds Generated by Intra-oral Behaviors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Mohammad Khair Nahhas, Nicolas Gerig, Jens Christoph Türp, Philippe Cattin, Elisabeth Wilhelm, and Georg Rauter

Contents

xv

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis of Movement Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 Cinzia Amici, Valter Cappellini, Federica Ragni, Raffaele Formicola, Alberto Borboni, Barbara Piovanelli, Stefano Negrini, and Gabriele Candiani Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

Rehabilitation Robotics

Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks Thiago Alves1(B) , Rog´erio Sales Gon¸calves1 , and Giuseppe Carbone2 1

2

Federal University of Uberlandia, Uberlandia, Brazil [email protected] Department of Mechanical, Energy and Management Engineering, Universita della Calabria, Rende, Italy

Abstract. Rehabilitation training is the most effective way to reduce motor impairments in post-stroke patients. Cable-driven robots have ideal characteristics for stroke rehabilitation and bimanual rehabilitation can transfer training skills to the activities of daily living. However, rehabilitation often presents problems with motivation and patient involvement since the therapy exercises are often monotonous and repetitive. The serious games aim to provide an interactive experience and generate a high level of motivation in patients. Accordingly, this paper presents a serious games approach in combination with a cable-driven robot for unilateral and bilateral/bimanual rehabilitation. Experimental tests are reported with 15 healthy subjects. They use a specifically developed cable-driven robot in combination with a serious game approach for bimanual rehabilitation exercises. The performed tests are described and discussed to show the level of user acceptance and engagement that is achievable with the proposed solution.

1

Introduction

Stroke is a leading cause of disability and it leaves a significant number of individuals with motor and cognitive deficits [1]. The paralysis of the upper limb is the most frequent consequence of brain injury [2]. Rehabilitation training is the most effective way to reduce motor impairments in stroke patients [3]. Rehabilitation movements can be classified as unilateral when using only the affected limb (paretic side) or bilateral when using both sides of the body. Bimanual movements are a specific type of bilateral movements in which there is simultaneous use of both hands in a coupled way [4]. Most rehabilitation therapies, conventional or assisted by technology, focus on the most affected limb, neglecting bimanual activities. Sometimes bimanual training does not yield a superior primary outcome, but it shows benefits such as increased daily use of the paretic side and recovery from other activities [2,5,6]. Robotics is attracting significant interest with novel rehabilitation assistive solutions. Several innovative designs are proposed, for example, the cable driven c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 3–11, 2022. https://doi.org/10.1007/978-3-030-76147-9_1

4

T. Alves et al.

robotic solutions. This design solution can be easily portable, and feasible for self-training at home for patients with stroke [7]. An effective treatment of stroke should be early, intensive, and repetitive, which can lead to problems with motivation and patient involvement since conventional rehabilitation therapy is often monotonous and repetitive. A new paradigm is emerging in the field of rehabilitation, characterized by the systematic use of computer games, called serious games. Serious games aim to provide a truly interactive experience, generating a high level of motivation in patients and softening the notion of effort during the exercises [8]. In this way, robotic therapy with serious games is often more intensive, more encouraging, and more motivating for patients than it is possible with regular therapy [8–12]. Among the available robot architectures, cable-driven robots have several promising features including a lightweight structure with small moving parts, and a relatively large workspace. They can be intrinsically safe due to cable flexibility, allowing safe manipulation close to humans. Additionally, the mechanical structure is easy to assemble/disassemble and be reconfigured to perform different therapies. They have low cost and simple maintenance [10], which are an important characteristic of self-administrated treatment by the patient at home as reported for example in [7]. This paper presents a cable-driven robot (CAR) to be applied in stroke rehabilitation and serious games developed to it. First, the design of the CAR for unilateral and bimanual rehabilitation is introduced. Then the serious games developed are described. Finally, the tests with healthy subjects are presented.

2

Design of the Cable-Driven Robot

The Cable-Actuated Robot (CAR) is a up to 2 degrees of freedom (DOF) cabledriven robotic structure with an end-effector, handle/strap or handlebar depending on the setup, Fig. 1. The structure consists of an aluminum fixed frame, DC motors, load cells, and rotary encoders. The control is a position proportional and it is performed by a microcontroller and a full bridge motor driver. The results of the validation performed with this device showed high accuracy and repeatability levels [11]. This device has 3 main possible setups: (I) unilateral with 1 DOF, (II) unilateral with 2 DOF and (III) bilateral/bimanual with 2 DOF. In setup I, Fig. 1(a), this device is used unilaterally and the movement possible is only translation over y-axis. In setup II, Fig. 1(b), the movements performed are translation over x-axis and translation over y-axis, which combined allow for planar trajectories. In setup III, Fig. 1(c), this device is used bilaterally, and there is coupling between the two sides of the human body for bimanual therapies. The movements that can be performed in this setup are rotation over z-axis and translation over y-axis. For rotational movements it is necessary to decrease the length of one cable while increasing the length of the other. Decreasing the left cable length and increasing the right one the bar rotates clockwise (CW). Similarly, decreasing the right and increasing the left one the bar rotates counterclockwise (CCW).

Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks

5

Fig. 1. CAR setups: (a) 1 DOF unilateral, (b) 2 DOF unilateral; (c) 2 DOF bimanual

3

Serious Games Development for Rehabilitation

Serious games motivate the patient to perform the rehabilitation exercises in a more attractive and pleasant way. These games can provide an environment of competitiveness, where visual and auditory rewards, as well as scores, can be provided indicating the patient’s progress [9,13]. Patients who experienced serious game in stroke rehabilitation had similar efficacy in improving arm function compared to conventional therapy and reported greater interest and fun during training [6,14,15]. Several serious games were developed for the CAR device for its different setups and will be shown in this section. The Rehab Basketball, Fig. 2, was designed for the CAR in setup I. This game is very simple and was completely developed in MATLAB for initial validation. The movement amplitude can be changed by the therapist in each session according to the patient capacity. The game goal is to take the ball to the basket by performing vertical movements. The force measured by the load cell is shown in an indicator, alerting unsatisfactory participation for the patient, and increasing score by a lower value. This game also applies an assist-as-needed control where the device can assist the patient make the movement if participation is not enough. The Square Apple game, Fig. 3(a), was designed for the CAR in setup II. This game has also been fully developed in MATLAB. Square Apple follows the characteristics of a game with motion/virtual reality control, where a hand is moved on a plane to pick up apples and take them to a basket, using as input the user’s movements up/down and left/right on the effector/handle. Each session of the game is terminated when the user picks up all the apples present in this and deposits them in the basket. The Grabbing Apple is an update from the previous game to the Unity platform, allowing for interface enhancement and visual effects. Also, in a variant

6

T. Alves et al.

of this game, Fig. 3(b), apples appear individually and in sequence, allowing the patient to focus on only one apple at time and also allowing assist-as-needed control to be applied.

Fig. 2. Main interface of the game Rehab Basketball

Fig. 3. Main screen of the games: (a) Square Apple and (b) Grabbing Apple

The motor functions of stroke patients are often related and evaluated by circular design/tracing tasks [15,16]. The Round Pizza game, Fig. 4, was also designed for setup II, but it uses a circular trajectory. In this game a cutter is moved in order to cut the pizza dough in the roundest possible way, using as input the patient’s movements (up/down and left/right) in the handle in order to perform a circular motion. After the game is over, the circularity pattern can be analyzed. The “size” of the pizza can be changed generating different levels of difficulty/workspace area. The patient’s evolution can be followed by comparing the ideal circle with the pattern of the circle “drawn” in each session. Preliminary tests performed with the circular trajectory achieved an average error of 0.3489 cm (2.33%) indicating a good accuracy [17]. The Motorcycle game, Fig. 5, was designed for the CAR in setup III. Motorcycle follows the characteristics of a motion/virtual reality control game, in which a motorcycle must be driven down a track dodging obstacles and advancing

Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks

7

on its route. This game uses as input the rotation movements of the bimanual (CW/CCW) performed by the user on the handlebars of the device, Fig. 1(c). In this way, the CW/CCW movements performed on the device handlebar correspond, to the bike turning right/left in game, Figs. 5(b) and (c) respectively.

Fig. 4. Round Pizza game screen for circular trajectory

Fig. 5. Motorcycle game screen showing directions: (a) straight, (b) right, and (c) left

The MineCart game, Fig. 6, is also designed to be used with the bimanual CAR setup. In this game a wagon must be moved right/left to pick up crystals falling from the ceiling in a mine cave and uses as input the rotation movements of the bimanual (CW/CCW). In this way the handlebar clockwise movement corresponds to the right wagon movement and the counterclockwise to the left. In these games, the forces generated with each arm are measured independently by load cells. Subjects are encouraged to use both upper limbs. If he tries to use just the less impaired arm, a warning (movement hampered proportional to less impaired arm use) reminds him not to do so by disrupting his ability to easily rotate the handlebar and leading to awareness of strong arm use. This prevents the learned non-use, when patients no longer use the weaker side and concentrate their efforts on the nonparetic side. As the aim is the opposite, the patient is forced to also use the opposite arm to rotate the bar. The score of the game MineCart is proportional to the number of crystals caught. A hidden score for the player/patient counts the lost crystals and the

8

T. Alves et al.

Fig. 6. MineCart game screen

balance of the last five crystals (caught/lost). Based on this, is determined if assistance is required and if so, its level. The assistance acts by initially facilitating and, after, even performing the movement by itself at the crystal direction, according to the assist level. Additionally, movements at the wrong direction are more difficult to perform.

4

Results

In order to perform a clinical trial, this project was submitted, and approved, to the Research Ethics Committee (REC) of the Federal University of Uberlˆandia under the CAAE number 00914818.5.0000.5152. The subjects were 15 healthy participants, 11 males and 4 females; and were, 19 to 30 years old; average of 22.4 ± 2.8 years old. These participants performed upper limb exercises using the MineCart game. Each participant should achieve an average score of 200 points (approximately 200 repetitions). After the tests with healthy participants using the MineCart game, motivation was measured with an Intrinsic Motivation Inventory (IMI). IMI [18] is a multidimensional questionnaire that provides qualitative information on the content and level of motivation a participant experience during an intervention. It is scored on a 7-point Likert scale, ranging from “nothing true” to “very true” (“strongly disagree/agree”). The indices in each IMI category performed with healthy subjects are presented in Fig. 7. A Game Engagement/Experience Questionnaire (GEQ) [19] was applied to capture the player’s experience based on various items such as fun, frustration, and challenge. In the GEQ performed, the average score (1 to 5) for “Interface” was 4.4, “Gameplay” and “Fun” was 4.2. The game difficulty was considered average (2.9). The game was not considered “Frustrating” (1.6) and the overall score was 4.5.

5

Discussion

In the healthy subjects IMI score close to 6 were observed in the categories Interest/Satisfaction (5.9 ± 0.9), Effort/Importance (5.8 ± 0.8), Value/Usefulness

Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks

9

Fig. 7. Intrinsic Motivation Inventory for healthy subjects

(6.1 ± 0.8) and Trust/Relationship (5.9 ± 1.2), indicating a high approval degree. The Perceived Competence (5.1 ± 0.6) indicated that participants felt competent/skilled to a moderate level when performing the activity. Pressure/Tension (2.8 ± 1.3) was a negative predictor and indicated that they were not nervous, tense or anxious during the exercises.

6

Conclusion

Rehabilitation training is the most effective way to reduce motor impairments in stroke patients. The use of robotic devices like cable-driven robots allows a greater intensity of rehabilitation therapies. Training with serious games is as effective as conventional therapy in improving the arm’s function when applied at the same intensity. Besides that, patients using serious games reported positive experiences such as greater interest and fun. This paper presented a 2 DOF cable-driven robot for stroke rehabilitation which allow bimanual training and therefore can improve the performance in ADL. Several serious games were developed for this device for its unilateral and bimanual setup. A test made with 15 healthy subjects showed a high approval degree and positive motivation in an IMI applied. These tests consisted in robotic device exercises with a specific game developed (MineCart). Thus, using the robotic rehabilitation and the serious games we expect that patients can perform training at home, or even in the rehabilitation clinic, without relying on the therapist’s availability, possibly allowing an increased training dose in relation to supervised treatment and, consequently, a greater improvement in dexterity. Acknowledgments. The authors thank UFU, FAPEMIG, CNPQ, and CAPES Finance Code 001 for the partial financial support to this work.

10

T. Alves et al.

References 1. WHO: World Health Organization (2021). https://www.who.int/. Accessed 3 Feb 2021 2. Tappeiner, L., Ottaviano, E., Husty, M.L.: A cable-driven robot for upper limb rehabilitation inspired by the mirror therapy. In: Zeghloul, S., Romdhane, L., Laribi, M.A. (eds.) Computational Kinematics. MMS, vol. 50, pp. 174–181. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-60867-9 20 3. Hatem, S.M., Saussez, G., Della Faille, M., et al.: Rehabilitation of motor function after stroke: a multiple systematic review focused on techniques to stimulate upper extremity recovery. Front. Hum. Neurosci. 10, 1–22 (2016) 4. Malabet, H.G., et al.: Symmetric motions for bimanual rehabilitation. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, Taipei, Taiwan, pp. 5133–5138 (2010) 5. Eggers, O.: Occupational Therapy in the Treatment of Adult Hemiplegia. Heinemann, London (1984) 6. Cardoso, L.R.L., et al.: Upper limb rehabilitation through bicycle controlling. In: 24th ABCM International Congress of Mechanical Engineering, Brazil (2017) 7. Nijenhuis, S.M., et al.: Feasibility study into self-administered training at home using an arm and hand device with motivational gaming environment in chronic stroke. J. Neuroeng. Rehabil. 12, 89 (2015) 8. Proen¸ca, J.P., Quqresma, C., Vieira, P.: Serious games for upper limb rehabilitation: a systematic review. Disabil. Rehabil.: Assistive Technol. 13(1), 95–100 (2018) 9. Appel, V.C.R., et al.: Classifying emotions in rehabilitation robotics based on facial skin temperature. In: 5th IEEE RAS EMBS International Conference BioRob, p. 276 (2014) 10. Gon¸calves, R.S., Carvalho, J.C.M., Ribeiro, J.F., Salim, V.V.: Cable-driven robot for upper and lower limbs rehabilitation. In: Handbook of Research on Advancements in Robotics and Mechatronics, 1st edn., pp. 284–315. IGI Global (2015) 11. Alves, T., et al.: Assist-as-needed control in a cable-actuated robot for human joints rehabilitation. J. Mech. Eng. Biomech. 3, 57–62 (2019) 12. Hern´ andez-Mart´ınez, E.E., Ceccarelli, M., Carbone, G., L´ opez-Caj´ un, C.S., J´ auregui-Correa, J.C.: Characterization of a cable-based parallel mechanism for measurement purposes. Mech. Based Des. Struct. Mach. 38(1), 25–49 (2010) 13. Caurin, G.A., et al.: Adaptative strategy for multi-user robotic rehabilitation games. In: IEEE Engineering in Medicine and Biology Society, Boston, pp. 1395– 1398 (2011) 14. Kottink, A.I.R., Prange, G.B.K., et al.: Gaming and conventional exercises for improvement of arm function after stroke: a randomized controlled pilot study. Games Health J. Res. Dev. Clin. Appl. 3(3), 184–191 (2014) 15. Prange, G.B., Kottink, A.I.R., Buurke, J.H., et al.: The effect of arm support combined with rehabilitation games on upper-extremity function in subacute stroke: a randomized controlled trial. Neurorehabil. Neural Repair 29(2), 174–182 (2015) 16. Dipietro, L., Krebs, H.I., Fasoli, S.E., et al.: Changing motor synergies in chronic stroke. J. Neurophysiol. 98(2), 757–768 (2007) 17. Alves, T., et al.: Cable-driven robots for circular trajectory exercises in rehabilitation. In: 25th ABCM International Congress of Mechanical Engineering, Uberlˆ andia, Brazil (2019)

Serious Games Strategies with Cable-Driven Robots for Rehabilitation Tasks

11

18. IMI: Intrinsic Motivation Inventory (IMI) (2018). http://selfdeterminationtheory. org/intrinsic-motivation-inventory. Accessed 13 Feb 2020 19. IJsselsteijn, W.A., et al.: Measuring the experience of digital game enjoyment. In: 6th MB Conference, Maastricht, Netherlands, pp. 88–89 (2008)

A Cable-Robot System for Promoting Healthy Postural Stability and Lower-Limb Biomechanics in Gait Rehabilitation Carl A. Nelson(B) Department of Mechanical and Materials Engineering, University of Nebraska-Lincoln, Lincoln, NE, USA [email protected]

Abstract. A number of cable-based systems have been developed for gait rehabilitation, but one aspect which remains relatively under-addressed is medial-lateral postural stability, which can be exacerbated when using certain types of gait rehabilitation equipment. In this paper, a cable system is presented for promoting healthy medial-lateral postural stability patterns and integrating with existing gait rehabilitation devices, as well as providing direct assistance at the joints of the lower limb. Keywords: Gait rehabilitation · Postural stability · Cable robot

1 Introduction Postural stability is a critical component of healthy gait. In normal walking, the center of mass oscillates medial-laterally as weight bearing shifts from one foot to the other [1], and descriptive statistics on sway data captured over multiple steps can be used to characterize whether gait is normal or pathological in this regard. In the context of rehabilitation, force stimuli (medial-lateral or otherwise) can be used to either perturb (challenge) postural sway during gait, or to guide an individual towards a nominal healthy sway pattern. Robotic devices have been used in various aspects of gait rehabilitation, including gait assessment [2] and gait therapy [3]. Examples include exoskeletons [4], soft exo-suits [5], body-weight support systems [6, 7], advanced treadmills [8], and end-effector-type robotic systems [9]. Cable-driven robotic systems [10] are of particular interest due to their potential for low inertia, high payload, relative simplicity, and reconfigurability. The use of cable systems in lower extremity rehabilitation has mainly focused on actuation of the leg segments [11–17], rather than postural intervention involving the trunk; some of these [11–13] are restricted to sagittal-plane motion. Only rarely [18] does one see whole-body cable systems, although in some cases cable systems are used for balance perturbation (challenge) tasks [2]. Although a cable-based platform for both measurement and stimulus of postural stability in both anterior-posterior and mediallateral directions has recently been developed [19], it is designed for sitting and standing rather than for walking. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 12–18, 2022. https://doi.org/10.1007/978-3-030-76147-9_2

A Cable-Robot System for Promoting Healthy Postural Stability

13

The issue of medial-lateral sway control is particularly important when using endeffector systems such as the elliptical-based ICARE (Intelligently Controlled Assistive Rehabilitation Elliptical) [20]. In contrast to treadmill or over-ground walking, these systems maintain a constant state of double-foot contact. Although the effect is reduced when coupled with body-weight support systems, there can be a tendency to not fully shift one’s weight medial-laterally as would be the case during normal gait. This, in turn, can affect the amount of heel lift during late stance, altering the subsequent joint angles in the knee and hip and disrupting the learning of normal joint angle patterns (see Fig. 1).

Fig. 1. ICARE system, and ankle angle deficiency with respect to normative data from an OpenSim model of normal walking.

To promote healthy patterns of postural sway during gait rehabilitation and encourage proper joint angle biomechanics in the lower limb, in this paper we present a cablebased system that provides stimulus to the trunk (medial-laterally) as well as the knee and ankle (sagittal-plane motions). We place special emphasis on synchronization of the medial-lateral stimulus with gait frequency, simplicity of the overall layout, and avoiding over-constraint. However, the system is also easily adapted for measurement and/or stimulus in the anterior/posterior direction (e.g., for simulating start/stop of gait).

2 Methods The design presented here combines elements of different varieties of cable systems for rehabilitation. Specifically, for sagittal-plane limb motions, we adopt the exo-suit approach [5] with Bowden-type cable actuation routed along the limbs, and for trunk motion we use an architecture more reminiscent of a cable-driven planar parallel robot. Notably, we are not interested in fully constraining the pose of the leg (or of any body part), but rather in providing corrective forces to encourage improved gait in specific areas that tend to be problematic. The general layout is shown in Fig. 2. The structure is designed to straddle an existing gait device such as the ICARE. There is one pair of cables for medial-lateral stimulus, a separate set of cables along the anterior/posterior direction for knee and/or ankle stimulus, and a waist harness to which these all attach. The medial-lateral cables meet at

14

C. A. Nelson

Fig. 2. Structural layout of cable system: top view (left) and front view (right). Cables are shown as dashed lines, with the anterior-posterior cables in green and the medial-lateral cables in red. Dark gray represents structure, whereas light gray indicates where structure could be absent for patient access. The medial-lateral actuator is shown at a hypothetical location on the frame in line with the cable routing. (Pulleys for cable routing not shown).

Fig. 3. Harness layout (top), showing posterior (P) cable routing to lower limbs, and adjustable attachment points for anterior (A) and medial-lateral (ML) cables as a function of the patient’s girth. Cable tensions applied to the harness (bottom and right), showing that floating actuators allow equilibrium in the A-P direction, while the fixed ML actuator unit produces net force |T3 -T4 | in the ML direction (the values of T3 and T4 being dependent on the relative position of the patient and on cable tensioner properties).

an actuation unit to form a closed loop (harness + cables + actuator), and each anterior cable is paired with a posterior joint-input cable through an actuator to form its own closed loop in a similar fashion. Whereas the medial-lateral actuator is fixed such that the trunk motion is stimulated with respect to a fixed reference frame, the anterior/posterior cable pair actuators “float” to permit a small amount of forward/backward motion during gait. This also establishes force equilibrium in the anterior/posterior direction without the need for sensing and control (see Fig. 3). Of course, if net force stimulus (e.g., perturbation for simulating start/stop of gait) is desired, a cable loop can be added along the anterior/posterior alignment with a fixed actuator to control this functionality. Each

A Cable-Robot System for Promoting Healthy Postural Stability

15

cable passes through a tensioner to provide mechanical compliance in the system for safety. The end-effector type rehabilitation systems with which this device is designed to be used [9, 20] typically generate a cyclic motion; for example, the ICARE system [20] has a continuously rotating crankshaft in the rear of the elliptical mechanism, and one full rotation of the crankshaft corresponds to one full gait cycle. To drive and synchronize the medial-lateral stimulus with the gait cycle, this crankshaft (or equivalent, dependent on the specifics of the gait rehabilitation system) is coupled to the actuation unit of the medial-lateral cable loop. To convert this continuous rotation to cyclic oscillation of the medial-lateral cables, the actuation unit consists of a Cardan gear with a Scotch yoke mechanism (see Fig. 4). Either the Cardan gear or the Scotch yoke can convert continuous rotation to oscillating straight-line motion. By using the Cardan gear for rotational-to-linear conversion and controlling the orientation of the Cardan gear mechanism with respect to the Scotch yoke, this provides an adjustable-stroke rotational-to-linear oscillatory motion conversion decomposed along the line of cable action, requiring no additional actuator for medial-lateral stimulus.

Fig. 4. Representation of a Cardan gear (left) and Cardan gear with Scotch yoke (right). The gear carrier is shown as a heavy line, and the planet is half the size of the ring gear.

3 Results Referring to the notation shown in Fig. 4, the line of the carrier link and the radius of the planet gear passing through the special trace point form an isosceles triangle such that the trace-point displacement is s = (2r)cosθ = Rcosθ

(1)

where the planet radius r is half the ring gear radius R, and θ is the angular position of the carrier input. Rotating the ring gear by an angle α causes the displacement to be projected onto the line of motion of the Scotch yoke: s = Rcosθcosα

(2)

16

C. A. Nelson

with the input θ coupled to the crankshaft of the gait device, the medial-lateral motion is synchronized with the gait cycle and retains a sinusoidal form. The amount of sway during normal walking can vary in the range of approximately + /−5 to 10 cm [21], so this dictates a ring gear radius of at least 10 cm. Since the amplitude is s max = Rcosα

(3)

the therapist/clinician can adjust the input as needed for a specific patient by setting the orientation α of the ring gear (e.g., manually, or through a motorized spur gear coupling, worm gear coupling, etc.). Although this medial-lateral actuation unit is shown in Fig. 2 as being located on the front or rear of the structure, with a system such as the ICARE it may make more sense to mount it such that the gear carrier can be directly coupled to the crankshaft (i.e., on the side of the gait device); mounting can vary depending on the gait device with which this cable system is used, but using modular extrusion framing in its construction can facilitate flexible placement of components such as this actuation unit, the various pulleys, etc. Since the system is cable-based, these design details are inherently more flexible. The cable tensions during actuation depend on the extent to which the cable input motion is matched by output motion of the harness worn by the patient (see Fig. 3). In a worst-case scenario, in which the patient is unable to move at all or a cable is snagged, the cable tensioners should accommodate the full range of cable displacement (i.e., + /−s max ). This allows sizing of the tensioner. As illustrated in Fig. 5, a standard two-pulley tensioner with a torsion spring can achieve a net cable displacement of twice the center distance (2c) by undergoing a rotation φ of 180°. Therefore, the center distance should be at least half the range of required displacement, or c = R (since s max = R when α = 0), or at least 10 cm as noted previously. The spring can be selected based on the moments typically produced in the frontal plane during walking, which are on the order of 100 Nm [1]. Approximating the height at which horizontal cable loads are applied to the waist harness as 1 m, cable tensions should be at least 100 N. Considering a torsion spring without pretension (zero moment at φ = 0), and assuming small pulley-size effects, the approximate tension relationship is kφ = cTsinφ

(4)

Then, with c = 0.1 m, an approximate range of torsion spring stiffness to achieve cable tensions up to 100 N is 0.5 < k < 1 Nm. Such components are readily available.

4 Discussion Since exo-suit design is established in other literature [5], we do not elaborate here on the details of the actuation in the anterior-posterior direction. However, standard actuators and similar cable tensioning methods can be used. Since postural moments in the sagittal plane are smaller in magnitude than those in the frontal plane [1], identical tensioners can be used throughout the system if ankle/knee joint torques are not required; otherwise, the anterior-posterior tensioners would need a larger stiffness to accommodate the somewhat larger tension range required for driving these joints.

A Cable-Robot System for Promoting Healthy Postural Stability

17

25

20

Tc/k

15

10

5

0 0

0.5

1

1.5

2

2.5

3

3.5

phi (radians)

Fig. 5. Cable tensioner and tension relationship.

Although Eqs. (1–3) address synchronization of the medial-lateral stimulus with the gait cycle in elliptical or similar gait devices, the cable system could also be used with treadmills or other devices in which the motion is continuous rather than cyclic. In this case, rather than coupling mechanically to the gait device, the input to the medial-lateral synchronizer (Fig. 4) would need to be driven by a separate motor based on gait events determined through sensing.

5 Conclusion A cable robot design has been presented which allows exo-suit-style assistance to the joint of the lower limb while promoting normal medial-lateral postural sway during walking on stationary gait rehabilitation devices. This design approach differs from the classical fully-constrained parallel rehabilitation robot design in several respects, most notably in its intended use as an accessory to stationary rehabilitation equipment rather than as the primary driver of limb motion. Elements of novelty of this system include the inclusion of medial-lateral stimulus, a variable-stroke synchronization mechanism for driving the medial-lateral stimulus in alignment with the gait cycle, and the blend of a planar parallel cable robot for trunk motion with exo-suit elements for the lower limb. Future work will include testing and validation of the system using a prototype with human subjects. Acknowledgments. The author appreciates help from Zvonimir Pusnik with creation of figures.

References 1. Winter, D.A.: Human balance and posture control during standing and walking. Gait Posture 3, 193–214 (1995) 2. Shirota, C., van Asseldonk, E., Matjaˇci´c, Z., et al.: Robot-supported assessment of balance in standing and walking. J. Neuroeng. Rehabil. 14, 80 (2017)

18

C. A. Nelson

3. Hobbs, B., Artemiadis, P.: A review of robot-assisted lower-limb stroke therapy: unexplored paths and future directions in gait rehabilitation. Front. Neurorobot. 14, 19 (2020) 4. Murray, S., Ha, K., Hartigan, C., Goldfarb, M.: An assistive control approach for a lowerlimb exoskeleton to facilitate recovery of walking following stroke. IEEE Trans. Neural Syst. Rehabil. Eng. 23(3), 441–449 (2015) 5. Panizzolo, F.A., Galiana, I., Asbeck, A.T., et al.: A biologically-inspired multi-joint soft exosuit that can reduce the energy cost of loaded walking. J. Neuroeng. Rehabil. 13(1), 1–14 (2016) 6. Hidler, J., Brennan, D., Black, I., Nichols, D., Brady, K., Nef, T.: ZeroG: overground gait and balance training system. J. Rehabil. Res. Dev. 48(4), 287–298 (2011) 7. Vallery, H., et al.: Multidirectional transparent support for overground gait training. In: 2013 IEEE 13th International Conference on Rehabilitation Robotics (ICORR), Seattle, WA, USA, pp. 1–7 (2013) 8. Cattin, E., et al.: SENLY: a novel robotic platform for fall risk prevention. In: IEEE International Conference on Robotics and Automation (Workshop Paper) (2010) 9. Stolle, C.J., Nelson, C.A., Burnfield, J.M., Buster, T.W.: Improved design of a gait rehabilitation robot. In: Husty, M., Hofbaur, M. (eds.) MESROB 2016. MMS, vol. 48, pp. 31–44. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-59972-4_3 10. Xiong, H., Diao, X.: A review of cable-driven rehabilitation devices. Disabil. Rehabil.: Assist. Technol. 15(8), 885–897 (2020) 11. Wu, M., Hornby, T.G., Landry, J.M., Roth, H., Schmit, B.D.: A cable-driven locomotor training system for restoration of gait in human SCI. Gait Posture 33, 256–260 (2011) 12. Gharatappeh, S., Abbasnejad, G., Yoon, J., Lee, H.: Control of cable-driven parallel robot for gait rehabilitation. In: 12th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), Goyang, South Korea, 28–30 October 2015, pp. 377–381 (2015) 13. Jin, X., Cui, X., Agrawal, S.K.: Design of a cable-driven active leg exoskeleton (C-ALEX) and gait training experiments with human subjects. In: 2015 IEEE International Conference on Robotics and Automation (ICRA), Seattle, WA, pp. 5578–5583 (2015) 14. Alamdari, A., Krovi, V.N.: Robotic Physical Exercise and System (ROPES): a cable-driven robotic rehabilitation system for lower-extremity motor therapy. In: 2015 ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC-CIE), Paper Number DETC2015-46393 (2015) 15. Bennour, S., Harshe, M., Romdhane, L., Merlet, J.-P.: A new experimental setup based on a parallel cable robot for analysis and control of human motion. Comput. Methods Biomech. Biomed. Eng. 14(Supplement 1), 83–85 (2011) 16. Harshe, M., Merlet, J.-P., Daney, D., Bennour, S.: A multi-sensors system for human motion measurement: preliminary setup. In: 13th IFToMM World Congress, Guanajuato, Mexico (2011) 17. Lamine, H., Romdhane, L., Bennour, S.: Parametric dynamic analysis of walking within a cable-based gait trainer. Robotica 37, 1225–1239 (2019) 18. Surdilovic, D., Bernhardt, R.: STRING-MAN: a new wire robot for gait rehabilitation. In: 2004 IEEE International Conference on Robotics and Automation (ICRA), New Orleans, LA, USA, 28–30 April 2004, pp. 2031–2036 (2004) 19. Khan, M.I.: Trunk rehabilitation using cable-driven robotic systems. PhD dissertation, Columbia University (2019) 20. Nelson, C.A., Burnfield, J.M., Shu, Y., Buster, T.W., Taylor, A.P., Graham, A.: Modified elliptical machine motor-drive design for assistive gait rehabilitation. ASME J. Med. Devices 5(2), 021001 (2011) 21. Wang, F., Skubic, M., Abbott, C., Keller, J.M.: 32nd Annual International Conference of the IEEE EMBS, Buenos Aires, Argentina, 31 August–4 September 2010, pp. 2225–2229 (2010)

Designing a Robotized System for Rehabilitation Taking into Account Anthropological Data of Patients Artem Voloshkin1(B) , Gregory Dubrovin2 , Anna Nozdracheva3 Larisa Rybak1 , Santhakumar Mohan4 , and Giuseppe Carbone5

,

1 Belgorod State Technological University Named After V.G. Shukhov, Belgorod, Russia

[email protected] 2 Kursk State Medical University, Kursk, Russia 3 Gamaleya National Research Center for Epidemiology and Microbiology, Moscow, Russia 4 Indian Institute of Technology Palakkad, Palakkad, India 5 University of Calabria, 87036 Rende, Calabria, Italy

Abstract. The paper discusses a robotic system (RS) for the rehabilitation of the lower limbs of patients. The system consists of an active parallel 3-PRRR robot and a passive part of the RRRR orthoses. A 3D model of a passive RRRR orthosis’s joint structure was developed, taking into account patients’ fit with different anthropological data. The requirements for the body mass index of patients are formulated to optimize the design parameters. This made it possible to determine the rehabilitation system’s workspace, taking into account the minimum and maximum regulation of the passive orthosis joints based on optimization methods. The simulation results are presented. Keywords: Mechanotherapy · Parallel robots · Orthosis position · Kinematic analysis

1 Introduction Robotic mechanotherapy is a relatively new direction in the rehabilitation of patients. Such systems make it possible to ensure the correct execution of the exercise and carry out movements with feedback. Depending on the therapist’s settings, the system either provides an auxiliary force or provides resistance. Such systems have the problem that they must either be adaptable to each patient or be represented by a wide range of names. Various companies solve this problem based on the direction of rehabilitation [1–3]. It is possible to create a system suitable for every patient, but the cost of development and production will not allow for mass use. Erigo, Lokomat, Lokohelp, Rehabot, GaitTrainer, Lopes, and other robotic devices are used to restore the entire lower limb’s locomotive function. The main movements of the lower limb are carried out in the sagittal plane. However, the possibility of additional rotational and transverse movements for the hip joint has a synergistic effect on the restoration of movement in the main plane. Training apparatus) for passive joint development (Continues Passive Motion - CPM therapy) © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 19–26, 2022. https://doi.org/10.1007/978-3-030-76147-9_3

20

A. Voloshkin et al.

can be divided into devices for one specific joint and mechanical therapy of the entire limb. A statically balanced passive orthosis model should synchronize movements in the hip, knee, and ankle joints in a given volume. Depending on the model, the range of motion may differ, but on average, it has the following meanings: for the hip joint, passive flexion-extension within 100°/0/0°, in the knee joint - 125°/0/0°, in the ankle joint - 50°/0/30° [4].

2 Development of a Passive Orthosis Model Work [5] describes the process of developing a rehabilitation system using a 3-PRRR robot (Fig. 1), only taking into account the patient’s height, without taking into account the bodily constitution. This system has an active mechanism of a parallel structure with three kinematic chains united by a central platform (end effector) and a passive orthosis with successively located joints in the hip, knee, and ankle joints. Rehabilitation requires precise positioning of the orthosis pivots relative to the patient’s joint. Therefore, the patient’s fit and restraint system must be carefully studied. Let us consider the scheme for determining the patient’s movements by the method of neutral zero position. Each joint has its own neutral position (zero position) due to maximum muscle relaxation. Three numbers characterize movement in the joint in one of the normal planes, denoting the extreme points of the amplitude of movement and the zero position between them, i.e., neutral position. The zero position for the hip joint is the position of the thigh in the frontal plane parallel to the axis of the body with the patella facing forward. The limb rotation axis in the hip joint passes through the femoral head center (Fig. 2).

Fig. 1. 3D model of a robotic system for rehabilitation: 1 - active 3-PRRR mechanism, 2 - passive RRRR orthosis.

In rehabilitation tasks, it is important to ensure that the orthosis rotational joints and the patient’s joints’ axes coincide. However, the hip joint has three axes of rotation: lateral and medial rotation occurs relative to the vertical axis, extension and flexion occurs relative to the transverse axis, and abduction and adduction occurs relative to the sagittal axis. In our case, the rotational joints will be installed in the transverse and sagittal axes, and the position of the limb relative to the vertical axis will be fixed with an orthosis. Since the axes pass through the hip joint center, the joints are installed on a rigid frame (Fig. 3). The intersection point of the transverse and sagittal axes should coincide

Designing a Robotized System for Rehabilitation

21

Fig. 2. Axes of rotation of the hip joint.

with the center of the hip joint. A sagittal adjustable part and an adjustable transverse part are provided to achieve this task, changing the lengths of which allows you to accurately determine the correct position of the orthosis relative to the patient, as shown in Fig. 4. During rehabilitation, it is necessary to take into account the anthropometric data of various patients. This becomes possible due to the movement of the seat relative to the X axis and a change in the height of the orthosis mount relative to the Z axis. As a result, optimal positioning of the orthosis relative to the patient’s limb is ensured.

a

b

Fig. 3. Passive RRRR orthosis: a) 1 - transverse joint, 2 - transverse axis, 3 - sagittal adjustable part, 4 - joint, 5 - sagittal axis, 6 - adjustable transverse part, 7 - adjustable frame: b) scheme of positioning the orthosis relative to the patient: 1 - transverse joint, 2 - transverse axis, 3 - sagittal adjustable part, 4 - sagittal plane joint.

Positioning the orthosis relative to the patient is important, but changing the attachment point changes the parallel mechanism’s workspace. The area of regulation must be limited under the anthropometric data of the patients. We will restrict them according to the Body Mass Index (BMI) from 16 kg/m2 to 35 kg/m2 . BMI is calculated using the formula BMI = mh · h mkg2 , where m is body weight in kilograms, h is height in meters. Since the BMI of patients who can undergo rehabilitation on this system, taking into account the height from 160 to 190 cm, is known, it becomes possible to determine the adjustment range of the orthosis attachment point to ensure a free fit of patients with an intertrochanteric distance from 26 to 40. Therefore, the minimum distance from the chair’s center to the transverse joint will be 13 cm, and the maximum distance, taking into account the error, is assumed to be 25 cm. Therefore, the adjustment range will be

22

A. Voloshkin et al.

12 cm. Based on these parameters of adjusting the orthosis joints’ point of attachment, the workspace of the end effector of the 3-PRRR robot changes. The method of nonuniform coverings was applied to determine the required workspace [6]. Let us compose a system of equations for the connection of the passive mechanism, describing the position of the point P - the center of the rotational joint of the ankle joint, depending on the angles of rotation ψ and α in the hip joint and the angle of rotation γ = 180 − θ in the knee joint (Fig. 4).

Fig. 4. Design diagram of a passive RRRR orthosis.

The coordinates of the center of the joint P are as follows:   ⎧ ⎨ xP = sinψ Lthing  cosα + Lcrus cosβ  y = −cosψ Lthing cosα + Lcrus cosβ ⎩ P zP = −Lthing sinα − Lcrus sinβ

(1)

where β = α + γ, Lthing is the length of the EF link, Lcrus is the length of the FP link. Let LPE = Lthigh cosα + Lcrus cosβ be the projection of the EP link onto the XOY plane. We will use the non-uniform coverage method for this purpose, which is described in detail in [7, 8] for other types of parallel robots. The workspace covering is a collection of ndimensional boxes bounded by the intervals of variable values. We will use the methods of interval analysis described in detail in [8]. Let us define the set of values of coordinates xP , yP , zP when changing the angles ψ, α, γ. System (1) includes six variables, which is a problem of dimension 26 . We reduce it to dimension 24 to reduce the dimension of the problem. We set the intervals of values that are guaranteed to include the ranges of admissible values: 

 (2) ZP := ZP , ZP = ZP ≤ zP ≤ ZP , 

 X P := X P , X P = X P ≤ xP ≤ X P ,

(3)



 Y P := Y P , Y P = Y P ≤ yP ≤ Y P

(4)

and interval B according to clinical data. The intervals of joints A ,   ,   are determined and compared with clinical data - intervals A, , . It also defines the LPE interval for the X P and Y P intervals and the LPE interval for the ZP and B intervals. There is a solution to the system (1) in the case of the intersection of the intervals L PE and L PE . The

Designing a Robotized System for Rehabilitation

23

lower bound of the intervals corresponds to the minimum values of the angles required for rehabilitation. The upper bound to the maximum. That is:  

A := A, A = A ≤ α ≤ A ; (5) 

  := ,  =  ≤ γ ≤  ;

(6)



  := ,  =  ≤ ψ ≤  ;

(7)



 β = α + γ, that isB := B, B = A +  ≤ β ≤ A +  .

(8)

We obtain the system of equations  ∩   = ∅, A ∩ A = ∅,  ∩   = ∅,  PE ∩ L PE  = ∅. If this system is met, then for the intervals X P , Y P , ZP and B there is at least one point included in the workspace. Using the data obtained in the calculations, we determine the required workspace with the following geometric parameters: A = [−20◦ , 10◦ ], Γ =[ − 60◦ , 0◦ ], Ψ =[0◦ , 25◦ ], B =[ − 80◦ , 10◦ ], Lthigh = Lcrus = 450 mm, a = 100 mm. As a result, the workspace’s overall dimensions were X - 383 mm, Y - 450 mm, Z - 759 mm (Fig. 5). The projections of the constructed full shell of the three-dimensional workspace are shown in Fig. 5a, 5b and 5c. L

a

b

c

Fig. 5. Projections of the shell of the workspace of the passive RRRR orthosis

3 Analysis of the Workspace of the Active 3-PRRR Mechanism Taking into Account the Patient’s Position Diagram of the active 3-PRRR mechanism is presented on Fig. 6. The development of the active mechanism was carried out following the provision of the workspace of the passive orthosis. Let us write down the intervals that describe the ranges of changes in the coordinates of the point P: ⎡ ⎤ xP (9) P = ⎣ yP ⎦ zP

24

A. Voloshkin et al.

Fig. 6. Diagram of the active 3-PRRR mechanism.

Determine the values of the coordinates of the points Di : ⎤ ⎤ ⎡ ⎡ √ √ ⎡ ⎤ xP xP + 23 R xP − 23 R ⎥ ⎥ ⎢ ⎢ D1 = ⎣ yP + R2 ⎦, D2 = ⎣ yP + R2 ⎦, D3 = ⎣ yP − R ⎦· zP zP zP The drive coordinates qi are defined as √ 3 R R, q2 = yD2 = yP + , q3 = zD3 = zP , q1 = xD1 = xP + 2 2

(10)

(11)

The workspace of the active 3-PRRR mechanism is a parallelepiped, the overall dimensions of which are determined by the ranges of the drive coordinates. Their minimum values should be chosen based on the fact that the active manipulator’s workspace should completely include the workspace of the passive orthosis (Fig. 7). Let us determine the influence of the range of regulation of the passive orthosis on the active manipulator’s workspace. The shape and size of the passive orthosis’s workspace will not change depending on the position of the attachment point. However, regulation affects its position relative to the active manipulator. Changing this position is a plane-parallel displacement relative to the X, Y, Z axes is presented on Fig. 8. In accordance with various anthropometric data of patients for the BMI range from 16 kg/m2 to 35 kg/m2 , the displacement of the entire workspace relative to the Y axis is 120 mm (Fig. 8a), relative to the X axis will be 100 mm (Fig. 8b) and along the axis Z 100 mm (Fig. 8c). As a result, the required volume of the workspace, taking into account the orthosis adjustment, is increased by 55.4%. The final calculation of the links lengths will be performed taking into account the possible intersections of the links of the mechanisms based on the methods of multicriteria optimization.

Designing a Robotized System for Rehabilitation

25

Fig. 7. Workspace: 1 - the required workspace of the active 3-PRRR mechanism taking into account the passive orthosis workspace, 2 - the workspace of the passive RRRR orthosis.

a

b

c

Fig. 8. Geometric definition of the work area. a - increase the workspace along the Y axis, b increase the workspace along the X axis, c - increase the workspace along the Z axis.

4 Conclusions The studies were carried out to confirm the complexity of the development of a universal robotic system for rehabilitation; ensuring the patient’s correct seating is a key aspect in carrying out rehabilitation exercises. However, it is unreasonably difficult to provide a landing for patients of any complexion. In this case, it is advisable to divide this system into three classes. The 1st class will include the vast majority of patients with a body mass index from 16 to 35 kg/m2 . The 2nd class will have a small workspace and provide rehabilitation for patients with BMI up to 16 kg/m2 ; The 3rd grade will be sent to rehabilitation with a body mass index of 35 kg/m2 . Thus, it is possible for a system of the 2nd class to reduce the required metal consumption, engine power, and the required workspace, thereby reducing the cost. The 1st class system will be aimed at the maximum number of patients and will cover rehabilitation institutions’ needs in 70% of cases. The 3rd class will have more power, and the fit’s adjustment will consider the patient’s completeness. The obviously more powerful system can have a greater gear ratio, thereby reducing the maximum speed and providing the required power. The cost of these robotic systems should ensure the mass use of robotic rehabilitation.

26

A. Voloshkin et al.

Acknowledgments. This work was supported by the Russian Science Foundation, the agreement number 19–19-00692.

References 1. Hidler, J.M., Wall, A.E.: Alterations in muscle activation patterns during robotic-assisted walking. Clin. Biomech. 20(2), 184–193 (2005) 2. Lam, T., Andershitz, M., Dietz, V.: Contribution of feedback and feed forward strategies to locomotor adaptations. J Neurophysiol. 95, 766–773 (2006) 3. Mirbagheri, M.M., Tsao, C., Pelosin, E., Rymer, WZ.: Therapeutic effects of robotic-assisted locomotor training on neuromuscular properties. In: IEEE 9th International Conference on Rehabilitation Robotics (ICORR), Chicago USA, pp. 561–564 (2005) 4. Baranova, E.A., Bredikhina, Y., Kabachkova, A.V., Kalinnikova, Y., Pashkov, V.K.: Modern approaches to robotic mechanotherapy with elements of biocontrol and telemedicine to restore lost motor functions. Bull. Tomsk State Univ. 433, 127–134 (2018) 5. Voloshkin, A.A., Virabyan, L.G., Mamaev, V.A., Kholoshevskaya, L.R.: Design of mechanisms of a robotic system for rehabilitation of the lower jimbs. In: International Conference of Young Scientists and Students ToPME 2020, vol. 703, pp. 584–591 (2020) 6. Malyshev, D.I., Nozdracheva, A.V., Kholoshevskaya, L.R.: Identification of the parallel 3PRRR manipulator parameters considering the workspace boundaries and the passive orthosis movements. J. Phys. Conf. Series 1582(1), 012062 (2020) 7. Evtushenko, Y., Posypkin, M., Rybak, L., Turkin, A.: Approximating a solution set of nonlinear inequalities. J. Global Optim. 71(1), 129–145 (2017). https://doi.org/10.1007/s10898-0170576-z 8. Jaulin, L.: Applied interval analysis: with examples in parameter and state estimation, robust control and robotics. Springer, New York (2001). https://doi.org/10.1007/978-1-4471-0249-6

Design Optimization and Dynamic Control of a 3-d.O.F. Planar Cable-Driven Parallel Robot for Upper Limb Rehabilitation Ferdaws Ennaiem1,2(B) , Hanen El Golli2 , Abdelbadiâ Chaker2 , Med Amine Laribi1 , Juan Sandoval1 , Sami Bennour2 , Abdelfattah Mlika2 , Lotfi Romdhane2,3 , and Saïd Zeghloul1 1 Department of GMSC, Pprime Institute CNRS, ENSMA, University of Poitiers, UPR 3346

Poitiers, France {ferdaws.ennaiem,med.amine.laribi,juan.sandoval, said.zeghloul}@univ-poitiers.fr 2 Mechanical Laboratory of Sousse (LMS), National Engineering School of Sousse, University of Sousse, 4000 Sousse, Tunisia {hanen.elgolli,abdelbadia.chaker,sami.bennour.meca, abdelfattah.mlika}@eniso.u-sousse.tn, [email protected] 3 American University of Sharjah, PO Box 26666, Sharjah, UAE

Abstract. Assistive robotic rehabilitation can be classified into two groups, the passive mode where the robot moves the patient’s affected member along the desired movement, and the assisted as needed mode, where the subject performs the exercises by himself and the robot interacts only to guarantee the accuracy of the movement. This paper proposes an optimal design of a 3-DoF planar cabledriven parallel robot intended for upper limb rehabilitation as well as the torque control strategies adopted for the assisted as needed and the passive rehabilitation modes. In this context, a motion capture system was carried out for the desired trajectories and the measured data were analyzed in order to extract the robot prescribed workspace. Keywords: Passive mode · Assisted as needed mode · Optimal design · Planar cable-driven parallel robot · Torque control strategies

1 Introduction Physical rehabilitation is the treatment aiming to improve or regain a patient’s functional abilities that allow him to perform, by self, activities of daily living. Two strategies of rehabilitation are used. The conventional rehabilitation is based on the repetition of a set of exercises with the manual assistance of the physiotherapist. This type of rehabilitation does not provide the needed accuracy and it is time and labor consuming. The second method is the robotic rehabilitation [1, 2]. Robots can assist continuously and accurately the subject’s movement. They also provide the set of measurements needed to evaluate the patient’s progress. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 27–37, 2022. https://doi.org/10.1007/978-3-030-76147-9_4

28

F. Ennaiem et al.

Cable-driven parallel robots (CDPRs) are among the best candidates for rehabilitation applications thanks to their characteristics namely, low inertia, large workspace, and high velocity [3]. Classified as parallel robots, they are composed of a fixed base and a mobile platform related by cables. The pose of the end-effector can be controlled by changing the cable lengths. Lately, this kind of robot has been widely used in several domains such as rehabilitation [4, 5] and industrial applications [6]. Several design approaches have been used in the literature in order to formulate optimization problems and lead to the optimal cable robot design [7–10]. In this paper, the goal is to improve the CDPR behavior by handling simultaneously the cable tensions, the elastic stiffness, and the dexterity. The design parameters are set in a way to consider the real mechanical structure with no simplifying assumptions. Besides the kinematic behavior of CDPRs, several control strategies have been developed for upper limb rehabilitation for the passive assistance [11] and the assisted as needed [12, 13] modes. Moreover, multiple control algorithms have been proposed in the aim of controlling accurately the end-effector motion. These algorithms integrate PID controllers [14], adaptive PD controllers [15], neuron PI-feedforward controllers [16], and the fuzzy controllers [17].

2 Methods 2.1 Experimental Protocol The upper limb rehabilitation consists of assisting the affected patient’s member along a desired trajectory. In this paper, two exercises were prescribed, where the patient moves his hand along circular trajectories. In order to have a natural smooth movement, the gestures of a healthy subject were analyzed and considered as the robot prescribed trajectories. The data needed to track the motion were recorded using a Qualisys motion capture system with five markers fixed on the subject’s hand, as illustrated in Fig. 1. The prescribed robot trajectories and the corresponding orientations are presented in Fig. 2. The first trajectory (Fig. 2 (a)) is composed of 331 points and the second one ((Fig. 2 (b)) is formed by 256 points.

Fig. 1. Motion capture setup: (a) markers positioning, (b) global and local frames.

2.2 Planar CDPR Modeling A planar CDPR with 4 actuated cables, shown in Fig. 3, is considered in this paper. The cables are attached to the fixed structure of the robot at points called cable exit points

Design Optimization and Dynamic Control of a 3-d.O.F. Planar

29

Fig. 2. Prescribed trajectories (a) 8-trajectory, (b) circular trajectory, (c) hand orientation recorded for the movement (a), and (d) hand orientation recorded for the movement (b).

(M1 , M2 , M3 , and M4 ) and to the moving platform at the attachment points (B1 , B2 , B3 , and B4 ), called anchor points. The position of the points Mi changes along the pulley’s circumference according to the end-effector location. The points A1 , A2 , A3 , and A4 represent the center of each actuated pulley. Referring to Fig. 3, FG (OG , YG , ZG ) and FL (OL , YL , ZL ) are the global frame attached to the robot base and the local frame attached to the platform, respectively. The origin OL is coincident with the marker H3 (see Fig. 1). The orientation of the platform with respect to the global frame is represented by the rotation matrix Rφ , whose expression is given by Eq. (1).

Fig. 3. Geometric description of a planar CDPR.

30

F. Ennaiem et al.



cos  − sin  Rφ = sin  cos 

 (1)

 denotes the rotation angle between the global and the local frames. n = [n1 , n2 , n3 , n4 ] is the matrix containing the unit vectors ni along the four cables. It is defined as follows: ⎡ ⎤ − sin (ψ1 + θ1 ) sin (ψ2 − θ2 ) sin (ψ3 + θ3 ) − sin (ψ4 − θ4 ) ⎦ (2) n=⎣ − cos (ψ1 + θ1 ) cos (ψ4 − θ4 ) − cos (ψ2 − θ2 ) cos (ψ3 + θ3 ) where ψi and θi are the angles between Bi Mi and ZG and between Bi Ai and Bi Mi , respectively, expressed as given by Eqs. (3–4). The length of each cable Li is given by Eq. (5) where R is the pulley radius. R is considered to be fixed since the cable radius was neglected (is equal to 0.3 mm) and so the variation caused by the cable coiling was not taken into account. The Jacobian matrix is defined as given by Eq. (6). |Aix − Bix | ) |Aiy − Biy |  R θi = tan−1 Li

Li = Mi Bi = Ai Bi 2 − R2 ψi = tan−1 (

 JT =

(3) (4) (5)

 n2 n3 n4 n1 (Rφ × OL B1 )∧n1 (Rφ × OL B2 )∧n2 (Rφ × OL B3 )∧n3 (Rφ × OL B4 )∧n4 (6)

The dynamic model is developed by applying the Newton-Euler equation, with the assumption of neglecting the cables mass and the dynamics of the pulleys. The general dynamic model of the CDPR can be divided into two expressions, the dynamic of the moving platform, and the dynamic of the actuators. The complete dynamic model is obtained by coupling them through the cable tensions Tc as illustrated in the following equations:    F  (7) = M χ¨ +CP χ = JT Tc + fg + Fext/OL M



mp I3∗3 03∗3 03∗1 where M = is the mass matrix, C = is the 03∗3 Rφ Ip RφT ω × Rφ Ip RφT ω  T Coriolis matrix, χ = xy is the end-effector pose vector, fg is the gravity force, Fext/OL are the external forces applied on the end-effector, mp denotes the mass of the moving platform, Ip is the inertial matrix of the moving platform written in its mass center OL and ω is the platform’s angular velocity. The tension distribution is determined with

Design Optimization and Dynamic Control of a 3-d.O.F. Planar

31

maintaining positive cable tensions. This solution is obtained using the Moore-Penrose pseudo-inverse.    +   Tc = Tp + Th = − JT (8) M χ¨ +C(χ, χ˙ )χ˙ + fg + Fext/OL + λNull JT where Tp is the particular solution, Th is the homogenous solution and λ is an arbitrary scalar. The dynamic model of Actuators is used to compute the required torques τm as follows: τm = Im q¨ +RTc + ffr

(9)   where Im is a matrix containing the actuators moments of inertia, q¨ = R−1 J χ¨ +J˙ χ˙ is the angular acceleration of the actuators, R = diag(r1 , r2 , r3 , r4 ), with ri is the radius of the ith pulley, and ffr is the actuator friction forces. 2.3 Formulation of the Optimization Problem A fully constrained planar CDPR with 3 DoF and 4 cables is considered. The mobile platform has a square shape. In the literature, some authors consider the cable exit points position as a design vector parameters [7, 18]. Such a choice can lead to loss of accuracy of the whole system since cables are wound around pulleys, and thus, the location of the exit points changes according to the mobile platform position. In this paper, the vector I, given by Eq. (10), gathering the position of points A1 , A2 , A3 , and A4 (the center of each pulley, determined using the variables ai and bi ), and the side length of the mobile platform, a, form the design vector (see Fig. 1). I = [a1 , b1 , a2 , b2 , a3 , b3 , a4 , b4 , a]

(10)

The first criterion is to obtain a structure that allows the minimum tension in each cable along the prescribed trajectories, which leads to low energy consumption. This criterion is mathematically formulated as follows:   4 i=1 Tcm (i) min (11) 4 where Tcm

⎤ ⎡  n n n n Tc1(j) Tc2(j) Tc3(j) Tc4(j) j=1 j=1 j=1 j=1 ⎦ =⎣ n· max Tc1(j) n· max Tc2(j) n· max Tc3(j) n· max Tc4(j) j=1..n

j=1..n

j=1..n

(12)

j=1..n

Tci denotes the tension of cable i and n is the number of points in each trajectory. The second criterion aims to maximize the dexterity of the robot over its workspace. The index used to evaluate this criterion, given by Eq. (13), is defined as the ratio of the largest singular value of the Jacobian matrix to the smallest. DI −1 takes values between 0 and 1. When it is close to 1, the pose of the platform is far from a singular configuration. DI −1 =

σmin σmax

(13)

32

F. Ennaiem et al.

The third criterion consists of maximizing the elastic stiffness of the robot over the workspace. The considered elastic stiffness index, given by Eq. (14), is defined as the ratio of the largest eigenvalue of the cable stiffness matrix, by the smallest [19]. ESI −1 takes values between 0 and 1. When it is close to 1, the stiffness distribution of the robot will be uniform. ESI −1 =

λmin λmax

(14)

A set of constraints were imposed in order to guarantee better robot performance. Since cables can only pull, the tensions must be positive to avoid slack cables. Tcmax is chosen according to the mechanical characteristics of the actuators. Thus Tc (i), the tension of cable i, must satisfy the constraint (15). 0 < Tcmin ≤ Tc (i) ≤ Tcmax

(15)

The potential collisions are also taken into consideration, during the movement along the prescribed trajectories, the mobile platform must avoid interference with cables. As shown in Fig. 3, the transmission angle αi is defined as the angle between the vectors ni and mi . It is computed as mentioned in Eq. (16). For each pose, this value must be greater than αlim .  mi −1 ni . (16) αi = cos mi where m1 = B1 B4 , m2 = B2 B3 , m3 = B3 B2 , m4 = B4 B1 . The optimization purpose is to find the optimal design vector which minimizes the objective function F, defined as a weighted sum of the three criteria introduced above. The optimization problem is formulated as follows: ⎧ min(F(I )) ⎨ Subject to (17) 0 < Tcmin ≤ Tc (i) ≤ Tcmax ⎩ αi < αlim In order to simplify the optimization process, the mono-objective problem formulation is considered using the coefficients β1 , β2 and β3 , whose sum is equal to 1. These factors define a weight for each criterion. The problem constraints were handled using a penalty approach. This method was adopted in our previous work [8]. F is then expressed as follows: F(I) = β1

4     1     Tcm (i) + β2 1 − ESIg−1  + β3 1 − DIg−1  + ℘1 + ℘2 4

(18)

i=1

where  ℘1 =

0 if Tcmin ≤ T c (i) ≤ Tcmax  0 otherwise

(19)

Design Optimization and Dynamic Control of a 3-d.O.F. Planar

 ℘2 =

33

0 if αi < αlim  0 otherwise

(20)

 is a large scalar, ℘k∈{1,2} represent the penalty functions, ESIg−1 and DIg−1 are the mean values of ESI −1 and DI −1 , respectively.

3 Results and Discussion 3.1 Robot Optimal Design The optimization process was performed using the PSO algorithm [20] using as parameters the values summarized in Table 1. Table 1. Optimization parameters. Parameter

Value

Parameter

Value

Mobile platform weight [kg]

0.5

Tcmax [N]

15

Cable radius [mm]

0.3

Tcmin [N]

0.5

Population size

250

αlim [°]

2

Pulley radius [mm]

37

β 1 , β2 , β3

0.33

The optimal design vector I∗ obtained is given in Eq. (21). The results of the optimization criteria are illustrated in Fig. 4. I∗ = [−0.375, −0.375, 0.375, −0.21, 0.375, 0.353, −0.375, 0.179, 0.1]

(a)

(b)

(21)

(c)

Fig. 4. (a) Dexterity, (b) Elastic stiffness, (c) Maximal cable tensions distributions.

The dexterity, the elastic stiffness and the maximal cable tensions computed for the optimal robot structure vary in [0.65, 0.74], [0.28, 0.5], and [0.5 N, 1.54 N], respectively.

34

F. Ennaiem et al.

3.2 Torque Control Strategies The interaction with the patient’s limb is quantified based on his disability level. During the first phase, the subject remains passive while the robot moves the affected member along the desired trajectory. In an advanced state, where the patient regains some of his motor abilities, the robot can be reconfigured to only assisted the motion when the patient performs the exercises inaccurately. It computes the external efforts exerted by the patient on the moving platform in order to apply the required forces Freq , given by Eq. (22), allowing to reduce the generated position error. A Matlab code based on the PID controller, whose law is given by Eq. (23), and the calculated torques was established to perform the simulation of the two modes of rehabilitation. Its bloc diagrams are given in Fig. 5. For the assisted as needed mode, the patient can produce a resistive effort that can change the robot’s trajectory. Thus, we suppose that the force applied by the patient opposite to the performed trajectory.

Fig. 5. Block diagram of the passive and the assisted as needed modes. IKM refers to the inverse kinematic model and IDM to the inverse dynamic model.

  Freq = Kp χd − χ − Kd χ˙ d  PID = −K χ¨ +Kp (t) + Ki 0

t

  d (t)

t  dt  + Kd dt

(22) (23)

where Kp , Ki , Kd are the gains of the proportional, integral and derivative terms respectively, and (t) represents the error between the current robot position χ and the desired robot position χd . The comparison between the desired and the actual trajectories presents an error that attains 8 × 10−3 m. This position error of the two trajectories for both modes is shown in Fig. 6. As can be observed, the PID controller and the applied Freq reduce the tracking error. Figure 7 illustrates the tension of each cable computed for the two modes along the circular trajectory.

Design Optimization and Dynamic Control of a 3-d.O.F. Planar

35

Fig. 6. The position error of the (a) 8-trajectory, and (b) circular trajectory for the passive mode, and the (c) 8-trajectory, and (d) circular trajectory for the assisted as needed mode.

Fig. 7. Cable tensions computed for both modes along the circular trajectory.

36

F. Ennaiem et al.

4 Conclusion The main objective of this paper was to firstly, find the optimal design vector of a planar cable-driven parallel robot for upper limb rehabilitation and secondly, to establish the torque control strategies corresponding to the assisted as needed and passive assistance modes. Along the prescribed trajectories, identified using the motion capture system, the cable tensions, the elastic stiffness, and the dexterity were optimized while satisfying a set of constraints. The torque control methods allowing the patient’s upper limb rehabilitation in the passive and the assisted as needed modes were presented. Acknowledgments. This work was financially supported by the “PHC Utique” program of the French Ministry of Foreign Affairs and Ministry of higher education, research and innovation and the Tunisian Ministry of higher education and scientific research in the CMCU project number 19G1121.

References 1. Lum, P.S., et al.: MIME robotic device for upper-limb neurorehabilitation in subacute stroke subjects: a follow-up study. J. Rehabil. Res. Dev. 43(5), 631 (2006) 2. Nef, T., Riener, R.: ARMin - design of a novel arm rehabilitation robot. In: 9th International Conference on Rehabilitation Robotics, ICORR 2005, pp. 57–60 (2005) 3. Qian, S., Zi, B., Shang, W.-W., Xu, Q.-S.: A review on cable-driven parallel robots. Chin. J. Mech. Eng. 31(1), 66 (2018) 4. Mao, Y., Agrawal, S.K.: Design of a cable-driven arm exoskeleton (CAREX) for neural rehabilitation. IEEE Trans. Robot. 28(4), 922–931 (2012) 5. Tappeiner, L., Ottaviano, E., Husty, M.: A cable-driven robot for upper limb rehabilitation inspired by the mirror therapy. In: Zeghloul, S., Romdhane, L., Laribi, M. (eds.) Computational Kinematics. Mechanisms and Machine Science, vol. 50, pp. 174–181. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-60867-9_20 6. Pott, A., et al.: IPAnema: a family of cable-driven parallel robots for industrial applications. In: Bruckmann, T., Pott, A. (eds.) Cable-Driven Parallel Robots. Springer, Berlin, Heidelberg, 119–134 (2013). https://doi.org/10.1007/978-3-642-31988-4_8 7. Gagliardini, L., et al.: Optimal design of cable-driven parallel robots for large industrial structures. In: 2014 IEEE International Conference on Robotics and Automation (ICRA), Hong Kong, China, pp. 5744–5749 (2014) 8. Ennaiem, F., et al.: Sensitivity based selection of an optimal cable-driven parallel robot design for rehabilitation purposes. Robotics 10(1), 7 (2021) 9. Hussein, H., Santos, J.C., Gouttefarde, M.: Geometric optimization of a large scale CDPR operating on a building facade. In: 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Madrid, pp. 5117–5124 (2018) 10. Fattah, A., Agrawal, S.K.: On the design of cable-suspended planar parallel robots. J. Mech. Des. 127(5), 1021–1028 (2005) 11. Barbazza, L., Zanotto, D., Rosati, G., Agrawal, S.K.: Design and optimal control of an underactuated cable-driven micro-macro robot, 3766 (2017) 12. Fortin-Côté, A., Cardou, P., Gosselin, C.: An admittance control scheme for haptic interfaces based on cable-driven parallel mechanisms. In: IEEE International Conference on Robotics and Automation, pp. 819–825 (2014)

Design Optimization and Dynamic Control of a 3-d.O.F. Planar

37

13. Wu, Q., Wang, X., Chen, B., Wu, H.: Development of a minimal-intervention-based admittance control strategy for upper extremity rehabilitation exoskeleton. IEEE Trans. Syst. Man Cybern. Syst. 48(6), 1005–1016 (2018) 14. Carignan, C.R., Naylor, M.P., Roderick, S.N.: Controlling shoulder impedance in a rehabilitation arm exoskeleton. In: Proceedings of the IEEE International Conference on Robotics and Automation, pp. 2453–2458 (2008) 15. Frisoli, A., Sotgiu, E., Procopio, C., Bergamasco, M., Rossi, B., Chisari, C.: Design and implementation of a training strategy in chronic stroke with an arm robotic exoskeleton. In: IEEE International Conference on Rehabilitation Robotics, p. 22275708 (2011) 16. Jiang, X.Z., et al.: Position control of a rehabilitation robotic joint based on neuron proportionintegral and feedforward control. J. Comput. Nonlinear Dyn. 7(2), 1–6 (2012) 17. Li, Z., et al.: Fuzzy approximation-based adaptive backstepping control of an exoskeleton for human upper limbs. IEEE Trans. Fuzzy Syst. 23(3), 555–566 (2014) 18. Li, Y., Xu, Q.: GA-based multi-objective optimal design of a planar 3-DOF cable-driven parallel manipulator. In: 2006 IEEE International Conference on Robotics and Biomimetics, Kunming, China, pp. 1360–1365 (2006) 19. Abdolshah, S., et al.: Optimizing stiffness and dexterity of planar adaptive cable-driven parallel robots. J. Mech. Rob. 9(3), 031004 (2017) 20. Eberhart, R., James, K.: A new optimizer using particle swarm theory. In: Proceedings of the Sixth International Symposium on Micro Machine and Human Science, MHS’95. IEEE (1995)

Novel Design of the ParReEx-Elbow Parallel Robot for the Rehabilitation of Brachial Monoparesis Bogdan Gherman1 , Paul Tucan1 , Calin Vaida1 , Giuseppe Carbone2 , and Doina Pisla1(B) 1 Technical University of Cluj-Napoca, Cluj-Napoca, Romania

[email protected] 2 University of Calabria, Arcavacata, Italy

[email protected]

Abstract. The paper presents the novel design of the ParReEx-elbow robotic system for the rehabilitation of the upper limb, namely the elbow flexion-extension and pronation-supination. Initially designed for post-stroke rehabilitation, the robot is now fitted for brachial monoparesis treatment, a manifestation of a broad range of neurologic diseases. Following the critical feedback and positive outcomes of previously performed clinical trials an improved design has been proposed. Validation simulations (FEM analysis and dynamic simulation) using the novel CAD robot model are presented, the results showing that the robot can be efficiently used for the proposed medical task. Keywords: Parallel robot · Upper limb rehabilitation · Design · FEM-analysis · Dynamic simulation

1 Introduction Neurological disorders are acknowledged to represent a major challenge in the 21st century, being an important cause of disability and even death. Brachial monoparesis (BM), as the inability to control properly one’s upper limb, has various causes: most of the times it is the result of a previous injury of the brachial plexus (a network plexus of nerves extending from the spinal cord, through the cervico-axillary canal in the neck), which directly controls the upper limb functions, other times it is caused by birth-injuries in infants [1]. Pure brachial monoparesis is actually a rare disease, without other neurologic deficits. BM can be caused by damages of the central or peripheral nervous system after strokes, craniocerebral and spinal trauma, multiple sclerosis, brain tumors, radiculopathies, plexopathies, neuropathies, etc. [2]. A viable treatment option consists in physical rehabilitation leading to the development of new nervous tracts, replacing the damaged ones. Several such robots have been previously developed in [3–8], most of them for the rehabilitation of the upper limb for post-stroke patients. Studies [9] have shown that these robots can be used with good results for the treatment of the BM. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 38–45, 2022. https://doi.org/10.1007/978-3-030-76147-9_5

Novel Design of the ParReEx-Elbow Parallel Robot

39

The robotic system presented in this paper has been initially developed for the rehabilitation of the upper limb for post-stroke patients, but the initial clinical trials [10] have demonstrated its potential for other diseases, including BM. Fitting the robot for BM rehabilitation is related to increased motion ranges and slightly different training exercises in terms of speed, duration and interaction modalities, all of which requiring an improved design. In Sect. 2, the paper presents the assessment of the ParReEx-elbow robotic system following the results obtained in initial clinical trials and its redesign approach to mitigate the challenges met during these trials. Section 3 presents some simulation results that validate the improved design of the medical robot, followed by discussions (Sect. 4) and conclusions (Sect. 5).

2 Methods 2.1 ParReEx-Elbow Robotic System The ParReEx-elbow parallel robotic system for the upper limb rehabilitation has been extensively presented in [11, 12] and [13] (Fig. 1).

Fig. 1. ParReEx-elbow kinematic scheme [12]

ParReEx-elbow has 2 DoFs and achieves the elbow flexion and pronation-supination motions. Each motion is achieved by one active joint: q1 achieves the pronationsupination through the kinematic joint consisting of the universal joint Rc and the gear transmission G (connected through the element b), the active anchor being positioned at the patient’s wrist (the mobile coordinate system X’Y’Z’). The elbow flexion is achieved using the q2 active joint and the element c, connected to the element b through the passive revolute joint R2 . Figure 2 shows the initial CAD model of the ParReEx-elbow robotic system setup, used in the initial clinical trials. The main components are: the active anchor (the larger spur gear crown, 1:4 ratio), along with other two components designed to increase the patient’s comfort: the ergonomic handle (which secures the patient’s hand, using a rubber strap) and the forearm support. The passive anchor is where the patient positions his/hers arm, in the proximity of the elbow.

40

B. Gherman et al.

Fig. 2. ParReEx-elbow robotic system: a. robotic system layout; b. detail

2.2 Initial Clinical Trials ParReEx-elbow robotic system has been involved in a series of clinical trials [10] at the Municipal Cluj-Napoca Clinical Hospital in Cluj-Napoca, Romania based on the IRB approval. Although the medical outcomes were positive [10], there have been a number of issues that pointed out to a few required improvements. 1. The active anchor (namely the larger spur gear bearing) would jam occasionally. This is mainly due to two reasons: large friction forces (both the bearing and the spur gear have been made of ABS - plastic) and extended periods of use. Actually, the experimental model of ParReEx-elbow has never been designed to work 8 h a day for such a long period. Initial tests, performed in laboratory conditions did not show significant issues due to friction that would lead to such behavior. 2. In the initial design, all adjustable elements (Fig. 2) have been made of ABS, which during dry runs have shown some degree of elasticity. The registered maximum deformation of the robot elements during training exercises has been of 4.5 mm. This issue has led to a certain degree of apparent insecurity and frailty, not highly appreciated by the user. 3. The initial design was intended to make ParReEx-elbow able to treat only the right hand of the patient, which may be the main drawback of this architecture. Of course, with some effort, motor q1 could be repositioned on the robot frame, but this would take too much time and would require special tools and knowledge, making it highly inefficient. Other small issues solved at the beginning of the clinical trials include: masking the control box with plexiglas, building a solid active anchor point for the patient’s forearm using sterilizable materials, building a swiveling passive anchor for the arm support. Figure 3 shows a series of snapshots of the ParReEx-elbow robotic system during training exercises.

Novel Design of the ParReEx-Elbow Parallel Robot

41

Fig. 3. ParReEx-elbow – snapshots taken during training sessions

2.3 The Novel Robot Design Improving the design of ParReEx-elbow has been required in order to: mitigate the issues presented in Subsect. 2.2 and prepare it for the treatment of BM. Figure 4 presents the ParReEx-elbow design optimization approach. For an improved design the authors have started from the medical task and anthropometric parameters [14] (motion ranges, velocities, limb length, active and passive anchors position, etc.). Each issue discussed in Sect. 2.2 will be addressed separately. The design has been analyzed after each step (from 1 to 3), performing both structural and dynamic simulation. 1. In order to reduce friction, the hand holder (including the large spur gear) has been modified and radial ball-bearings have been introduced. The hand holder has been enlarged and stainless steel rings have been used (in pairs of two on each side of the hand holder, to reduce friction on the horizontal and vertical planes), together with rollers (polyamide with ball-bearings to reduce both friction and wear – 18 pieces) – Fig. 5. Additionally, the shaft transmitting the motion from the q1 motor has also been modified and ball-bearings have been added. 2. To reduce the elasticity of the structure and support the new, more massive design, the supporting structure of the hand holder has been modified and drastically improved (Fig. 5). 3. In order to make ParReEx-elbow a bimanual rehabilitation robotic system, the q1 motor has been attached to the robot arm actuated by motor q2 using an ABS 3D printed support. The universal joint has been removed and the robot can easily be used for the right or left hand just by turning the q2 motor with roughly 180° clockwise and the hand holder (using the q1 motor) with 180°. In addition, protective cases have been designed, to isolate all moving components with which the patient should not get into contact with during training exercises (rollers, rotating shafts, motor q1 , etc.).

42

B. Gherman et al.

Fig. 4. Optimization approach

Fig. 5. The ParReEx-elbow robot redesign

3 Results The proposed design is analyzed from the structural point of view to check for significant deformations of its elements. A dynamic analysis is also performed to investigate if the two motors are able to efficiently drive the new design. A FEM analysis has been performed using Siemens NX Pre/Post Module [15]. The following static loads have been used (Fig. 6): 5 N on the ergonomic handle, 100 N on the hand support and 10 N on the forearm support, all applied on the OZ axis (against gravitation) to the maximum values for regular patients: 1.5 kg for the forearm and 0.585 kg for the hand in the case of a 90 kg male according to Plagenhoef [16]. The results presented in Fig. 7 show a maximum displacement less than 1 mm, which fully satisfies the proposed requirements. For the dynamic analysis friction coefficients have been introduced, according to the pairs of used materials (polyamide-steel) for which µdynamic = 0.04 [17]. Table 1 presents the motion parameters and the torque characteristics of the two actuating motors

Novel Design of the ParReEx-Elbow Parallel Robot

43

Fig. 6. The location and direction of applied load for FEM analysis

Fig. 7. The ParReEx-elbow FEM analysis

[18]. The obtained mean torque value for the simulation of the q1 motor is 4 Nm, while for the q2 motor: 11.7 Nm (Fig. 8). Table 1. Simulation parameters and ParReEx-elbow motors torque characteristics Motion amplitude

Frequency

Nominal torque

Peak torque

Flexion-extension

90°

45°/s

134 Nm

200 Nm

Pronation-supination

180°

20°/s

6.1 Nm

15.3 Nm

Fig. 8. The ParReEx-elbow dynamic analysis for the motion parameters presented in Table 1 and a time period of 30 s

44

B. Gherman et al.

4 Discussion The ParReEx-elbow robotic system has been initially designed for post-stroke rehabilitation of patients with upper limb impairments. The initial clinical tests have shown that it can be used to improve the health of patients with other neuromotor impairments, including BM. The ParReEx-elbow novel design has been developed following the experimental data collected during the clinical trials. The improvements targeted: q1 motor occasional jamming, reducing the frailty feeling given by a large elasticity of the ABS components and the bimanual function. Judging from the results presented in Figs. 7 and 8, the proposed target has been reached. Accuracy in rehabilitation is not a huge issue, so the displacement within the ergonomic handle and the forearm support (Fig. 7) should have little to no influence upon the training exercises. One may say that there are too many components used to support the upper limb active anchor (the handle, the wrist and forearm support), but these elements are all adjustable and relinquish the designer from the requirement of using an adjustable robot arm (as in the previous design – Fig. 2). Bimanual operation is a “must-have” feature for both economic and space-related issues, which is why ParReEx-elbow has been redesigned to reduce the footprint of the robot, to increase the chances of market adoption. The results obtained within the achieved FEM and dynamic simulations, prove that the redesigned robotic system can be build and used for further clinical trials, expected to start at the end of 2021.

5 Conclusions The aim of this paper is to present the novel design of the ParReEx-elbow rehabilitation robot for the elbow flexion-extension and pronation-supination. The robot design has been improved targeting the brachial monoparesis treatment following a first set of clinical trials. A set of simulations (FEM analysis and dynamic simulations) have proved that the new design is suitable for the medical task. An upgraded control system suitable for bimanual operation and the implementation of multiple control modalities (mirroring and active-assistive) will lead to a robotic system prepared for BM treatment within the next clinical trials. Acknowledgments. This work was supported by a grant of the Romanian Ministry of Research and Innovation, CCCDI – UEFISCDI, project number PN-III-P2-2.1-PED-20193022/546PED/2020 (NeuroAssist) and the project POCU/380/6/13/123927 – ANTREDOC, “Entrepreneurial competencies and excellence research in doctoral and postdoctoral studies programs”, project co-funded from the European Social Fund through the Human Capital Operational Program 2014–2020.

References 1. Jang, S.H., Seo, Y.S.: Complete monoplegia due to limb-kinetic apraxia in a patient with traumatic brain injury: a case report. Medicine 99(49), e22452 (2020)

Novel Design of the ParReEx-Elbow Parallel Robot

45

2. Birua, G.J.: Acute monoplegia due to spinal injury. Indian J. Neurosurg. 08(02), 127–129 (2019) 3. Weber, L.M., Stein, J.: The use of robots in stroke rehabilitation: a narrative review. NeuroRehabilitation 43(1), 99–110 (2018) 4. Zimmermann, Y., et al.: ANYexo: a versatile and dynamic upper-limb rehabilitation robot. IEEE Robot. Autom. Lett. 4(4), 3649–3656 (2019) 5. Zhang, L., Guo, S., Sun, Q.: Development and assist-as-needed control of an end-effector upper limb rehabilitation robot. Appl. Sci. 10(19), 6684 (2020) 6. Morone, G., Cocchi, I., Paolucci, S., Iosa, M.: Robot-assisted therapy for arm recovery for stroke patients: state of the art and clinical implication. Expert Rev. Med. Devices 17(3), 223–233 (2020) 7. Vaida, C., et al.: Systematic design of a parallel robotic system for lower limb rehabilitation. IEEE Access 8, 34522–34537 (2020) 8. Plitea, N., et al.: Innovative development of parallel robots and microrobots. Acta Tech. Napocensis Ser. Appl. Math. 5(49), 5–26 (2006) 9. Qassim, H.M., Wan Hasan, W.Z.: A review on upper limb rehabilitation robots. Appl. Sci. 10(19), 6976 (2020) 10. Major, Z., et al.: The impact of robotic rehabilitation on the motor system in neurological diseases. A multimodal neurophysiological approach. Int. J. Environ. Res. Public Health 17(18), 6557 (2020) 11. Birlescu, I., et al.: On the singularities of a parallel robotic system used for elbow and wrist rehabilitation. In: Lenarcic, J., Parenti-Castelli, V. (eds.) ARK 2018. SPAR, vol. 8, pp. 203– 211. Springer, Cham (2019). https://doi.org/10.1007/978-3-319-93188-3_24 12. Tucan, P., et al.: Fuzzy logic-based risk assessment of a parallel robot for elbow and wrist rehabilitation. Int. J. Environ. Res. Public Health 17(2), 654 (2020) 13. Gherman, B., et al.: Parallel robotic system for upper limb medical recovery. Patent No. 132234/14.06.2017 14. Carbone, G., Gherman, B., Ulinici, I., Vaida, C., Pisla, D.: Design issues for an inherently safe robotic rehabilitation device. In: Ferraresi, C., Quaglia, G. (eds.) RAAD 2017. MMS, vol. 49, pp. 1025–1032. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-612768_110 15. Siemens: https://www.plm.automation.siemens.com/global/en/products/nx/. (2021) 16. Plagenhoef, S., et al.: Anatomical data for analyzing human motion. Res. Q. Exerc. Sport 54(2), 169–178 (1983) 17. Alaci, S., et al.: Study of the rolling friction coefficient between dissimilar materials through the motion of a conical pendulum. Materials 13(21), 5032 (2020) 18. B&R Industrial Automation GmbH: https://www.br-automation.com/. (2021)

Exoskeletons and Prostheses

Development of a New Knee Endoprosthesis and Finite Element Analysis of Contact Stresses Daniela Tarnita1(B) , Dan Calafeteanu2 , Ilie Dunitru1 , Alin Petcu1 , Marius Georgescu1 , and Danut Nicolae Tarnita2 1 University of Craiova, Craiova, Romania

[email protected] 2 University of Medicine and Pharmacy, Craiova, Romania

Abstract. In this paper, based on the virtual model of a classic knee prosthesis and on the clinical observations on the limitations of the range of motion of the prosthetic knee, the 3D virtual model of a new knee prosthesis is developed. This new prosthesis allows the development of lower contact stresses and an increase in the flexion-extension interval of the prosthetic knee. Finite element analysis (FEA) is applied in order to obtain the stress maps and the maximum von Mises stress obtained on the three components of the prosthesis and on the femur and tibia. Six models of prosthetic knee assembly were developed, obtained for three different varus angles and for both prostheses: the classic one (initial prosthesis) and the new one, (proposed prosthesis)). The results obtained lead to the conclusion that, with increasing varus angle, the von Mises stresses increase in each of three prosthetic components. Due to the curvatures and of added material in the new proposed prosthesis and due to the increase of the contact surface between the prosthesis elements, the contact stress decreases in the case of the new prosthesis compared to the classic one, which leads to improved biomechanics of the prosthetic knee. Keywords: Virtual knee prosthesis · Finite element method · Contact stress

1 Introduction The maximum value of flexion angle of human knee is about 160° [1, 2]. The flexion angle increases, from about 100°–120° and can vary between 111° to 165° in the case of certain activities, such as: squatting, kneeling, respectively, the cross-legged position adopted for religious activities [1, 3]; An increased knee flexion angle is very important for people working in construction and agriculture or gardening. Total knee arthroplasty (TKA) is defined as a reconstructive surgical intervention with prosthetic replacement of the articular components and bone sacrifice, aiming to restore the joint mobility and the normal functioning of muscles, ligaments, and other structures which contribute to joint movement performing. Clinical observations and specialists’ studies have led to the conclusion that most of current prostheses have only one radius of curvature of the femoral component (F.C.), and the flexion movements of the shank on the thigh are limited to an angle value of about 110–120°. A lot of studies which analyze TKA show © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 49–57, 2022. https://doi.org/10.1007/978-3-030-76147-9_6

50

D. Tarnita et al.

that patients on rare occasion flex the knee above 120° [4–8], very few studies showing cases of flexion >120° [9, 10]. Current surgical techniques and models of knee prosthesis have limitations in meeting the needs of patients who, for their daily activities, require a knee flexion of >120°. Over 50% of patients with prosthetic knees cannot kneel easily due to fear of affecting the prostheses. Preoperative, intraoperative and postoperative factors affect the range of knee flexion after TKA, [9, 11, 12]. In [13], the authors studied intrinsic biomechanical mechanisms that affect the reconstructed knee’s ability to achieve a flexion of >120°. The aim of this study is to elaborate the 3D virtual model of a new knee endoprosthesis, which allows the development of lower contact stresses and an increase of the range of knee flexion. Six different virtual models of prosthetic knee assemblies, considering 3 different varus angles and both knee prostheses: proposed prosthesis (P.P) and initial prosthesis (I.P) were elaborated and analyzed by using FEA.

2 Virtual Modeling of the New Knee Prostheses To obtain the 3D virtual model of the P.P., we used the preprocessor DesignModeler, which run under Ansys Workbench 15.07, frequently used in human biomechanics [14– 18]. Based on the geometry and dimensions of an existing physical prosthesis, type Genesis II, currently used in TKA, the virtual model of I.P. was developed and presented in detail in [17]. The virtual modeling of the P.P. (Fig. 1 a) was performed based on the virtual models of the three components of I.P.: polyethylene insert (P.I.), tibial prosthesis (T.P.) and femoral prosthesis (F.P.) (Fig. 1b), which were modified using advanced commands available in the DesignModeler preprocessor. Shape changes were made by adding material to all three components and by design three radii of curvature that allows an increased contact area between F.P and T.P. The P.P. offers the possibility of a large flexion angle, based on changing the shape of F.P. that has 3 radii of curvature, on: patellar zone, posterior zone (condylar zone) and the contact zone between F.P and T.P., and on posterior wall thickness increasing of the F.P. by 2.5 mm compared to the initial prosthesis. The design changes lead to the extension of the F.P. surface in the posterior area and, implicitly, to the increase of the articular contact area, even at large flexion angles. Advantages of Proposed Prosthesis Compared to Existing Prostheses • The new prosthesis is proposed especially for young people - in uncemented version, providing a more durable natural fixation, as well as for the elderly people- in the cemented version; • The design of the proposed prosthesis allows the restoration of the physiological capabilities of the prosthetic knee joint allowing an increased flexion angle, very close to healthy knee (150°); • For reducing the wear of the prosthesis components, the contact stress and pain, the design of the proposed prosthesis provides the increase of the contact surfaces, a very important aspect in the case of moderate flexion (walking) but especially in the case of a high flexion angle.

Development of a New Knee Endoprosthesis

51

a)

b) Fig. 1. Frontal view, Back view, Side view and Top view of virtual models of: a) I.P.; b) P.P.

3 FEA of Six Virtual Models of Knee Prosthetic Assemblies Both prostheses: P.P. and I.P. used for TKA, and three different varus angles (the angle formed by the mechanical axes of the tibia and femur bones) of the prosthetic knee joint are considered: 176°, 182° and 188°. Thus resulted 6 developed virtual models that will be analyzed with FEA. For all cases an anteroposterior tibial inclination equal to 0° is considered. For the discretization process, Solid186 hexahedral and Solid 187 tetrahedral, with medium nodes, with size equal to 1,5 mm (in the area of interest in our research) and 2,5 mm (for the other zones of the entire assembly of prosthetic knee) (Fig. 2) were used for obtaining the networks of nodes and finite elements used for analyses (Table 1). The limit and control conditions for FEA are similar for all six virtual assemblies: • the period for virtual analysis was set to 1 s; • the iterative method set for solver was “Preconditioned Conjugate Gradient”; • the degrees of freedom were set to: translation of the femur against the tibia on Zdirection, Fig. 3 a); femur rotation on the tibia head on Y-direction; tibia rotation around the ankle on Y-direction, Fig. 3b); • the characteristics of materials [19] are assigned to each component (Table 2); • on Z-direction a force of 2400N is uniformly distributed on the femur head (Fig. 3 c). In precedent papers [20, 21], to evaluating prosthetic knee joints, values equal 3 to 4 times body weight (BW) were adopted for loading force. We adopted a load equal to 3 times normal BW (800 N), namely 2400 N.

52

D. Tarnita et al.

. a)

b)

c)

d)

Fig. 2. a) Lateral view of cut femur and tibia, prepared for TKA; b) local view of discretized knee joint– P.P. assembly (back view); c) isometric view and side view of P.P.

Table 1. Network of nodes and finite elements for the six virtual assemblies.

176o 182o 188o

Proposed prosthesis (P.P.) No. of Nodes No. of Elements 298 959 92 070 296 696 92 131 291 291 90 137

a

Initial prosthesis (I.P.) No. of Nodes No. of Elements 351 324 107 128 347 939 106 853 338 678 103 666

b

c

Fig. 3. Location of: a) ankle rotation; b) hip rotation and translation; c) area of the load force

Development of a New Knee Endoprosthesis

53

Table 2. Characteristics of materials [19] Geometry

Young’s modulus [MPa]

Poisson’s ratio

Femur

17 600

0.3

Tibia

12 500

0.3

F.P

210 000

0.3

T.P

210 000

0.475

P.I

1 100

0.42

4 Results Six FEA analyses were performed for the six virtual models of knee joint-prosthesis assemblies in order to obtain the distribution and maximum von Mises stresses developed on femur and tibia bones, as well as on the three components of the prosthesis: F.P, T.P., P.I. The stress diagrams obtained in femur, tibia, F.P., T.P. and P.I. for a varus angle of 176° and a loading force of 2400 N, for both types of knee prosthetic assemblies (using I.P. and, respectively, P.P.) can be seen in Fig. 4 and Fig. 5.

Fig. 4. Von Mises stress distribution for: a) P.P.- knee joint assembly; b) femur bone; c) tibia bone at 176° varus angle and a loading force of 2400 N.

Similar stress distribution diagrams were obtained for the other four analyzed prosthetic knee assemblies. The maximum von Mises stress obtained for the six analyzed cases are presented in Table 3 and they are plotted in Fig. 6.

54

D. Tarnita et al.

a)

b)

c)

I.P. a)

b)

c)

P.P. Fig. 5. Von Mises stress distribution in: initial prosthesis and proposed prosthesis (top view– first raw; bottom view–second raw) for varus angle of 176°: a) P.I.; b) T.P.; c) F.P.

Table 3. Maximum von Mises stress [MPa] obtained for 6 prosthetic knee assemblies for a loading force equal to 2400 N Varus angle

P.P I.P

1760 1820 1880 1760 1820 1880

P.I. Stress P.T. Stress P.F. Stress Femur Stress Tibia Stress 45.12 41.33 43.79 25.03 32.11 48.68 44.36 47.12 27.73 34.21 51.63 47.43 50.15 30.83 36.71 53.78 50.12 51.84 31.29 42.14 57.11 53.17 55.14 33.31 44.75 60.74 56.45 58.53 35.36 47.51

Development of a New Knee Endoprosthesis

55

20 18 16 14 12 176

182 P.P POLI. Stress

188

176

P.T. Stress

182

188

I.P P.F. Stress

Fig. 6. Plots of maximum von Mises stress developed in prosthetic components

5 Discussions The compared results lead to the conclusion that von Mises stress increases in each component of the prosthesis, with increasing varus angle. Higher values are developed in P.I., slightly lower values appear in F.P. and, the smallest are recorded in T.P. Compared to the results obtained from I.P. analysis, P.P ensures the decrease of contact stress in the prosthetic human knee joint, in all prosthesis components, and also on the femur and tibia by 15% to 20%. The results obtained show good behavior in the P.P. model. In the literature, the average contact area of a normal knee varies in the range [765– 1150] mm2 [22]. The contact surface sizes of most knee prostheses are in the range [80–250] mm2 function of flexion angle, loads and design. The Genesis II prosthesis is constructed with a contact surface equal to approximately 95 mm2 for 0° in flexion, and high contact stress, to 110 mm2 for 15° flexion angle with smaller contact stress [22]. As a result, we can consider that an analysis of the von Mises stresses developed in virtual prostheses for 0° in flexion will also cover the situations corresponding to other flexion angles during walking. For P.P., by modifying the geometric design, the contact area increased to about 160 mm2 , with about 30%, comparing with the contact area on I.P., leading to approximatively 15–20% lower contact stress than those developed on I.P. The maximum values of contact stresses reported in Table 3, higher than 45 MPa, were recorded on small areas (approximately 10–20 mm2 ), by comparing with the whole contact surface, which is bigger The average values range in [25–35] MPa. The values and distribution of contact stresses are comparable to those reported in [22]. In the present study, the results are limited to the orthostatic position, therefore, to a static load and stress. Future work will investigate the dynamic behavior of the prosthetic knee and the effect of flexion angle variation on contact stresses and on P.P. geometry. At the same time, experimental evaluations on the physical model of the proposed prosthesis will be performed to demonstrate the increase of the flexion angle of the prosthetic joint.

56

D. Tarnita et al.

6 Conclusions In this study, the virtual model of a new knee endoprosthesis was elaborated and FEA was applied on six virtual assemblies of prosthetic knee to obtain stress distributions and maximum von Mises stress and compare them. The values are slower in the P.Pknee joint assemblies than in I.P- knee joint assemblies. Parameterized 3D models offer the possibility of optimizing prostheses, but also other types of implants, by changes in geometry, material or contact area, which lead to the knee biomechanics improvement, or of other studied human joints. Acknowledgements. This research is supported by the funds of Project 546/2020, PN-IIIP2–2.1-PED-2019–3022, “Innovative modular robotic system for medical recovery of brachial monoparesis - NeuroAssist” funded by UEFISCDI.

References 1. Mulholland, S.J., Wyss, U.P.: Activities of daily living in non-Western cultures: range of motion requirements for hip and knee joint implants. Int. J. Rehab. Res. 24, 191–198 (2001) 2. Nagura, T., Dyrby, C.O., Alexander, E.J., Andriacchi, T.P.: Mechanical loads at the knee joint during deep flexion. J. Orthop. Res. 20, 881–887 (2002) 3. Li, G., et al.: Knee kinematics with a high-flexion posterior stabilized total knee prosthesis: an in vitro robotic experimental investigation. JBJS 86, 1721–1729 (2004) 4. Anouchi, Y., et al.: Range of motion in total knee replacement. Clin. Orthop. 331, 87–92 (1996) 5. D’Lima, D.D., Trice, M., Urquhart, A.G., Colwell, C.W., Jr.: Comparison between the kinematics of fixed and rotating bearing knee prostheses. Clin Orthop. 380, 151–158 (2000) 6. Li, G., et al.: Biomechanics of posterior-substituting total knee arthroplasty: an in vitro study. Clin. Orthop. 404, 214–239 (2002) 7. Tarnita, D., Pisla, D., Geonea, I., et al.: Static and dynamic analysis of osteoarthritic and orthotic human knee. J. Bionic. Eng. 16, 514–525 (2019) 8. Vaida, C., Birlescu, I., Pisla, A., et al.: Systematic design of a parallel robotic system for lower limb rehabilitation. IEEE Access 8, 34522–34537 (2020) 9. Bellemans, J., et al.: Fluoroscopic analysis of the kinematics of deep flexion in total knee arthroplasty. Influence of posterior condylar offset. J. Bone Joint Surg. Br. 84, 50–53 (2002) 10. Hartford, J.M., Banit, D., et al.: Radiographic analysis of low contact stress meniscal bearing total knee replacements. J. Bone Joint Surg. Am. 83, 229–264 (2001) 11. Kawamura, H., Bourne, R.B.: Factors affecting range of flexion after total knee arthroplasty. J. Orthop Sci. 6, 248–300 (2001) 12. Lizaur, A., et al.: Preoperative factors influencing the range of movement after total knee arthroplasty for severe osteoarthritis. J. Bone Joint Surg. Br. 79, 626–635 (1997) 13. Li, G., Zayontz, S., et al.: Kinematics of the knee at high flexion angles: an in vitro investigation. J. Orthop. Res. 22, 90–95 (2004) 14. Tarnita, D., Catana, M., Tarnita, D.N.: Contributions on the modeling and simulation of the human knee joint with applications to the robotic structures. In: Rodi´c, A., Pisla, D., Bleuler, H. (eds.) New Trends in Medical and Service Robots. MMS, vol. 20, pp. 283–297. Springer, Cham (2014). https://doi.org/10.1007/978-3-319-05431-5_19

Development of a New Knee Endoprosthesis

57

15. Vidal, A., Lesso, R., Rodríguez, R., et al.: Analysis, simulation and prediction of contact stresses in articular cartilage of the knee joint. WIT Trans. Biomed. Health, Model. Med. Biol. VII, 12, 55–64 (2007) 16. Tarnita, D., et al.: Effects of malalignment angle on the contact stress of knee prosthesis compts, using finite element method. Rom. J. Morphol. Embryol. 58(3), 831–836 (2017) 17. Tarnita, D., Calafeteanu, D., et al.: Development of a three-dimensional finite element knee prosthesis model. Appl. Mech. Mater. 150–155 (2016) 18. Calafeteanu, D., Tarnita, D., Catana, M., Tarnita, D.: Influences of varus tilt on the stresses in human prosthetic knee joint. Appl. Mech. Mater. 143–148 (2016) 19. Chantarapanich, N., et al.: A finite element study of stress distributions în normal and osteoarthritic knee joints. J. Med. Assoc. Thai. 92(Suppl 6), S97–103 (2009) 20. Walker, P.S.: Requirements for successful total knee replacements. Des. Considerations Orthop. Clin. North Am. 20(1), 15–29 (1989) 21. Morrison, J.B.: The mechanics of the knee joint in relation to normal walking. J. Biomech. 3(1), 51–61 (1970) 22. Szivek, J.A., Anderson, P.L., Benjamin, J.B.: Average and peak contact stress distribution evaluation of total knee arthroplasties. J. Arthroplasty 11(8), 952–963 (1996)

Observer Based Sliding Mode Control for a Knee Exoskeleton Yujie Su1 , Wuxiang Zhang1,2(B) , and Xilung Ding1,2 1

School of Mechanical Engineering and Automation, Beihang University, Beijing 100191, China 2 Beijing Advanced Innovation Center for Biomedical Engineering, Beihang University, Beijing 100191, China [email protected]

Abstract. To solve the problems in estimation and controlling caused by disturbance, an observer based sliding mode control method for an anthropomorphically designed knee exoskeleton is proposed in this study. Firstly, a sliding mode disturbance observer was designed to estimate the disturbance, including the internal disturbance caused by uncertainty in the dynamic model and external disturbance exerted by a human. Then, impedance control was used to achieve human-robot interaction control. Lastly, a sliding mode controller was designed to drive the exoskeleton. Experiments with disturbance were designed to verify the performance of the sliding mode disturbance observer and the robustness of the sliding mode controller. The results showed that the proposed method had a better performance in estimating and controlling in the presence of unknown disturbance. Keywords: Sliding mode disturbance observer · Impedance control Anthropomorphically designed knee exoskeleton

1

·

Introduction

Exoskeletons are wearable robots that can assist people in endurance, dexterity, and strength [1], and recently have become the focused area in robotics. Compliance is a critical topic in designing exoskeleton systems, and many researchers had tried to improve the compliance using anthropomorphically designed configurations and compliant actuators. As for the control, researchers focus on maintaining a proper human-robot interaction. Some control strategies have been proposed, and they can be categorized into position-based control, force-based control, and bio-signals-based control. Impedance control is a widely used force-based control strategy in exoskeleton systems. It is a hybrid force-position control in which the interaction force and position are adjusted simultaneously. However, the performance of impedance control is proven to decline in the presence of disturbance [2]. If there were c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 58–69, 2022. https://doi.org/10.1007/978-3-030-76147-9_7

Observer Based Sliding Mode Control for a Knee Exoskeleton

59

nonlinearity and uncertainty in the exoskeleton model, the performance declines and will affect the compliance. Adaptive impedance control was proposed to deal with this problem, which combined the impedance control and adaptive control method. Weiming et al. [3] used a fuzzy logic controller to adjust the parameters of impedance model in real-time. Rui et al. [4] proposed a dynamic identification method based on reinforcement learning to eliminate the uncertainty. For intention estimation, some sensor-less methods have been proposed recently to overcome expensive cost and constrained applicable places problems in sensory methods [5]. SAC (sensitivity amplification control) control strategy was used in BLEEX [6]. BLEEX could track the wearer’s movement without measuring the interaction force/torque. Mohammed et al. [7] proposed an observer based control method for a knee exoskeleton. A disturbance observer was used to estimate the torque of the wearer. Liang et al. proposed a torque estimation method based on the dynamic model of the exoskeleton and motor. Disturbance observer was then used to estimate the joint torque [8]. The sensor-less estimation methods aforementioned rely highly on the model accuracy. In most researches, the models and parameters are assumed to be accurate, but in some cases, the model is hard to establish accurately. Besides, nonlinear disturbance observer (NDO) has strict requirements on the disturbance [9], and it is difficult to design the parameters. Thus, eliminating the effect of the disturbance on the estimation needs to be further studied. This paper focuses on the control of an anthropomorphically designed knee exoskeleton, which is designed to fit the instantaneous changing centers of rotation (ICR) of the human knee, and thus a complex configuration was adopted. In this paper, the knee exoskeleton is considered to assist people in sitting positions, used for rehabilitation training and strength augmentation. Firstly, the exoskeleton system’s integrated model was established, and disturbance was included in the model. Then the controller was designed. A sliding mode disturbance observer (SMDO) was designed to estimate the interaction torque. Compared with NDO, SMDO has fewer requirements on the disturbance and is easier to design. Later, an impedance model was established to generate the desired trajectory. After that, sliding mode control (SMC) was used to drive the exoskeleton. Lastly, experiments were conducted to verify the effectiveness of the proposed control method.

2 2.1

The Exoskeleton System and Modeling Mechanical Configuration of the Exoskeleton

A six-bar mechanism consisting of two single-degree-of-freedom four-bar mechanisms was adopted to fit the ICR’s shape to realize the anthropomorphic design. The design and optimization are presented in [10]. The exoskeleton’s mechanical structure is shown in Fig. 1(a) and the schematic diagram is shown in Fig. 1(b). The four-bar mechanism OACB in Fig. 1(b) was designed to fit the ICR of the human knee joint. On the other hand, the four-bar mechanism F EDBAO was

60

Y. Su et al.

to improve the transmission performance. SEA (series elastic actuator) is used as the actuator, and the structure is shown in Fig. 2(a).

(a)

(b)

Fig. 1. (a) CAD view of exoskeleton and (b) Schematic diagram of exoskeleton (c) CAD view of SEA and schematic diagram of SEA

2.2

Identification and Estimation of the Dynamic Model

As analyzed in [10], the relations between different angles (θi , i = (2, 3, ..., 7)) are implicit and nonlinear, so the derivation of the dynamic model is challenging and time-consuming. For simplification, a linearized model is adopted, and it could be written as: A(s) θo (1) τe = B(s) where τ and θo are the input and output of the model, and A(s) and B(s) are polynomials, and the orders and coefficients are obtained through model identification and parameter estimation. To obtain the model and parameters, firstly, a series of random torque τ were applied to the exoskeleton and recorded the output θo . Try models with different orders and performed curve fitting to find out the most suitable model and parameters. Lastly, we chose a second-order model for its simplicity and acceptable errors. And it could be written as: τe = λ1 θ¨o + λ1 θ˙o + λ3 θo + λ4 + de ,

|de | < σd

(2)

Observer Based Sliding Mode Control for a Knee Exoskeleton

61

in which θo , θ˙o , θ¨o are angle, angular velocity and angular acceleration of the exoskeleton. And λ1 , λ2 , λ3 and λ4 are constant parameters. de is the error between the accurate model and will be compensated by adaptive control method. The parameters are shown in Table 1. Table 1. System parameters Parameter ks

cs

λ1

λ2

λ3

λ4

N · m/(rad/s) kg · m2 /rad2 N · m/(rad/s) N · m/rad N · m

Unit

N · m/rad

Value

1.8752 × 10−3 9.9171 × 10−5 1.3323

33.1814

−5.3736

σd N ·m

−3.6145 1.352

After identification, experiments were performed for validation. The exoskeleton moved in three different frequencies, and the movement and torque were recorded. The actual torque, modeling torque and their errors are shown in Fig. 2. It can be seen that the actual torques and the modeling torques were closed, and the errors are small and limited, indicating that the parameters were identified precisely and the simplification of the model was suitable.

(a)

(c)

(b)

(d)

(e)

(f)

Fig. 2. (a) Comparison of the actual torque and modeling torque, 1.5 Hz, (b) Comparison of the actual torque and modeling torque, 1 Hz, (c) Comparison of the actual torque and modeling torque, 0.75 Hz, (d) Error between actual torque and modeling torque in three frequencies

62

2.3

Y. Su et al.

Integrated Model of Exoskeleton and SEA

The schematic diagram of SEA is shown in Fig. 2(b). The torsional spring is considered as a spring and a damper in parallel so we could obtain the torque generated by the spring: τs = ks (θm − θo ) + cs (θ˙m − θ˙o )

(3)

where ks is the stiffness and cs is the damping and the value are shown in Table 1. τs stands for the torque generated by the torsional spring. And the motor and harmonic drive are modeled as one element, and θm stands for their output angle. As for the control of SEA, velocity source control is used, and thus the θ˙m is the control input. Replacing it with the u, Eq. (3) could be rewritten as: τs = cs u − cs θ˙o + ks (θm − θo )

(4)

Let τs = τe , the integrated model of the exoskeleton and SEA could be obtained: λ1 θ¨o + λ2 θ˙o + λ3 θo + λ4 + de = cs u − cs θ˙o + ks (θm − θo )

(5)

So the Eq. (5) could be simplified as follows: λ1 θ¨o = cs u + F − de

(6)

where F = ks θm − (ks + λ3 )θo − (cs + λ2 )θ˙o − λ4 .

3

Observer Based Impedance Control

The controller structure is shown in Fig. 3, and it consists of three parts: the sliding mode disturbance observer (SMDO), the sliding mode controller (SMC), and the impedance controller. The SMDO is designed to estimate the interaction torque (τi ). The impedance controller is designed to generate the reference motion trajectory. Finally, SMC is designed to drive the exoskeleton in the presence of disturbance, including human effort and uncertainty of the dynamic model. 3.1

Sliding Mode Disturbance Observer

Lemma 1. Given the perturbed nonlinear differential equation  1 2 x(t) ˙ + ω1 |x| sgn(x(t)) + ω2 sgn(x(τ ))dτ = ξ(t), ξ(t) ≤ |C|

(7)

where ξ(t) is the unknown disturbance, and C is the upper bound of the derivative of the disturbance. x(t) is √ the state to be solved, and ω1 and ω2 are constant. ˙ Providing that ω1 ≤ 1.5 C and ω2 ≤ 1.1C, then x(t) and it derivative x(t) 7.6x(0) converge to zero in finite time tr , where tr ≤ ω2 −C .

Observer Based Sliding Mode Control for a Knee Exoskeleton

63

According to [11], a spuertwisting algorithm could be used to design the SMDO. The interaction torque (τi ) will be introduced when a subject is wearing the exoskeleton, and thus the disturbance becomes a combination of model uncertainty and interaction torque. Rewrite the dynamic model of the exoskeleton system as follows: (8) λ1 θ¨o = cs u + F − D where D = de + τi . Define the state variable z and the sliding surface sd and the SMDO is designed as follows: ⎧ sd = θ˙o − z ⎪ ⎪ ⎪ ⎨ ˆ z˙ = f (θo ) + g(θo )u + D (9)  ⎪ ⎪ 1 ⎪ ˆ 2 ⎩ D = ω1 |sd | sgn(sd ) + ω2 sgn(sd ) Note that z and θ˙o are the angular velocity estimated by the observer and ˆ is the disturbance actual value, respectively, and sd stands for their error. D estimated by the observer. And the sgn(s) is the signum function. And the derivate of the sliding surface sd follows: ˆ =D−D ˆ s˙ d = θ¨o − z˙ = f + gu + D − f − gu − D Substituting Eq. (9) into Eq. (10), we can obtain:  1 s˙ d + ω1 |sd | 2 sgn(sd ) + ω2 sgn(sd (τ ))dτ = D

(10)

(11)

According to Lemma 1, sd and its derivative s˙ d will converge to zero in finite ˜ =D ˆ − D = 0, indicating that time if derivative of D is bounded. So we have D the SMDO could estimate the combining disturbance precisely. The disturbance torque estimated was considered as the interaction torque τˆi , and it follows: ˆ = d e + τi τˆi = D

(12)

ˆ D ˜ is the total disturNote 1. We use an approximation here and let τˆi = D. ˆ approaches to bance of the internal and external disturbance. The error of D zeros, and the SMC will also compensate the residual. τˆi stands for the estimated interaction torque, and it is only used in impedance control to generated the desired trajectory. So we can get the error of the interaction torque the observer estimates, which is: (13) τ˜i = τˆi − τi → de ≤ σd It could be seen that the residual of the interaction torque estimated by the observer will approach the error of dynamic model.

64

Y. Su et al.

3.2

SMDO Based Sliding Mode Control

To design the SMDO based SMC, the following sliding surface is considered: s = c1 e + e˙

(14)

where e is the error between the actual position and the reference trajectory so e = θo − θd , c1 > 0 and θd is the desired trajectory. Then the control law could be designed as followed: u=

1 ˆ − ηs − σd sign(s)) (λ1 θ¨d − F − c1 e˙ − D cs

(15)

where η > 0. And the stability of the SMDO based SMC controller is then analyzed. Proposition 1. For the approximate dynamic model of the knee exoskeleton system with the controller above, assume that the estimation error τˆi is bounded, then the trajectory of the system can be driven onto the sliding surface. Proof. Choose the Lyapunov function as V = 12 s2 , and by it differentiating it, we can get: 1 V˙ = ss˙ = c1 se˙ + s (cs u + F + D) − θ¨d (16) λ1 Substituting (8) and (15) into equation(16), we can obtain: ˆ − ηs − σd sign(s)) (D − D 1 ˜ η V˙ = s = s(D − σd sign(s)) − s2 ≤ 0 λ1 λ1 λ1

(17)

Therefore, the Lyapunov function is decreasing and the stability is then proved. 3.3

Impedance Control

Impedance control is used for human robot interaction control. An impedance model is designed and used to generate the reference trajectory according to the interaction torque. A linear model is designed as follows: 1 θd = τi Jd s2 + Bd s + Kd

(18)

where Jd , Bd , and Kd represent the inertia, damping, and stiffness of the impedance between human and exoskeleton. The impedance model proposed determines the exoskeleton’s assistance level should provide [12] and should be designed according to the wearer’s mobility. The parameters in the impedance model are chosen and optimized according to [13,14].

Observer Based Sliding Mode Control for a Knee Exoskeleton

4 4.1

65

Setup and Protocol of Experiments Verification of the SMDO Based SMC

Experiments were conducted to evaluate the performance of the SMDO based SMC and NDO based SMC. The NDO based SMC was designed as [7]. The exoskeleton was controlled by the two different controllers to track velocity trajectory with and without external disturbance force. The external disturbance force was generated by a spring connected to the exoskeleton and the platform, as shown in Fig. 4. The spring’s deflection changed when the exoskeleton moved, and thus the force exerted on the exoskeleton varied. The actual angular velocity of the two controllers was recorded and then compared. 4.2

Verification of the SMC Combined with Impedance Control

To verify the effectiveness of the SMC combined with impedance control, wearing experiments were conducted on the artificial leg as shown in Fig. 5. In this experiment, the artificial leg and the exoskeleton were connected by bandages. The artificial leg was driven by an external force exerted by humans and moved actively. Moreover, the exoskeleton moved following the trajectory generated by the impudence model. Two Inertial Measurement Units (IMU) were attached to the artificial leg’s thigh and shank to measure the knee joint’s angle. Both SMDO based SMC and NDO based SMC were included in the experiments. Besides, an experiment on a human was also conducted, and the setup was shown in Fig. 8(a). The healthy human subject sat on the platform, and his leg was also connected to the exoskeleton, and IMU was used to measure the knee angle. However, in this experiment, only SMDO based SMC was tested.

Fig. 3. Controller Structure

5 5.1

Fig. 4. External force/ torque

Fig. 5. Artificial leg wearing the knee exoskeleton

Experimental Results and Analysis Performance and Robustness SMDO Based SMC

The first experiment was to verify the two controllers’ performance under disturbance, which included internal and external disturbance. In the experiment, a

66

Y. Su et al.

given sinusoidal trajectory 5 Hz was used as the reference velocity trajectory, and two control methods were used to control the exoskeleton to track the reference velocity reference. The angular velocity was recorded, analyzed, and compared, and the results are shown in Fig. 6. It can be seen from Fig. 6(a) and Fig. 6(b) that both SMDO based SMC and NDO based SMC can track the desired velocity trajectory stably with and without external disturbance, which shows the stability of the adaptive SMC. By comparing the tracking errors in Fig. 6(c) and Fig. 6(d), it can be seen that the tracking error increased and began to chatter with external disturbance, but they were still bound in small ranges. Additionally, the RMS of the error form SMDO based SMC is bigger without external disturbance while smaller with external disturbance, indicating that SMDO based SMC control performs better when an unknown disturbance exists.

(a)

(b)

(c)

(d)

Fig. 6. Comparison of the tracking precision without external disturbance torque: (a) Tracking results without external disturbance, (b) Tracking result with external disturbance, (c) Tracking errors without external disturbance (RMS, SMDO based SMC: 0.2672, NDO based SMC: 0.1909), (d) Tracking errors with external disturbance (RMS, SMDO based SMC: 0.2507, NDO based SMC: 0.3070)

5.2

Performance of SMC Combined with Impedance Control

In the artificial leg experiment, the knee and exoskeleton angles were recorded, and they were shown in Fig. 7. It can be seen from the results that both SMC combined with impedance control were able to shadow the movements of the artificial leg, indicating that both disturbance observers could estimate the interaction torque and the impedance model was suitable. Besides, by comparing the tracking error and the RMS, one can also notice that SMDO based SMC has smaller tracking errors, indicating that SMDO based SMC has a better performance on estimating and controlling under unknown disturbance. By comparing the two controllers’ errors from the two experiments, it can be seen that SMDO based SMC performs better when the external disturbance exists. ⎧ x = cos(t) ⎪ ⎪ ⎨ X1 (x) = y = sin(t) ⎪ x ⎪ ⎩z = y

Observer Based Sliding Mode Control for a Knee Exoskeleton

67

The result from the human experiment is shown in Fig. 8(c). The tracking results from Fig. 8(b) shows that the exoskeleton can track human movement stably, which is consistent with the results from artificial leg experiments. The tracking error in Fig. 8(c) shows the tracking error is bounded, and it is beyond 10% of the amplitude. The magnitude and RMS of the error are a litter larger than those in the artificial leg experiments. That was because the human leg moved faster, and the motion range was also bigger. The tracking performance indicates that the proposed SMDO based SMC controller is a suitable solution in estimating and controlling for our anthropomorphically designed knee exoskeleton.

(a)

(b)

(c)

(d)

Fig. 7. Comparison of the tracking precisions of experiment on artificial leg: (a) Tracking result of SMDO based SMC, (b) Tracking result of NDO based SMC, (c) Tracking error of SMDO based SMC(RMS: 0.0426), (d) Tracking error of NDO based SMC (RMS: 0.0662)

(a)

(b)

(c)

Fig. 8. Set up and result of human experiment. (a) Experiment setup, (b) Tracking result of SMDO based SMC, (c) Tracking error of SMDO based SMC(RMS: 0.072)

6

Conclusion

In this study, a disturbance observer based sliding mode controller was designed for human-robot interaction control of an anthropomorphically designed knee

68

Y. Su et al.

exoskeleton. A sliding mode disturbance controller was used to estimate the interaction torque and external disturbance. Then an impedance controller was used to generate the desired trajectory according to the impedance model. Lastly, a sliding mode controller was used to drive the exoskeleton to move accurately under disturbance. The results from the velocity tracking and wearing show that the tracking error with external disturbance of the proposed SMDO based SMC is smaller, indicating that it has a better performance on estimating and controlling when external disturbance exists. Furthermore, the wearing experiment result shows that the proposed observer based impedance control was able to track the wearer’s motion. Acknowledgment. This research is supported by the National Natural Science Foundation of China (NSFC) under grant No. 91848104, the National Key R&D Program of China under grant No.2016YFE0105000.

References 1. Zhang, L., Liu, G., Han, B., Wang, Z., Li, H., Jiao, Y.: Assistive devices of human knee joint: a review. Robot. Auton. Syst. 125, 103394 (2020) 2. Moghadam, M.M., Shahi, H., Yousefi-Koma, A.: An improvement on impedance control performance of an exoskeleton suit in the presence of uncertainty. In: 2015 3rd RSI International Conference on Robotics and Mechatronics (ICROM), pp. 412–417. IEEE (2015) 3. Mi, W., Zhang, T.: Fuzzy variable impedance adaptive robust control algorithm of exoskeleton robots. In: 2019 Chinese Control Conference (CCC), pp. 4302–4307. IEEE (2019) 4. Huang, R., Cheng, H., Guo, H.L.: Learning virtual impedance for control of a human-coupled lower exoskeleton. J. Univ. Electron. Sci. Technol. China 47(3), 321–329 (2018) 5. Ugurlu, B., Nishimura, M., Hyodo, K., Kawanishi, M., Narikiyo, T.: A framework for sensorless torque estimation and control in wearable exoskeletons. In: 2012 12th IEEE International Workshop on Advanced Motion Control (AMC), pp. 1–7. IEEE (2012) 6. Kazerooni, H., Racine, J.-L., Huang, L., Steger, R.: On the control of the berkeley lower extremity exoskeleton (bleex). In: Proceedings of the 2005 IEEE International Conference on Robotics and Automation, pp. 4353–4360. IEEE (2005) 7. Mohammed, S., Huo, W., Huang, J., Rifa¨ı, H., Amirat, Y.: Nonlinear disturbance observer based sliding mode control of a human-driven knee joint orthosis. Robot. Auton. Syst. 75, 41–49 (2016) 8. Liang, C., Hsiao, T., Hsiao, C.C.: Joint torque estimation of a powered exoskeleton under compliance control loop. In: 2018 International Automatic Control Conference (CACS), pp. 1–6. IEEE (2018) 9. Chen, W.H., Ballance, D.J., Gawthrop, P.J., O’Reilly, J.: A nonlinear disturbance observer for robotic manipulators. IEEE Trans. Industr. Electron. 47(4), 932–938 (2000) 10. Xiao, B., Shao, Y., Zhang, W.: Design and optimization of single-degree-of-freedom six-bar mechanisms for knee joint of lower extremity exoskeleton robot. In: 2019 IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 2861– 2866. IEEE (2019)

Observer Based Sliding Mode Control for a Knee Exoskeleton

69

11. Hall, C.E., Shtessel, Y.B.: Sliding mode disturbance observer-based control for a reusable launch vehicle. J. Guidance Control Dyn. 29(6), 1315–1328 (2006) 12. Huo, W., Mohammed, S., Amirat, Y., Kong, K.: Active impedance control of a lower limb exoskeleton to assist sit-to-stand movement. In: 2016 IEEE International Conference on Robotics and Automation (ICRA), pp. 3530–3536. IEEE (2016) 13. Tran, H.T., Cheng, H., Duong, M.K., Zheng, H.: Fuzzy-based impedance regulation for control of the coupled human-exoskeleton system. In: 2014 IEEE International Conference on Robotics and Biomimetics (ROBIO 2014), pp. 986–992. IEEE (2014) 14. Huo, W., Mohammed, S., Amirat, Y.: Observer-based active impedance control of a knee-joint assistive orthosis. In: 2015 IEEE International Conference on Rehabilitation Robotics (ICORR), pp. 313–318. IEEE (2015)

Design and Motion Simulation of a New Exoskeleton Leg Mechanism Ionut Geonea(B) , Cristian Copilus, i, Sorin Dumitru, and Adrian Sorin Ros, ca University of Craiova, Craiova, Romania [email protected]

Abstract. In this paper, we will present a new constructive solution of a robotic system for the rehabilitation of people with locomotor disabilities. In the first part of the paper, we performed an experimental analysis of human gait, with help a Biometrics Data Acquisition System. This step is necessary in order to perform a comparative analysis between human gait and developed exoskeleton movement. In a first stage, we presented a constructive solution for the mechanism that shapes the legs of the exoskeleton. This solution is based on a closed kinematic chain and is driven by a single motor. Based on the mechanism kinematic scheme, the virtual prototype model developed in the SolidWorks CAD environment is presented. The virtual prototype is used to perform a motion simulation in ADAMS dynamic analysis software. In this way, we performed a comparative analyse of the motions of healthy human subjects in relation to the motions realized by the designed exoskeleton. Keywords: Human leg · Motion assistance · Exoskeleton robot · Simulation

1 Introduction Recently, research on the development of robotic systems for the rehabilitation of people with locomotor disabilities has gained momentum. In these studies, the progress made by robotic rehabilitation systems, methods and means of locomotor rehabilitation, as well as future research directions are evaluated [1–3]. Depending on the rehabilitation principle used, active robotic rehabilitation systems can be classified into five main categories, as follows [4]: 1) treadmill gait trainers; 2) foot-plate-based gait trainers; 3) mobile robotics-based devices; 4) stationary gait trainers; 5) ankle rehabilitation systems, as active foot orthoses. An exoskeleton is defined as an external system, usually a robotic system, which is carried by a human subject [5]. Humans wear exoskeletons for two reasons: to assist walking in case of people with disabilities and to increase the ability to carry heavier loads e.g., in the case of soldiers [6]. Recently, these exoskeletons have been developed in the case of industrial applications, being worn by operators in the automotive industry, especially the exoskeletons for the upper limbs [7]. In summary, the exoskeletons’ main purpose is to provide supplementary forces to people with impaired leg function. Multi joint exoskeletons presented in [8] and most of recent exoskeletons are based on the principle to use a predefined gait trajectory control. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 70–77, 2022. https://doi.org/10.1007/978-3-030-76147-9_8

Design and Motion Simulation of a New Exoskeleton Leg Mechanism

71

Robotic devices have the potential to make therapy reasonably priced and therefore accessible to patients and for longer time [9]. However, currently, most used principles in robot-assisted rehabilitation therapies are expensive and use a robotic rehabilitation device on a treadmill. This has the advantage that patients with disabilities are suspended with belts at a higher frame, thus maintaining balance. In this paper, we will present a new exoskeleton constructive solution driven by a single engine, designed to help human locomotion.

2 Analysis of Human Gait In this paragraph, we present an experimental human gait characterization. For that purpose, the integrated equipment Biometrics for the human biomechanics experimental analysis was used [10, 11]. Biometrics allows the simultaneous collection of biomechanical data of kinematic and dynamic nature through electrogoniometers, accelerometers, force platforms, EMG sensors, contact pressure sensors and other types of sensors. The data collection process is straight forward (see Fig. 1). The mounted electrogoniometers on the subject transmit the information received via cables to the DataLog, device that converts the received signal and transmits it via Bluetooth to the PC.

Fig.1. Human participant with attached goniometer type sensors.

The data received by the computer is converted by the Biometrics software into diagrams. The gait analysis was performed on ten healthy human participants. We performed statistical data processing, in order to obtain the average cycle. Figure 2 shows the average extension flexion angle during a walking cycle for the hip and knee joints of the lower limb for the participants analysed. The results obtained for the average cycles of the leg joint angles are useful to compare them with the motion performed by the exoskeleton robotic system. From the obtained data presented in Fig. 2, it is observed that the amplitude of the angle in the knee joint is 62° and 28° for the hip joint. Figure 3 shows the ground reaction forces obtained with the help of force platforms, for a participant with a mass of 71 kg and 1.68 m height. The first graph corresponds to the right leg GRF and the next graphs for the following five steps that are performed. It is observed that the maximum value of the ground reaction force (GRF) is 1.22 times higher than the gravitational force of the participant.

72

I. Geonea et al.

Hip joint angle

20 10 0

-10

0

20

40

60

Gait cycle [%]

Knee joint angle

80 Angle [deg]

Angle [deg]

30

80

100

60 40 20 0

0

20

40

60

80

100

Gait cycle [%]

Force [daN]

Fig. 2. The average variation cycle for the flexion-extension angle corresponding to the analysed participants.

Fig. 3. Variation graphs for ground reaction forces measured by force platforms.

3 Design of a New Exoskeleton Leg Mechanism In this section, we present the CAD model of an exoskeleton designed to assist human locomotion for rehabilitation purposes. The previous results obtained by the research team from the University of Craiova on this research direction are presented in the articles [8, 9]. These articles present human leg motion assistance mechanisms, based on Chebyshev-type and pantograph-type kinematic chains. New structural solutions are presented in Fig. 4. Compared to previous achievements, the newly developed solutions provide laws of motion that accurately reproduce the movement of human participants, aspect presented in the next paragraph. A first structural solution is presented, in Fig. 4, a), which is based on parallelogram kinematic chains and comprise 9 links and 14 rotational joints. This ensures the assistance of the movement of the hip and knee joints. The human hip and knee are represented by the joints F and L in the structure of the mechanism. The second structural solution is based on 11 links and 16 kinematic joints (see Fig. 4, b)). In this solution, we enable the actuation of all three joints of the leg, namely, the hip, which is materialized by the joint F, the knee by the joint M, and the ankle by the joint N. This solution complements the previous solution by providing assistance also to the ankle joint. The solution of Fig. 4, c) is based on the kinematic chain of a pantograph mechanism and comprises 9 links and 13 rotational joints. We also intend to present results in

Design and Motion Simulation of a New Exoskeleton Leg Mechanism

73

further studies for this original solution. The calculation of the degrees of mobility with Chebyshev’s formula for planar kinematic chains indicates that all three proposed solutions of mechanisms for the leg of an exoskeleton have 1 DOF so only one motor is required for device actuation.

6

F E 4

G

F H J

I

3

D

A

C

B 2

1

5

7 K 8 9

M a)

b)

c)

Fig. 4. New structural solutions proposed for human locomotion assistance mechanisms.

In this paper, we will focus on studying, solution presented in Fig. 4, a), from a kinematic and dynamic point of view. This mechanism incorporates the kinematic chain of the human leg, excluding the ankle joint. The trajectory of the point M obtained by simulation in ADAMS, is a closed, ovoid curve (see Fig. 6). It is a low-cost solution, because it can be driven by a single motor, (see Fig. 5). We modelled the experimental prototype of an exoskeleton, in which the legs are designed based on the kinematic scheme shown in Fig. 4, a). We performed a kinematic simulation in ADAMS, where we considered the joints A, E, and F connected to the base by rotational connections. The purpose of this simulation is to obtain the trajectory made by the point M, which is the ankle joint. The resulting trajectory is a closed curve that has an ovoid shape, (see Fig. 6). This trajectory is realized during human walking gait, as presented in similar research articles [12]. We made a complete assembly of the exoskeleton, which is shown in Fig. 5. It comprises the two mechanisms for the right and left leg, an upper frame (10) on which the elements 6, 3 and 1 are connected by bolts. On the upper frame, which will be fixed to the pelvis of the human participant with disabilities, a motor is mounted (11), which will drive the shaft (12) at the end of which are mounted the elements (1) of the mechanism corresponding to the right and left leg, through a chain transmission (13).

74

I. Geonea et al.

10 11 12 13

Fig. 5. CAD model of the exoskeleton assembly.

4 Virtual Simulation of Exoskeleton Type Robotic System In this paragraph we present the results obtained in the dynamic gait simulation for the exoskeleton worn by a virtual mannequin. Human gait is a special case of modelling a multibody system, in which the impact, friction and slip occurs. All these phenomena observed during the contact between the foot and the ground, must be introduced in the contact model, in order to have a modelling as close as possible to reality. In some multibody models for bipedal walking, the contact between the foot and the ground is modelled by the introduction of additional kinematic joints [13]. In the type of model, we want to build, the contact type model is used by defining an impact function. Several studies for the foot-floor interface are presented in the literature. Valiant [14] has published studies on the dynamic characteristics of the plantar surface of the foot. These studies were developed by Meglan [15], who studied the loading-deformation parameters of the foot with the floor. Meglan developed a general equation valid for calculating the vertical reaction force of the foot with the ground, using optimization techniques, to frame the data obtained by Valiant, in a polynomial equation. The equation obtained by Meglan is: F(x, v) = (4.538e10)x4 + (9.719e10)x4 v

(1)

where: F is the resulting force, x is the penetration depth and v is the penetration velocity. The ADAMS modelling tools do not allow the introduction of such a polynomial contact function, but instead offer the possibility to introduce a cubic function (with the IMPACT method for defining contact):        2 F(x, v) = kx + c x d · 3 − 2x d v (2) where: k is a linear constant, (model of a linear spring), c is the damping coefficient and d is the depth of penetration. The approximate values of the contact coefficients, by the impact method, used in ADAMS, according to Meglan’s relations are: k = 1.0E + 08 N/mm; c = 3000 N/mm/sec and d = 0.1 mm [14]. Taking into account the above,

Design and Motion Simulation of a New Exoskeleton Leg Mechanism

75

we defined the contact between the foot and the ground, by selecting the two solid components of the sole and the floor, as shown in Fig. 6. The virtual model of the human mannequin was developed in Solid Works. The mannequin limb proportions are designed respecting the average dimensions of humans. The masses for the anatomical segments were defined based on indications from the literature [14]. The contact model parameters, inserted in the ADAMS dynamic model uses the following parameters: contact stiffness, contact force exponent, damping, penetration depth and friction coefficients [16], with values indicated in Fig. 6. We also defined friction coefficients in the exoskeleton joints. The angular variation for the left and right knee joints are presented in Fig. 7. Also, the ADAMS computed hip joint angular variation are presented in Fig. 8. The computed displacements of ankle joint are presented in Fig. 9. The values resulting from the simulation are close as a variation range with those obtained for normal walking, presented in paragraph 2. Also, from the dynamic analysis in ADAMS are obtained the reaction forces with ground, presented in Fig. 10. The obtained GRF are almost identical with the experimental one from Fig. 3. The operating torque of the motor has a maximum value of 32 Nm.

Fig. 6. Adams defined contact parameters.

Fig. 7. Computed flexion-extension angle of the exoskeleton knee joints.

76

I. Geonea et al.

Fig. 8. Computed flexion-extension angle of the exoskeleton hip joints.

Fig. 9. Computed displacement of the ankle joint of the exoskeleton of the left leg: (a) on X-axis; (b) on Y-axis.

Fig. 10. Computed foot ground reaction forces.

5 Conclusions In this paper, we presented an original structural solution for a mechanism that ensures similar movements as the human leg. Based on these structural solutions we designed a virtual model, whose movement we analysed using ADAMS. We performed this simulation in ADAMS for the case when the exoskeleton is worn by a virtual mannequin. It can be seen from Fig. 7 that the angle of variation in the knee joint reaches 55°, a value close to that achieved in human gait. For the hip joint, the variation angle made by the exoskeleton is 30°, as is presented in Fig. 8, a value similar to measurements in human participants. From the dynamic analysis we obtained the interaction forces between the foot of the virtual mannequin and the ground. These are shown in Fig. 10. Their variations show maximums at contact with the ground, and a second maximum on the graph is recorded when the foot is detached from the ground, similar to the experimental results shown in the Fig. 3. The maximum values of 900 N, are comparable to the total weight of the virtual exoskeleton-mannequin assembly. The results show that

Design and Motion Simulation of a New Exoskeleton Leg Mechanism

77

the designed exoskeleton achieves motions that faithfully reproduce human gait in knee joint angles and ankle motion and shape and magnitude of ground reaction forces.

References 1. Anama, K., Al-Jumaily, A.A.: Active exoskeleton control systems: state of the art. Procedia Eng. 41, 988–994 (2012) 2. Young, A.J., Ferris, D.P.: State-of-the-art and future directions for lower limb robotic exoskeletons. IEEE Trans. Neural Syst. Rehabil. Eng. 25(2), 171–182 (2017). https://doi.org/10.1109/ TNSRE.2016.2521160 3. Lajeunesse, V., Vincent, C., Routhier, F., Careau, E., Michaud, F.: Exoskeletons’ design and usefulness evidence according to a systematic review of lower limb exoskeletons used for functional mobility by people with spinal cord injury. Disabil. Rehabil. Assist. Technol. 11(7), 535–547 (2016). https://doi.org/10.3109/17483107.2015 4. Díaz, I., Gil, J.J., Sánchez, E.: Lower-limb robotic rehabilitation: literature review and challenges. J. Robot, 11 (2011). https://doi.org/10.1155/2011/759764 5. Louie, D.R., Eng, J.J.: Powered robotic exoskeletons in post-stroke rehabilitation of gait: a scoping review. J Neuroeng. Rehabil. 13(1), 1 (2016). https://doi.org/10.1186/s12984-0160162-5 6. Viteckova, S., Kutilek, P., Jirina, M.: Wearable lower limb robotics: a review. Biocybernetics Biomed. Eng. 33, 96–105 (2013) 7. Tingfang, Y., Cempinia, M., et al.: Review of assistive strategies in powered lower-limb orthoses and exoskeletons. Robot. Auton. Syst. 64, 120–136 (2015) 8. Geonea, I.D., Tarnita, D.: Design and evaluation of a new exoskeleton for gait rehabilitation. Mech. Sci. 8, 307–321 (2017). https://doi.org/10.5194/ms-8-307-2017 9. Geonea, I., Dumitru, N., Tarnita, D., Rinderu, P.: Design and kinematics of a new leg exoskeleton for human motion assistance. In: IFToMM World Congress on Mechanism and Machine Science, pp. 165–174. Springer, Cham (2019).https://doi.org/10.1007/978-3-03020131-9_17 10. Tarni¸ta˘ , D., Geonea, I., Petcu, A., Tarni¸ta˘ , D.-N.: Experimental characterization of human walking on stairs applied to humanoid dynamics. In: Rodi´c, A., Borangiu, T. (eds.) RAAD 2016. AISC, vol. 540, pp. 293–301. Springer, Cham (2017). https://doi.org/10.1007/978-3319-49058-8_32 11. Tarnita, D.: Wearable sensors used for human gait analysis. Rom. J. Morphol. Embryol. 57(2), 373–382 (2016) 12. Wang, M., Ceccarelli, M., Carbone, G.: A feasibility study on the design and walking operation of a biped locomotor via dynamic simulation. Front. Mech. Eng. 11(2), 144–158 (2016). https://doi.org/10.1007/s11465-016-0391-0 13. Wojtyra, M.: dynamical analysis of human walking. In: 15th European ADAMS Users, Conference Technical Papers, Rome, Italy (2000) 14. Patton, J.L.: Forward dynamic modeling of human locomotion. Ph.D. thesis, Michigan State University (1993) 15. Kecskemethy, A.: Integrating efficient kinematics in biomechanics of human motions. Procedia IUTAM 2, 86–92 (2011). https://doi.org/10.1016/j.piutam.2011.04.009 16. Wang, M.F., Ceccarelli, M., Carbone, G.: Experimental tests on operation performance of a LARM leg mechanism with 3-DOF parallel architecture. Mech. Sci. 6, 1–8 (2015). https:// doi.org/10.5194/ms-6-1-2015

Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton Mattia Vincenzo Lo Piccolo1 , Giovanni Gerardo Muscolo2(B) , and Carlo Ferraresi1 1 DIMEAS-Department of Mechanical and Aerospace Engineering, Politecnico di Torino,

Turin, Italy 2 Department of Computer Science, University of Verona, Verona, Italy

[email protected]

Abstract. This paper presents a preliminary study on a novel upper body exoskeleton conceived for human effort reduction in uncomfortable operations. The exoskeleton is constituted by a light wearable structure energized by two McKibben pneumatic artificial muscles, which could be used in passive (such as springs) or active mode (such as actuators). In this work, the modelling and the design of the exoskeleton is presented. The performance of the exoskeleton is evaluated considering two commercial McKibben muscles used in passive mode. The results underline that the exoskeleton can be adapted to compensate the weight of the upper limbs and a load in the hands of an operator under defined working conditions. Keywords: Fluidic muscle · McKibben pneumatic actuator · Pneumatic exoskeleton · Artificial muscle · Upper body exoskeleton

1 Introduction Over the past two decades, upper body exoskeletons have been developed for many applications ranging from heavy and light lifting [1, 2], haptic and fundamental research on haptic devices [3], bilateral tele-robotics [4], rehabilitation and physical robotic therapy [5] and military and defense applications [6]. Any type of wearable robot should provide safety, comfort, and effective human-machine interaction [7]. Many categories of exoskeletons with different final objectives are developed [8]: i.e., as support to provide physical therapy sessions in the clinical setting, as rehabilitation cases, or in controlled environments such as a physiotherapy clinic. Some of these are designed to allow movement to patients who have completely lost their body mobility [9]. In these cases, the exoskeleton is often able to partially replace the lost human movements, and a robust control should be provided to the wearable system in order to allow the full stability and safety to the user. Today, industrial sector is one of the most important field of application for exoskeleton devices, for assisting operators and allowing them to perform their activities more easily and with longer duration [2, 10]. The biggest challenge is to reduce the weight and the power absorbed by these devices. On the other hand, the exoskeleton adds considerable inertia to the user’s movements: in the case of active exoskeletons, the © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 78–85, 2022. https://doi.org/10.1007/978-3-030-76147-9_9

Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton

79

added inertia must be overcome increasing the power supplied to the electric actuators; in the case of passive exoskeletons, the added inertia must be overcome by the user [11]. A broad overview of upper limb exoskeletons is presented in [8]. In this article, several disadvantages of using pneumatic actuators to energize the device are highlighted, although this is especially true in the case of active exoskeletons. Robotic degrees of freedom are not the same as the human bone model, due to the biological misalignment that reaches up to some centimeters during normal movement [12] and can cause discomfort, non-linearity in movement or even injury [7]. In literature, the effects of discomfort, caused by devices in which these misalignments have not been considered, are frequently underlined [13, 14]. In the case of rehabilitative robotics, for example, the misalignments between wearable robots and their users have proved to be responsible for less effective treatment [15, 16]. A possible solution to these problems is applying active soft insoles [11], which show great results in reducing the impedance and allowing a more natural movement to the user who wears it. Due to the effect of the arm’s position on the shoulder, ergonomists recommend that workers sitting at a desk or table try to position their arms with 20° (or less) of abduction and 25° (or less) of flexion. Workers who are required to hold their arms in a sustained position over the head, for operations that can take a long time, are particularly sensitive to degenerative tendinitis in biceps and supraspinatus [17]. The glenohumeral joint supports the compressive forces that are estimated to reach 50% of body weight [17]. These elements place very high demands on the compatibility of the device with its operators and on its comfort and ergonomics. In this work, the authors present an exoskeleton whose kinematic structure is independent from the osteoarticular structure of the user; therefore, the motion of the user’s limb has no addictive constraints. The energy to support the gravitational load of the arms’ weight is provided by compressed air, stored in two McKibben muscles (one for each arm) that act as pneumatic springs. The paper presents the functional design of the exoskeleton and the evaluation of its possible performances. The results show that the device, despite its simple design, around a definite working condition is able to significantly reduce the human effort.

2 Functional Design of the Upper Body Exoskeleton In this research work, the individual’s physiognomic data of the standard man have been used [18, 19]. For an average mass of 70 kg and an average height of 1.72 m, the mass of the limb segments are 1.96 kg for the upper arm, 1.12 kg for the forearm, and 0.42 kg for the hand [20]. When the limb is extended horizontally, gravity creates a large moment on the shoulder joint that must be contrasted by contraction of the shoulder muscles. The graphs of Fig. 1 show the value of the shoulder torque separately varying the shoulder angle θ1 (above) and the elbow angle θ2 (below). Angle θ1 represents mainly a flexion motion, but also a limited abduction may occur during the use of the exoskeleton, therefore it is defined as the “elevation angle”. In the above graph, the elbow is maintained extended (θ2 = 0°) while θ1 varies from 0° to 120°. Due to the variation of the weight’s arm with the angle, the value of the moment at the shoulder increases as

80

M. V. Lo Piccolo et al.

θ1 increases until reaching the horizontal limb condition (θ1 = 90°), while it decreases for values greater than 90°. In the below graph, the arm is kept horizontal (θ1 = 90°) while forearm flexes from 0° to 150°. For sake of simplicity, the condition of the above graph, i.e. with the elbow angle kept constant at 0°, was assumed for the performance verification of the device, being the most onerous. Figure 2 shows the wearable exoskeleton in a first 3D CAD model (on the left) and in a sagittal scheme with its main parameters (on the right). The exoskeleton is conceived as a simple and light structure worn by the user using three belts (two fixed on the shoulders and one on the waist). Two pneumatic muscles (one for each arm) are fixed to the structure at their lower ends, while the upper end of each actuator is attached to a wire, which is obliged to pass on pulleys and is connected to a bracelet fixed to the individual’s arm. Parameters a, c, α and β are geometrical characteristics of the device, b is the distance between the last pulley and the attachment point of the wire on the bracelet, θ1 is the elevation angle; r is the lever arm of the pneumatic muscle tension with respect to the shoulder joint.

Fig. 1. Shoulder torque due to gravity with elbow angle θ2 = 0° while varying shoulder angle θ1 (above), and with shoulder angle θ1 = 90° while varying elbow angle θ2 (below).

Fig. 2. Scheme of the exoskeleton: a, c, α and β are design data, b is the free length of the actuator tendon, r is the lever arm of the actuator tension.

Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton

Considering the triangle a, b, c, the length b can be calculated as:  b = a2 + c2 + 2accos(θ1 + β − α);

81

(1)

Given the semi perimeter (S) of the triangle, its area (A) can be calculated with the Heron’s formula:  A = S(S − a)(S − b)(S − c); (2) Hence the lever arm (r) of the actuator’s force with respect to the shoulder joint is: r = 2A/b;

(3)

The exoskeleton has been conceived to help a worker during operations that have to be performed with its arms in elevated position. It may be employed in two operating configurations: 1) active exoskeleton: the pressure in pneumatic muscles is regulated in real-time by an automatic control system; in this configuration, also an external supply source is needed; 2) passive exoskeleton: the pneumatic muscles are pressurized at a given value and are used as pre-loaded springs; in this configuration, the system is fully passive, and no external supply and control are needed. Since this research is aimed at realizing a device as simple as possible, the passive configuration has been considered.

3 Pneumatic Muscles The pneumatic artificial muscles (PAMs) are driven by a compressed gas, in our case air, which acts on an inflatable internal bladder sheathed by a twisted double-wave braided mesh which contracts its longitudinal length as it expands in a radial direction. During contraction, the artificial muscle generates a tensile force, which will be able to pull a tendon element. The most popular pneumatic muscle is the McKibben muscle, but different types are described in literature [21–24]. Several models were proposed to describe the static behavior of the McKibben muscle. For mechanical characterization of such muscles, the tensile force is reported as a function of the contraction ratio. Typically, the tensile force decreases with increasing contraction under constant pressure. The relationship between force and pressure is almost linear at constant extension, by adjusting the system pressure it is possible to vary the muscle length in isotonic condition [25–30]. In order to verify the effectiveness of pneumatic muscles in our passive exoskeleton prototype, some simulations have been done considering commercial McKibben pneumatic muscles produced by Shadow Robot Company, UK. In base of the required performance of the exoskeleton, two McKibben muscles of 20 mm and 30 mm diameter have been chosen. The nominal characteristics of the selected PAMs are shown in Fig. 3, left [31].

4 Simulations In [32], the authors underline how severe shoulder flexion (or abduction), is predictive of chronic or recurrent shoulder disorders. This is not in contrast with the research presented

82

M. V. Lo Piccolo et al.

in [33], where the authors underline to redesign jobs in order to minimize the elevation of arms. For these reasons, the authors decided to design the exoskeleton in order to compensate the shoulder torque of the user in the more critical elevation angle of above 90°. The geometrical model of Eqs. (1)-(3) requires the definition of parameters a, c, α and β (see Fig. 2). In order to find an optimal solution, simulations have been performed with different combinations of those parameters. It was decided to explore the behavior of the exoskeleton imposing to start from the elevation angle θ1 at 90° and with θ2 = 0. Figure 3 (right) shows the results of a simulation performed on the user shown in Fig. 2 with and without a load of 1 kg in the hand. The graphs report the torques on the shoulder joint due to gravity (black, continuous line), including a load of 1 kg in the hand (black, dotted line), and created by the Shadow PAM with diameter 20 mm and 30 mm, and with several supply pressures (2, 4 and 6 bar). The parameters used in the sketch of the Fig. 2 are: a = 0.15 m, c = 0.2 m, α = 80°, β = 14°. Including 1 kg load the moment increases, but this may be compensated by the same PAM varying only pressure.

Fig. 3. Left: nominal characteristic of shadow PAM with diameter 20 mm and 30 mm [31]; right: torques on the shoulder joint due to gravity (black, continuous line), including a load of 1 kg in the hand (black, dotted line), and created by the Shadow pneumatic artificial muscle with diameter of 20 mm, 30 mm (a = 0.15 m, c = 0.2 m, α = 80°, β = 14°).

Figure 4 shows the torques on the shoulder joint due to gravity (black, continuous line), including a load of 1 kg in the hand (black, dotted line), and created by the Shadow pneumatic artificial muscle with diameter of 20 mm (at pressure of 2 bar) varying respectively the parameter a from 0.05 m to 0.25 m and angle α from 20° to 80° (see Fig. 2). It can be noted by the graphs how the variation of α has a more relevant influence on the compensation of the user’s effort. In particular, if α could be varied from 80° to 20°, the effort of the user may be completely compensated in the studied range of the elevation angle from 90° to 115°.

Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton

83

Fig. 4. Torques on the shoulder joint due to gravity (black, continuous line), including a load of 1 kg in the hand (black, dotted line), and created by the shadow PAM with diameter of 20 mm, pressure of 2 bar and length a varying from 0.05 m to 0.25 m (left) (c = 0.2 m, α = 80°, β = 14°) and angle α varying from 20° to 90° (right) (a = 0.15 m, c = 0.2 m, β = 14°).

5 Conclusion The exoskeleton here presented uses the McKibben muscles in a passive way, therefore representing a very simple, cheap and light device to reduce the effort on the shoulder joint due to the weight of the upper limbs when the user must perform operations with arms in elevated position. The graphs comparing the torque at the shoulder created by gravity with the one generated in opposition by the PAM show that, by adequately adjusting the preload pressure in the muscle, it is possible to reach a condition of complete balance in a nominal working position, which in our project is included in a range of the elevation angle from 90 to 120°. The preload pressure can be selected based on the anthropometric characteristics of the operator and also on the presence of a tool in the hand. It is important to note that the PAM is used in the device as a pneumatic spring but, unlike the mechanical springs, the variation of the preload pressure determines a variation of the real stiffness of the PAM, which is thus customized to the application in a more effective way. Around the working position, the operator may freely vary the elevation angle with a limited effort. The structure of the exoskeleton can also allow free abduction movements to improve ergonomics. The present work is a first feasibility study. In our analyses we modified only elevation angle θ1 , while fixing elbow angle θ2 = 0° for the sake of simplicity. Future work will simulate various working conditions, with defined sequences of movements that will affect both joints of the limb, aimed at optimal exploiting of the actuators’ characteristics and to optimize the mechanical structure of the device and the transmission system towards the limb.

84

M. V. Lo Piccolo et al.

References 1. Perry, J., Rosen, J., Burns, S.: Upper-limb powered exoskeleton design. IEEE-ASME Trans. Mechatron. 12, 408–417 (2007) 2. Panero, E., Muscolo, G.G., Pastorelli, S., Gastaldi, L.: Model based analysis of trunk exoskeleton for human efforts reduction. In: Berns, K., Görges, D. (eds.) RAAD 2019. AISC, vol. 980, pp. 410–418. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-19648-6_47 3. Muscolo, G.G., Marcheschi, S., Fontana, M., Bergamasco, M.: Dynamics modeling of human–machine control interface for underwater teleoperation. Robotica. 1–15 (2020).https:// doi.org/10.1017/S0263574720000624 4. Schiele, A., De Bartolomei, M., Van Der Helm, F.: Towards intuitive control of space robots: a ground development facility with exoskeleton. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Beijing, pp. 1396–1401 (2006) 5. Riener, A., Nef, T., Colombo, G.: Robot-aided neurorehabilitation of the upper extremities. Med. Biolocial Eng. Comput. 43, 2–10 (2005) 6. Zoss, A., Kazerooni, H.: Design of an electrically actuated lower extremity exoskeleton. Adv. Robot. 20(9), 967–988 (2006) 7. Schiele, A.: Ergonomics of exoskeletons: objective performance metrics. In: The 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems, St. Louis, USA, October 2009 8. Trono, G., Nicolì, A., Muscolo, G.G.: Sustainable compliant physical interaction in a bipedwheeled wearable machine. Front. Mech. Eng. 6, 581626 (2020). https://doi.org/10.3389/ fmech.2020.581626 9. Gull, M.A., Bai, S., Bak, T.: A review on design of upper limb exoskeletons. Robotics 9, 16 (2020). https://doi.org/10.3390/robotics9010016 10. de Looze, M.P., Krause, F., O’Sullivan, L.W.: The potential and acceptance of exoskeletons in industry. In: Gonzalez-Vargas, J., Ibáñez, J., Contreras-Vidal, J.L., van der Kooij, H., Pons, J.L. (eds.) Wearable robotics: Challenges and trends. BB, vol. 16, pp. 195–199. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-46532-6_32 11. Wehner, M., et al.: Design and evaluation of a lightweight soft exosuit for gait assistance. In: IEEE International Conference on Robotics and Automation (2012) 12. Nef, T., Riener, R.: Shoulder actuation mechanisms for arm rehabilitation exoskeletons. In: 2008 2nd IEEE RAS and EMBS International Conference on Biomedical Robotics and Biomechatronics. IEEE (2008) 13. Colombo, G., Joerg, M., Dietz, V.: Driven gait orthosis to do locomotor training of paraplegic patients. In: International Conference of the 22nd Annual EMBS, Chicago, IL, USA, 23–28 July 2000 14. Zoccali, A., Muscolo, G.G.: Comfort perception analysis of human models interfacing with novel biped-wheeled-exoskeletons. In: Rauter, G., Cattin, P.C., Zam, A., Riener, R., Carbone, G., Pisla, D. (eds.) MESROB 2020. MMS, vol. 93, pp. 21–28. Springer, Cham (2021). https:// doi.org/10.1007/978-3-030-58104-6_3 15. Hidler, J.M., Wall, A.E.: Alterations in muscle activation patterns during robotic-assisted walking. Clin. Biomech. 20, 184–193 (2005) 16. Neckel, N.D., Blonien, N., Nichols, D., Hidler, J.M.: Abnormal joint torque patterns exhibited by chronik stroke subjects while walking with a prescribed physiological gait pattern. J. Neuro Eng. Rehabil. 5, 19 (2008). https://doi.org/10.1186/1743-0003-5-19 17. Hall, Susan J. “Linear Kinematics of Human Movement.“ Basic Biomechanics. 6th Edition ed. New York, USA: McGraw Hill (2012): 319–354 18. Stani, M.M., Goehler, C.M.: Reproducing human arm motion using a kinematically coupled humanoid shoulder–elbow complex. Appl. Bionics Biomech. 5(4), 175–185 (2008)

Use of Pneumatic Artificial Muscles in a Passive Upper Body Exoskeleton

85

19. Muscolo, G.G., Caldwell, D., Cannella, F.: Biomechanics of human locomotion with constraints to design flexible-wheeled biped robots. In: 2017 IEEE International Conference on Advanced Intelligent Mechatronics (AIM) (IEEE), pp. 1273–1278 (2017) 20. Yang, L., Grooten, W.J.A., Forsman, M.: An iPhone application for upper arm posture and movement measurements. Appl. Ergon. 65, 492–500 (2017) 21. Chou, C.-P., Hannaford, B.: Measurement and modeling of McKibben pneumatic artificial muscles. Trans. Robot. Autom. 12, 90–102 (1996) 22. Daerden, F.: Conception and realization of pleated pneumatic artificial muscles and their use as compliant actuation elements. Vrije Universiteit Brussel, p. 176 (1999) 23. Mižáková, J., Piteˇl, J., Tóthová, M.: Pneumatic artificial muscle as actuator in mechatronic system. Appl. Mech. Mater. 460, 81–90 (2014) 24. Ferraresi, C., Franco, W., Manuello, B.A.: Flexible pneumatic actuator: a comparison between the McKibben and the straight fibres muscle. J. Robot. Mechatron. 13(1), 56–63 (2001) 25. Tóthová, M., Piteˇl, J., Hošovský, A., Sárosi, J.: Numerical approximation of static characteristics of McKibben pneumatic artificial muscle. Int. J. Math. Comput. Simul. 9, 228–233 (2015) 26. Sárosi, J.: New approximation algorithm for the force of fluidic muscles. In: 2012 7th IEEE International Symposium on Applied Computational Intelligence and Informatics (SACI), pp. 229–233. IEEE, May 2012 27. Ferraresi, C., Franco, W., Manuello Bertetto, A.: Modelisation and characterization of a pneumatic muscle actuator for non conventional robotics. In: 7th International Workshop on Robotics in Alpe-Adria-Danube Region. RAAD (1998) 28. Franco, W., Maffiodo, D., De Benedictis, C., Ferraresi, C.: Use of McKibben muscle in a haptic interface. Robotics 8(1), 13 (2019) 29. De Benedictis, C., Franco, W., Maffiodo, D., Ferraresi, C.: Hand rehabilitation device actuated by a pneumatic muscle. Mech. Mach. Sci. 67, 102–111 (2019) 30. Franco, W., Maffiodo, D., De Benedictis, C., Ferraresi, C.: Dynamic modeling and experimental validation of a haptic finger based on a McKibben muscle. Mech. Mach. Sci. 66, 251–259 (2019) 31. Visited on October 2020. http://lars.mec.ua.pt/public/LAR%20Projects/RobotActuation/ 2002_MarcoMelo_VascoQuinteiro/Projecto/Artigos_net/net/air%20muscles/Shadow%20R obot%20Company%20Air%20Muscles.htm 32. Punnett, L., Fine, L.J., Keyserling, W.M., Herrin, G.D., Chaffin, D.B.: Shoulder disorders and postural stress in automobile assembly work. Scand. J. Work Environ. Health, 283–291 (2000) 33. Svendsen, S.W., Bonde, J.P., Mathiassen, S.E., Stengaard-Pedersen, K., Frich, L.H.: Work related shoulder disorders: quantitative exposure-response relations with reference to arm posture. Occup. Environ. Med. 61(10), 844–853 (2004)

Surgical Robotics and Micro Manipulation

Laser-Induced Breakdown Spectroscopy Combined with Artificial Neural Network for Pre-carbonization Detection in Laserosteotomy Ferda Canbaz1(B) , Hamed Abbasi1 , Yakub A. Bayhaqi1 , Philippe C. Cattin2 , and Azhar Zam1(B) 1 Biomedical Laser and Optics Group (BLOG), Department of Biomedical Engineering,

University of Basel, 4123 Allschwil, Switzerland {ferda.canbaz,azhar.zam}@unibas.ch 2 Center for Medical Image Analysis and Navigation (CIAN), Department of Biomedical Engineering, University of Basel, 4123 Allschwil, Switzerland

Abstract. To obtain efficient laser ablation in bone, dehydration, early carbonization and carbonization need to be avoided. Achieving this can only be provided by using an automated control of the ablation laser and irrigation system. As a preliminary study, we demonstrated a laser-induced breakdown spectroscopy based early carbonization detection system by analyzing carbonized bone tissues. Carbonization of bone samples was generated in a controlled way, by applying different number of Er:YAG pulses (0–25) at different locations on bone sample. To detect number of applied pulses, leading to the detection of carbonization level, we used a feed-forward Artificial Neural Network (ANN) with multi-layer perceptron structure. The results of the ANN were compared with the actual label, and R-squared of 0.85, 0.88, 0.86, 0.83, and 0.84 (0.85 on average) were achieved.

1 Introduction Laserosteotomy provides high-precision, fast healing, and less tissue loss compared to the conventional methods [1–3]. Since it is a contactless method, risk of bacterial contamination is also minimized [4]. To achieve efficient laser ablation, different laser sources have been utilized such as: Ti:sapphire at 800 nm [5], carbon dioxide (CO2 ) at 10.4 µm [6], harmonics of neodymium-doped ytterbium aluminum garnet (Nd:YAG) at 532 nm and 1064 nm [7–9], holmium (Ho:YAG) at 2.1 µm [10, 11], and erbium (Er:YAG) at 2.94 µm [12, 13] lasers. Effect of several operating regimes were also investigated providing pulses from milliseconds to femtoseconds at different operating wavelengths. Depending on the laser tissue interaction mechanism, the risk of damaging the surrounding tissue and tissue removal rate vary. Relatively long pulse durations (µs range) leading to photo-thermal interaction have been used to obtain high amount of tissue removal [14–19]. Here, the laser pulse transfers its energy to the sample surface leading to a rapid temperature increase and eventual evaporation [18]. Due to the strong absorption of both water and hydroxyapatite (main components of bone) around 3 µm, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 89–96, 2022. https://doi.org/10.1007/978-3-030-76147-9_10

90

F. Canbaz et al.

flash-lamp pumped Er:YAG lasers (operating at 2.94 µm) were widely used in bonecutting experiments providing high efficiencies [20, 21]. However, with these lasers, due to the relatively long pulse durations, thermal confinement cannot be provided, hence the surrounding tissue also heats up. To avoid thermal damage to the surrounding tissue, water-based irrigation systems have usually been employed [17, 19, 22]. This leads to an optimization problem. Since the absorption of water is strong around 3 µm, the accumulation of water on the sample surface must be avoided not to block the laser beam [23]. On the other hand, lack of water on the bone surface may lead to damaging the ablation area or the surrounding tissue [24, 25]. The most probable damage is carbonization, which can also be described as burning the tissue, leading to delay healing. Hence, to avoid carbonization without reducing the ablation efficiency, a feedback system is necessary to detect the early carbonization or dehydration on the bone sample and adjust the amount of water on the surface or stop the ablation laser [22, 26, 27]. In this study, we used an Echelle spectrometer-based laser-induced breakdown spectroscopy (LIBS) setup to detect carbonization with a pig femur bone. A microsecond Er:YAG and a nanosecond frequency-doubled Nd:YAG lasers were utilized to ablate and analyze the regions of interest, respectively. Different number of Er:YAG pulses were applied to the femur bone surface. Later, at each point, we collected spectra using the LIBS system. Then, in the analysis, a feed-forward artificial neural network with multi-layer perceptron (MLP) structure was utilized. We, finally, investigated the level of carbonization of the tissue by estimating the number of applied pulses.

2 Experimental Setup and Results Figures 1(a), 1(b), and 1(c) show ablation, detection, and analysis setups, respectively. Note that, in this study, two separate setups were utilized for ablation and detection, however, combining these two setups is possible by placing a dichroic mirror to align the lasers coaxially [28]. In ablation experiments, we used a flash-lamp pumped Er:YAG laser producing 280 µs pulses at 2.94 µm. Repetition frequency of the laser could be adjusted from 1 to 20 Hz. At 1 Hz, the output energy of the laser was as high as 900 mJ. The laser beam was sent on the sample without any focusing optics providing only dehydration or carbonization without ablating. The beam diameter of the laser was 4 mm. Figures 1(b) and 1(c) show the detection system including an excitation laser and a home-built spectrometer. Here we utilized a frequency-doubled Nd:YAG laser producing 5 ns pulses at 532 nm to induce plasma light on the sample surface. The output energy of the Nd:YAG laser varies from 1 to 200 mJ at the repetition frequencies of 0.2–20 Hz. The laser beam was focused with a microscope objective (LMH-5X-532, 5X, numerical aperture (NA) = 0.13, Thorlabs). The induced plasma light was collected with the same objective then sent through a beam splitter which was used to both block the laser beam and transmit the plasma light. As can be seen from the Fig. 1(b), the plasma light (purple) was then focused into a collection fiber with a 50 µm of core diameter. The numerical aperture (NA) of the fiber is 0.22. The other facet of the fiber is placed in the middle of a home-build Echelle spectrometer with an F number of 3 [29]. Figure 1(c) shows the detailed Solidworks design of the spectrometer consisting two 2-inches of

Laser-Induced Breakdown Spectroscopy Combined with ANN

91

Fig. 1. (a) Er:YAG as the ablation laser operating at 2.94 µm and a bone sample which was placed in front of the laser without any focusing elements. (b) Schematic of the carbonization detection setup including an Nd:YAG laser, focusing, and collection optics. (c) CAD design of the spectrometer set-up. (Inset) a typical Echellogram which was collected by using an LED lamp.

parabolic off-axis mirrors (PM1 and PM2), an Echelle grating (EG), another grating (G) to disperse the beam in the y-direction, an imaging lens, and an ICCD camera. The beam delivered by the fiber was initially reflected by PM1 on to the EG which was placed at the blazing angle to disperse different wavelengths in the x-direction. The EG delivers the beam to PM2 by travelling slightly above the fiber. The PM2 reflects the beam on the second grating (G) to disperse different wavelengths in the y-direction. Lastly, imaging lens focuses the beam into the ICCD camera pixels. In the inset of Fig. 1(c), a typical Echellogram of a white light LED source can be seen. In the ablation experiments, initially, we placed the bone sample in front of the Er:YAG laser. At different positions on bone sample, we send different number of pulses (from 0 to 25, with an increment of 1 pulse) on to the bone sample without using any irrigation system. Here, the reason of not using any irrigation system is to directly relate

92

F. Canbaz et al.

Fig. 2. Bone sample after the ablation experiments. Within each grid on the sample, number of applied pulses was different.

the number of pulses and degree of carbonization on the sample surface. As can be seen from Fig. 2, visual damage on the bone surface was already occurred after 2 pulses (3rd grid from bottom left in Fig. 2). To detect carbonization or dehydration, we then placed the sample at the focal plane of the detection system. In the experiments, the output energy was set to be 100 mJ to collect all the information from the sample surface. At each ablation position, we collected 70 spectra. Note that after 70 laser shots, the surface layer was slightly ablated (a whitish bone became visible after ablation of yellowish/brownish damaged surface) with Nd:YAG laser, so no further data were collected to ensure that the Nd:YAG laser pulse was within an area with similar thermal damage level. The Q-switched pulsed output of the Nd:YAG laser was electronically synchronized with the ICCD camera. The camera was collected each spectrum with a delay of 1 µs with respect to the Qswitched pulse to avoid Bremsstrahlung emission which occurs within the first ~1 µs. The gate of the camera was kept open for 1 ms to gather enough data for analysis. From 26 different positions on the bone samples, in total, we collected 1820 spectra. Then the collected spectra were equally distributed to 5-fold datasets.

Fig. 3. Schematic of the feed-forward Artificial Neural Network (ANN) with Multi-Layer Perceptron (MLP) structure.

A feed-forward Artificial Neural Network (ANN) with Multi-Layer Perceptron (MLP) structure was used to predict the number of pulses to which was sent to the

Laser-Induced Breakdown Spectroscopy Combined with ANN

93

Fig. 4. (a) to (e): Cross-validated results of the developed regression model to investigate the number of pulses which have been sent to the bone without irrigation. R-squared of 0.85, 0.88, 0.86, 0.83, and 0.84 (0.85 on average) were achieved.

bone without any irrigation system. The ANN was composed of 4 fully connected hidden layers. A schematic of the developed ANN structure can be seen from Fig. 3. The ANN input layer was the spectrum data of localized plasma. Following the input layers, the hidden layers’ sizes were set gradually to 1024, 512, 256, and 128 neurons, respectively. The sigmoid activation function was applied to all of the hidden layers. The ANN was then trained using the mean-squared error optimizer to solve the regression problem and defined the output layer’s size to one neuron. Dropout regularizer with a low rate of 0.1 was applied to the hidden layers to avoid overfitting. We implemented and trained the ANN models in Keras, with a TensorFlow backend [30]. The training was done on a workstation equipped with an Intel Xeon E5620 processor and an NVIDIA GTX 1080 Ti GPU. All models were trained with 1000 epochs and a batch size of 32 samples to fit the GPU’s memory capacity. The ANN model was then cross validated using fivefold data. In each case, the predicted results were compared with the actual label, and R-squared of 0.85, 0.88, 0.86, 0.83, and 0.84 (0.85 on average) were achieved (Figs. 4(a)–(e)).

3 Discussion and Conclusions Laserosteotomes using Er:YAG laser could only be clinically applied if they do not carbonize the tissue. This can only be ensured with an irrigation system which was controlled by a feedback mechanism to control the amount of water on the region of interest. This leads to less thermal damage, increase the ablation speed, and accelerate healing. State-of-the-art for determining the irrigation parameters is to apply water and air flow rates based on pre-performed experiments without monitoring the bone surface in real-time. Combination of the feedback and ablation systems is necessary mainly because of the inhomogeneous structure of bone and variation between patient to patient. Recently, we demonstrated that a LIBS-based detection system can classify

94

F. Canbaz et al.

fully carbonized bone from not carbonized one. While such a classification can avoid further carbonization, even localized carbonization on the surface of the bone can still prolong the healing process. There are several stages before full carbonization occurs, such as dehydration and early carbonization. The detection of these early signs makes prospective carbonization prediction possible. This study aims at finding a regression between all steps of laser-induced thermal damage. Therefore, the spectrum of localized plasma generated in the ablation spot, together with the number of shots sent to the bone without irrigation was recorded using a home-built fiber-coupled Echelle spectrometer. According to the results, a feed forward artificial neural network (ANN) provided the best estimation for the number of applied pulses on the bone sample. This study demonstrates the detection of different stages of carbonization on a pig bone sample. We used a LIBS setup to detect carbonization using pig femur bone samples. We sent different number of Er:YAG pulses (0–25) to the femur bone surface at different positions. By using the ANN, we estimated the number of applied pulses on the bone sample. Then the predicted results were compared with the actual label, and R-squared of 0.85, 0.88, 0.86, 0.83, and 0.84 (0.85 on average) were achieved. Future work contains the combination of the ablation and detection setups to collect data in the real-time. Moreover, an addition of an irrigation system is required to hydrate the sample surface. This would change the experimental conditions. To obtain the optimum performance in this case, the regressor needs to be retrained. Note that, to perform the ablation without any carbonization, the feedback system needs to provide either a “stop” or a “go” message within the two consecutive laser pulses (~100 ms). Acknowledgments. The authors gratefully acknowledge funding of the Werner Siemens Foundation through the Minimally Invasive Robot-Assisted Computer-guided LaserosteotomE (MIRACLE) project.

References 1. Baek, K.W., et al.: A comparative investigation of bone surface after cutting with mechanical tools and Er: YAG laser. Lasers Surg. Med. 47(5), 426–432 (2015) 2. Zhang, Y., Wang, C., Zhou, S., Jiang, W., Liu, Z., Xu, L.: A comparison review on orthopedic surgery using piezosurgery and conventional tools. Proc. CIRP 65, 99–104 (2017) 3. Beltrán Bernal, L.M., Abbasi, H., Zam, A.: Laser in bone surgery. In: Stübinger, S., Klämpfl, F., Schmidt, M., Zeilhofer, H.-F. (eds.) Lasers in Oral and Maxillofacial Surgery, pp. 99–109. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-29604-9_9 4. Eyrich, G.K.: Laser-osteotomy induced changes in bone. Med. Laser Appl. 20(1), 25–36 (2005) 5. Neev, J., Da Silva, L.B., Feit, M.D., Perry, M.D., Rubenchik, A.M., Stuart, B.C.: Ultrashort pulse lasers for hard tissue ablation. IEEE J. Sel. Top. Quant. 2(4), 790–800 (1996) 6. Gonzalez, C., Vandemerwe, W.P., Smith, M., Reinisch, L.: Comparison of the erbiumyttrium-aluminum-garnet and carbon-dioxide lasers for invitro bone and cartilage ablation. Laryngoscope 100(1), 14–17 (1990) 7. Tulea, C.-A., Caron, J., Gehlich, N., Lenenbach, A., Noll, R., Loosen, P.: Laser cutting of bone tissue under bulk water with a pulsed ps-laser at 532 nm, p. 9. SPIE (2015)

Laser-Induced Breakdown Spectroscopy Combined with ANN

95

8. Abbasi, H., Rauter, G., Guzman, R., Cattin, P.C., Zam, A.: Plasma plume expansion dynamics in nanosecond Nd: YAG laserosteotome. In: High-Speed Biomedical Imaging and Spectroscopy III: Toward Big Data Instrumentation and Management, p. 1050513. International Society for Optics and Photonics (2018) 9. Abbasi, H., Rauter, G., Guzman, R., Cattin, P.C., Zam, A.: Differentiation of femur bone from surrounding soft tissue using laser-induced breakdown spectroscopy as a feedback system for smart laserosteotomy, vol. 10685. SPIE Photonics Europe. SPIE (2018) 10. Zhang, X., Chen, C., Chen, F., Zhan, Z., Xie, S., Ye, Q.: In vitro investigation on Ho: YAG laser-assisted bone ablation underwater. Lasers Med. Sci. 31(5), 891–898 (2016). https://doi. org/10.1007/s10103-016-1931-x 11. Hafez, M.I., Coombs, R.R.H., Zhou, S., McCarthy, I.D.: Ablation of bone, cartilage, and facet joint capsule using Ho: YAG laser. J. Clin. Laser Med. Surg. 20(5), 251–255 (2002). https:// doi.org/10.1089/10445470260420759 12. Hibst, R., Keller, U.: Experimental studies of the application of the Er: YAG laser on dental hard substances. 1. Measurement of the ablation rate. Laser Surg. Med. 9(4), 338–344 (1989). https://doi.org/10.1002/lsm.1900090405 13. Bernal, L.M.B., Canbaz, F., Droneau, A., Friederich, N.F., Cattin, P.C., Zam, A.: Optimizing deep bone ablation by means of a microsecond Er: YAG laser and a novel water microjet irrigation system. Biomed. Opt. Express 11(12), 7253–7272 (2020). https://doi.org/10.1364/ Boe.408914 14. Stübinger, S.: Advances in bone surgery: the Er: YAG laser in oral surgery and implant dentistry. Clin. Cosmet. Investig. Dent. 2, 47–62 (2010) 15. Visuri, S.R., Walsh Jr, J.T., Wigdor, H.A.: Erbium laser ablation of dental hard tissue: effect of water cooling. Lasers Surg. Med. 18(3), 294–300 (1996). https://doi.org/10.1002/(SICI)10969101(1996)18:33.0.CO;2-6 16. Papadaki, M., Doukas, A., Farinelli, W.A., Kaban, L., Troulis, M.: Vertical ramus osteotomy with Er: YAG laser: a feasibility study. Int. J. Oral Maxillofac. Surg. 36(12), 1193–1197 (2007). https://doi.org/10.1016/j.ijom.2007.05.019 17. Kuscer, L., Diaci, J.: Measurements of erbium laser-ablation efficiency in hard dental tissues under different water cooling conditions. J. Biomed. Opt. 18(10), 108002 (2013). https://doi. org/10.1117/1.Jbo.18.10.108002 18. Tuchin, V.: Tissue optics and photonics: light-tissue interaction II. J. Biomed. Photon. Eng. 2(3), 030201 (2016) 19. Bernal, L.M.B., et al.: Performance of Er: YAG laser ablation of hard bone under different irrigation water cooling conditions. In: Optical Interactions with Tissue and Cells XXIX 2018, p. 104920B. International Society for Optics and Photonics (2018) 20. Rupprecht, S., Tangermann-Gerk, K., Wiltfang, J., Neukam, F.W., Schlegel, A.: Sensor-based laser ablation for tissue specific cutting: an experimental study. Lasers Med. Sci. 19(2), 81–88 (2004) 21. Peavy, G.M., Reinisch, L., Payne, J.T., Venugopalan, V.: Comparison of cortical bone ablations by using infrared laser wavelengths 2.9 to 9.2 µm. Lasers Surg. Med. 25(5), 421–434 (1999). 10.1002/(SICI)1096-9101(1999)25:53.0.CO;2-J 22. Kang, H.W., Rizoiu, I., Welch, A.J.: Hard tissue ablation with a spray-assisted mid-IR laser. Phys. Med. Biol. 52(24), 7243 (2007) 23. Walsh, J.T., Deutsch, T.F.: Er: YAG laser ablation of tissue - measurement of ablation rates. Laser Surg. Med. 9(4), 327–337 (1989). https://doi.org/10.1002/lsm.1900090404 24. Buchelt, M., et al.: Erb: YAG and Hol: YAG laser osteotomy: the effect of laser ablation on bone healing. Lasers Surg. Med. 15(4), 373–381 (1994). https://doi.org/10.1002/lsm.190015 0407

96

F. Canbaz et al.

25. Abbasi, H., Beltrán, L., Rauter, G., Guzman, R., Cattin, P.C., Zam, A.: Effect of cooling water on ablation in Er: YAG laserosteotome of hard bone. In: Third International Conference on Applications of Optics and Photonics 2017, p. 104531I. International Society for Optics and Photonics (2017) 26. Abbasi, H., Rauter, G., Guzman, R., Cattin, P.C., Zam, A.: Laser-induced breakdown spectroscopy as a potential tool for autocarbonization detection in laserosteotomy. J. Biomed. Opt. 23(7), 071206 (2018) 27. Beltran, L., Abbasi, H., Rauter, G., Friederich, N., Cattin, P., Zam, A.: Effect of laser pulse duration on ablation efficiency of hard bone in microseconds regime. In: Third International Conference on Applications of Optics and Photonics 2017, p. 104531S. International Society for Optics and Photonics (2017) 28. Abbasi, H., et al.: Combined Nd: YAG and Er: YAG lasers for real-time closed-loop tissuespecific laser osteotomy. Biomed. Opt. Express 11(4), 1790–1807 (2020). https://doi.org/10. 1364/Boe.385862 29. Abbasi, H., Rauter, G., Guzman, R., Cattin, P.C., Zam, A.: Design and implementation of a compact high-throughput Echelle spectrometer using off-the-shelf off-axis parabolic mirrors for analysis of biological samples by LIBS (Conference Presentation). In: Optical Methods for Inspection, Characterization, and Imaging of Biomaterials IV 2019, p. 1106003. International Society for Optics and Photonics (2019) 30. Chollet, F., et al.: Keras. GitHub (2015). https://github.com/fchollet/keras

Development and Evaluation of a Force-Sensitive Flexure-Based Microgripper Concept C´edric Duverney1(B) , Mohamed Ali El Bahi1 , Nicolas Gerig1 , Philippe C. Cattin2 , and Georg Rauter1 1

BIROMED-Lab, Department of Biomedical Engineering, University of Basel, 4123 Allschwil, Switzerland [email protected] 2 CIAN, Department of Biomedical Engineering, University of Basel, 4123 Allschwil, Switzerland https://biromed.dbe.unibas.ch

Abstract. Dimensions of mechanical and electronic parts are steadily reduced due to miniaturization of technical devices, such as smartphones, robots, and medical implants. Size reduction increases complexity of part manufacturing and especially of device assembly. When assembling prototypes of such devices, flexible, teleoperated microassembly systems can help to achieve the required assembly precision and to reduce damages to parts. This paper presents a concept of a microgripper for such a microassembly system. It is based on a flexure structure, largely eliminating mechanical play in the design and allowing for a straightforward implementation of force sensing capabilities through strain gauges. We developed a first prototype and demonstrated functionality of the gripping structure. Our grip force sensing yielded promising results. Structural fragility and parasitic lateral gripper finger movements have been identified as key limitations of the current design. These will be addressed in subsequent iterations by different material choices and improvements to the finger geometry. Keywords: Micromanipulation

1

· Force sensing · Flexure-based design

Introduction

A global trend towards miniaturization has been observed to be one of the main driving forces behind developments in consumer electronics and medical technology [18]. Given the increasingly small size of their components, the assembly of such miniaturized devices with dimensions between 100 µm and 10 mm is becoming more challenging. For large production batches, highly specialized and automated assembly lines can be used. For unique prototype devices however, the development and programming of such assembly lines is not economically viable. The need for flexible, precise, and intuitive assembly solutions gave rise to c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 97–106, 2022. https://doi.org/10.1007/978-3-030-76147-9_11

98

C. Duverney et al.

recent developments of teleoperated microassembly systems combining robotic miniature assembly stations with mechanically decoupled, remote user interfaces [1,12]. First such systems became commercially available [7]. Due to mechanical separation of assembly and user side, interaction forces occurring between the microparts and the assembly station are no longer inherently transmitted to the user’s hands. This generally renders the teleoperated assembly of microparts much more challenging than manual assembly due to the increasingly counterintuitive nature of physics at the microscale given the predominance of surface forces such as surface tension over volume forces such as gravity [15]. These surface forces typically are unknown to novice users, as they are outweighed by gravitational and inertial forces at the larger, familiar scales of everyday objects. The addition of force feedback to a microassembly system [1,11] would enable the user to learn this dynamic environment, leading to a higher quality and more intuitive assembly process. Benefits of added haptic feedback in terms of object characterization have been demonstrated in the context of minimally invasive surgery [10], indicating the critical importance of this technique for restoring intuition in teleoperation at small scales. One of the core components of a microassembly system is a microgripper or a general micromanipulation instrument to grasp, move, and place the target microparts. Universal microgrippers are effective for the assembly of a range of microsystems without the need of time-consuming tool exchanges, because their structures have been optimized towards different part geometries and sizes [4]. This paradigm stands in contrast to robotic grippers used in industrial applications, which tend to be highly specialized. More versatile and dexterous gripper structures are used in minimally invasive surgical devices [6] but tend to be difficult to further miniaturize. Moreover, most industrial and surgical gripping devices consist of rigid links and mechanical joints and will exhibit mechanical play, which will limit their performance when manipulating small-sized parts. A possibility to circumvent these issues with traditional gripping devices in microassembly is to rely on flexure-based structures that inherently display no mechanical play [17,19]. These compliant grippers further allow to realize soft grasping, which is particularly suited for manipulating fragile objects and structures [9] and hence offer promising solutions also for surgical applications [5]. As flexure-based structures can often be manufactured monolithically and do not require small-scale functional mechanical parts such as bearings or screws, they tend to be simpler to miniaturize. These structures can directly be used to measure gripping forces based on their elastic deformation if combined for instance with strain gauges, capacitive elements, or cameras [2,13]. Flexure-based microgrippers often feature two-dimensional geometries of two fingers manufactured with techniques such as wire-cut EDM or photolithography [13]. While this design choice leads to cost-effective monolithic structures, grasping of certain object geometries can be challenging, for instance when attempting to grasp and hold a screw in a fixed orientation. For such cylindrical objects, a three-dimensional design with three fingers might be more suitable. In this work, we present our first concept design and realization of such a three-finger

Development of a Force-Sensitive Flexure-Based Microgripper Concept

99

flexure-based microgripper with integrated force-sensing capabilities. The proposed microgripper is intended to be used within a teleoperated microassembly system to manipulate and assemble standard and custom mechanical and electronic parts as for example found in miniature robots or medical devices. Targeted part dimensions lie in the range of 100 µm to 10 mm and assembly takes place under standard ambient conditions. A gripper opening range of 0 mm to 11 mm and a gripping force of up to 30 mN are required to grasp and lift the largest and heaviest objects against gravity. Measurement of the gripping force acting along the motion directions of the gripper fingers is required to provide the user with haptic feedback. With this design, we aim at largely eliminating mechanical play usually present in standard robotic grippers to improve handling of smallsized parts. We further aim at providing a comparably large gripper opening range to reduce the number of necessary tool changes when manipulating parts of varying sizes. Lastly, we aim at providing a simple, low-cost, and easily fabricable gripper design as a test bench for future investigations on gripper finger topology as well as actuation and force sensing principles.

2

Methods

Firstly, we employed a linkage model-based design process to generate our proposed gripper structure. Secondly, we realized our first prototype gripper and designed a matching calibration and evaluation setup. Thirdly, we performed a quantitative evaluation of the implemented force-sensing capabilities. 2.1

Development Process

We employed the design procedure illustrated in Fig. 1 and based on Ref. [17] to generate the structure of our targeted flexure-based gripper: Firstly, we defined parallel finger motion as the desired motion model. In contrast to rotary finger motion, parallel finger motion ensures that gripping forces will always lie in planes perpendicular to all finger contact planes, reducing the risk of object slippage. Secondly, we designed a linkage model consisting only of rigid links and mechanical joints implementing the desired motion model and the gripping force Fgrip through an actuator force Factu . Thirdly, we derived an initial design of a flexure-based gripper finger structure from the mechanical linkage model taking into account dimensional constraints arising from the intended target application. This monolithic structure consisted of thicker segments mimicking the rigid links of the linkage model, and of thin segments allowing for some motion in the areas of the mechanical joints. Fourthly, we improved the joint design with finite element analysis towards a better stress distribution to avoid plastic deformation. The resulting s-joint presented in Fig. 2 was subsequently integrated in the initial finger design, yielding the evaluated finger design which was utilized in the proposed microgripper concept.

100

C. Duverney et al.

Fig. 1. Design process of the gripper finger from the desired parallel motion model to the print-ready evaluated 3D design. Initial and evaluated designs both have a thickness of 2 mm.

Fig. 2. FEM-derived stress distribution (von Mises) in ABS polymer i- and s-joints for identical input forces Fin and identical resulting joint rotations of 10◦ . The s-joint geometry produces a wider spread of the stress distribution, while largely reducing the maximum stress value.

2.2

Microgripper Prototype and Measurement Setup

The proposed microgripper concept is presented in Fig. 3. All parts were FDM 3D-printed in ABS polymer. Three flexure-based fingers as described in Sec. 2.1 were printed with layers parallel to the drawing plane in Fig. 1 and were circularly arranged on a rigid gripper base. The actuation force was provided by a DC motor (RE 25 24V, Maxon Motor, Sachseln, Switzerland) and transmitted to the gripper fingers by means of a lead screw (pitch 0.4 mm). One strain gauge (SS.060.40.2500PM, HAPTICA, Busto Arsizio, Italy) was affixed to each gripper finger with standard super glue and connected to a Wheatstone bridge circuit (equal resistances of 2500 Ω, supply voltage of 5 V) generating an analogue voltage corresponding to the current gripper finger deformation. The voltage was digitized with a resolution of 24 bits (EL3602 with measurement range ±1.25 V, Beckhoff Automation, Verl, Germany). A six-axis force/torque (F/T) sensor (Nano17 with calibration SI-12-0.12, ATI Industrial Automation, Apex,

Development of a Force-Sensitive Flexure-Based Microgripper Concept

101

NC, USA) was used to provide ground truth gripping force data for one of the three gripper fingers.

Fig. 3. CAD rendering (bottom left corner, strain gauges and fasteners not shown) and photograph (main image) of the proposed microgripper concept with 1 flexure-based fingers ( 1’ evaluated finger), 2 gripper base, 3 translation clamp, 4 central nut and lead screw, 5 DC motor, 6 strain gauges, 7 strain gauge electronics, and 8 six-axis F/T sensor. The force measurement schematic (top right corner) depicts the interaction between the evaluated flexure-based finger and the F/T sensor for strain gauge calibration and evaluation.

The microgripper concept was evaluated using a setup as sketched in Fig. 4. A control cabinet based on an industrial PC (CX2020, Beckhoff Automation, Verl, Germany) running a real-time automation software (TwinCAT 3.1, Beckhoff Automation, Verl, Germany) has been used in conjunction with a motor driver (MAXPOS 50/5, Maxon Motor, Sachseln, Switzerland), an F/T sensor interface (ECATOEM, ATI Industrial Automation, Apex, NC, USA), and Wheatstone bridge circuits to control the gripper. A development PC (Elitebook 840 G4, Hewlett-Packard, Palo Alto, CA, USA) was used to program and monitor the control cabinet. Only one of the three gripper fingers was calibrated and evaluated with the F/T sensor. The interaction between the evaluated finger and the F/T sensor is sketched in Fig. 3 (top right corner).

102

C. Duverney et al.

Fig. 4. System components and their interconnections: the a development PC is used to program and monitor the b control cabinet, which in turn is connected to the c motor driver interfacing the d DC motor, the e strain gauge electronics interfacing the f strain gauges, and the g F/T sensor module interfacing the h F/T sensor to calibrate and evaluate the i microgripper.

2.3

Evaluation

Two test series were performed. Each consisted of ten operation cycles of closing and opening the gripper. Closing was performed until a normal gripping force of at least 0.03 N was measured at the F/T sensor, at which point the gripper was opened again by a predefined displacement of three motor revolutions. The first test series was used to linearly calibrate the strain gauge force Fsg = a ∗ Usg + b, with Usg the recorded strain gauge voltage and a = −1.985 N/V , b = 0.184 N calibration parameters, by fitting Fsg to the recorded F/T sensor force FFT . The second test series was then used to evaluate the strain gauge force measurements obtained with the determined calibration parameters a, b with respect to the F/T sensor force measurements.

3

Results

The fabricated prototype gripper was able to perform the desired type of finger motion, did not plastically deform or break, and could cover an opening range between 0 mm and 12.65 mm. A gripping force of up to 42.5 mN was recorded with the F/T sensor. Comparison of the forces obtained with the calibrated strain gauge and with the F/T reference sensor showed matching of force rise, peaks, and fall along the time axis, as well as similar matching force amplitudes (Fig. 5). From the F/T sensor zeroing on (red arrow in Fig. 5), the deviation of Fsg relative to FFT showed an absolute mean of 1.4 mN and a standard deviation of 3 mN. In the ten regions with zero force, a positive drift of Fsg was observed, as well as a negative offset in zero force regions that was increasing over consecutive opening-closing cycles from approximately 3 mN (first cycle) to 5 mN (tenth cycle).

Development of a Force-Sensitive Flexure-Based Microgripper Concept

103

Fig. 5. Comparison of the force measurements provided by the F/T sensor (FFT ) and by the strain gauge of the evaluated finger (Fsg ). Two out of the ten operation cycles performed are depicted. The motor angular position (αDC ), measured in axis revolutions [rev], was increased to close the gripper until a force threshold of 0.03 N was measured with the F/T sensor, at which point the gripper was opened again. The jump in FFT visible at around 3.5 s (red arrow) corresponds to the initial zeroing of the F/T sensor.

4

Discussion

Regarding the gripper structure, the achieved prototype and the applied linkage model-based design process worked and fulfilled the stated requirements on gripping force (up to 30 mN) and gripper opening range (0 mm to 11 mm). The FEM-based joint improvement step allowed to avoid any plastic deformation of the gripper fingers during operation. However, a further improvement step should be added to the design process to reduce the overall bulkiness of the gripper structure. Moreover, the selected s-joint geometry was generated purely based on insights gained from experimentation with and FEM analysis of a range of simple i-joint geometries. More potentially feasible joint geometries should be considered and compared to the selected s-joint geometry, as other choices might allow for a more compact gripper structure while maintaining the current deflection range. The presented three-finger architecture was selected as it was expected to be effective in particular for common cylindrical microparts such as screws, pins, and gears. The selected ABS polymer is not ideal, because it led to a generally fragile structure with parasitic lateral motions of the fingers and to slippage of the grasped objects due to its surface properties. Both of these effects could be identified with the naked eye during normal gripper operation, especially when switching from closing to opening motion. This indicates that the rotation of the driving lead screw needs to be better decoupled from the gripper structure so as not to induce lateral finger motions. These issues could be alleviated by fabricating the gripper fingers from a more robust, linear elastic material, by modifying the finger geometry to provide better lateral rigidity, and by coating the finger tips with an appropriate material providing

104

C. Duverney et al.

better grip. Further investigations in this direction are required, as both the choice of finger tip material and geometry heavily influence how strongly small grasped objects are influenced by surface forces, and hence how easily they can be released by the gripper. Should object releasing be identified as an issue, active release strategies will need to be considered [16]. Should object slippage remain an issue even with improved finger geometry and material, automated slippage detection methods will need to be investigated, for example based on additional shear force [3] or vision [8] sensors. Although functionality of the force transmission concept could be demonstrated, slight mechanical play was identified within the translation clamp, highlighting the need for a modified mechanism. Relying on a piezoelectric actuator instead of the current oversized DC motor, a solution frequently implemented in flexure-based microgrippers [14], might prove beneficial. If a piezoelectric actuator is used, the presented linkage model might need to be adapted to account for its travel range different from the current DC motor and lead screw combination while maintaining the required gripper opening range. Finally, the integration, placement, fixation, and electrical connection of the strain gauges requires improvement to obtain a more robust, durable finger design. Regarding the force measurements, the experimental data (Fig. 5) demonstrated that strain gauges offer an easy and cost-effective option to obtain satisfactory gripping force measurements, even though the underlying ABS polymer did not offer an ideal, smooth fixation surface and standard super glue was used to affix the sensors. A maximum positive overshoot of Fsg relative to FFT of 14 mN and a maximum negative overshoot of −21.1 mN were observed around the force peaks. We do not expect these short-duration overshoots to be a problem, however they indicate the need for further investigations into suitable signal filtering. The slight positive drift of the strain gauge measurement visible in zero force regions is likely due to the viscoelastic properties exhibited by ABS polymers. These properties lead to a damping of the final stages of the elastic return movement of the gripper finger as the gripping force vanishes, resulting in the observed damped response of Fsg . The increasingly large negative offset of Fsg visible across consecutive zero force regions is likely rooted in the same effects and is expected to largely disappear after no force has been applied to the finger for a sufficiently long time. The resulting force measurement inaccuracies when no object is grasped should not pose any problem in the intended microassembly application. These effects can be attenuated by fabricating the gripper fingers from another material with negligible viscoelastic properties, such as a suitable aluminum alloy. Based on this first prototype’s force measurement results on the gripper fingers, we do not expect that force sensing will remain the major challenge for realization of the planned microassembly application. Instead, we consider finding appropriate force and motion scaling factors between gripper and haptic input device and the realization of a stable closed-loop force feedback control as our next key challenges for the project.

Development of a Force-Sensitive Flexure-Based Microgripper Concept

5

105

Conclusion

We could successfully apply a linkage model-based design process to realize our first FDM 3D-printed prototype gripper fabricated out of ABS polymer. The prototype gripper demonstrated functionality of both the gripping mechanism and the force measurement principle. We assume that our gripper structure can further be optimized to realize desired gripping forces while increasing rigidity to reduce parasitic lateral motions of the gripper fingers. The overall bulkiness of the gripper can be reduced by using a small-sized piezoelectric actuator. The current design was cost-effective, allowed for fast design iterations, and yielded satisfactory force sensing results despite simple manufacturing techniques. Acknowledgment. The authors gratefully acknowledge funding of this work by the Werner Siemens Foundation through the MIRACLE project.

References 1. Bolopion, A., R´egnier, S.: A review of haptic feedback teleoperation systems for micromanipulation and microassembly. IEEE Trans. Autom. Sci. Eng. 10(3), 496– 502 (2013). https://doi.org/10.1109/TASE.2013.2245122 2. Boudaoud, M., Regnier, S.: An overview on gripping force measurement at the micro and nano-scales using two-fingered microrobotic systems. Int. J. Adv. Rob. Syst. 11(3), 45 (2014). https://doi.org/10.5772/57571 3. Canepa, G., Petrigliano, R., Campanella, M., De Rossi, D.: Detection of incipient object slippage by skin-like sensing and neural network processing. IEEE Trans. Syst. Man Cybern. Part B (Cybern.) 28(3), 348–356 (1998). https://doi.org/10. 1109/3477.678629 4. Carbone, G.: Grasping in Robotics, vol. 10. Springer, Heidelberg (2012). https:// doi.org/10.1007/978-1-4471-4664-3 5. Dearden, J., Grames, C., Jensen, B.D., Magleby, S.P., Howell, L.L.: Inverted L-arm gripper compliant mechanism. J. Med. Devices 11(3) (2017). https://doi.org/10. 1115/1.4036336 6. Dikaiakos, G., Tzemanaki, A., Pipe, A.G., Dogramadzi, S.: Mechatronic implementation in minimally invasive surgical instruments. In: 5th IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics, pp. 357– 362. IEEE (2014). https://doi.org/10.1109/BIOROB.2014.6913802 7. FemtoTools AG, Buchs, Switzerland: Ft-mta03 micromechanical testing and assembly system, November 2016. https://www.femtotools.com/products/ft-mta03/ description. Accessed 25 Jan 2021 8. Ito, Y., Kim, Y., Obinata, G.: Robust slippage degree estimation based on reference update of vision-based tactile sensor. IEEE Sens. J. 11(9), 2037–2047 (2011). https://doi.org/10.1109/JSEN.2010.2104316 ´ ˇ Handroos, H.: A novel simple, adaptive, and 9. Milojevi´c, A., Linß, S., Cojbaˇ si´c, Z, versatile soft-robotic compliant two-finger gripper with an inherently gentle touch. J. Mech. Robot. 13(1), 011015 (2021). https://doi.org/10.1115/1.4048752 10. Moradi Dalvand, M., Shirinzadeh, B., Nahavandi, S., Smith, J.: Effects of realistic force feedback in a robotic assisted minimally invasive surgery system. Minimally Invasive Ther. Allied Technol. 23(3), 127–135 (2014). https://doi.org/10. 3109/13645706.2013.867886

106

C. Duverney et al.

11. Pacchierotti, C., Scheggi, S., Prattichizzo, D., Misra, S.: Haptic feedback for microrobotics applications: a review. Front. Robot. AI 3, 53 (2016). https://doi.org/10. 3389/frobt.2016.00053 12. Probst, M., H¨ urzeler, C., Borer, R., Nelson, B.J.: A microassembly system for the flexible assembly of hybrid robotic mems devices. Int. J. Optomechatron. 3(2), 69–90 (2009). https://doi.org/10.1080/15599610902894592 13. Reddy, A.N., Maheshwari, N., Sahu, D.K., Ananthasuresh, G.: Miniature compliant grippers with vision-based force sensing. IEEE Trans. Rob. 26(5), 867–877 (2010). https://doi.org/10.1109/TRO.2010.2056210 14. Sun, X., et al.: A novel flexure-based microgripper with double amplification mechanisms for micro/nano manipulation. Rev. Sci. Instrum. 84(8), 085002 (2013). https://doi.org/10.1063/1.4817695 15. Wautelet, M.: Scaling laws in the macro-, micro-and nanoworlds. Eur. J. Phys. 22(6), 601 (2001). https://doi.org/10.1088/0143-0807/22/6/305 16. Weck, M., Peschke, C.: Equipment technology for flexible and automated microassembly. Microsyst. Technol. 10(3), 241–246 (2004). https://doi.org/10.1007/ s00542-003-0360-5 17. Wenjie, C., Wei, L.: Design of a flexure-based gripper used in optical fiber handling. In: 2004 IEEE Conference on Robotics, Automation and Mechatronics, vol. 1, pp. 83–88. IEEE (2004). https://doi.org/10.1109/RAMECH.2004.1438896 18. Wong, H., Iwai, H.: The road to miniaturization. Phys. World 18(9), 40 (2005). https://doi.org/10.1088/2058-7058/18/9/31 19. Xu, Q.: Design and development of a novel compliant gripper with integrated position and grasping/interaction force sensing. IEEE Trans. Autom. Sci. Eng. 14(3), 1415–1428 (2015). https://doi.org/10.1109/TASE.2015.2469108

Universal Mechanical Interface for Surgical Telemanipulation Using Conventional Instruments Max B. Schäfer(B) , Gerrit R. Friedrich, and Peter P. Pott Institute of Medical Device Technology, University of Stuttgart, Stuttgart, Germany [email protected]

Abstract. Minimally invasive surgery has been indispensable in the healthcare sector for many years due to crucial advantages compared to open surgery. Subsequently, there has been a strong emergence of robot-assisted surgery to overcome remaining difficulties of conventional minimally invasive surgery, such as poor ergonomics and limited dexterity. However, the financial hurdle impedes a widespread use of systems for robot-assisted interventions and shows an apparent need for cost efficient solutions. The use of standard components for robotassisted surgery, such as conventional minimally invasive instruments, can help to address this issue. In this paper, a modular interface is presented which enables the usage of conventional minimally invasive instruments in a robotic telemanipulation setup. The interface allows to mount and actuate conventional instruments and thus enables the access to existing clinical processes, such as reprocessing and sterilization. The evaluation of the interface demonstrates the feasibility of the concept, however, further effort is required to enable the use of instruments with extended dexterity due to wrist articulation. Keywords: Robot-assisted surgery · Minimally invasive surgery · Telemanipulation · Laparoscopic instruments

1 Introduction Advantages such as reduced blood loss and a shorter postoperative hospital stay make minimally invasive surgery (MIS) indispensable in today’s healthcare [1]. However, operating with long, rigid, and slender instruments through small incisions also leads to some limitations and difficulties [2]. In order to remedy the disadvantages of MIS and still maintain its advantages over conventional interventions, robot-assisted surgery (RAS) systems have been in use for several years [3, 4]. As a subcategory of RAS, teleoperated MIS provides numerous advantages compared to conventional MIS, such as improved ergonomics, intuitive handling, tremor filtration or the scalability of motions [5, 6]. However, there is a financial hurdle which impedes the widespread use of these systems. Accordingly, there is a need for cost effective solutions. Aiming on a high level of integration of RAS systems into existing hospital workflows, the presented work focuses on the use of conventional instruments in teleoperated © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 107–114, 2022. https://doi.org/10.1007/978-3-030-76147-9_12

108

M. B. Schäfer et al.

MIS. Making use of existing infrastructure, such as already established procedures of reprocessing, might greatly benefit the ease of integration and reduce operating costs. Moreover, using off-the-shelf standard components for the whole telemanipulation setup is a cost-efficient and versatile approach for research around teleoperation in RAS, and could potentially also lead to a more widespread use of RAS systems in the operating room for various applications [7]. Other RAS systems in the research phase, but also commercially available systems, mostly use specifically designed instruments. The daVinci RAS system (Intuitive Surgical, Inc., Sunnyvale, CA, USA) for example uses dedicated designed instruments with a cable-based actuation [8, 9]. CMR Surgical (CMR Surgical Ltd., Cambridge, UK) also utilizes its own cable-based instrumentation for the Versius Surgical System [10]. The Senhance system (TransEnterix, Inc., Morrisville, NC, USA) uses custom-designed instruments with rigid linkages [11]. Only the BROCA system was identified to be compatible with conventional laparoscopic instruments [12]. In this work, a mechanical interface between an industrial robotic arm and conventional instruments for MIS is presented. Forming the manipulator-side of a master-slavetelemanipulation setup, the robotic arm and the surgical instruments are controlled by a mimetic input device. In the following, the interface is presented, characterized, and discussed aiming on a versatile and modular design for use of diverse instrument types.

2 Methods Telemanipulation System. As the manipulator of the master-slave telemanipulation system, a seven degree of freedom (DOF) articulated robotic arm (Panda, Franka Emika GmbH, Munich, Germany) was used. Featuring a payload of 3 kg in a range of motion which is sufficient for the targeted telemanipulation setup, the robotic arm is suitable for the guidance of a surgical instrument. Due to its kinematic redundancy, the robotic arm is capable of achieving any end effector pose with different joint configurations. This is especially beneficial for collaborative work with medical staff, additional robotic arms, or other equipment in the limited space around the surgical situs. Via Ethernet (TCP/IP), the input device and the control unit of the robotic arm are connected to a workstation running the control algorithms [13]. Instruments for Minimally Invasive Surgery. In this work, an interface for mounting and actuation of conventional instruments for MIS at the robotic end effector is presented. Therefore, rigid instruments with different kind of end effectors, but without so called wrist articulation (Fig. 1a) are considered. The majority of available instruments follow a similar design regarding the actuation of their end effectors. Despite the fact that various instruments from different manufacturers can be actuated with the presented working principle, the interface is designed particularly for instruments of the CLICKline by Karl Storz (KARL STORZ SE & Co. KG, Tuttlingen, Germany). The Karl Storz 33310 MD Kelly grasping forceps was used for evaluation of the interface. The interface to the robot was developed in regard to the design of the proximal end of the instruments shaft (see Fig. 1b). The instrument was originally actuated by a scissors handle via a simple click connection. Key elements of the shaft end are the annular groove for

Universal Mechanical Interface for Surgical Telemanipulation

109

axial fixation, the longitudinal groove for rotational fixation, the connecting rod, which allows for gripping movements and the hollow shaft as the housing of the connecting rod. Furthermore, the chosen instruments allow for monopolar high-frequency surgery which is also considered in the presented interface.

Joint without wrist articulation

Longitudinal Groove Annular Groove

Shaft Forceps

a)

Connecting Rod

Shaft

b)

Fig. 1. Grasping forceps of the Karl Storz instrument without wrist articulation (a) and proximal end of the Karl Storz instrument shaft (b).

Prototyping. Mechanical parts were manufactured using a Prusa i3 MK3S 3D printer (Prusa Research s.r.o., Praque, CZ) and polylactic acid (PLA) filament. To provide durable screw connections, metal thread inserts were melted in the PLA-components using a soldering iron. Electromechanic Components. For the actuation of the instruments end effector, a linear actuator, consisting of a NEMA 14 stepper motor and a conventional linear spindle drive (Nanotec Electronic GmbH & Co KG, Feldkirchen, Germany), and an A4988 stepper motor driver (AllegroSystems LLC, Manchester, NH, USA) were used. For closed-loop position feedback of the drive-unit, an OPTEK OPB 606A reflection light sensor (TT Electronics plc, Woking, UK) was set up. In order to determine the pose of the instruments end effector relatively to the robot’s end effector, an inertial measuring unit (IMU) GY-521 with a MPU-6050 chip (InvenSense Inc., San Jose, CA, USA) was integrated. All electronic components are controlled by an Arduino Nano V3 (Arduino LLC, MA, USA) with an ATmega328 microcontroller (Microchip Technology Inc., Chandler, AZ, USA). System Evaluation. Gripping forces were measured with a spring balance covering a measuring range from 0 N to 30 N. Onset latency of the drive-unit and the control algorithms was evaluated by visual measurement with a high-speed camera (MiniVis EoSens, High Speed Vision GmbH, Ettlingen, Germany) with a frame rate of 5000 frames per second. Validation of the monopolar high-frequency cutting abilities of the instrument, guided by the interface, was done by cutting and coagulation of a piece of chicken muscle tissue.

3 System Design and Results The mechanical interface system for conventional laparoscopic instruments consists of an adapter-unit supporting the instruments, a drive-unit containing the actuation components and the IMU, a gimbal joint to increase the workspace of the setup, and a

110

M. B. Schäfer et al.

control-unit. Due to the fully passive gimbal joint, the robot can only exert forces in three dimensions and torque around the end effectors rotational axis onto the instrument. Accordingly, the trocar point is used as a counter bearing. Figure 2 gives an overview of the complete setup. In the following sections, a brief description of the named units and their main functionalities is given, followed by the system characteristics.

Fig. 2. Complete setup of the interface and an instrument for minimally invasive surgery attached to the articulated robotic arm.

Adapter-Unit. The adapter-unit is designed specifically for the Karl Storz instruments stated above (see Fig. 3). However, different instruments can be used by slightly modifying the adapter unit. To attach an instrument to the adapter, the whole unit can be pivoted around a bolt, allowing the connection rod of the instrument to be fully inserted. The longitudinal groove in the instrument shaft is used to fix the rotational position while the annular groove is used for the click-in connection when inserting the shaft into the adapter (see Fig. 1b and Fig. 3a). To remove the instrument, a spring-based release button has to be pressed. Further, to enable monopolar high-frequency cutting, a connection pin is provided to allow connection of a medical high-frequency generator. The assembly of the adapter-unit to the rest of the interface can be released by removing the bolt. This enables to equip the setup with further adapters for instruments from other manufacturers and also represents a safety release of the instrument in case the robot is out of power during an intervention. Drive-Unit. Main component of the drive-unit is a linear actuator consisting of a stepper motor and a linear spindle (see Fig. 4). The spindle drives the nut and thus pushes the linear glider which transmits the movement to the ballhead of the connection rod. The overall displacement of the ballhead is 2 mm. For a closed-loop position feedback of the closure angle of the instruments end effector, the position of the linear glider is measured. This is realized by a combination of a photoelectric barrier sensor for an initial homing movement and an optical encoder. The optical encoder (see Fig. 4) consists of a reflection light sensor and a grey scale progression ranging from black to white. Since the interface contains a completely passive gimbal joint, it is necessary to measure the two angles of this joint to obtain the pose (position and orientation) of the

Universal Mechanical Interface for Surgical Telemanipulation a)

High-frequency Connection Pin

111

b)

Instrument Shaft

Connecting Rod Annular Groove Longitudinal Groove

Release Button Bolt Connection

Fig. 3. Cross section of the adapter-unit with the instrument (a) and adapter printed in PLA (b). a)

Stepper Motor

b)

Linear Glider Bolt Connection

Linear Spindle Spindle Nut Optical Encoder Fork for Ballhead

Fig. 4. Cross section of the drive-unit (a) and drive-unit printed in PLA (b).

instruments end effector. The instruments end effector pose can then be calculated from the pose of the robotic end effector and the gimbal joint angles. Therefore, an IMU is placed in the drive-unit. Its MPU-6050 sensor chip fuses the data of the respective rotation rate sensor and the acceleration sensor and thus reduces occurring sensor drift. Control-Unit. Main component of the control-unit is an Arduino Nano and the driver for the stepper motor. The housing provides mounting for all modules, ventilation slots, and a connection interface for the DC power plug, a mini USB-connection to flash the Arduino, a LAN-connection for fast data transmission, and a 10-pin socket to access free digital and analog pins of the Arduino. Features and Results. The overall dimensions of the interface are 336 mm length and a maximum diameter of 75 mm. Including the instrument and the power and data cables, the interface has a mass of 692 g. The linear actuator is able to generate a force of 160 N up to a linear speed of 12 mm/s, at 24 V voltage and 1.2 A current. Considering the reduction ratio of the gripping mechanism of the surgical instrument and the frictional loss, a gripping force of 25.75 N was finally measured at the tip of the Kelly grasping forceps. A complete closure of the grasping forceps is enabled in less than 200 ms while still providing the grasping force stated above. In full step operation, an angular resolution of 0.34° over the complete movement range of 68° of the forceps is achieved. The response onset latency of the drive-unit to an input signal was measured to be below 2.5 ms.

112

M. B. Schäfer et al.

Sterility and Safety Concept. To enable usage of the interface without exposing the electromechanic system to a sterilization process, a sterility concept using medical drapes is proposed. Therefore, robot and interface are covered in a long and slender drape. The drape provides a puncture section through which the instrument can be clicked into the adapter-unit. Furthermore, to provide a hardware-based safety feature in addition to the safety installations in the robots control algorithm, a quick-release is enabled by a bolt connection, enabling the removal of the instrument from the interface in case of an emergency situation (Fig. 3b). Figure 5 depicts the fully assembled interface where on the right side the connection to the robotic arm is enabled and on the left side the instrument can be inserted.

Gimbal Joint

Control-Unit Adapter-Unit

Drive-Unit

Fig. 5. Setup of the interface with adapter-unit, drive-unit, gimbal joint and control-unit.

4 Discussion This project aims on the use of conventional instruments for telemanipulated RAS. With the presented interface, a compact solution with various integrated features is given. In this first step, the approach focuses on stiff instruments without wrist articulation. However, to fully deploy the advantages of telemanipulated instead of conventional MIS, the additional DOF of articulated wrist instruments are indispensable [14]. Though, the presented interface is a reliable proof-of-concept and reveals the possibility to use conventional instruments for the targeted application. Furthermore, a compatibility of the interface to varying instrument designs of different manufacturers is required. The adapter-unit of the presented prototype was indeed designed for the Karl Storz instruments, but the modular design of the interface allows fast and easy change of the adapter-unit. Due to the universally designed clutch between the linear actuator and the connecting rod of the instrument, only the adapter-unit has to be designed instrument-specific. For the evaluation of the grasping forceps actuation, a simple linear potentiometer was used as an input device. The position of the potentiometer was mapped to the

Universal Mechanical Interface for Surgical Telemanipulation

113

closure angle of the forceps. This position control of the forceps should be enhanced by a more application-oriented control strategy such as a cascaded position-force-control to also enable the possibility of varying the interaction forces between the forceps and the grasped object. With the presented prototype, all required properties and functionalities, such as grasping force, closure time, onset latency, overall dimensions, and weight and also the feature of performing high-frequency cutting were achieved. However, to push the design from the prototype level to a more robust setup for extensive testing, a few parts should be manufactured in aluminum material instead of printed PLA. This applies for example to the linear guide of the drive-unit since the printed parts showed considerable friction and wearout. Furthermore, the electronic components should be combined on a common printed circuit board instead of being manually soldered to multiple veroboards. This could help to drastically reduce the size of the control unit. Last but not least, the modular design of the interface enables the implementation of additional segments, such as a force and torque measurement unit, by just placing it between drive-unit and gimbal joint or between gimbal joint and control-unit. Since the setup is made for research purposes, this strongly supports customization and modification of the interface.

5 Conclusion In this paper, a compact and modular interface for the use of conventional rigid instruments for teleoperated robotic MIS is presented. It enables the investigation of various research questions around telemanipulation in the medical field and depicts a first solution for the deployment of off-the-shelf products and existing processes in the clinical environment. Crucial challenges are the enhancement of using instruments with additional wrist articulation and to apply appropriate control algorithms to the drive-unit. Acknowledgment. The authors want to thank the company Karl Storz (KARL STORZ SE & Co. KG, Tuttlingen, Germany) for supporting this project by providing surgical instruments.

References 1. Tonutti, M., Elson, D.S., Yang, G.-Z., Darzi, A.W., Sodergren, M.H.: The role of technology in minimally invasive surgery: state of the art, recent developments and future directions. Postgrad. Med. J. 93, 159–167 (2017) 2. Gallagher, A.G., McClure, N., McGuigan, J., Ritchie, K., Sheehy, N.P.: An ergonomic analysis of the fulcrum effect in the acquisition of endoscopic skills. Endoscopy 30(7), 617–620 (1998) 3. Cole, A.P., Trinh, Q.-D., Sood, A., Menon, M.: The rise of robotic surgery in the new millennium. J. Urol. 197(2S), 213–215 (2017) 4. Ranev, D., Teixeira, J.: History of computer-assisted surgery. Surg. Clin. North Am. 100(2), 209–218 (2020) 5. Wee, I.J.Y., Kuo, L.-J., Ngu, J.C.-Y.: A systematic review of the true benefit of robotic surgery: ergonomics. Int. J. Med. Robot. Comput. Assist. Surg. 16(4), 2113 (2020)

114

M. B. Schäfer et al.

6. van der Schatte Olivier, R.H., van’t Hullenaar, C.D.P., Ruurda, J.P., Broeders, I.A.M.J.: Ergonomics, user comfort, and performance in standard and robot-assisted laparoscopic surgery. Surg. Endosc. 23(6), 1365–1371 (2009) 7. Schäfer, M.B., Stewart, K.W., Pott, P.P.: Industrial robots for teleoperated surgery – a systematic review of existing approaches. Curr. Direct. Biomed. Eng. 5(1), 153–156 (2019) 8. Sung, G.T., Gill, I.S.: Robotic laparoscopic surgery: a comparison of the da Vinci and Zeus systems. Urology 58(6), 893–898 (2001) 9. Freschi, C., Ferrari, V., Melfi, F., Ferrari, M., Mosca, F., Cuschieri, A.: Technical review of the da Vinci surgical telemanipulator. Int. J. Med. Robot. Comput. Assist. Surg. 9(4), 396–406 (2013) 10. Hares, L., Roberts, P., Marshall, K., Slack, M.: Using end-user feedback to optimize the design of the Versius Surgical System, a new robot-assisted device for use in minimal access surgery. BMJ Surg. Interv. Health Technol. 1, 1 (2019) 11. Albrecht, S.: Entwicklung aktuierter Instrumente mit multiplen Freiheitsgraden für Telemanipulatoren der minimal-invasiven Chirurgie, Technische Universität Berlin Dissertation, Berlin (2018) 12. Universidad de Córdoba: Nuevo concepto de Robótica Quirúrgica en Córdoba (2016) 13. Schäfer, M.B., Stewart, K.W., Losch, N., Pott, P.P.: Assessment of a commercial virtual reality controller for telemanipulation of an articulated robotic arm. In: 2020 8th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob), pp. 860– 865. IEEE (2020) 14. Anderson, P.L., Lathrop, R.A., Webster, R.J., III.: Robot-like dexterity without computers and motors: a review of hand-held laparoscopic instruments with wrist-like tip articulation. Expert Rev. Med. Devices 13(7), 661–672 (2016)

Volume Rendering-Based Patient Registration for Extended Reality 1(B) ˙ Marek Zelechowski , Bal´ azs Faludi1 , Georg Rauter2 , and Philippe C. Cattin1 1

CIAN, Department of Biomedical Engineering, University of Basel, Gewerbestrasse 14, 4123 Allschwil, Switzerland [email protected] 2 BIROMED-Lab, Department of Biomedical Engineering, University of Basel, Gewerbestrasse 14, 4123 Allschwil, Switzerland

Abstract. Modern computer-assisted surgery systems require a patient registration functionality to align the preoperative patient data with the patient on the operating room (OR) table. State-of-the-art systems use a two-stage approach where a handful of landmarks known in the preoperative data set are first pointed at on the patient, followed by some surface landmarks where the corresponding points are not known beforehand. The methods employed are, for example, iterative closest point (ICP), to align the digitized surface landmarks on the patient with a surface mesh of the target structure segmented during surgical planning. With the processing power of modern graphics processing units (GPUs), there is a desire to use direct volume-rendered images of a patient’s preoperative data during the planning stage of the intervention. This would greatly simplify the planning, as manual segmentation of anatomical structures would not be needed anymore. However, the current patient registration approaches rely on such segmented structures and can not yet deal with volume rendered data. This paper proposes a two-stage registration method similar to the classical referencing approaches but that uses volume rendered images instead of the common surface meshes. We quantitatively evaluated the proposed concept on a phantom model and could show its applicability even when segmented anatomical structures are not available. This approach shows great potential for reducing the time between image acquisition and operation for preplanned, orthopedic interventions.

1

Introduction

Medical imaging technologies have been widely adopted in clinical routine since their invention. They became an important diagnostic and surgical planning tool, quickly making their way into the operating room (OR) [6]. Image-guided surgery is now a standard approach in a variety of procedures, allowing the surgeon to precisely navigate a patient body, enabling the safe usage of minimally-invasive techniques that greatly reduce recovery time and possible adverse effects [13,15]. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 115–124, 2022. https://doi.org/10.1007/978-3-030-76147-9_13

116

˙ M. Zelechowski et al.

One of the biggest challenges of image-guided surgery is patient registration, which aims to find the transform between the patient and the preoperative data set as the reference frame. Only a successful registration enables surgical navigation, which is at the core of computer-assisted interventions, effectively enabling the surgeon to locate the target tissues and reach them safely and accurately. Despite some clear advantages of using navigation systems in the OR, many surgeons limit their use to basic video feedback and rely on their own skills and expertise to navigate a patient’s body [5]. There could be many reasons for this. However, a need for time-consuming preprocessing of the data (i.e. manual segmentation), the X-ray exposure (for intraoperative data acquisition), and the cumbersomeness of the system among others definitely restrict navigation systems usage [11]. As graphics processing units are becoming more and more powerful, they enable direct volume rendering techniques to be utilized for volumetric data visualization at a frame rate appropriate for virtual and augmented reality (VR/AR) [1,9]. This approach does not rely on a manual segmentation step, but uses a predefined transfer function (threshold segmentation): color and opacity are assigned to each voxel in the volumetric space. Nowadays, many surgical planning and intraoperative visualization software packages for virtual and augmented reality rely on computationally efficient surface models [2]. However, this introduces additional steps to the workflow, often requiring manual or semiautomatic segmentation of the data [4]. Excluding the mesh generation step makes volume rendering a far more suitable approach to be used in scenarios when time is of the essence. Reducing the time between image acquisition and displaying is crucial both in emergencies and particularly when performing intraoperative imaging (e.g. cone-beam computer tomography). With the recent hardware developments in VR/AR, we have seen a gradual adoption of these systems in clinics. While not yet widespread, VR has proven to be a very intuitive and flexible tool for surgical planning, especially in complex orthopedic or spine cases [1,7]. At relatively low cost, augmented reality, in particular, holds promise for seamless and intuitive visualization of imaging data, e.g. through virtual projections overlaid onto a patient [2,12]. To address some of the aforementioned challenges and to register a patient, in this paper we present a visualization system for VR and AR based on a twostep registration method. Our two-step method combines the Kabsch algorithm and a distance map generated from the volumetric data set taking into account also the transfer function. By utilizing real-time volume rendering and a predefined transfer function, we eliminate the need to manually segment the target structures for each case, thus minimizing the time between the imaging and the display step. We validate our method using a computer tomography (CT) data set in virtual reality. The accuracy measurements were performed using a professional, IR-based optical tracking system. The results show that in our target application - osteotomy - the proposed approach outperforms the standard Kabsch algorithm, without increasing the workload on the surgeon, nor requiring preprocessing of the data for patient registration.

Volume Rendering-Based Patient Registration for Extended Reality

2

117

Methods

Patient registration is very challenging in a cluttered OR environment. Therefore, we aimed for our method to be easily implemented in the surgical workflow, regardless of tracking technology used and at a minimal increase of load on the staff. On top of this, we wanted to make sure that no preprocessing was necessary, which otherwise might prevent the method from being used with intraoperative imaging data. 2.1

Volume Rendering for Virtual and Augmented Reality

To visualize 3D medical images, we used a developer version SpectoVR (Diffuse GmbH, Switzerland) software, which has built-in VR and AR support [1,8]. This application employs a high-performance volume rendering engine, which makes it suitable for virtual reality, where low frame rates may cause discomfort and nausea for the user. The software can load any volumetric data sets in DICOM format through a desktop application, while the VR and AR support is provided by an accompanying Unity application (Unity Technologies, USA). Thanks to direct volume rendering, there is no need for a complex, manual segmentation of the data, and the desired tissue types are displayed based on a predefined transfer function, through which the Hounsfield units are mapped to a selected color and opacity. Among other optimizations, the software precomputes a 3D distance map based on the applied transfer function. For each voxel in the volume, this map stores the distance to the closest visible voxel in the volume and can be later accessed in the application. Due to performance reasons, the resolution of the distance map is set to 1/4 of the data resolution. To improve the accuracy of our method, we increased the resolution to 1/1 at the cost of slightly longer loading times. This allowed us to efficiently implement a snap-to-surface feature and automatically find the matching points of the surface of the tissue in the two-step method. 2.2

Two-Step Approach

Our registration method consists of two steps: (1) the initial registration using paired point sets fed to the Kabsch algorithm, and (2) the fine registration where corresponding points are automatically calculated and optimized from a distance map. Additionally, during the marker placement in the first step, the user can be assisted with a snap-to-surface feature. This workflow is similar to the classical patient referencing approaches but the two steps are adapted such that they work with volumetric instead of surface data. (1) Initial registration: The selection of suitable landmarks on the volumetric data set can be performed either during the preoperative planning phase or intraoperatively. The users are then presented with the medical data set either in VR or AR, and with a tracked pointing device they select a minimum of 4 points on the surface of the rendered voxel object (see Fig. 1) at clearly distinguishable

118

˙ M. Zelechowski et al.

anatomical landmarks. The pointer also serves as a manipulator that can be used to move and rotate the data set freely in the VR/AR space. For user’s convenience, up-scaling of the model is also possible. Despite these facilitations, placement of landmarks on the surface of virtual objects without any assistance can be challenging [10]. To address this issue, we implemented a snap-to-surface function that moves a placed marker to the nearest point on the surface by using the distance map (see Fig. 1, right) initially calculated to speed-up the volume rendering procedure. After the selection of at least 4 points on the data set in VR/AR, the corresponding points are selected on the real patient. From these points are then selected on the real patient and from these two sets, an initial alignment transform is calculated by minimizing the mean square error between point pairs. To achieve this in a robust fashion, we employ the well-known Kabsch algorithm. (2) Fine registration: The initial alignment is necessary to proceed with the second step, where the surgeon draws one or multiple lines on the target surface of the patient. These 3D lines are then discretized into points matching the resolution of the data set. The algorithm uses these surface points as starting positions to find the closest voxels of the volumetric rendered surface by accessing the distance map of the loaded medical data set. Since high frame rates are a necessity for a comfortable use of a VR system, the algorithm progresses one step each rendered frame (90 frames per second (fps) for HTC Vive VR headset,

Fig. 1. Left: a view from the VR headset with voxel markers placed on the skull. Right: a cross section of the distance map of the skull. Brighter values correspond to a higher distance to the closest voxel.

Volume Rendering-Based Patient Registration for Extended Reality

119

60 fps for MagicLeap One), until it reaches the border of the target tissue (the value of the voxel in the distance map equals 0). This prevents any decrease in performance, which might affect user well-being. Once the surface point is found, the matching pair is added to a markers list. The process is parallel and runs for each of the drawn points. The Kabsch algorithm is then applied to the new set of pairs, and a second alignment transform is computed, further reducing the root mean squared deviation (RMSD). 2.3

Experimental Setup

To quantitatively assess the accuracy of our method, we used the Qualisys motion capture system (Qualisys AB, Sweden) with the Qualisys Track Manager 2020.2 (build 5880) (QTM) software. A total of six cameras were used in the setup and the calibration was done using a 300 mm wand. The locations of passive IR markers attached to the VR controller were defined in the QTM software as a rigid body. The system reported a residual error of the tracked body of 0.153 mm. Augmented reality is the preferred technology to be used in the OR, as it does not trap the surgeon in a virtual world without a connection to the patient. It must be noted that there are no commercially available AR systems with an accuracy and a reliability comparable to the 6◦ of freedom (DoF) Lighthouse tracking technology that is utilized by VR systems (Valve Corp., USA). The AR headset at our disposal, Magic Leap One (Magic Leap, Inc., USA), uses magnetic

Fig. 2. A custom-designed pointing attachment for the HTC vive controller.

120

˙ M. Zelechowski et al.

tracking combined with inertial sensors that gives a very rough estimation of the controller position. This prevents us from performing the two-step registration process, as the location of the physical pointing device in augmented reality is not accurate enough. Considering the above, in our validation experiment we used the HTC Vive Pro VR system (HTC Corp., Taiwan) connected to a midrange gaming laptop (Intel i7 9750H, NVidia RTX2070 GPU, 16 GB RAM). The headset renders the image at 1440 × 1600 pixels per eye at a 90 Hz screen refresh rate and has a controller for interacting with virtual objects. The localization technology that Valve uses for their VR system uses an infrared based external tracking system, which does not suffer from the aforementioned problems with controller tracking. It has been reported that that SteamVR Tracking can achieve a submillimeter accuracy in a non-laboratory environment [3]. As the controllers provided with the VR headset do not have a physical spot that would allow for precise pointing, we designed and 3D-printed a custom attachment (see Fig. 2) based on the existing model [14]. The pointing tip is in front of the controller, at a distance of 92.5 mm from the controller model origin. The entire registration process can be also performed using the Magic Leap AR headset and SteamVR controller by aligning the VR and AR coordinate systems through a built-in image recognition at the cost of an additional alignment error. In Fig. 3 we preview the of alignment performed on the VR system on the Magic Leap AR headset (NVIDIA®Parker SOC CPU, NVIDIA PascalTM GPU, 8 GB RAM). 2.4

Data Set

We used a CT image of a skull of the didactic skeleton “Oscar” (Erler-Zimmer GmbH, Germany) that was acquired with a Phoenix Nanotom scanner (General Electric, USA). The data set we obtained and downsampled to 550 × 750 ×  650 voxels (voxel size s = 0.296 0.296 0.296 mm). The original volume size of   162.80 222.00 192.40 mm was scaled to a more suitable size for examination. For the experimental setup, the skull was fixed on a table with SteamVR tracking stations located on 4 sides, approx. 2 m away from the object. We performed a manual registration of the CT data set using Qualisys tracking data and the classic Kabsch algorithm, with a Fiducial Registration Error (FRE) of 1.789 mm. After aligning the VR and Qualisys coordinate systems (F RE = 0.226 mm), we acquired a ground truth transform of the data set in the VR space. 2.5

Testing Scenarios

We asked the users to perform patient registration in three scenarios: a) using only the classic Kabsch algorithm with points manually placed in the virtual space and pointed at on the patient; b) using the Kabsch algorithm with snapping to the surface enabled in VR; c) using the two-step method, i.e. Kabsch algorithm with snapping to surface for the initial alignment, followed by additional

Volume Rendering-Based Patient Registration for Extended Reality

121

Fig. 3. Results of the final registration of the CT data of the skull overlaid on the phantom model. Photographs taken through the see-through display of the Magic Leap One AR headset (Magic Leap, Inc., USA) with the VR and AR coordinate systems aligned (F RE = 2.23 mm). Left: projection at half intensity with a cut-off in the middle. Right: model displayed fully, at the maximum intensity.

surface point registration. A total of six non-clinicians, with varied experience using VR, took part in our validation study. First, each user, equipped with the VR headset and the controller/pointer, was allowed to examine the data set in a virtual examining room. The participants could operate the controller in one of three manners: a voxel interaction mode, a marker placement mode in the virtual space and a marker placement mode in the real-world. In the voxel interaction mode, the user had the ability to move around the room and interact with the voxel object by changing the scale and rotating or changing its position. Switching to the marker placement tool allowed the creation, positioning and removal of markers in the respective space. When ready, the user selected at least 4 arbitrary points on the surface of the virtual object. Next, after taking off the headset, the user selected the corresponding points on the physical patient model while seeing a preview of the virtual points on a monitor. Once the initial alignment was done, the participant drew lines, i.e. digitized surface points, on the patient model, for it to be used in the fine registration step. The registration transform was then refined and applied to align the voxel data with the patient (see Fig. 3). After each trial, the final transform of the voxel object was compared with the ground truth position and orientation of the patient obtained using the medical grade Qualisys tracking system. We report the positional and rotational errors of registration for each scenario.

122

3

˙ M. Zelechowski et al.

Results

To estimate and quantify the algorithm performance, we analyzed the positional and rotational offsets relative to the ground truth transform obtained using the Qualisys tracking system. We report the positional and rotational error values obtained from the algorithm experimental trials in Fig. 4. A Two-Stage Registration. With the classic Kabsch algorithm, the reported deviation from the target position equals 13.73 ± 3.19 mm on average, while with the surface snapping feature enabled, it improved to 10.64 ± 2.12 mm. Utilizing the snap-to-surface feature in this step resulted in a statistically significant improvement to the positioning in comparison to the classic Kabsch method (p < 0.05, Mann-Whitney test). By additionally using the digitized surface points in the fine registration step, we reduced the positional error even further to 8.77 ± 1.26 mm, which is significantly lower than the values from the initial alignment (p < 0.001, Mann-Whitney test). The right-hand side of Fig. 4 shows the rotational offset from the ground truth transform. We observed no significant difference between the classic Kabsch and our proposed methods using the Mann-Whitney test. The average rotational errors were 6.21 ± 3.47◦ , 5.16 ± 3.81◦ and 7.17 ± 3.95◦ for the classic Kabsch, with snapping feature enabled and two-step method, respectively.

Fig. 4. Comparison of the positional (left) and rotational (right) errors for three tested scenarios. Data represents median (line), 25–75 percentile (box) and min-max range (bars), ****p < 0.0001, *p < 0.05.

4

Discussion

Image guidance is of significant help to the surgeon and a crucial part of many medical interventions. Thanks to advances in GPU technology, direct volume rendering has become a very efficient and fast visualization method, even for large volumetric data sets. At the same time, developments in augmented and

Volume Rendering-Based Patient Registration for Extended Reality

123

virtual reality have enabled healthcare professionals to get a better understanding of the patient anatomical structures, especially in abnormal cases. In this work, we leverage these two technologies and simplify the patient registration process to make it possible to use our method in intraoperative scenarios. We believe that the reduction in the time it takes to acquire and display the data, and the fact that the technology requires little manual effort, are major factors facilitating the use of image-guided techniques in the OR. Our implementation of the two-step method shows the potential of utilizing real-time volume rendering in image-guided surgery. We demonstrated the effective usage of the distance map, which is normally used for rendering optimization; therefore, no additional calculations are required. Registration tasks require a highly accurate tracking for the pointing tool, and using a consumer-grade VR tracking system may affect the performance of the entire system. The results obtained for the classic Kabsch algorithm reflect this very well. Considering these results as a base for comparison with our method, we have demonstrated a reduction in positional error and an overall improvement of consistency among users. Acknowledging this, our system is highly modular, and any component is interchangeable, therefore less accurate, not calibrated tracking systems can be replaced with a medical-grade one. The same principle applies to the visualization hardware, and the currently used VR headset can be switched with an AR device. The automatic generation of voxel surface points in the second step, results in a very limited amount of additional work for the surgeon, which increases the chances of surgeons adopting the technology. We believe that the current implementation, developed with laser osteotomy in mind, can be extended to other image-guided procedures by utilizing more refined transfer functions.

5

Conclusions

In this paper, we demonstrated the feasibility of using direct volume rendering techniques in patient registration process. The method we proposed does not require any mesh generation or a complex, manual segmentation. The snapping-to-surface feature we implemented that relies on the distance map improves the registration through improved marker placement. As such, the feature may also be applied in surgical planning tools based on volume rendering for annotation and measuring tasks, thereby increasing the accuracy of measurements. While the proposed method requires some actions from the personnel in the operating room, it mostly eliminates the time-consuming preprocessing of the imaging data. This significant optimization in the workflow might even enable planning on intraoperative data or of complex trauma cases where time is critical.

124

˙ M. Zelechowski et al.

References 1. Croci, D.M., et al.: Novel patient-specific 3d-virtual reality visualisation software (spectovr) for the planning of spine surgery: a case series of eight patients. BMJ Innovations 6(4), 215–219 (2020) 2. Gerup, J., Soerensen, C.B., Dieckmann, P.: Augmented reality and mixed reality for healthcare education beyond surgery: an integrative review. Int. J. Med. Educ. 11, 1–18 (2020) 3. Groves, L.A., Carnahan, P., Allen, D.R., Adam, R., Peters, T.M., Chen, E.C.: Accuracy assessment for the co-registration between optical and vive head-mounted display tracking. Int. J. Comput. Assist. Radiol. Surg. 14(7), 1207–1215 (2019). https://doi.org/10.1007/s11548-019-01992-4 4. Gu, W., Shah, K., Knopf, J., Navab, N., Unberath, M.: Feasibility of image-based augmented reality guidance of total shoulder arthroplasty using microsoft hololens 1. Comput. Methods Biomech. Biomed. Eng. Imaging Vis. 9(3), 1–10 (2020) 5. J¨ odicke, A., Ottenhausen, M., Lenarz, T.: Clinical use of navigation in lateral skull base surgery: results of a multispecialty national survey among skull base surgeons in Germany. J. Neurol. Surg. Part B 79(06), 545–553 (2018) 6. Keereweer, S., et al.: Optical image-guided surgery-where do we stand? Mol. Imag. Biol. 13(2), 199–207 (2011). https://doi.org/10.1007/s11307-010-0373-2 7. Kim, Y., Kim, H., Kim, Y.O.: Virtual reality and augmented reality in plastic surgery: a review. Arch. Plast. Surg. 44(3), 179 (2017) 8. Knecht, S., Brantner, P., Cattin, P., Tobler, D., K¨ uhne, M., Sticherling, C.: Stateof-the-art multimodality approach to assist ablations in complex anatomies-from 3d printing to virtual reality. Pacing Clin. Electrophysiol. 42(1), 101–103 (2019) 9. Maloca, P.M., et al.: Validation of virtual reality orbitometry bridges digital and physical worlds. Sci. Rep. 10(1), 1–9 (2020) 10. Ren, J., Patel, R.V., McIsaac, K.A., Guiraudon, G., Peters, T.M.: Dynamic 3-d virtual fixtures for minimally invasive beating heart procedures. IEEE Trans. Med. Imaging 27(8), 1061–1070 (2008) 11. Ringel, F., Villard, J., Ryang, Y.-M., Meyer, B.: Navigation, robotics, and intraoperative imaging in spinal surgery. In: Schramm, J. (ed.) Advances and Technical Standards in Neurosurgery. ATSN, vol. 41, pp. 3–22. Springer, Cham (2014). https://doi.org/10.1007/978-3-319-01830-0 1 12. Schneider, C., et al.: Evaluation of an augmented reality based image guidance system for laparoscopic liver surgery. HPB 21, S671–S672 (2019) 13. Thaeter, M., Kobbe, P., Verhaven, E., Pape, H.C.: Minimally invasive techniques in orthopedic trauma. Curr. Trauma Rep. 2(4), 232–237 (2016). https://doi.org/ 10.1007/s40719-016-0066-7 14. Thingiverse.com, T.: Vive system blocker (saber guard) (2018). https://www. thingiverse.com/thing:3005388 15. Yin, W., Ma, J., Liao, Y.: Surgical treatment of brainstem cavernous malformations with three basic skull base approaches and minimally invasive techniques: observations in 20 patients. Transl. Neurosci. Clin. 3(2), 74–83 (2017)

Towards Robotic Surgery for Cartilage Replacement: A Review on Cartilage Defects Philipp Krenn1(B) , Manuela Eugster2 , Esther I. Zoller2 , Niklaus F. Friederich2,3 , and Georg Rauter2 1

Kantonsspital St. Gallen, Rorschacher Strasse 95, 9001 St. Gallen, Switzerland [email protected] 2 BIROMED-Lab, Department of Biomedical Engineering, University of Basel, Gewerbestrasse 14, 4123 Allschwil, Switzerland 3 University Hospital Basel, Petersgraben 4, 4031 Basel, Switzerland Abstract. Precise knowledge of the size and location of hyaline cartilage defects is essential for the development of robotic tools for cartilage repair surgery in the knee. The German Cartilage Registry indicates that about 45% of the defects are at the medial femoral condyle and nearly 30% at the patella. Our literature analysis confirms that the femur, especially the medial side, is the most frequent location for cartilage defects in the knee but that defects in the retropatellar region are common as well. Surgery methods to treat cartilage defects can be differentiated by the type of surgical approach. Most defects are eligible for minimal-invasive arthroscopic treatment. Accordingly developed instruments should reach the femoral and retropatellar region. Regarding cartilage defect size a range from 0.02 cm2 to 50 cm2 was reported in literature. Keywords: Cartilage replacement surgery Chondral defect knee

1

· Robotic knee surgery ·

Introduction

Hyaline cartilage defects occur in 63% of patients undergoing knee arthroscopy [1]. Current treatment options for such defects range from palliation (i.e., debridement), reparation (i.e., marrow stimulation techniques) and restoration (i.e., autologous chondrocyte implantation (ACI), matrix-induceed ACI (MACI) or osteochondral grafting) to resurfacing (i.e., focal defect resurfacing or joint arthroplasty). Restoration techniques are usually preferred for larger lesions or when other less invasive techniques have not been able to alleviate the patient’s symptoms [2]. For ACI, a biopsy of cartilage is taken and cultured ex vivo. In a second intervention, the injured cartilage is removed and the site covered with an impermeable periosteal patch. Thereafter, the cells will be injected under the periosteal patch and the patch sealed [2]. In contrast to ACI, for MACI the harvested chondrocytes are seeded for culturing. During the second intervention, the c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 125–136, 2022. https://doi.org/10.1007/978-3-030-76147-9_14

126

P. Krenn et al.

cultured membrane gets cropped according to the defect geometry and attached to the defect location with a sealer [3]. In contrast to ACI and MACI, osteochondral grafting only requires one surgical intervention. An osteochondral graft of healthy cartilage is transplanted to the defect site either from a low-stress joint region of the patient (autograft) or from the joint of a deceased donor (allograft) [2,4]. While osteochondral allograft transplantation and ACI are usually performed by an arthrotomy, osteochondral autograft transplantation (OAT) and MACI can be performed arthroscopically [2,3]. The arthroscopic dissection of cartilage defects is technically demanding [5] and measurement of the defect size for subsequent evaluation of the graft size in MACI is less accurate than via arthrotomy [3]. For microfracture, the defect is initially prepared which means that damaged cartilage is removed to create stable vertical walls of healthy cartilage. Then the calcified layer is removed. Several perforations of the subchondral bone are made with awls. As a result, marrow components enter the defect site and reparative tissue is built up [6]. The choice of surgery performed to treat cartilage defects depend on various criteria which are anatomic factors like leg axis, defect size and location. In pathological leg axis a correction should be performed before cartilage surgery. Moreover orthopedic surgeons should consider the physical demands of every patient. Patients should be in the focus of treatment and not radiological findings on magnetic resonance imaging (MRI). Low demand patient in advanced age may be better treated with implantation of prothesis, while prothestic joint replacement should be postponed as far as possible in younger and active patients. Furthermore previous treatment should be considered before planning surgery. Over the last years, more and more robotic instruments have been developed for the treatment of several diseases [7]. Robotic devices have the potential to simplify the preparation of defects. Advantages of using robotic devices for cartilage restoration procedures include the decrease of invasiveness, enabling a digital workflow (accurate preparation of the injured cartilage region and appropriate preparation of the healthy graft) [8]. Robotic tools have the potential to improve the preparation of the injured region and the integration of the healthy graft, and thus the cartilage regeneration process. Nevertheless, the requirements for such robotic tools have to be clarified before development (Fig. 1). The purpose of this study is to review and summarize information on the size, location and frequency of cartilage defects in the knee.

2

Methods

Systematic review of the literature on cartilage defects in the knee was performed using PubMed. PubMed is a search engine for publications in medicine and biomedical engineering. The following keywords were used for the search: “cartilage defect knee”, “chondral lesion knee” and “knee cartilage repair” for human subjects (cartilage AND defect AND knee AND (location OR (frequency OR frequencies)); chondral AND lesion AND knee AND (location OR (frequency

Robotic Surgery for Cartilage Replacement

127

Fig. 1. Based on the specifications of the defect and the patient’s physical demand, different treatment options exist. The treatments can be subdivided into treatments that are performed arthroscopically and treatments that require an arthrotomy. Arthroscopy is the less invasive treatment option. We aim to develop a robotic tool that allows performing cartilage restoration in a more precise and less invasive manner than nowadays.

OR frequencies)); knee AND cartilage AND repair AND (location OR (frequency OR frequencies))). Studies from 01/01/1995–31/12/2019 were included. This resulted in following number of matches: cartilage defect knee: 80, chondral lesion knee: 45, knee cartilage repair: 161. Both studies without precise information on defect location and/or size and studies that contained data on artificially created cartilage defects were excluded. Further, articles that did not contain new data and case reports were also excluded. Exclusion criteria resulted in 31 included studies for the term “cartilage defect knee”, in 16 for “chondral lesion knee” and in 32 studies for “knee cartilage repair”. Considering that some studies were matching with more than one term, the final amount of different articles was 62. Of all these 62 articles data concerning cartilage defect location, frequency and size were summarized.

3

Results

Due to the fact, that in five different cases multiple data sets of one author were considered for this study, there is a multiple count of patients. This affects following studies: the four studies by Ebert [9–12], the five studies by Niemeyer [13–17], two studies by Pestka [18,19], two studies by Tirico [20,21] as well as the two studies by Wang [22,23]. The used search terms led to a very heterogeneous result. Some only included patients who were treated by surgery, while other studies did not focus on surgery. Mostly, a direct comparison of the results (number of cases, patients, knees) was not possible, caused by the authors using different ways of counting their cases. While some authors preferred to count patients, others used the number of knees and some counted the number of cartilage defects. Therefore, we decided to convert the obtained data into number of cartilage defects. Furthermore, one study included both cartilage defects of the knee and defects of the talus [24]. In that special case we subtracted the number of talar cartilage defects.

128

P. Krenn et al.

Fig. 2. Overview of cartilage defect sizes. The minimal (min), maximal (max), mean (mean) and standard deviation (std) of the cartilage defect size are visualized for the reviewed publications. The minimal and maximal value are represented with a vertical line. The mean value is marked with a star and for standard deviation a box is used. Not all data were normally distributed.

Robotic Surgery for Cartilage Replacement

3.1

129

Defect Size

While many studies had no minimum or maximum size as an inclusion criteria some studies only included defects of a certain size, such as >1 cm2 [25,26], >2 cm2 [27,28], ≥2 cm2 [22,29,30], 3 cm2 [15,32]. Wang et al. defined a minimum width of 10 mm [23] as an inclusion criteria, whereas Knutsen et al. used the description “relatively large” [33]. However, not only the inclusion criteria regarding size varied also the method of measurement. Some had measured the size on MRI images whereas others measured it during surgery, open or by arthroscopy. The smallest defect size described was 0.02 cm2 [34] and the largest 50 cm2 [21]. Some included studies mentioned only the overall defect size whereas other studies distinguished the defect size of different subgroups. These subgroups were anatomic regions [35,36], anatomic regions for each gender [30], age groups (children and adults) [17] and clinical information like anterior cruciate ligament intact knees and knees with a reconstructed anterior cruciate ligament [22], traumatic and degenerative cartilage defects [37], failure and non-failure of microfracture [31]. Other sub-groups consisted of the method to measure the defects [38] or different operative treatments [26,27,29,32,39,40]. Including all information the lowest mean of cartilage defects was found in the subgroup lateral defects of the study from Everhart, 0.47 cm2 [36] and the highest mean in the subgroup male with trochlea defects of the study from Kreuz, 9.66 cm2 [30]. An overview of the different defect sizes can be found in Fig. 2. 3.2

Defect Locations

To facilitate comparison of the different studies seven regions were determined and all data was matched to one of these regions. The regions are medial femur, lateral femur, medial tibia, lateral tibia, trochlea, patella and unspecific. In four studies no data concerning defect location was mentioned [33,38,41,42]. The femur region could not be divided in medial and lateral side in seven studies [16,17,21,43–46]. In six studies the tibia region was not divided into medial side and lateral side [21,28,34,45–47]. Three studies differentiated the location only in three regions, medial, lateral and patella with trochlea region [11,48,49]. Another study showed similar described regions: medial, lateral and patella [50]. In the study of de Windt no differentiation of lateral femur and lateral tibia could be made [51]. The majority of studies that had included every anatomic region in the knee showed that more than half of the defects were on the femoral side. On the femur the medial side is more commonly affected. Exceptions to this are the studies by Provencher et al., Ringler et al., Niethammer et al., Scillia et al., Jones et al., and Nakamura et al. [17,46,52–55]. Six studies had a relatively high percentage of defects on the tibial side: Everhart et al. (45%) [36], Ebert et al. (30%) [9], Nakamura et al. (46%) [55], Tandogan et al. (26%) [56], Jones et al. (31%) [54], Fok et al. (29%) [57]. Three of the included studies had solely patients with defects at the patella and trochlea [23,37,58]. These studies showed that the distribution of these two regions were 69–70% for patella cartilage defects and 26–31% for trochlea defects. In Fig. 3 we show the distribution of defect locations.

130

P. Krenn et al.

Fig. 3. Location of cartilage defects. The total number of defects is specified in brackets. The following locations were provided: Medial femoral condyle (MFC), lateral femoral condyle (LFC), medial tibia (MT), lateral tibia (LT), trochlea (T), patella (P). Areas with dashed lines mean that two regions are combined.

Robotic Surgery for Cartilage Replacement

4

131

Discussion

In general, it is difficult to compare the different studies concerning size or location, as many studies focus on a specific topic or patient collective. Furthermore, not every included study had a focus on cartilage defects. Some studies mentioned cartilage defects as additive information. The heterogeneity of the studies caused a wide range for the defect sizes. What can be clearly seen in this summary is that most of the defects are on the femoral side, specifically on the medial condyle. Another option would have been to focus only on data from cartilage defects in need of treatment, which would have resulted in a more homogeneous data set. Knowledge about the most frequent location of cartilage defects and size of defects is important for the future development of robotic instruments in knee surgery, as the developed instruments should reach the femoral condyle area and cover a wide range of defect sizes. To define the required degrees of freedom or the range of motion for the instruments, information about the intraarticular space is required in addition to the location and size of the defect. It should be considered that the accessibility of the different anatomical regions depends on the knee position during surgery (knee flexion angle), which itself depends on the defect position. Thus, the workspace available for the surgical instrument depends on the defect location and the anatomy of the patient. To derive such quantitative requirements for robotic instruments for minimally invasive knee surgery, the intraarticular space inside the knee joint was evaluated for performing bone cuts in unicondylar knee arthroplasty (UKA) by Eugster et al. [59]. They concluded that an instrument thickness of 5–8 mm could be feasible for femoral bone cuts with a robotic tool for UKA [59]. We assume that a similar tool size would also be feasible for a robotic tool for cartilage surgery. Robotic assisted treatment should be used to expand the field of application of arthroscopy, because arthroscopic treatment has a lower morbidity compared to an open treatment [60]. Until today the main limitation of arthroscopy is to reach the posterior region of the condyles and the patella [61]. Based on the available data no quantitative conclusion concerning the degree of freedom of robotic instruments can be made. Cartilage defects can be located in every compartment of the knee joint. Therefore, we would suggest that a robotic instrument for cartilage repair should allow positioning of the instrument tip in all six degrees of freedom. We see a possible workflow of robotic knee arthroscopy for cartilage defect repair as follows: For safety reasons, the incision is performed by the surgeon. After incisions in the medial and lateral soft spot, the robotic instruments can be placed into the joint. The surgical task then continues robotically. It is advisable to use one incision for the arthroscope similar to standard arthroscopy for visualization and the second incision for the working instruments. To clean the borders of cartilage defects a milling tool might be used. The milling tools could have different sizes which would help handling the different defect sizes as precisely as possible. In case a membrane has to be used to repair the defect, the suturing instruments should have a pressure sensor to avoid damage to the surrounding healthy cartilage. With regard to choosing the appropriate size of the instruments to be developed, one could be

132

P. Krenn et al.

guided by conventional arthroscopic instruments. In general it is an advantage for the application of robotic assisted cartilage surgery that MRI is commonly used before cartilage surgery. It facilitates careful planning before surgery.

5

Conclusion

This literature analysis shows the majority of cartilage defects are located on the femur, especially on the medial side. For the development of new instruments it is important to know the distribution of cartilage defects. We show that the size of the defects are highly variable. Nowadays smaller cartilage defects are treated by microfracture. As this is not a difficult surgical technique, robotic surgery tools are not expected to be routinely used for this technique. It is more likely that robotic assisted surgery in the future will play an important role in osteochondral autograft transfer and autologous chondrocyte implantation. Therefore in the development for robotic devices it can be predicted, that it will be important to focus on relatively large defects 4 cm2 . Acknowledgement. We gratefully acknowledge funding of the Werner Siemens Foundation through the MIRACLE project.

References 1. Curl, W.W., Krome, J., Gordon, E.S., Rushing, J., Smith, B.P., Poehling, G.G.: Cartilage injuries: a review of 31,516 knee arthroscopies. Arthrosc. J. Arthrosc. Relat. Surg. 13(4), 456–460 (1997) 2. Cole, B.J., Lee, S.J.: Complex knee reconstruction: articular cartilage treatment options. Arthroscopy 19(10), 1–10 (2003) 3. Willers, C., Partsalis, T., Zheng, M.H.: Articular cartilage repair: procedures versus products. Expert Rev. Med. Devices 4(3), 373–392 (2007) 4. Richter, D.L., Tanksley, J.A., Miller, M.D.: Osteochondral autograft transplantation: a review of the surgical technique and outcomes. Sports Med. Arthrosc. Rev. 24(2), 74–78 (2016) 5. McNickle, A.G., Provencher, M.T., Cole, B.J.: Overview of existing cartilage repair technology. Sports Med. Arthrosc. Rev. 16(4), 196–201 (2008) 6. Watts, E.: Articular cartilage defects of knee (2020). https://www.orthobullets. com/knee-and-sports/3133/articular-cartilage-defects-of-knee. Accessed 24 Jan 2021 7. Leal Ghezzi, T., Campos Corleta, O.: 30 years of robotic surgery. World J. Surg. 40(10), 2550–2557 (2016). https://doi.org/10.1007/s00268-016-3543-9 8. Lipskas, J., Deep, K., Yao, W.: Robotic-assisted 3D bio-printing for repairing bone and cartilage defects through a minimally invasive approach. Sci. Rep. 9(1), 1–9 (2019) 9. Ebert, J.R., Fallon, M., Ackland, T.R., Wood, D.J., Janes, G.C.: Arthroscopic matrix-induced autologous chondrocyte implantation: 2-year outcomes. Arthrosc. J. Arthrosc. Relat. Surg. 28(7), 952–964 (2012) 10. Ebert, J.R., Smith, A., Edwards, P.K., Hambly, K., Wood, D.J., Ackland, T.R.: Factors predictive of outcome 5 years after matrix-induced autologous chondrocyte implantation in the tibiofemoral joint. Am. J. Sports Med. 41(6), 1245–1254 (2013)

Robotic Surgery for Cartilage Replacement

133

11. Ebert, J.R., Smith, A., Fallon, M., Wood, D.J., Ackland, T.R.: Degree of preoperative subchondral bone edema is not associated with pain and graft outcomes after matrix-induced autologous chondrocyte implantation. Am. J. Sports Med. 42(11), 2689–2698 (2014) 12. Ebert, J.R., Fallon, M., Wood, D.J., Janes, G.C.: A prospective clinical and radiological evaluation at 5 years after arthroscopic matrix-induced autologous chondrocyte implantation. Am. J. Sports Med. 45(1), 59–69 (2017) 13. Niemeyer, P., et al.: Autologous chondrocyte implantation for the treatment of retropatellar cartilage defects: clinical results referred to defect localisation. Arch. Orthop. Trauma Surg. 128(11), 1223–1231 (2008). https://doi.org/10.1007/ s00402-007-0413-9 14. Niemeyer, P., et al.: Characteristic complications after autologous chondrocyte implantation for cartilage defects of the knee joint. Am. J. Sports Med. 36(11), 2091–2099 (2008) 15. Niemeyer, P., Pestka, J.M., Erggelet, C., Steinwachs, M., Salzmann, G.M., S¨ udkamp, N.P.: Comparison of arthroscopic and open assessment of size and grade of cartilage defects of the knee. Arthrosc. J. Arthrosc. Relat. Surg. 27(1), 46–51 (2011) 16. Niemeyer, P., Pestka, J.M., Salzmann, G.M., S¨ udkamp, N.P., Schmal, H.: Influence of cell quality on clinical outcome after autologous chondrocyte implantation. Am. J. Sports Med. 40(3), 556–561 (2012) 17. Niethammer, T.R., Holzgruber, M., G¨ ulecy¨ uz, M.F., Weber, P., Pietschmann, M.F., M¨ uller, P.E.: Matrix based autologous chondrocyte implantation in children and adolescents: a match paired analysis in a follow-up over three years postoperation. Int. Orthop. 41(2), 343–350 (2016). https://doi.org/10.1007/s00264016-3321-1 18. Pestka, J.M., Bode, G., Salzmann, G., S¨ udkamp, N.P., Niemeyer, P.: Clinical outcome of autologous chondrocyte implantation for failed microfracture treatment of full-thickness cartilage defects of the knee joint. Am. J. Sports Med. 40(2), 325–331 (2012) 19. Pestka, J.M., Feucht, M.J., Porichis, S., Bode, G., S¨ udkamp, N.P., Niemeyer, P.: Return to sports activity and work after autologous chondrocyte implantation of the knee: which factors influence outcomes? Am. J. Sports Med. 44(2), 370–377 (2016) 20. T´ırico, L.E., McCauley, J.C., Pulido, P.A., Bugbee, W.D.: Does anterior cruciate ligament reconstruction affect the outcome of osteochondral allograft transplantation? A matched cohort study with a mean follow-up of 6 years. Am. J. Sports Med. 46(8), 1836–1843 (2018) 21. T´ırico, L.E., McCauley, J.C., Pulido, P.A., Demange, M.K., Bugbee, W.D.: Is patient satisfaction associated with clinical outcomes after osteochondral allograft transplantation in the knee? Am. J. Sports Med. 47(1), 82–87 (2019) 22. Wang, D., et al.: Similar outcomes after osteochondral allograft transplantation in anterior cruciate ligament-intact and-reconstructed knees: a comparative matchedgroup analysis with minimum 2-year follow-up. Arthrosc. J. Arthrosc. Relat. Surg. 33(12), 2198–2207 (2017) 23. Wang, T., et al.: Patellofemoral cartilage lesions treated with particulated juvenile allograft cartilage: a prospective study with minimum 2-year clinical and magnetic resonance imaging outcomes. Arthrosc. J. Arthrosc. Relat. Surg. 34(5), 1498–1505 (2018)

134

P. Krenn et al.

24. Angele, P., Fritz, J., Albrecht, D., Koh, J., Zellner, J.: Defect type, localization and marker gene expression determines early adverse events of matrix-associated autologous chondrocyte implantation. Injury 46, S2–S9 (2015) 25. Van Rossom, S., Khatib, N., Holt, C., Van Assche, D., Jonkers, I.: Subjects with medial and lateral tibiofemoral articular cartilage defects do not alter compartmental loading during walking. Clin. Biomech. 60, 149–156 (2018) 26. Bartlett, W., et al.: Autologous chondrocyte implantation versus matrix-induced autologous chondrocyte implantation for osteochondral defects of the knee: a prospective, randomised study. J. Bone Joint Surg. 87(5), 640–645 (2005) 27. Minas, T., Ogura, T., Headrick, J., Bryant, T.: Autologous chondrocyte implantation “sandwich” technique compared with autologous bone grafting for deep osteochondral lesions in the knee. Am. J. Sports Med. 46(2), 322–332 (2018) 28. Albrecht, C., Tichy, B., Zak, L., Aldrian, S., N¨ urnberger, S., Marlovits, S.: Influence of cell differentiation and il-1β expression on clinical outcomes after matrixassociated chondrocyte transplantation. Am. J. Sports Med. 42(1), 59–69 (2014) 29. Akgun, I., et al.: Matrix-induced autologous mesenchymal stem cell implantation versus matrix-induced autologous chondrocyte implantation in the treatment of chondral defects of the knee: a 2-year randomized study. Arch. Orthop. Trauma Surg. 135(2), 251–263 (2014). https://doi.org/10.1007/s00402-014-2136-z 30. Kreuz, P.C., et al.: Influence of sex on the outcome of autologous chondrocyte implantation in chondral defects of the knee. Am. J. Sports Med. 41(7), 1541– 1548 (2013) 31. Salzmann, G.M., Sah, B., S¨ udkamp, N.P., Niemeyer, P.: Reoperative characteristics after microfracture of knee cartilage lesions in 454 patients. Knee Surg. Sports Traumatol. Arthrosc. 21(2), 365–371 (2013). https://doi.org/10.1007/s00167-0121973-y 32. Niemeyer, P., et al.: First-generation versus second-generation autologous chondrocyte implantation for treatment of cartilage defects of the knee: a matched-pair analysis on long-term clinical outcome. Int. Orthop. 38(10), 2065–2070 (2014). https://doi.org/10.1007/s00264-014-2368-0 33. Knutsen, G., et al.: A randomized multicenter trial comparing autologous chondrocyte implantation with microfracture: long-term follow-up at 14 to 15 years. JBJS 98(16), 1332–1339 (2016) 34. Wright, K.T., Kuiper, J.H., Richardson, J.B., Gallacher, P., Roberts, S.: The absence of detectable ADAMTS-4 (aggrecanase-1) activity in synovial fluid is a predictive indicator of autologous chondrocyte implantation success. Am. J. Sports Med. 45(8), 1806–1814 (2017) 35. Pietschmann, M.F., et al.: The incidence and clinical relevance of graft hypertrophy after matrix-based autologous chondrocyte implantation. Am. J. Sports Med. 40(1), 68–74 (2012) 36. Everhart, J.S., Abouljoud, M.M., Flanigan, D.C.: Role of full-thickness cartilage defects in knee osteoarthritis (OA) incidence and progression: data from the OA initiative. J. Orthop. Res. 37(1), 77–83 (2019) 37. Mehl, J., et al.: Degenerative isolated cartilage defects of the patellofemoral joint are associated with more severe symptoms compared to trauma-related defects: results of the German Cartilage Registry (KnorpelRegister DGOU). Knee Surg. Sports Traumatol. Arthrosc. 27(2), 580–589 (2018). https://doi.org/10.1007/ s00167-018-5184-z 38. Selmi, T.A.S., et al.: Autologous chondrocyte implantation in a novel alginateagarose hydrogel: outcome at two years. J. Bone Joint Surg. 90(5), 597–604 (2008)

Robotic Surgery for Cartilage Replacement

135

39. Dasar, U., Gursoy, S., Akkaya, M., Algin, O., Isik, C., Bozkurt, M.: Microfracture technique versus carbon fibre rod implantation for treatment of knee articular cartilage lesions. J. Orthop. Surg. 24(2), 188–193 (2016) ˇ et al.: Evaluation of native hyaline cartilage and repair tissue after two 40. Zb` yn ˇ, S., cartilage repair surgery techniques with 23Na MR imaging at 7 T: initial experience. Osteoarthr. Cartil. 20(8), 837–845 (2012) 41. Niemeyer, P., Feucht, M.J., Fritz, J., Albrecht, D., Spahn, G., Angele, P.: Cartilage repair surgery for full-thickness defects of the knee in Germany: indications and epidemiological data from the German Cartilage Registry (KnorpelRegister DGOU). Arch. Orthop. Trauma Surg. 136(7), 891–897 (2016). https://doi.org/10. 1007/s00402-016-2453-5 42. Minzlaff, P., Feucht, M.J., Saier, T., Cotic, M., Plath, J.E., Imhoff, A.B., Hinterwimmer, S.: Can young and active patients participate in sports after osteochondral autologous transfer combined with valgus high tibial osteotomy? Knee Surg. Sports Traumatol. Arthrosc. 24(5), 1594–1600 (2014). https://doi.org/10. 1007/s00167-014-3447-x 43. Delcogliano, A., Caporaso, A., Menghi, A., Rinonapoli, G., Chiossi, S.: Results of autologous osteochondral grafts in chondral lesions of the knee. Minerva Chir. 57(3), 273–281 (2002) 44. Nielsen, E.S., McCauley, J.C., Pulido, P.A., Bugbee, W.D.: Return to sport and recreational activity after osteochondral allograft transplantation in the knee. Am. J. Sports Med. 45(7), 1608–1614 (2017) 45. Behery, O.A., Harris, J.D., Karnes, J.M., Siston, R.A., Flanigan, D.C.: Factors influencing the outcome of autologous chondrocyte implantation: a systematic review. J. Knee Surg. 26(03), 203–212 (2013) 46. Scillia, A.J., et al.: Return to play after chondroplasty of the knee in national football league athletes. Am. J. Sports Med. 43(3), 663–668 (2015) 47. Mithoefer, K., Venugopal, V., Manaqibwala, M.: Incidence, degree, and clinical effect of subchondral bone overgrowth after microfracture in the knee. Am. J. Sports Med. 44(8), 2057–2063 (2016) 48. Hjermundrud, V., Bjune, T.K., Risberg, M.A., Engebretsen, L., ˚ Arøen, A.: Fullthickness cartilage lesion do not affect knee function in patients with ACL injury. Knee Surg. Sports Traumatol. Arthrosc. 18(3), 298–303 (2010). https://doi.org/ 10.1007/s00167-009-0894-x 49. Ciccotti, M.C., et al.: The prevalence of articular cartilage changes in the knee joint in patients undergoing arthroscopy for meniscal pathology. Arthroscopy 28(10), 1437–1444 (2012) 50. Anderson, A.F., Anderson, C.N.: Correlation of meniscal and articular cartilage injuries in children and adolescents with timing of anterior cruciate ligament reconstruction. Am. J. Sports Med. 43(2), 275–281 (2015) 51. de Windt, T.S., Bekkers, J.E., Creemers, L.B., Dhert, W.J., Saris, D.B.: Patient profiling in cartilage regeneration: prognostic factors determining success of treatment for cartilage defects. Am. J. Sports Med. 37(1), 58–62 (2009) 52. Provencher, M.T., et al.: Symptomatic focal knee chondral injuries in national football league combine players are associated with poorer performance and less volume of play. Arthroscopy 34(3), 671–677 (2018) 53. Ringler, M.D., Shotts, E.E., Collins, M.S., Howe, B.M.: Intra-articular pathology associated with isolated posterior cruciate ligament injury on MRI. Skeletal Radiol. 45(12), 1695–1703 (2016). https://doi.org/10.1007/s00256-016-2495-3

136

P. Krenn et al.

54. Jones, S., Caplan, N., Gibson, A.S.C., Kader, N., Kader, D.: Interactions between severity and location of chondral lesions and meniscal tears found at arthroscopy. Knee Surg. Sports Traumatol. Arthrosc. 19(10), 1699–1703 (2011). https://doi. org/10.1007/s00167-011-1470-8 55. Nakamura, N., Horibe, S., Toritsuka, Y., Mitsuoka, T., Natsu-ume, T., Yoneda, K., Hamada, M., Tanaka, Y., Boorman, R.S., Yoshikawa, H., et al.: The locationspecific healing response of damaged articular cartilage after ACL reconstruction: short-term follow-up. Knee Surg. Sports Traumatol. Arthrosc. 16(9), 843–848 (2008). https://doi.org/10.1007/s00167-008-0565-3 56. Tandogan, R.N., et al.: Analysis of meniscal and chondral lesions accompanying anterior cruciate ligament tears: relationship with age, time from injury, and level of sport. Knee Surg. Sports Traumatol. Arthrosc. 12(4), 262–270 (2004). https:// doi.org/10.1007/s00167-003-0398-z 57. Fok, A.W., Yau, W.: Delay in ACL reconstruction is associated with more severe and painful meniscal and chondral injuries. Knee Surg. Sports Traumatol. Arthrosc. 21(4), 928–933 (2013). https://doi.org/10.1007/s00167-012-2027-1 58. Franzone, J.M., Vitale, M.A., Stein, B.E.S., Ahmad, C.S.: Is there an association between chronicity of patellar instability and patellofemoral cartilage lesions? An arthroscopic assessment of chondral injury. J. Knee Surg. 25(05), 411–416 (2012) 59. Eugster, M., et al.: Quantitative evaluation of the thickness of the available manipulation volume inside the knee joint capsule for minimally invasive robotic unicondylar knee arthroplasty. IEEE Transactions on Biomedical Engineering (2020) 60. Quaresma, M.B., Portela, J., Soares do Brito, J.: Open versus arthroscopic surgery for diffuse tenosynovial giant-cell tumours of the knee: a systematic review. EFORT Open Rev. 5(6), 339–346 (2020) 61. Steinwachs, M.R., Waibl, B., Mumme, M.: Arthroscopic treatment of cartilage lesions with microfracture and BST-CarGel. Arthrosc. Tech. 3(3), e399–e402 (2014)

Nursing Robotics and Human Performance Evaluation

Investigating the First Robotic Nurses: Humanoid Robot Nightingale and Partners for COVID-19 Preventive Design Esyin Chew1(B)

, Pei Lee Lee2 , Jiaji Yang1 , and Shuyang Hu1

1 EUREKA Robotics Lab, Cardiff School of Technologies, Cardiff Metropolitan University,

Cardiff, UK [email protected] 2 HSC Hospital, Menara HSC, 5-1, 187, Jalan Ampang, Taman U Thant, 50450 Kuala Lumpur, Malaysia

Abstract. Responding to these global COVID-19 changes for daily healthcare services clinic, while maintaining safe social distancing, the paper reports the human-centred iterative design with real-fields feasibility inquiries to investigate the first robotic nurse and her partners in Wales. The research adapted the ancient Eastern human nature of seven emotions and six biological wills for the selection criteria and novel design principles for the care robots. We report the preliminary work for integrating, customising, implementing and evaluating three novel robotic nurses: Robot Nightingale, Robot Almeida and Robot Eureka in a care home and a hospital. Bionic Scenarios Definition with 5 merging principles are extracted from the Feasibility Inquiries 1–3. Limitations are discussed from the stakeholders’ experiences. Our research has no intension to replace human nurses, but a thoughtful feasibility and interdisciplinary study for bionic robotic nurses for conventional engineers’ and practitioners’ references.

1 Background of the First Robotic Nurse in Wales Highly infectious diseases such as COVID-19, H1N1 and MERS are major challenges to global health and significant risks to social-economic stability. Once an outbreak has occurred, the employment of Personal Protective Equipment (PPE) and disinfection measures to manage the epidemic becomes paramount. The World Health Organisation (WHO) calls for global researchers’ contribution to address these issues, including the best ways to protect health care, home care and hospitality workers [1]. In early 2020, the UK National Health Service (NHS) staff urged that they are being put at risk during the epidemic because of a lack of PPEs, and almost a year later the situation remain stressful, although the PPEs supplies are sufficient; nevertheless, “due to the inability to isolate patients, infections in staff and “asymptomatic carriage” [2, 3]. The pandemic has introduced massive challenges for daily healthcare services and clinical operations, while maintaining safe social distancing. Responding to these global challenges, various robots have enabled clinical and non-clinical care, swab sampling, medicine and healthcare tele-presence, medicines and meals deliveries to COVID-19 wards [4–8]. However, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 139–146, 2022. https://doi.org/10.1007/978-3-030-76147-9_15

140

E. Chew et al.

most literatures remain either conceptual design or trials with high cost robots, i.e. Pepper [9] or CRUZR [10] that many organisations could not afford for a large scale of deployment [5, 11–13]. To address this timely challenge, EUREKA Robotics Centre [14] and its global strategic collaborators aim to iteratively design-test-enhanced affordable and bionic robotic nurse. The paper reports the first iterative design with real-fields feasibility inquiries for the first robotic nurse and her partners in Wales.

2 Research Methods for Robot Nightingale and Partners 2.1 Human-Centred Design (HCD) and Interdisciplinary Methods There are conventional mechanical engineering-designed industrial robots [15]. However, some of these industrial robots have limited impact to social and public sector that is focusing in human care. Therefore, Social Robots and Medical Robots e increasingly gain attention from the research community, in oppose to the traditional industrial robotics. Buchanan asserts that HCD is about the experiences of human-being, to “expand the meaning of the term ‘product’ from a physical artifact to an engagement with human beings, the diverse branches of design become clearer, revealing the logical pattern” [16]. Grounding on Buchanan’s view, this is an interdisciplinary qualitative research with intercultural dialogue (care home, hospital, general practitioners and communitybased user experience), where the social sciences and anthropology have helped to change our robotics design from static mechanics to dynamic, bionic and culturally situated [16], aiming for HCD design towards both aesthetics and features enhancement. Themes inform the design methods with strategic impacts and those themes bridge the fields of HCD and robotics innovation: (1) Users’ problem and bionic solutions identification, (2) Bionic scenarios definition, (3) Goals: merging of extracted principles for bionic solution reframing, and (4) Themes of design principles for the next interactive deployment and enhancement [17]; as discussed in the following sections. 2.2 Users’ Problem and Bionic Solutions Identification Most people working in healthcare and social care perceive the industry as “high human touch”. There are criticism of “inhuman” to use robots in care services [18]. On the contrary, people do not simply see that technology can make healthcare work more efficient and subtract the manual repetitive efforts so that one has even more time for the human touch [19]. We define bionic robotic nurses as bio-inspired design to assist nurses in conducting repetitive tasks or challenging and high risks works so that the human’s strength, abilities, energy can be greatly enhanced. Mimicking the full taste of life, the HCD robotic nurses attempt to adapt the ancient human nature definition of seven emotions (joy, anger, sadness, fear, love, dislike, like) and six biological wills (live, death, eyes, ears, mouth, nose) [20, 21]. A HCD robotic nurse design shall embrace the above seven emotions and the below six biological wills for a better human-robot companionship [22]: 1. To live with polygenic capabilities (High EQ Intelligence): to recognise human multi-dimension emotion like gestures, face expression and motions, and never show impatience or angriness against patients or co-workers.

Investigating the First Robotic Nurses

141

2. To “fear of death” (Behavioral Decision-making Intelligence): maximised degree of freedom for flexi-behaviours, neural network motion control algorithms, obstacles avoidance and automated charging ability, paired with charging station. 3. To “see with robotic eyes” (Vision intelligence): motion captures, facial expressions and accurate object recognition, and respond with deep learning. 4. To “hear with robotic ears”: leading language engine with more microphone arrays to analyse sound, waves and voice sources, precisely tracking with 360° all direction pickup within safe distancing. 5. To interact with mouth (Speech Intelligence): multi-lingual languages library and emotional speech synthesis from the vision and listening. 6. To “smell with robotic noses”: multi-dimensional sensors for powerful nursing ability to perceives (collect data from) the surrounding world. After a year of robotic model review and comparative industrial visits, three robots are selected based on to the assessment criteria, i.e. affordability, bionic capabilities and HCD aesthetics (see Table 1). We decided to discharge other current robotic nurses such as Pepper [5], Cruz [10], Moxi [23], Cira-03[24] and Yumi [25] due to under-satisfying of the selection criteria above. The details evaluation and critical analysis and our novel and in-house developed technical bionic platforms and APIs with deep learning, are reported in a separate journal article under review. 2.3 Bionic Scenarios Definition This paper focuses in a continued work of blended-HCD case study for customising, implementing and evaluating three robotic nurses, namely: Robot Nightingale, name after the researchers visited the British Museum Florence Nightingale in London, memorising the first British nurse and the founder of modern nursing. To promote more “women in robotics”, Florence Nightingale is a healthcare pioneer, a female icon, and a leader whom we hope her legacy lives on today via our bionic designed of Robot Nightingale from 2020 [26], the Florence’s bicentenary (she was born on 12th May 1820). Partner 1: Robot Almeida [27], name after the British Virologist, June Almeida who first discovered coronaviruses but with little recognition [28]. Partner 2: Robot Eureka [27], name after the EUREKA Robotics Lab, as a service robot for enhanced Human-Robot-Interactions (HRI) and companionship [22].

142

E. Chew et al.

Table 1. The Capabilities, Goals and Processes for Robot Nightingale, Robot EUREKA and Robot Almeida HCD research Robots

Capabilities and Principles (Why)

Goals and Processes (When, Where and How)

Almeida [27]

Bionic little robot with flexi-movements and is associated with wireless healthcare meters, monitoring temperature, blood pressure/oxygen, weights; recording, analysing data with personalised response. Robot Almeida has a complete 7 emotions but limited 4 wishes (1, 3, 4 & 5) with 8 intelligent sensors

A low cost home-healthcare robot for telemedicine healthcare initial measurement meters via wireless communication Feasibility Inquiry 1: Qualitative purposeful sampling for doctors and nurses from two hospitals and General Practitioners (GP) through design and product demo (June – Aug 2020)

Eureka [27]

Bionic middle-size robotic nurse with 1 woofer, 2 cameras and one 3D sensors, 7 mics, 8 tactile sensors, gyroscope, 14 obstacles detections, and 2 human body induction with customized robotics apps for FAQ, easy personalised songs and dances, visitors and staff faces recognition and registration. A complete 7 emotions and 2 wishes (1, 3) but limited wishes 2 4, 5 and 6

A low cost hospital and care home service robot for well-being activities, remote patients/residents visits and medicine/food delivery Feasibility Inquiry 2: Qualitative purposeful sampling for field experience: a hospital and a care home for public engagement in 2020 summer, Christmas and 2021 new year with 7 staff and their residents /patients (June, Dec 2020 - Feb 2021)

Nightingale [26] Bionic humanoid robotic nurse move autonomously (or by remote control) in 360 degrees, perform emotion/age recognition with rich body languages (i.e. 25 emotion expressions, head turning, arms lifting, and maneuver astutely (or by remote control). Robot Nightingale speaks multilingual and able to communicate with Robots Almeida for initial healthcare check, with customisable robotics apps Adapting the ancient Eastern human nature of seven emotions and six biological wills and In the aesthetics appearance, this robot has a complete 7 emotions and 6 wishes: 25 kinds of emotion expressions and 368 kinds of emotion languages to makes HRI enjoyable

An affordable hospital and care home robot for HCD nursing care. and perform receptionist’ tasks, ward round with remote video observation and bi-direction free video talk through mobile phone control for suspected or confirmed-COVID-19 patients, guided well-being exercises, facilities tour, FAQs, telemedicine and customised interactivities Feasibility Inquiry 3: Qualitative expert sampling for professional group (1 director of nurse, 3 consultants, 2 general practices, and their patients through physical or video inquiries) to review Robot Nightingale’s capabilities: the robotic telemedicine, with remote video observation and bi-direction free video talk through mobile phone control for suspected-confirmed-COVID-19 patients. (Nov 2020 - Dec 2021)

Investigating the First Robotic Nurses

143

3 Constructive Results and Future Work We deployed and demonstrated robotic nurses who attempt to express the seven emotions and six biological wills with bionic eyes and gesture expressions, in respond to the vision capturing. All feedback from the stakeholders above are collected qualitatively via face-to-face focus group discussion, one-to-one inquiry, and in the form of written emails and WhatsApps. The feedback is identically positive from the directors, staff, patients and residents from two hospitals and one care home. The third and fourth HCD and interdisciplinary methods are the “Goals: merging of extracted principles for bionic solution reframing” and “Themes of design principles for the next interactive deployment” as follows, with five merging thematic principles extracted from the Feasibility Inquiries 1–3: (1) Confirming responses with user acceptance and “high human-touch”: from patients, residents and staff acceptance for the aesthetically friendly robotic nurse as assistants [29]: “…amazing, the residents have really enjoyed having her.” “kids would love it.” “The staff are in love and one of our residents has cried with joy listening to her play.” “The robot truly make the difference to our residents especially during this Christmas period, and we are extremely grateful!” “we would like to provide matched funding.” “We want to run a Farewell Party for Robot EUREKA (at the end of the inquiries)” (2) Affordance and cost-effectiveness: Insufficient nurses to provide personalised care for one patient in long hours. Robots can certainly help in sharing some heavy loads of human manpower: “I always have the issues for staffing, especially nurses on shift and during the COVID-19 self-isolation rules. Robotic nurses are certainly helpful not to take away their jobs, but to release them to perform core healthcare works. “The robot enables safe social distancing and tackle the COVID-19 infectious risks.” “Robot Almeida and Eureka are affordable and alerting when patients or visitors fall outside of optimal range of temperatures or blood pressure.” (3) Limitations and next iterative design of the robotic nurses: the size of the robot Almeida is too small for adults but suitable for toddlers. Its flexible movement and behavior caused some challenges in healthcare data check. The bi-lingual EnglishWelsh menu and Epidemic FAQs on Robot Almeida are interesting but its limited battery life is not practical. It is also inadequate for a fluent and free communication, especially when the network is poor – a better cloud speech engine with enhance deep learning and 5G is required. Eureka has the helpful function of HD projector for activities and distraction procedures, whereas Nightingale has the best bionic features but with no HD projector. Integrated or combined features are desirable. Both of them have automatic charging stations but only workable within the limited connectivity area.

144

E. Chew et al.

(4) Safeguard of patient data confidentiality and specialist equipment in the robots intervention, during the known healthcare data collection by Robot Almeida, and automatic data collection from the sensors on Robot Eureka and Robot Nightingale. To consider patients-equipment-contactless redesign for Robot Almeida, requiring further blueprint in clinical setting and safeguard of the equipment such as Robot Nightingale and Eureka. (5) Make a difference to people and organisations: Something different and specialty to engage patients and in addition, it can be considered as an enhanced corporate marketing value. “It is great to have to make something difference to my nurses and patients.” “Robots will make patients and nurses feel safe during the COVID-19 pandemic.” “Good marketing tool for a private university, to be identical.”

4 Conclusion Aiming to collaborate with global partners for investigating HCD healthcare robotics into the real fields, this paper reports the preliminary design and pilot of a long-term care research for global preventive design with three models of robotic nurses, and for widening participation and users’ acceptance. Before we move on too quickly to the question of whether or not healthcare sectors ought to take these robotic nurses, we should first examine whether or not they can do to assist human what conventional designs seem only too willing to assume. Our findings suggest that (1) bionic scenario definitions leading to the five initial thematic principles for further investigation; and (2) bio-inspired HCD design for robotic nurses is essential to safeguard human dignity and human rights. Will the rise of Robot Nightingale be an avenging angel, disruptive innovator or controversial catalyst for nursing ethics? A strong message from this research is that Robotic nurses have never intended to replace nurses, but a helping pair of hands, eyes, ears, mouth and nose, an extension of nurses with 7 emotions and 6 biological wills with increasing human-touch. This project established the feasibility of using robotics for COVID19 preventive design in multiple care environments, for epidemic prevention design. Acknowledgement. This work is funded by the 2020 Cardiff Met’s Get Started Grant and in the process of applying for the Welsh Government SMART Expertise 2021–2022. Special appreciation to our strategic robotics partners in the UK and China to provide hardware provision, and granted EUREKA Robotics Lab as the sole distributorship in the UK. Thanks to all the research participants from healthcare sectors who provided their honest experiences and constructive perceptions, contributing in the research and adventurous journey during COVID-19.

References 1. WHO: World experts and funders set priorities for COVID-19 research, World Health Organization (2020). https://bit.ly/39vRpjj 2. BBCa: Coronavirus: NHS staff ‘at risk’ over lack of protective gear (2020). https://www.bbc. co.uk/news/uk-51950276

Investigating the First Robotic Nurses

145

3. BBCb: Covid: ‘Significant’ number caught virus in Bedfordshire hospitals (2020). https:// www.bbc.co.uk/news/uk-england-beds-bucks-herts-55948920 4. Kimmig, R., Verheijen, R., Rudnicki, M. and SERGS Council: Robot assisted surgery during the COVID-19 pandemic, especially for gynecological cancer: a statement of the Society of European Robotic Gynaecological Surgery (SERGS) J. Gynecol. Oncol. 31(3), e59 (2020). https://doi.org/10.3802/jgo.2020.31.e59 5. Shen, Y., et al.: Robots under COVID-19 pandemic: a comprehensive survey. IEEE Access 9, 1590–1615 (2020) 6. Wang, S., Wang, K., Tang, R., Qiao, J., Liu, H., Hou, Z.-G.: Design of a low-cost miniature robot to assist the COVID-19 nasopharyngeal swab sampling. In: IEEE Transactions on Medical Robotics and Bionics. https://doi.org/10.1109/TMRB.2020.3036461 7. Cuthbertson, A.: Coronavirus: ‘Little Peanut’ Robot Delivers Food to People Inquarantine in China. Independent New (2020). https://bit.ly/39sxBgO 8. Cardona, M., Cortez, F., Palacios, A., Cerros, K.: Mobile robots application against Covid-19 pandemic. In: 2020 IEEE ANDESCON, Quito, Ecuador, pp. 1–5 (2020). https://doi.org/10. 1109/ANDESCON50619.2020.9272072 9. Cruzr: UBTECH Robotics (2021). https://www.ubtrobot.com/products/cruzr?ls=en 10. Pepper: Softbank (2021). https://www.softbankrobotics.com/emea/en/industries/healthcare 11. Yang, G.: Keep healthcare workers safe: application of teleoperated robot in isolation ward for COVID-19 prevention and control. Chin. J. Mech. Eng. 33(1), 1–4 (2020). https://doi.org/ 10.1186/s10033-020-00464-0 12. Rane, K.P.: Design and development of low cost humanoid robot with thermal temperature scanner for COVID-19 virus preliminary identification. Int. J. Adv. Trends Comput. Sci. Eng. 9(3) (2020). https://doi.org/10.30534/ijatcse/2020/153932020 13. Kaiser, M.S., Al Mamun, S., Mahmud, M., Tania, M.H.: Healthcare robots to combat COVID19. In: Santosh, K., Joshi, A. (eds.) COVID-19: Prediction, Decision-Making, and its Impacts. LNDECT, vol. 60, pp. 83–97. Springer, Singapore (2021). https://doi.org/10.1007/978-98115-9682-7_10 14. EUREKA: EUREKA Robotics Centre, one of the UK 11th Specialist Centre, nested at Cardiff School of Technologies, Wales (2021). www.cardiffmet.ac.uk/eureka 15. Robots: IEEE’s Guides to the World of Robotics (2021). https://robots.ieee.org/robots/ 16. Buchanan, R.: Human dignity and human rights: thoughts on the principles of human-centered design. MIT Press J. 17(3), 35–39 (2001). https://doi.org/10.1162/074793601750357178 17. Tan, N., Sun, Z., Elara, M.R., Nishann, B., et al.: A system-of-systems bio-inspired design process: conceptual design and physical prototype of a reconfigurable robot capable of multimodal locomotion. J. Front. Neurorobot. 13, 78 (2019) 18. Alberti, B.F.: Robots to be introduced in UK care homes to allay loneliness – that’s inhuman. The Conversation (2020). https://bit.ly/39pQErJ 19. Hourbon, G.: Building ventures to improve the quality of later life, Zinc (2021). https://bit. ly/3tx2okT 20. Nylan, M.: The Five “Confucian” Classics (Liji), Yale University Press (2001) 21. Legge, J.: Sacred Books of the East, vol. 27 and 28. Oxford University Press, Oxford (1885) 22. Yang, J., Chew, E., Liu, P.: Service humanoid robotics: review and design of a novel bioniccompanionship framework. In: Chew, E., et al. (eds.) RiTA 2020. LNME, pp. 185–194. Springer, Singapore (2021). https://doi.org/10.1007/978-981-16-4803-8_20 23. Moxi: Moxi (2021). Diligent Robotics Inc. https://diligentrobots.com/moxi 24. Cira-03 Robot: Egyptian inventor trials robot that can test for COVID-19, Swissinfo, Swiss Broadcasting Corporation (2021). https://www.swissinfo.ch/eng/reuters/egyptian-inventortrials-robot-that-can-test-for-covid-19/46185198 25. Engelking, C.: Meet YuMi: A Robot Nurse Built to Make the Rounds, Discover Magazine (2019). https://bit.ly/39tEIW0

146

E. Chew et al.

26. Robot Nightingale: EUREKA Robotics Lab, Cardiff School of Technologies (2020). https:// twitter.com/cardiffmet/status/1253306449742180353?s=20. https://twitter.com/EsyinChew/ status/1314263909369630722?s=20. https://twitter.com/EsyinChew/status/120737876395 4237440?s=20 27. Robot Almeida and Robot Eureka: EUREKA Robotics Lab, Cardiff School of Technologies (2020). https://twitter.com/eurekarobot/status/1275132365778694145?s=20. https://twitter.com/EsyinChew/status/1275181734619987968?s=20. https://twitter.com/eur ekarobot/status/1236416154266144769?s=20 28. Combs, S.: History and Culture: Coronavirus Coverage - June Almeida, National Geography (2020). https://on.natgeo.com/39tyUMj 29. EUREKA Twitter: Snippet of experience of Robotic nurses in a care sector (2021). https://twitter.com/eurekarobot/status/1352657275358425088?s=20. https://twitter.com/eur ekarobot/status/1275132365778694145?s=20

Impact of Ear Occlusion on In-Ear Sounds Generated by Intra-oral Behaviors Mohammad Khair Nahhas1(B) , Nicolas Gerig1 , Jens Christoph T¨ urp2 , 3 4 1 Philippe Cattin , Elisabeth Wilhelm , and Georg Rauter 1

BIROMED-Lab, Department of Biomedical Engineering, University of Basel, 4123 Allschwil, Switzerland [email protected] 2 UZB, Department of Oral Health and Medicine, University Center for Dental Medicine Basel, Mattenstrasse 40, 4058 Basel, Switzerland 3 CIAN, Department of Biomedical Engineering, University of Basel, 4123 Allschwil, Switzerland 4 Engineering and Technology Institute Groningen, Faculty of Science and Engineering, University of Groningen, Groningen, The Netherlands https://biromed.dbe.unibas.ch

Abstract. We conducted a case study with one volunteer and a recording setup to detect sounds induced by the actions: jaw clenching, tooth grinding, reading, eating, and drinking. The setup consisted of two in-ear microphones, where the left ear was semi-occluded with a commercially available earpiece and the right ear was occluded with a mouldable silicon ear piece. Investigations in the time and frequency domains demonstrated that for behaviors such as eating, tooth grinding, and reading, sounds could be recorded with both sensors. For jaw clenching, however, occluding the ear with a mouldable piece was necessary to enable its detection. This can be attributed to the fact that the mouldable ear piece sealed the ear canal and isolated it from the environment, resulting in a detectable change in pressure. In conclusion, our work suggests that detecting behaviors such as eating, grinding, reading with a semi-occluded ear is possible, whereas, behaviors such as clenching require the complete occlusion of the ear if the activity should be easily detectable. Nevertheless, the latter approach may limit real-world applicability because it hinders the hearing capabilities. Keywords: Acoustic signals Ear occlusion

1

· Physiological bio-markers · Hearables ·

Introduction

Our body produces sounds that can act as markers to detect, measure, or asses impaired or abnormal physiological processes. For instance, listening to the heart Elisabeth Wilhelm and Georg Rauter both authors contributed equally. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022  G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 147–154, 2022. https://doi.org/10.1007/978-3-030-76147-9_16

148

M. K. Nahhas et al.

beat for a relatively short period with a stethoscope is a well established noninvasive method to detect health problems [1]. Wearable devices allow new possibilities for health monitoring [2]. They allow the monitoring of various physiological bio-markers over a longer period. Different research groups used wearable devices to detect health problems with microphones: knee osteoarthritis [3] or irritable bowl syndrome [4]. Wearable devices that are worn in the ear or around the ear which are equipped with various sensors are called: Hearables [5]. They have been utilized by various research groups to monitor sound based bio-markers, for example, sounds made while eating have been monitored from the ear to detect dietary behaviors [6]. Also, breathing rate and heart beat rate monitoring via the earcanal have also been studied [7]. We are interested in detecting the parafunctional orofacial behavior: bruxism. It is defined as “masticatory muscle activities that occur during sleep (characterized as rhythmic or non-rhythmic) and wakefulness (characterized by repetitive or sustained tooth contact and/or by bracing or thrusting of the mandible)” [8]. From the literature, bruxism induces various health implications such as teeth wear, muscles hypertrophy, toothache, and various other complications [9]. Bruxism-induced sounds can be produced by continuous tooth grinding or jaw clenching. They can propagate across the head and reach the earcanal. Therefore, we are firstly, interested in developing a framework to detect sounds generated by various oral behaviors such as bruxism from the ear. One of the important requirements to detect such sounds is to reduce the environmental noise. Various approaches have been explored in the literature, either by adding a microphone directed towards the environment to pick up ambient noise that can be later deducted form the in-ear signal [10]. Alternatively, tightly occluding the ear may reduce environmental noise as well. The aim of this case study is to investigate the possibility to detect various intra-oral behaviors-induced sounds from one semi-occluded and one tightly occluded earcanals.

2 2.1

Methods Setup

The setup illustrated in Fig. 1 consisted of two bone conducting microphones (Sonion, Hoofddorp, Netherlands) to be worn in both ears. For the left ear (L), the microphone was mounted on a commercial solid earpiece positioning the microphone at the entrance of the ear canal. For the right ear (R), the microphone was placed inside the earcanal and was occluded using a mouldable silicon earplug. Thus, the occlusion of the right ear was tighter than that of the left ear. Each microphone was connected to a data acquisition device (Daq-L/R) (MCC, Bietigheim-Bissingen, Germany) that was connected to a PC via USB. The Daqs were set to acquire data at a sampling rate of 16 kHz with 18 bit resolution. In addition, a graphical user interface was developed in-house using the platform Unity (Unity Technologies, California, US) to act as a guide for the participant through the experiment. A push button was handed to the participant

Impact of Ear Occlusion on In-Ear Sounds Generated

149

Fig. 1. Experimental setup: graphical user interface (GUI), left ear sensor (L): semioccluded with a commercial solid earpiece, right ear sensor (R): occluded with a mouldable silicon earpiece, push button, left data acquisition unit (Daq-L) and right data acquisition unit (Daq-R), a PC to store, label, and filter the data.

to be pressed during certain periods. The Daqs provided the needed voltage for the microphones and the push button, one volt and five volts, respectively. Matlab 2019 (Mathworks, Massachusetts, US) was used to post-process the acquired data. 2.2

Protocol

This experiment was performed on one healthy volunteer (male, age = 40 years old), who agreed with the procedure of the study and data publication. The experiment consisted of 6 tasks that were performed in the following order: jaw clenching, tooth grinding, jaw clenching again, reading, eating, and water drinking. The participant was asked to sit in front of the computer screen for instructions through the experiment. The jaw clenching tasks consisted of four clenching periods five seconds each. The participant was asked to clench the jaw with maximum bearable pressure. The tooth grinding task consisted of four grinding periods ten seconds each. Ten seconds pauses were introduced between active periods during the clenching and grinding tasks. The reading task consisted of reciting the paragraph “The North Wind and the Sun” [11]. The eating task consisted of two periods, where the participant was asked to eat a cracker and a fruit. During the final task, the participant had to drink a glass of water. 2.3

Data Processing

The data from both microphones of the second clenching task were normalized with respect to the maximum data point then were passed first through a low pass-FIR-filter (300 Hz) followed by a high pass-FIR-filter (55 Hz). In addition, spectral subtraction was introduced to reduce the influence of the test environment noise to a minimum by taking the average of the first three seconds of one of the pause periods as a reference. Also, the spectral flux and the spectrograms

150

M. K. Nahhas et al.

were estimated for both sensors. The window size was 50 ms and the overlap was 50%. To quantify the difference between clenching and pause periods, the energy level for each period was estimated as follows: Ew =

Nw 

|xi |2

(1)

N 1  Ew N w=1

(2)

i=1

Ee =

where, Ew is the energy of a window, Nw is the window’s number of samples, xi is the acoustic amplitude at sample i, Ee is the energy of time period e, and N is the total number of windows within the specified period. The energy vector of each sensor was normalized with respect to its maximum value.

3

Results

Figure 2 illustrates the unfiltered outputs of both microphones for the whole experiment. Both sensors detected grinding, reading, and eating. The drinking task was only detectable from the occluded ear. In addition, both jaw clenching tasks were not visible from either outputs. Figure 3 illustrates the filtered sensors outputs for the second clenching task. The semi-occluded ear output does not show a difference between the clenching and pause periods, whereas, the occluded ear output shows that there is a detectable difference between the different clenching and pause periods.

Fig. 2. Unfiltered normalized acoustic signals of the complete experiment obtained from the semi-occluded left ear (Left) and occluded right ear (Right). The blue shaded areas depict the tasks, jaw clenching, tooth grinding, jaw clenching, reading, eating, and drinking. While the white sections are pause periods.

Impact of Ear Occlusion on In-Ear Sounds Generated

151

Fig. 3. Filtered acoustic signals of the second clenching task obtained from the semioccluded left ear (Left) and from the occluded right ear (Right). Where, P and C represent the sequence of pause and clenching episodes. Table 1. Energy level obtained from Eq. 2 for each period within the second clenching task. The energy vector of each sensor was normalized with respect to the sensor’s maximum value.

Additionally, the energy levels for the clenching and pausing periods were calculated using Eq. 2 and listed in Table 1. The differences in the energy between the clenching periods and the pausing periods were higher in the recordings obtained from the occluded ear compared to the semi-occluded ear. Figure 4 illustrates the spectrograms of the second clenching task filtered outputs. It illustrates that the power of the frequency range 55–300 Hz for the occluded ear was higher during the clenching periods compared to the pause periods. Whereas, the semi-occluded ear produced almost no visible difference. Finally, Fig. 5 illustrates the spectral flux of the filtered recordings. The amplitudes of the pause periods from the occluded ear were lower than that of the clenching periods. For the semi-occluded ear the amplitudes of the pause and the clenching periods were almost the same.

4

Discussion

The variation in the energy levels for the pause periods between the occluded ear and the semi-occluded ear listed in Table 1 could be related to the fact that the occluded ear was isolated from the environment, thus, reducing the noise level in the occluded earcanal. In addition, from the occluded ear, the

152

M. K. Nahhas et al.

Fig. 4. Spectrograms of the second clenching task after filtering obtained from the semioccluded left ear (Left) and from the occluded right ear (Right). The red rectangles define the clenching periods, whereas, pause periods are outside the rectangles. P and C represent the sequence of pause and clenching episodes.

Fig. 5. Normalized spectral flux for the second clenching task after filtering obtained from the semi-occluded ear (Left) and from the occluded right ear (Right). Where, P and C represent the sequence of pause and clenching episodes. The orange dots at the beginning and the end of certain clenching periods resemble the jumps that were observed in Fig. 3.

clenching periods have had higher energy than that of the pause periods. Three reasons could explain this observation. First, clenching activated the middle ear muscles which deformed the tympanic membrane, thus, changing the pressure inside the sealed earcanal. Also, clenching activates the masticatory muscles that generated sounds. These sounds could propagate to the earcanal via bone conduction. Another plausible reason could be the deformation of the earcanal walls due to clenching-induced movement inside the temporomandibular joint, thus, changing the pressure inside the earcanal. The jump at the beginning and the end of the clenching periods as illustrated in Fig. 3 could be attributed to click sounds being generated as the participant brought the upper and lower teeth together or was trying to separate them. The spectral flux illustrated in Fig. 5 supported the aforementioned explanation as similar jumps at the beginning and the end of certain clenching periods were observed. Also, tightly occluding the ear hindered the hearing capability of the volunteer. Thus, to detect behaviors with low acoustic energy such as jaw clenching

Impact of Ear Occlusion on In-Ear Sounds Generated

153

in real-world would be challenging. Accordingly, further investigations would be required to implement the optimal occluding approach to detect such low energy signals.

5

Conclusion

With this paper, we presented our experimental recording setup that successfully detected intra-oral behaviors such as jaw clenching, teeth grinding, eating, reading, and drinking. We also investigated the effects of either fully occluding or semi-occluding the ear on recording sounds produced by oral behavior. Behaviors such as eating, reading, and tooth grinding were detected from both sensors. However, detecting behaviors such as jaw clenching was not possible at the first glance from either sensors. Eliminating the ambient and electrical noise, revealed that the occluding the ear helped in the detection of jaw clenching. Nevertheless, an occluded ear hinders the user’s capabilities in hearing ambient sounds which might be safety relevant in daily life. Therefore, a transfer of the device and task identification from lab environment to the real-world requires further investigations to find the optimal occlusion concept.

Acknowledgement. We would like to express our gratitude for the generous financial support provided by the Werner Siemens Foundation. In addition, the author would like to thank Norbert Zenati member of CIAN at the Department of Biomedical Engineering, for his support with developing the graphical user interface.

References 1. Leng, S., Tan, R.S., Chai, K.T.C., Wang, C., Ghista, D., Zhong, L.: The electronic stethoscope. Biomed. Eng. Online 14, 66 (2015) 2. Seshadri, D.R., et al.: Wearable sensors for monitoring the physiological and biochemical profile of the athlete. NPJ Digital Med. 2, 72 (2019) 3. Schl¨ uter, D.K., et al.: Use of acoustic emission to identify novel candidate biomarkers for knee osteoarthritis (OA). PloS One 14, e0223711 (2019) 4. Du, X., Allwood, G., Webberley, K.M., Inderjeeth, A.-J., Osseiran, A., Marshall, B.J.: Noninvasive diagnosis of irritable bowel syndrome via bowel sound features: proof of concept. Clin. Transl. Gastroenterol. 10, e00017 (2019) 5. Mas`e, M., Micarelli, A., Strapazzon, G.: Hearables: new perspectives and pitfalls of in-ear devices for physiological monitoring. a scoping review. Front. Physiol. 11, 568886 (2020) 6. Amft, O., St¨ ager, M., Lukowicz, P., Tr¨ oster, G.: Analysis of chewing sounds for dietary monitoring. In: Beigl, M., Intille, S., Rekimoto, J., Tokuda, H. (eds.) UbiComp 2005. LNCS, vol. 3660, pp. 56–72. Springer, Heidelberg (2005). https://doi. org/10.1007/11551201 4 7. Martin, A., Voix, J.: In-ear audio wearable: measurement of heart and breathing rates for health and safety monitoring. IEEE Trans. Biomed. Eng. 65(6), 1256– 1263 (2018)

154

M. K. Nahhas et al.

8. Lobbezoo, F., et al.: International consensus on the assessment of bruxism: report of a work in progress. J. Oral Rehabil. 45(11), 837–844 (2018) 9. Lavigne, G., Khoury, S., Abe, S., Yamaguchi, T., Raphael, K.: Bruxism physiology and pathology: an overview for clinicians. J. Oral Rehabil. 35(7), 476–494 (2008) 10. P¨ aßler, S., Wolff, M., Fischer, W.-J.: Food intake monitoring: an acoustical approach to automated food intake activity detection and classification of consumed food. Physiol. Meas. 33, 1073–1093 (2012) 11. International Phonetic Association, & International Phonetic Association Staff.: Handbook of the International Phonetic Association: A Guide to the Use of the International Phonetic Alphabet. Cambridge University Press, Cambridge (1999)

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis of Movement Profiles Cinzia Amici1 , Valter Cappellini1 , Federica Ragni1(B) , Raffaele Formicola1 , Alberto Borboni1 , Barbara Piovanelli2 , Stefano Negrini3,4 , and Gabriele Candiani5 1 Department of Mechanical and Industrial Engineering, University of Brescia, Brescia, Italy

[email protected] 2 IRCCS Fondazione Don Carlo Gnocchi, Milan, Italy 3 Department of Biomedical, Surgical and Dental Sciences, University of Milan “La Statale”,

Milan, Italy 4 IRCCS Istituto Ortopedico Galeazzi, Milan, Italy 5 Institute for Electromagnetic Sensing of the Environment, CNR, Milan, Italy

Abstract. The importance of investigating trunk flexion-extension movements to assess the subject’s health state is well established in clinical practice. Several diagnostic tools are currently available, which mainly ground on subjective evaluations, or quantitative but invasive methods (e.g. radiography). Because of technological constraints, non-invasive instruments, like optoelectronic acquisition systems with passive skin optical markers, still provide data affected by not-negligible artefacts. Besides, an effective analysis should involve movements performed by the subject at a self-imposed velocity, introducing potential inter-and intra-subject variability in data. This paper presents the preliminary analysis of the movement profile of passive skin optical markers during a trunk flexion-extension task, based on a parametric identification process. According to the Asymmetric Gaussian Functions (AGFs) optimization strategy, an optimization procedure for the fitting of markers’ spatial displacement with different Gaussian functions is proposed and applied to a dataset of 29 healthy subjects, for 59 exercises. With the primary aim of investigating strength points and limits of single-and multi-components AGFs models as fitting functions, a descriptive statistical analysis is performed for both the methods on the fitting performance in different conditions, and for the single-component AGF case, on the estimated parametric coefficients as well.

1 Introduction The trunk movement plays an important role in the Activities of Daily Living (ADLs). In particular, it allows maintaining the postural equilibrium [1], compensating limb movements [2], elaborating anticipatory pattern movements [3], as well as improving endurance and isokinetic movement performances in sports and professional activities [4, 5]. Furthermore, the presence of abnormal trunk compensations and movement planning strategies represent a peculiar characteristic of different orthopedic and neurological pathologies [6–9]. For these reasons, different studies are present in literature to investigate the trunk movement [10, 11]. Specifically, since flexion-extension plays an important functional © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 155–163, 2022. https://doi.org/10.1007/978-3-030-76147-9_17

156

C. Amici et al.

role [2], its evaluation is fundamental in clinical practice for the correct assessment of the spine health state and diagnosis. Representing a standard for the measurement of kinematical and dynamical trunk performances, it is thus the most studied trunk movement. Several measurement methods have been applied to detect and measure movement performances, such as: isokinetic machines [12], Inertial Measurement Units (IMUs) [13], wearable devices [14] and the gold standard, i.e. optoelectronic systems [10]. Although different measurement systems are adopted, the kinematic of the trunk remains an open research field [10] with different potential applications, elaborating accurate, quantitative variables and range values to describe the trunk performances. For these reasons, in this work the flexion-extension movement of the trunk has been analyzed, employing an optoelectronic acquisition system for the measurement of this task in a population sample of 29 healthy subjects, for 59 movement repetitions, approaching the modelization. A synthetic modelization of the movement can be realized: i) with transformations [15] in different domains, such as Fourier Transformation in time, or Wavelets in timefrequency domain, ii) with interpolations, through piecewise functions [16] such as splines, trigonometric functions or other constructions, or iii) with interpolation of function series [17]. In particular, the first approach is well known in literature, as it allows a valid description of phenomena, but produces a considerable number of parameters, which needs to be synthetized with subsequent analyses. The second strategy is employed to describe accurately the phenomenon, but the set of output parameters is not easily associated to its physical meaning. The final option can produce a restricted set of parameters when the series is limited and thus can be easily connected to the physical meaning of the phenomenon. For this reason, in this work the third strategy has been adopted; furthermore, to the knowledge of the authors, this option was never applied in literature to the movement analysis of the trunk flexion-extension. Particularly, in this work, the suitability of Asymmetric Gaussian Functions (AGFs) as fitting functions for the analysis of trunk flexion-extension task is analyzed, comparing strength point and limits of single-and multi-components models in different conditions; this kind of approach is well adopted in other research fields [18–21], and its mathematics is well known and can drive to an easy treatment.

2 Methods 2.1 Acquisition Protocol For the study of the trunk flexion-extension movement, 29 adult healthy subjects (24 males, age 23.14 ± 1.43 years, height 1.73 ± 0.06 m, BMI 23.18 ± 3.00 kg/m2 ) were recruited. The study was approved by IRCCS Fondazione Don Carlo Gnocchi ethical committee; all procedures were conducted according to the Declaration of Helsinki, and the subjects gave their written informed consent to the study participation. The experimental setup was composed of an optoelectronic acquisition system with eight cameras and a set of 32 passive optical skin markers (hemispheric markers, diameter 5 mm), located by a trained operator on the landmark points corresponding to: i) the spinous process and transverse processes of the seventh cervical vertebra (C7), of the

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis

157

third, seventh and twelfth thoracic vertebrae (T3, T7, T12), and of the lumbar vertebrae from the first to the fifth (L1–L5), ii) the spinous process of the second sacrum vertebra (S2), iii) the antero-superior iliac spines (ASIS), and iv) the poster-superior iliac spines (PSIS). Starting from the orthostatic position, with arms at the sides and head straight forward, each subject was asked to perform a flexion of the trunk at a self-imposed, comfortable velocity, and to return to the starting position. The subjects were instructed to perform a continuous movement, i.e. avoiding as much as possible interruptions or hesitations once the movement has begun; 11 of the participants performed the flexion-extension task three times, 8 twice, and the remaining 10 once. 2.2 Theoretical Framework Among the possible functions suitable for the data fitting in an optimization process, Asymmetric Gaussian Functions (AGFs) were selected. Two models can be adopted for the fitting function definition, based on a single-or a multi-components approach. 2.2.1 Single-Component AGF In this model, the fitting function f (t) is defined as a single asymmetric gaussian function [18, 22]. The function, described in the Eqs. (1) and (2), is totally determined by the seven coefficients c1 , c2 , …c7 . f (t; c1 , c2 , . . . , c7 ) = c1 + c2 · g(t; c3 , c4 , . . . , c7 )

g(t; c3 , c4 , . . . , c7 ) =

⎧ ⎨

  c −t c7 − 3c e  6  ,t c 3 5 ⎩ − t−c c

e

4

≤ c3

(1)

(2)

, t > c3

2.2.2 Multi-components AGFs This model defines the fitting function f (t) as the linear combination of n asymmetric gaussian functions. As described in Eq. (3), given n components, the function is totally determined by (6·n + 1) coefficients.   f (t; c1 , c2 , . . . , c6n+1 ) = c1 + c6(n−1)+2 · g t; c6(n−1)+3 , . . . , c6(n−1)+7 (3) n

2.3 Data Analysis To allow a complete description of the movement, the vectorial contributes of the markers’ displacement along lateral, anterior and medial axes have been combined for each marker into the absolute position vector R. To make data comparable among subjects and among repetitions, for each acquisition a normalization factor has been defined as the mean distance between the quotes of the spinous process of C7 and of the center of

158

C. Amici et al.

gravity of the pelvis, identified by PSISs and ASISs makers, during the first second of sampling. All the R signals of the evaluated acquisition have been then normalized by this factor. After a first preliminary analysis of the markers’ motion profiles, the R signal generated by the spinous process of T7 has been identified as the most representative position profile of the task, suitable for the function study. For each acquisition, this selected signal underwent two independent optimization processes: i) a single-component AGF, and ii) a multi-components AGFs, with 3 and 4 Gaussian components. The cost function was defined as the absolute difference between signal to fit and fitting function, and the residual error was computed as the root mean square (rms) value of the cost function. As signal to fit, different cases were investigated, since considering the differential order, the nth derivative of the basic position profile, with n = 0, 1, 2, 3, was analyzed. Figure 1 depicts a block diagram of the optimization process.

Fig. 1. Block diagram of the pseudo-code for the optimization process, where sf represents the signal to fit, ff the fitting function, C the set of the coefficients, and res the residual error.

In order to assess the validity of the proposed approaches, a non-parametric statistical analysis has been performed considering the following quantities: values of the residual error of the fitting, number of iterations required to reach the optimum solution, number of employed Gaussian components, number of differential order. The parameters values were tested to investigate the statistical difference in the residual among the different combination of Gaussian components’ number nGauss and differential order nDiff ; Kruskal-Wallis test and the median check were performed on data coming both from single-and multi-components AGFs. Spearman correlation tests were conducted among Table 1. Correlation values convention. Values

0.00–0.19

0.20–0.39

0.40–0.59

0.60–0.79

0.80–1.00

Correlation

Very weak

Weak

Moderate

Strong

Very strong

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis

159

the seven coefficients c1 , c2 , … c7 coming from the single-component AGF, according to the scale in Table 1. Mann-Whitney-Wilcoxon test was performed, investigating the null hypothesis H0MWW of samples coming from populations with identical distributions, Kruskal-Wallis test was evaluated, and Spearman correlation tests were conducted between the two testing conditions, according to the scale in Table 1. A significance level at 10% was used.

3 Results Figure 2 collects an example of fitting results for an illustrative position profile; figure’s columns depict different cases of signal to fit (nDiff variability), whereas figure’s rows present different numbers of Gaussian components (nGauss variability). Focusing on the whole population dataset, the Kruskal-Wallis test performed on nGauss and nDiff supports the hypothesis that the analyzed data can be considered as a sample drown from a normally distributed population (Chi-squared = 71.381, p-value = 3.16e−16 , df = 2, and Chi-squared = 1225.1, p-value < 2.2e−16 , df = 3, respectively). Table 2 collects the results of the median check on nGauss and nDiff . Considering the single-component AGF optimization model, Table 3 depicts the Spearman correlation values obtained for the coefficients c1 , c2 , … c7 .

4 Discussion Although the trunk flexion-extension mainly acts along the subject’s sagittal plane, motion contributes along the medial axis should be also considered for a complete description of the movement, especially in clinical applications, for diagnostic purposes. For this reason, the evaluation of the R profiles has been preferred to the simplified analysis of the motion projected in the sagittal plane. For the data normalization, given the dimension of the adopted markers (diameter 5 mm) and the minimum accuracy of the optoelectronic acquisition system assured by the calibration process (maximum residual error ≤ 3 mm), a custom normalization factor, based on an acquisition-dependent anthropometric measurement of the spine, has been chosen instead of a generic subject’s high. Comparing single-and multi-components AGFs, the use of a single Gaussian function should be preferred where possible, since it requires the estimation of a reduced number of parameters in the optimization process, and the coefficients of the Gaussian function can be easily interpreted in light of the physical meaning of the signal. For instance, c6 and c7 , defining the width of left and right function half, provide information about the slope of the motion profile corresponding to the flexion and extension phases of the movement respectively. Their comparison can also allow an estimation of the subject’s ability to perform constant velocity movements, described by a profile with symmetric trends for flexion and extension phases [23]. Nevertheless, a single Gaussian function presents a limited flexibility (as number of degrees of freedom), which is inadequate when aiming to detect possible signal’s peaks or local maxima/minima. Since these features assume a crucial functional meaning for the clinician, the adoption of multi-components AGFs models become necessary for their analysis. On the other hand, increasing the number of

160

C. Amici et al.

Fig. 2. Examples of fitting results of the same position profile RT7 for different cases of signal to fit (table’s columns) and numbers of Gaussian components (table’s rows). For each combination, the upper subplot depicts the signal to fit (black), and the optimal fitted profile (red); in the lower subplot, the n Gaussian components g are depicted.

Gaussian components could generate overfitting phenomena, as the misleading position of the fourth component peak in the image of Fig. 2 with nGauss = 4, nDiff = 0 depicts. Besides, regardless of the number of adopted components, the piecewise definition of AGFs cannot assure the continuity of the resulting optimal profile, as the Gaussian components in the image of Fig. 2 with nGauss = 4, nDiff = 0 presents. Imposing the fitting at the nth derivative order of the function would guarantee the continuity of the

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis

161

Table 2. Values of the median check on residual error and iteration’s number for nGauss and nDiff . Residual error

Iterations’ number nDiff

nGauss

nDiff

0

1

2

3

0

1

0.022

0.041

0.209

2.601

1

83

33

34

60

3

0.003

0.015

0.154

2.017

3

352

159.5

117.5

150

4

0.003

0.021

0.133

2.137

4

521

252.5

172

164.5

nGauss

1

2

3

Table 3. Spearman correlation values across the coefficients c1 , c2 , … c7 for the singlecomponent AGF. The function to fit corresponds to the position signal. c1

c2

c3

c4

c5

c6

c7

c1

1.000

−0.176

0.344

0.210

0.051

0.232

0.187

c2

−0.176

1.000

−0.126

0.041

−0.385

0.070

−0.414

c3

0.344

−0.126

1.000

0.362

−0.070

0.530

0.158

c4

0. 210

0.041

0.362

1.000

0.247

0.437

−0.177

c5

0.051

−0.385

−0.070

0.247

1.000

−0.113

0.076

c6

0.232

0.070

0.530

0.437

−0.113

1.000

0.282

c7

0.187

−0.414

0.158

−0.177

0.076

0.282

1.000

motion law at nth -1 order, even if the results of the optimization process for our dataset demonstrated a rapidly degenerative trend of the fitting performance as nDiff increases. Likewise, the median check (Table 2) depicts significantly lower residual errors, i.e. better fitting performance, for null nDiff tests, and for multi-components AGFs, although with comparable values between 3 and 4 components cases. On the contrary, a significant difference can be detected comparing the number of iterations. Focusing on the single-component AGF optimization model, the analysis of the Spearman correlation values for the coefficients c1 , c2 , … c7 (Table 3) reveals at most a moderate positive correlation between c3 –c6 and c4 –c6 , and a moderate negative correlation between c2 –c7 supporting the validity of these coefficients for the function description.

5 Conclusion This paper presents and compares in different conditions the performance of singleand multi-components AGFs models for the definition of the fitting function in the optimization process of the motion profile of the trunk flexion-extension task. The singlecomponent model requires the estimation of a limited number of parameters and allows an easy physical interpretation of the obtained values but does not guarantee for a correct

162

C. Amici et al.

detection of all the features required for the correct assessment of the profile. The fitting residual error is reduced as the number of adopted Gaussian components increases, but the computational burden increases accordingly, introducing the risk for overfitting phenomena and misleading results interpretations. In conclusion, the modelization of the trunk flexion-extension profile through AGFs models presents good performances. These results could be useful in the future for further applications, like investigating the differences between the movement patterns of healthy and pathological subjects from a quantitative perspective. Acknowledgments. The authors warmly thank Stefano Piotti and Stefano Pedretti for the support in the data collection process.

References 1. Stapley, P.J., Pozzo, T., Cheron, G., Grishin, A.: Does the coordination between posture and movement during human whole-body reaching ensure center of mass stabilization? Exp. Brain Res. 129, 134–146 (1999). https://doi.org/10.1007/s002210050944 2. Massion, J.: Movement, posture and equilibrium: interaction and coordination. Prog. Neurobiol. 38, 35–56 (1992). https://doi.org/10.1016/0301-0082(92)90034-C 3. Pollet, J., et al.: A new method to detect differences in start behavioural conditions of anterior reaching activity. Gait Posture 74, 30 (2019). https://doi.org/10.1016/j.gaitpost.2019.07.487 4. Agopyan, A.: An analysis of movements with or without back bend of the trunk or large hip extension in 1st Juniors’ Rhythmic Gymnastics World Championship-2019. Is there injury risk for gymnasts? Int. J. Perform. Anal. Sport 21, 108–125 (2020). https://doi.org/10.1080/ 24748668.2020.1850038 5. Mrozek, A., et al.: Assessment of the functional movement screen test with the use of motion capture system by the example of trunk stability push-up exercise among adolescent female football players. Vibr. Phys. Syst. 31, 1–10 (2020) 6. Nikfekr, E., Kerr, K., Attfield, S., Playford, E.D.: Trunk movement in Parkinson’s disease during rising from seated position. Mov. Disord. 17, 274–282 (2002). https://doi.org/10.1002/ mds.10073 7. Noamani, A., Lemay, J.F., Musselman, K.E., Rouhani, H.: Postural control strategy after incomplete spinal cord injury: effect of sensory inputs on trunk-leg movement coordination. J. Neuroeng. Rehabil. 17, 1–12 (2020). https://doi.org/10.1186/s12984-020-00775-2 8. Seo, J.W., et al.: Principal characteristics of affected and unaffected side trunk movement and gait event parameters during hemiplegic stroke gait with IMU sensor. Sensors (Switzerland) 20, 1–10 (2020). https://doi.org/10.3390/s20247338 9. Mañago, M.M., Kline, P.W., Alvarez, E., Christiansen, C.L.: Trunk and pelvis movement compensation in people with multiple sclerosis: relationships to muscle function and gait performance outcomes. Gait Posture 78, 48–53 (2020). https://doi.org/10.1016/j.gaitpost.2020. 03.006 10. Negrini, S., et al.: Trunk motion analysis: a systematic review from a clinical and methodological perspective. Eur. J. Phys. Rehabil. Med. 52, 583–592 (2016) 11. Asgari, M., et al.: Trunk dynamic stability assessment for individuals with and without nonspecific low back pain during repetitive movement. Hum. Factors J. Hum. Factors Ergon. Soc. 1–14 (2020). https://doi.org/10.1177/0018720820939697

Trunk Flexion-Extension in Healthy Subjects: Preliminary Analysis

163

12. Amici, C., Ragni, F., Ghidoni, M., Fausti, D., Bissolotti, L., Tiboni, M.: Multi-sensor validation approach of an end-effector-based robot for the rehabilitation of the upper and lower limb. Electronics 9, 1751 (2020). https://doi.org/10.3390/electronics9111751 13. Negrini, S., et al.: Use of wearable inertial sensor in the assessment of timed-up-and-go test: influence of device placement on temporal variable estimation. In: Perego, P., Andreoni, G., Rizzo, G. (eds.) MobiHealth 2016. LNICSSITE, vol. 192, pp. 310–317. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-58877-3_40 14. Archetti, L., Ragni, F., Roby-Brami, A., Saint-Bauzel, L., Amici, C.: Inclusive human intention prediction with wearable sensors: machine learning techniques for the reaching task use case. In: Proceedings of 7th International Electronic Conference on Sensors and Applications, p. 8234. MDPI, Basel (2020) 15. Dehzangi, O., Taherisadr, M.: Human gait identification using two dimensional multiresolution analysis. Smart Heal. 19, 100167 (2021). https://doi.org/10.1016/j.smhl.2020. 100167 16. Matthew, R.P., Seko, S., Bailey, J., Bajcsy, R., Lotz, J.: Simple spline representation for identifying sit-to-stand strategies. In: 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 4097–4103. IEEE (2019) 17. Le Folgoc, L., Delingette, H., Criminisi, A., Ayache, N.: Sparse Bayesian registration of medical images for self-tuning of parameters and spatially adaptive parametrization of displacements. Med. Image Anal. 36, 79–97 (2017). https://doi.org/10.1016/j.media.2016. 09.008 18. Jönsson, P., Eklundh, L.: TIMESAT - a program for analyzing time-series of satellite sensor data. Comput. Geosci. 30, 833–845 (2004). https://doi.org/10.1016/j.cageo.2004.05.006 19. Tan, B., et al.: Modis vegetation phenology metrics estimated with an enhanced TIMESAT algorithm. J. Sel. Top. Appl. Earth Obs. Remote Sens. 4, 4 (2011) 20. Pompilio, L., Pedrazzi, G., Sgavetti, M., Cloutis, E.A., Craig, M.A., Roush, T.L.: Exponential Gaussian approach for spectral modeling: the EGO algorithm I. Band saturation. Icarus 201, 781–794 (2009). https://doi.org/10.1016/j.icarus.2009.01.022 21. Pompilio, L., Pedrazzi, G., Cloutis, E.A., Craig, M.A., Roush, T.L.: Exponential Gaussian approach for spectral modelling: the EGO algorithm II. Band asymmetry. Icarus 208, 811–823 (2010). https://doi.org/10.1016/j.icarus.2010.03.020 22. Jönsson, P., Eklundh, L.: Seasonality extraction by function fitting to time-series of satellite sensor data. IEEE Trans. Geosci. Remote Sens. 40, 1824–1832 (2002). https://doi.org/10. 1109/TGRS.2002.802519 23. Negrini, S., Piovanelli, B., Amici, C., Donzelli, S., Zaina, F.: Identification through movement analysis of chronic low back pain pathological spinal movements patterns and their sensibility to change during exercise treatment. In: ISSLS Annual Meeting, Kyoto (2019)

Author Index

A Abbasi, Hamed, 89 Alves, Thiago, 3 Amici, Cinzia, 155 B Bayhaqi, Yakub A., 89 Bennour, Sami, 27 Borboni, Alberto, 155 C Calafeteanu, Dan, 49 Canbaz, Ferda, 89 Candiani, Gabriele, 155 Cappellini, Valter, 155 Carbone, Giuseppe, 3, 19, 38 Cattin, Philippe, 147 Cattin, Philippe C., 89, 97, 115 Chaker, Abdelbadiâ, 27 Chew, Esyin, 139 Copilus, i, Cristian, 70 D Ding, Xilung, 58 Dubrovin, Gregory, 19 Dumitru, Sorin, 70 Dunitru, Ilie, 49 Duverney, Cédric, 97 E El Bahi, Mohamed Ali, 97 Ennaiem, Ferdaws, 27 Eugster, Manuela, 125

F Faludi, Balázs, 115 Ferraresi, Carlo, 78 Formicola, Raffaele, 155 Friederich, Niklaus F., 125 Friedrich, Gerrit R., 107 G Geonea, Ionut, 70 Georgescu, Marius, 49 Gerig, Nicolas, 97, 147 Gherman, Bogdan, 38 Golli, Hanen El, 27 Gonçalves, Rogério Sales, 3 H Hu, Shuyang, 139 K Krenn, Philipp, 125 L Laribi, Med Amine, 27 Lee, Pei Lee, 139 Lo Piccolo, Mattia Vincenzo, 78 M Mlika, Abdelfattah, 27 Mohan, Santhakumar, 19 Muscolo, Giovanni Gerardo, 78 N Nahhas, Mohammad Khair, 147 Negrini, Stefano, 155

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 G. Rauter et al. (Eds.): MESROB 2021, MMS 106, pp. 165–166, 2022. https://doi.org/10.1007/978-3-030-76147-9

166 Nelson, Carl A., 12 Nozdracheva, Anna, 19 P Petcu, Alin, 49 Piovanelli, Barbara, 155 Pisla, Doina, 38 Pott, Peter P., 107 R Ragni, Federica, 155 Rauter, Georg, 97, 115, 125, 147 Romdhane, Lotfi, 27 Ros, ca, Adrian Sorin, 70 Rybak, Larisa, 19 S Sandoval, Juan, 27 Schäfer, Max B., 107 Su, Yujie, 58

Author Index T Tarnita, Daniela, 49 Tarnita, Danut Nicolae, 49 Tucan, Paul, 38 Türp, Jens Christoph, 147 V Vaida, Calin, 38 Voloshkin, Artem, 19 W Wilhelm, Elisabeth, 147 Y Yang, Jiaji, 139 Z Zam, Azhar, 89 Zeghloul, Saïd, 27 ˙ Zelechowski, Marek, 115 Zhang, Wuxiang, 58 Zoller, Esther I., 125