Development of the Neuromechanics Evaluation Device (NED) for Subject-Specifc Lower Limb Modelling of Spinal Cord Injury

Imperial College of Science, Technology and Medicine Department of Bioengineering

Development of the neuromechanics evaluation device (NED) for subject-specifc lower limb modelling of spinal cord injury

Hsien-Yung, Huang

Submitted in part fulflment of the requirements for the degree of Doctor of Philosophy in Bioengineering of Imperial College London and the Diploma of Imperial College London, October 2018

October, 2018

This is to certify that the work in this thesis has been carried out at the Department of Bio- engineering, Imperial College of Science, Technology and Medicine and has not been previously submitted to any other university or technical institution for a degree or award. The thesis comprises only my original work, except where due acknowledgement is made in the text.

The copyright of this thesis rests with the author and is made available under a Creative Commons Attribution Non-Commercial No Derivatives licence. Researchers are free to copy, distribute or transmit the thesis on the condition that they attribute it, that they do not use it for commercial purposes and that they do not alter, transform or build upon it. For any reuse or redistribution, researchers must make clear to others the licence terms of this work.

2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

3 Abstract

Investigating human neuromechanics can be used to characterise motor impairments in neurologically afected individuals, and can give insight into how the brain controls movement. In particular, the lower limbs’ neuromechanics is critical to assess the balance control and mobility. However, there are still few experimental results on the neuromechanics of the lower limb. This is especially true of the hip, in part due to the difculty to quickly accelerate the heavy leg in a controlled manner. Furthermore, existing robotic interfaces for the lower limb typically constrain the joints motion and cannot provide the quick and smooth perturbations necessary to identify subject’s biomechanics. In this context, this thesis presents: i) a new robotic interface that have developed to measure lower-limb neuromechanics, ii) a systematic investigation of the hip viscoelasticity carried out with this interface, and iii) a biomechanical model that provides subject-specifc dynamic behaviour based on parameters from neuromechanics experiments.

The Neuromechanics Evaluation Device (NED) is an endpoint-based cable-driven robotic interface that can be used to measure the hip, knee and ankle joint neuromechanics. Subjects can take an upright posture, which is important for subjects with weak motor condition, while the interface moves the limb’s extremity over a large workspace without constraining the joint orientation. Rapid position displacements can be applied to the studied limb by the powerful actuator fxed to the ground. Mechanical evaluations showed that NED has a rigidity above 500N/m and viscosity below 50Ns/m. It is also able to produce fast perturbations (e.g. a displacement of 2cm within 230ms) without vibration, which can be used to identify lower-limb neuromechanics.

The ability of NED to carry out various neuromechanics measurements is illustrated in two experiments carried out with healthy subjects. First, a measurement of the maximum voluntary torque at the hip joint yielded values in line with reported estimates in the literature. Second, a systematic investigation of the hip joint viscoelasticity was carried out with 10 subjects. In line with previous fndings on the upper and lower limbs, hip stifness was found to monotonically increase with the applied force, with a slight dependence on the hip angle. These experiments exemplify how NED can be efciently used to characterise the lower-limb joint neuromechanics.

The thesis further presents a model of lower limb neuromechanics integrating subject-specifc parameters that can be identifed through experiments with a robotic interface. The proposed

4 model incorporates physiological parameters such as the torque-angle and the torque-angular velocity dependencies, as well as the joint viscoelasticity. The model was used to evaluate typical neuromechanics alterations caused by a spinal cord injury. In addition, the infuence of various neuromechanical joint parameters to postural control was evaluated in simulations.

5 6 Acknowledgements

I would like to thank my supervisor Professor Etienne Burdet who provided me with the chance to join the Symbitron project and develop the big robot NED. I would also like to thank my second supervisors Ildar, Arash and Andrew, who have always been supportive during the project and provides essential suggestions. The discussions and brainstormings we had made great changes to my future and I really cherish this experience.

Human Robotics Group has always been a big family to me. For all those who have been in this family: Alessandro, Alfredo, Atsushi, Audrey, Carlo, Consuelo, Elif, Elisabeth, Eugenie, Francesca, Franck, Gero, Ildar, Jin, Jonathan, Martina, Matjaz, Mike, Milena, Moritz, Nuria, Paul, Paulo, Pierre-Jean, Sarah, Sean, Sharah, Shou-han, Sofa, Xiaodong, Yanan. Thank you for the time we had together.

Additionally, I would like to thank Professor Dario Farina, to share the space with me to develop NED. Also, the Neuromechanics and Rehabilitation Technology Group had shared me light while we are all in a lab located down in the basement, without a window. For all those who have been in this lab: Alessandro, Andrea, Carina, Christos, Corrado, Deren, Emanuele, Emiliano, Emma, Federico, Giulia, Giuseppe, Gonthicha, Guglielmo, Irene, Ivan, Margherita, Mario, Markus, Markus, Martyna, Matteo, Sigrid, Silvia, Simone. The time we had together is memorable.

Many thanks to all my friends. Your passions and energy made me felt warm and welcomed in a foreign land called London. It is all of you who reminds me who I am and what should I become.

Most important of all, I would like to thank my family and my beloved Minmin for all the supports during my quest towards my PhD. Nothing could have happened without the support and guidance you gave me. I cannot accomplish until this stage without you all. This thesis is dedicated to you.

7 Acronym

BF Bicep Femoris

CAD Computer aided design

CNS Central nervous system

CPG Central pattern generators

DOF Degree of freedom

EMG Electromyography

HW model Hammerstein-Wiener model

MVC Maximum voluntary contraction

MVJT Maximum voluntary joint torque

MVIC maximal voluntary isometric contractions

NED Neuromechanics Evaluation Device

NRMSE Normalized root mean square error

PC Personal computer

PIC controller Proportionalintegralderivative controller

RF Rectus Femoris

ROM Range of motion

SCI Spinal cord injury

TA Tibialis Anterior

8 Contents

Abstract 4

Acknowledgements 7

Acronyms 7

1 Introduction 19

2 A lower limb Neuromechanics Evaluation Device (NED) 23

2.1 Overview ...... 23

2.2 Introduction ...... 24

2.2.1 Existing neuromechanics estimation devices ...... 24

2.2.2 Functional requirements ...... 26

2.3 Device design ...... 27

2.3.1 General description ...... 27

2.3.2 Cable transmission ...... 29

2.3.3 Control system ...... 31

2.3.4 Safety measure and ergonomics ...... 32

9 10 CONTENTS

2.4 System characterisation ...... 33

2.4.1 Kinematics and sensitivity analysis ...... 33

2.4.2 Spatial and temporal dependency of cable tension ...... 35

2.4.3 Cable temporal dependency ...... 37

2.4.4 Cable system modelling ...... 38

2.4.5 Cable Nonlinearities ...... 40

2.5 Validation ...... 41

2.5.1 Dummy leg mechanics ...... 41

2.5.2 Stifness estimation ...... 43

2.5.3 Optimal position perturbation to identify stifness ...... 44

2.6 Discussion ...... 47

3 Hip joint neuromechanics evaluation with cable driven robot NED 49

3.1 Maximal voluntary isometric contractions (MVIC) ...... 49

3.1.1 Experiment protocol ...... 50

3.1.2 Results ...... 52

3.1.3 Discussion ...... 54

3.2 The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity ...... 55

3.2.1 Literature ...... 55

3.2.2 Methods ...... 57

3.2.3 Results ...... 62 CONTENTS 11

3.2.4 Discussion ...... 68

4 Subject-specifc modelling and evaluation 71

4.1 Literature review ...... 72

4.1.1 Physiological alterations following spinal cord injury ...... 72

4.1.2 Hill’s type muscle model ...... 76

4.1.3 Inverted pendulum model ...... 78

4.2 Subject-specifc modelling ...... 80

4.2.1 Model description ...... 80

4.2.2 Method to identify the model’s parameters ...... 82

4.3 Model evaluation using a single-joint inverted pendulum simulation ...... 84

4.3.1 Model parameter ...... 84

4.3.2 Scenario description and evaluation methods ...... 86

4.3.3 Human balance simulation with subject-specifc modelling ...... 90

4.3.4 Parameter sensitivity analysis and robustness to muscle noise ...... 93

4.3.5 Balance experiment with LOPES ...... 100

4.3.6 Discussion ...... 100

5 Conclusion 102

Bibliography 104

A Motor selection 115 B Ethical approval 117

12 List of Tables

2.1 Characteristics of existing lower limb neuromechanics evaluation devices . . . . . 25

2.2 Past list ...... 30

3.1 Biographical information of the subject in MVIC experiment ...... 51

3.2 Biographical information of the subjects in hip joint impedance measurements . 59

3.3 Statistics of linear regression and mixed efect models ...... 69

4.1 Values of the model’s parameters and relations ...... 86

13 14 List of Figures

2.1 Description of the Neuromechanics Evaluation Device (NED) ...... 28

2.2 The control system of NED ...... 32

2.3 Laser safety system ...... 33

2.4 Of-plane motion induced error ...... 34

2.5 Cable sagging ...... 36

2.6 Temporal characteristics of NED ...... 38

2.7 Identifcation of NED as a linear second order system ...... 39

2.8 Identifcation of NED as a linear second order system ...... 40

2.9 Identifcation of the mechanics of a 18kg dummy leg ...... 42

2.10 Stifness identifcation of a spring with known elasticity using NED ...... 44

2.11 Estimation result of a pilot study with one subject ...... 46

3.1 EMG process and sample measurement of the maximal voluntary isometric contractions experiment ...... 52

3.2 Torque-angle relation of hip joint fexion and extension motion ...... 53

3.3 Sketch of Neuromechanics Evaluation Device (NED) and perturbation profle used to estimate the hip viscoelasticity ...... 58

15 3.4 Hip stifness results for all subjects and conditions ...... 60

3.5 Hip stifness measurement depends on the perturbation direction...... 63

3.6 Violin plots showing the probability density of force-level dependency and how it changes due to hip angle ...... 65

3.7 Model prediction accuracy comparison ...... 67

4.1 Central nervous system ...... 74

4.2 Hill’s muscle-tendon model ...... 76

4.3 Simplifed forward dynamics for human motion ...... 76

4.4 Inverted pendulum model ...... 79

4.5 Sketch and block diagram of the single-joint subject-specifc modelling ...... 80

4.6 Normalised torque-angular velocity and torque-angle dependency of both able- bodied and SCI subjects...... 85

4.7 Stability margin description ...... 87

4.8 Balance experiment with LOPES ...... 89

4.9 Time series data of both the perturbation experiment conducted with LOPES and model simulated output...... 91

4.10 Simulation results of balance maintaining task...... 92

4.11 Specifcation used for stability evaluation ...... 94

4.12 Single parameter sensitivity analysis of the modelling approach ...... 95

4.13 Multi-parameter sensitivity analysis ...... 97

4.14 Noise resulted in fuctuation on multi-parameter sensitivity analysis ...... 98

16 4.15 Comparison between subject-specifc modelling and simple inverted pendulum model ...... 99

17 18 Chapter 1

Introduction

While standing on solid tarmacadam or unstable foating wood panels, one uses diferent standing strategies to safely interact with the environment. Depending on the terrain, one can frmly place one’s leg upon rigid obstacles or carefully modulate the ankle rigidity to absorb the oscillatory motion of unstable objects. This ability to modulate the body dynamics in order to interact with the environment is achieved by adjusting the limb viscoelasticity. This visco-elasticity can be observed for example when the limb is perturbed from an unexpected disturbance, where it tends to return to the planned movement. The human nervous system can modify the muscles’ stifness and viscosity, and by combining several muscles it can control joint viscoelasticity and the mechanical interaction with the environment. Neurological diseases and injuries can afect the motor functions, and investigating patients’ joint viscoelasticity is often one of the most efcient ways to characterise their condition and physiology.

Human tends to use similar patterns of muscle activation without thinking, e.g. during level- ground walking. As is customary in fnding commonalities among human viscoelasticity aug- mentation, identifying contrast behaviour among diferent populations can thus be used to analyse specifc neurophysiological characteristics. For instance, human mainly uses ankle muscles to regulate body motion when experiencing minor or none external force perturbation (Winter, 1995; Winter et al., 1998). By comparing the balance recovery in old vs. young populations, Engelhart et al. (2016) found a higher ankle stifness in older individuals. As other measurable

19 20 Chapter 1. Introduction ankle characteristics were similar, one can conclude that ligament tissue properties may be stifer in elderly participants without needing an invasive inspection.

Investigating joint viscoelasticity can also provide valuable clinical insight and potentially lead to better health care. For example, Lorentzen et al. (2010) has recruited three types of spasticity afected patients (stroke, multiple sclerosis and spinal cord injury) and measured their ankle joint viscoelasticity. As spasticity is a pathological condition which causes uncontrolled excessive muscle contraction when the muscles are stretched vastly, it is expected to observe an increase in joint stifness or viscosity when the limbs are moved rapidly. By comparing the estimated stifness values with clinical diagnosis of spasticity, Lorentzen et al. (2010) found that up to 32% of patients were misdiagnosed as spastic patients and provided with the wrong medical treatment. Therefore, it is suggested that clinical diagnosis should be refned to provide accurate treatment. This example shows the potential in neuromechanics investigations among pathologically afected subjects.

Commonly, stifness and viscosity can be estimated by applying a mechanical disturbance on the limb and observing the resulting change of position and force. The disturbance can be provided in various forms including isokinetic limb motions (Lorentzen et al., 2010), multi-joint torque perturbations (Koopman et al., 2016) or position pulses (Sinkjaer and Magnussen, 1994; Mirbagheri et al., 2000; Burdet et al., 2000). A large body of experiments has estimated stifness and viscosity in the ankle (Sinkj´eret al., 1993; Kearney et al., 1997; Mirbagheri et al., 2000; Lorentzen et al., 2010) and knee joint (Ludvig et al., 2017). However, there is scarce knowledge of hip joint neuromechanics, perhaps, due to the difculty of carrying out suitable experiments with the heavy leg mass. Existing robotic interfaces to estimate viscoelasticity in the lower limb either require a sitting or lying position (Amankwah et al., 2004; Sinclair et al., 2006; Perell et al., 1996; Akman et al., 1999), or are not sufciently rigid to apply fast perturbations without causing non-negligible oscillations (L¨unenburger et al., 2005; Koopman et al., 2016; Meuleman et al., 2016; Farkhatdinov et al., 2019). In addition, all of these interfaces are afxed to the body and thus determine the joints around which the limb can move, while the joints’ anatomical axis generally varies with the posture (e.g. knee joint rotate and translate during locomotion). 21

In view of the limitations in previous devices to investigate the lower-limb neuromechanics, one goal of this thesis is to develop a dedicated robotic interface to investigate the hip, knee and ankle joints. This rigid interface should be able to provide controllable and consistent mechanical disturbance on the limb, while changes of position and force can be measured to estimate joint viscoelasticity. In comparison to the aforementioned interfaces, the device should not impose kinematic constraints to the leg during identifcations. Additionally, the participant should be able to naturally move the leg while in an upright posture as body dynamics may be diferent during lying or sitting. The description of the developed neuromechanics evaluation device (NED) is described in Chapter 2 along with mechanical evaluations and experimental validations.

After showing the potential in conducting neuromechanics estimations in Chapter 2, we would like to use NED to investigate how hip joint viscoelasticity changes due to factors such as: joint angle, limb force level, perturbation direction and in the dominant and non-dominant legs. Experiments carried out on the upper limb, knee and ankle suggested that joint viscoelasticity increases with the applied force level (Burdet et al., 2013; Mirbagheri et al., 2001; Dirnberger et al., 2012b). In Chapter 3 the factors afecting hip stifness are systematically investigated, along with an additional experiment involving the maximum voluntary limb contraction.

Spinal cord injured individuals exhibit neuromusculoskeletal deterioration (e.g. muscle atrophy (Shields, 2002; Dudley-Javoroski and Shields, 2008) or muscle type changes (Lotta et al., 1991; McDonald et al., 2005)) afecting the motor functions. It is a challenging task to model SCI patients’ locomotion due to the diversity and variability of neuromusculoskeletal deterioration (Sinclair et al., 2006). Actuated lower-limb exoskeletons are efective tools to regain patients’ walking ability (Yan et al., 2015; Marchal-Crespo and Reinkensmeyer, 2009), however, they do not consider the individual’s neurophysiological condition (Talaty et al., 2013; Kawamoto et al., 2010). The lack of knowledge of participants’ neurophysiology forced the exoskeleton control strategy to conduct additional safety limits without using the residual motor functions of the patient.

Device such as NED enables one to investigate the limb viscoelasticity and develop subject- 22 Chapter 1. Introduction specifc models of the lower-limb neuromechanics. Chapter 4 develops such a model for SCI which allows quantifying patients’ residual motor capacity and provides suggestions for exoskeleton control. Importantly, all model parameters are measurable through simple non-invasive experiments that can be conducted with NED or other devices. A series of standing simulations are carried out based on literature data and measurements to examine the balance recovery behaviour of the model under diferent motion scenarios (force perturbation and body tilting). Chapter 2

A lower limb Neuromechanics Evaluation Device (NED)

2.1 Overview

This chapter presents a versatile cable-driven robotic interface to investigate the single-joint joint neuromechanics of the hip, knee and ankle. This endpoint-based interface ofers highly dynamic interaction and accurate position control, as is typically required for neuromechanics identifcation. It can be used with the subject upright, corresponding to natural posture during walking or standing, and does not impose kinematic constraints on a joint, in contrast to existing interfaces. Mechanical evaluations demonstrated that the interface yields a rigidity above 500N/m with low viscosity. Tests with a rigid dummy leg and linear springs show that it can identify the mechanical impedance of a limb accurately. A smooth perturbation is developed and tested with a human subject, which can be used to estimate the hip neuromechanics.

Section 2.2.1 reviews existing devices and compares their features and limitations, which are addressed in a new system NED as described in Section 2.2.2. The system’s design and description are given in Section 2.3. The system’s characteristics are detailed in Section 2.4 and a preliminary investigation of the mechanical properties is given in Section 2.5. Further neuromechanics measurements and investigations are documented in Chapter 3.

23 24 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

2.2 Introduction

An accurate characterisation of lower limb neuromechanics is required to understand lower limb neurophysiology and to design appropriate control for a robotic walking aid. To identify the limbs neuromechanics, a rigid robotic interface equipped with powerful actuators is typically required to apply force/position disturbances while sensors record the resulting modifcation of position/force. Available robotic interfaces to identify the lower limb neuromechanics include motor-driven dynamometers (Sinclair et al., 2006; Hahn et al., 2011; Amankwah et al., 2004; Dirnberger et al., 2012b; de Araujo Ribeiro Alvares et al., 2015; Valovich-mcLeod et al., 2004) and gait rehabilitation exoskeleton devices (L¨unenburger et al., 2005; Koopman et al., 2016), whose characteristics are compared in Table 2.1.

2.2.1 Existing neuromechanics estimation devices

Motor-driven dynamometers are robotic interfaces that perform single joint rotations while the two limbs are fxed to the interface. It can be used for single joint identifcation and physical therapy, providing isotonic, isometric and isokinetic experimental conditions. Such robotic interfaces have been used to estimate the torque-angle relation of the ankle joint (Hahn et al., 2011), injury or disease induced increase in ankle joint stifness (Lorentzen et al., 2010), passive resistance torque increased at the knee (Perell et al., 1996) or at all joints (Akman et al., 1999) due to spinal cord injury (i.e. spasticity), and the diference in ankle range of motion (ROM) in individuals with cerebral palsy (De Gooijer-Van De Groep et al., 2013; Sloot et al., 2015). These interfaces however impose movement constraints on a joint, which may result in unnatural motions and accounts for variability in torque-angle identifcation results using single-joint motor-driven dynamometer relative to estimations from multi-joint inverse dynamics (Hahn et al., 2011). Furthermore, motor-driven dynamometers are not equipped with body weight support for hip joint measurements. Therefore, hip joint neuromechanics investigations with those interfaces could only involve healthy participants (Claiborne et al., 2009), or be used with impaired individuals only in postures not requiring weight bearing. 2.2. Introduction 25

Table 2.1: Characteristics of existing lower limb neuromechanics evaluation devices Ribeiro Alvares et al. et al. (2016); Meule- et al. (2009); Bieryla al. (1999) al. (2011); Dirnberger t papers oijer-Van De Groep et al. unenburger et al. (2005) relevan Hahn et et al. (2012a) De Go (2013); Sloot et al.Claiborne (2015) et al. (2009) de Araujo (2015); Perell et al. (1996) Akman et L¨ Koopman man et al. (2016) - di- di- di- t di- drive device characteristics Single-joint rect driveknee (hip, and ankle) Ankle join rect drive Single-joint rect driveknee (hip, and ankle) Single-joint rect driveknee (hip, and ankle) Single-joint direct Gait rehabilitation Gaititation rehabil research device and /s /s /s ◦ ◦ ◦ 560 450 500 − − − speed 1 unknown 0 0 unknown 0 . 89 m/s unknown max force/torque 500N/750Nm unknown 890N 678Nm unknown unknown 250N/60Nm & R Ferstl ersity of Twente & del 500H (Charracx (MOOG FCS Inc., (Hocoma AG, Switzer- 3 dynamometer (Biodex 770 Norm (Lumex Inc., GmbH, Germany) land) Moog, Netherlands) device name IsoMed2000 (D no name Biodex Medical Systems, USA) Cybex USA) Kin-Com mo Lokomat LOPES (Univ Netherlands) Co., TN, USA) 26 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

Gait rehabilitation robotic exoskeletons are interfaces afxed to the body. They can provide controlled gait assistance and be used to analyse neuromechanical factors such as spasticity and voluntary muscle force level (L¨unenburger et al., 2005), or joint impedance during leg swinging (Koopman et al., 2016). However, exoskeletons constrain the joints movement, and may induce non-negligible vibrations due to the difculty to design a rigid mechanical structure. Furthermore, they may not be suitable to develop specifc experiment protocols required to measure the lower limb neuromechanics. For example, the stifness estimation method of (Mirbagheri et al., 2000, 2001) relies on perturbation pulse trains of 1.72◦ amplitude and 150ms pulse width, which is challenging to implement on existing gait rehabilitation exoskeletons due to their limited rigidity and torque capabilities.

Until present, most existing devices focused on ankle or knee joint measurements, while devices targeting the hip joint could usually be used only for isokinetic motion (Claiborne et al., 2009; Bieryla et al., 2009) or for multi-joint torque perturbations (Koopman et al., 2016) that may limit the measurements accuracy. In view of these functional limitations, we developed a robotic interface that can be used to systematically investigate the single-joint neuromechanics of the hip, knee and ankle joint in a natural upright position, without constraining the targeted joint motion, and with negligible structural vibrations even during highly dynamic perturbations.

2.2.2 Functional requirements

We want to develop a versatile device with a powerful actuator, a rigid mechanical structure and suitable control, that can be used to implement various protocols, including isometric and isokinetic conditions as well as brief mechanical perturbations, in order to characterise the lower body joints neuromechanics. To implement isometric conditions, the device should be able to provide a strong standstill torque to resist the subject’s maximum voluntary contractions. The joint torque measurements obtained during a stair climbing experiment (Andriacchi et al., 1980) were used as a reference for our device’s actuator’s standstill torque. To perform isokinetic experiments and quantify velocity-dependent muscle or joint behaviour, our device should also be able to implement a full range of movements with constant velocity range from 2.3. Device design 27

0 to 250◦/s (Fornusek et al., 2007). Finally, the device should be able to realise the highly dynamic movements required for estimating mechanical joint impedance. For instance, evaluating ankle refexes as in (Mirbagheri et al., 2000, 2001) requires a fast ankle angle perturbation with 1.72◦ amplitude and 150ms pulse width. In order to perform similar controlled position perturbations on the hip joint, considering that a human leg contributes to 15-20% of the body weight (Winter, 2009), a torque up to 100Nm would be required for a 90kg subject.

Large forces and accelerations required to characterise the lower limb neuromechanics demand a powerful and thus heavy actuator. Therefore, this actuator should be rigidly fxed away from the moving limb and reliable motion transmission should be used. A pneumatic cylinder connection (Aoyagi et al., 2007) or a two-bar linkage transmission (Fujii et al., 2007) could be used in this purpose, which would however result in a limited workspace. A pretensioned cable transmission can provide actuation with low inertia and without backlash (Townsend, 1988), and is therefore selected for our interface.

The structure of the interface should be rigid, to minimise undesired vibrations and deforma- tions. An additional structure is required to support the subject in an upright posture such as for walking. This structure should be mechanically independent from the robotic interface in order to avoid the transmission of vibrations from the motor. Finally, to avoid constraining the hip and knee joints motion such as with the Lokomat (L¨unenburger et al., 2005) and LOPES (Koopman et al., 2016) exoskeletons, we developed an end-point based interface interacting with the extremity of the examined limb while the rest of the body can move freely.

2.3 Device design

2.3.1 General description

Considering all the design factors presented in the previous section, our solution included a cable driven device with a large actuator placed outside of the workspace transmitting power to the limb, while the subject is supported in a natural upright posture by an independent 28 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a Neuromechanics Evaluation Device (NED) adjustable pulley

F1 L F2 loadcell τm, θm

x motor b Hip experiment c Knee experiment

knee adjustable brace seat

d Capstan Tload motor

Thold φ rope

Figure 2.1: Description of the Neuromechanics Evaluation Device (NED). Panel (a) shows how the subject seated in a rigid chair with an open design allowing the leg movement. The motor force transmitted by the cable is recorded by load-cells placed on both sides of the ankle fxture, with F1 and F2 as measured front and rear force respectively. θ˙m and τm are the speed and torque at the motor, θ˙ the hip joint angular velocity,x ˙ the cable linear movement speed and L the measured leg length. Panel (b) and (c) show how NED can be used to carry out experiments to investigate the knee and hip neuromechanics. The front and back pulleys can be shifted along the rail to ensure a perpendicular cable interaction. Panel (d) shows a schematic of the interaction force between pulley and a rope. The relation between the holding force Thold and the loading Tload can be described as Equ. 2.1 with a contact angle of ϕ and a coefcient of friction µ. 2.3. Device design 29 structure. The developed Neuromechanics Evaluation Device (NED) is illustrated in Fig.2.1.

As shown in Fig.2.1a, the subject is half seated on a rigid chair (of length 0.55m, width 0.7m and height 1.5m) with one leg suspended in the workspace and attached to the system via a foot fxture. The leg is moved by the motor (AM8061, Beckhof, Germany) located at the bottom of the workspace, via a steel cable (7x7 galvanised steel with PVC coating). Two load-cells (TAS510, HT sensors, China) are placed between the extremities of the foot fxture and the cable to measure the respective interaction forces. The front and rear pulleys can be locked at diferent positions along the rail (3m in the horizontal direction and 1.5m vertically) in order to keep the cable perpendicular to the leg (as shown in Fig.2.1a) and tensed. The cable tension is adjusted by the turnbuckles placed in series with the cable. All structures are made with aluminum strut profles (40x40L, Bosch Rexroth, Germany) and bolted to the cement foor.

The open seat enables the experimenter to perform hip experiments at diferent knee angles as shown in Fig.2.1b. In this case the knee joint can be kept at a specifc joint angle using a knee brace (T-scope, Breg), which enables us to study the infuence of the knee angle. By adjusting the seat, it becomes also possible to study the knee neuromechanics as shown in Fig.2.1c. Table 2.2 lists NED’s components and their characteristics, including the selected actuator.

In order to select an actuator based upon our design criteria, motors produced by Beckhof, ETEL and Infranor were evaluated depending on their technical specifcations including motor peak torque, standstill torque, moment of inertia and possible gearbox reduction ratio. The actuator selection process is detailed in Appendix A.

2.3.2 Cable transmission

The cable transmission, using the Capstan efect, requires a sufciently large contact force between the cable and the pulley to prevent slippage. The Capstan equation

µϕ Tload = Thold e (2.1) 30 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

Table 2.2: Past list Component Product name Company Characteristics Motor AM8061-0L2A Beckhof 12.1Nm rated torque, 1.9kW rated power Gearbox AG2210- Beckhof Gear ratio 1:5 +LP120S-MF1-5- 1|1 Motor pulley N/A N/A motor pulley to ft with selected actuator (diameter: 5.5cm, length: 5.5cm, material: stainless steel S80) made with stainless steel (S80) Motor con- AX5112 Beckhof Rated output current 12A troller Controller PRO ECO 120W Weidmuller 120W, 24Vdc, 5A, Mrf part power supply no.1469480000 Resister EL3351 Beckhof 1 channel, 16bit resolution with mea- bridge input surement range ± 12V Compression TAS510 HT Sensor Capacity 150kg with combine error and tension ±0.3 % FS load cell Potentiometer EL3255 Beckhof 5 channel, 16bit resolution with mea- input surement error < ±0.5 % Industrial CX5130-0125 Beckhof Intel Atom E3827, dual-core, 1.75GHz computer computer PRO ECO 120W Weidmuller 120W, 24Vdc, 5A, Mrf part power supply no.1469480000 OS TwinCAT 3 Beckhof Safety relay Safety relay Phoenix con- 1 current path, DIN rail, 24VUs, Mrf PSR-MS35-1NO- tact part no.2904953 24DC-SC Host com- ROG Strix i7- ASUS Core I7-7700HQ, 8GB RAM, puter 7700HQ 1TB+128GBSSHD, NVIDIA GeForce GTX1050 4GB GDDR5, Windows 10 Cable 7x7 galvanised The wire rope 3.5mm overall diameter including steel wire rope shop PVC coating, minimum breaking loading of 438kg Turn buckle BZP steel turn- The wire rope DIN 1480 Specifcation buckle shop Knee orthosis T Scope premier Breg Drop locks allows the knee to post-op knee brace be locked at 5 diferent position −10◦, 0◦, 10◦, 20◦ and 30◦ EMG ampli- EMG-USB2+ OT Bioelet- 16-256 channels+16 auxilliary chan- fer tronica nels, sampling frequency 512-10240, CMRR> 95dB, Noise (RTI)<4µVRMS 2.3. Device design 31 defnes the force relation between both the cable and the contact area of a cylinder while pulling a leg forward (Fig.2.1d), where Tload is the front cable tension which bares larger loading, Thold the force required to hold the loading on pulley, µ the friction coefcient between the cable and pulley, and ϕ the contact angle.

The minimum contact angle ϕ to prevent slippage is calculated from the friction coefcient and the expected cable force on both sides of the pulley. Assuming a joint torque of 150Nm (Andriacchi et al., 1980) and a plastic-metal friction coefcient of 0.1-0.3, the minimum number of cable turns must be four. Conservatively, the cable was wounded fve times around the pulley.

2.3.3 Control system

The control architecture of NED is depicted in Fig.2.2. The supervisor computer provides the position command to the control computer (CX5130, Beckhof), which performs the real-time control and monitors the motor controller (AX5112, Beckhof) and motor (AM8061, Beckhof). The cable system transmits the motion. In the example of a position disturbance, the angular displacement of motor shaft ∆θm is monitored by the software limits to avoid over-stretching the leg. The two load-cells at the extremities of the ankle fxture record the interaction forces

F1, F2, which are fed back via a resister bridge unit. Muscle activity of the subject are recorded with surface electromyography (EMG) electrodes and fltered by an amplifer. Safety relays, which are controlled by a laser safety system and emergency buttons, are connected to the power supply of the motor controller in case of need. Both the control computer and motor controller are powered separately by a power supply (PRO ECO 120W, Weidmuller, Finland). The operating software environment TwinCAT plans and implements the fastest point-to-point motion with given speed, acceleration and jerk limits.

In addition to considering the feedback error, the motor controller uses three sensor channels. Two resister bridges (EL3351, Beckhof) are used to measure the interaction force between the device and the subject’s leg, and an analog channel (EL3255, Beckhof) can be used e.g. for measuring the leg motion. 32 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

Controlbsystem Emergency button Power Power supply supply Laserbsafety

Relay Emergency Perturbation Controlbcomputer Motor button Profiles Controller AX5112 Cable PID Loadbcell Subject system ΔD Δτ Supervisor ROM Motor computer Monitor AM8v61

Resisterbbridge Δθm F13bF2 EL3351

EMGbamplifier rawbEMG FilteredbEMG EMGOUSB2g

Figure 2.2: The control system of NED.

2.3.4 Safety measure and ergonomics

Safety is a critical factor for robotic interfaces which are in contact with the human body. There- fore, we implemented redundant hardware and software measures to ensure safety throughout our experiments. A safety system was developed to defne the allowable range of motion as shown in Fig.2.3. The laser box emits a signal to the photodiode in the receptor box, which controls two safety relays. If the laser beam is blocked by any obstacle, e.g. leg moving beyond the expected range, safety relays will shut down the motor controller. Second, software safety measures implemented in the motor controller shut down the power when a position, speed, acceleration or power/torque limit is reached. Specifc software limits defne the workspace in which the leg should move depending on the targeted experiment. Finally, the power supply of both the motor and motor controller will shut down if any of the three emergency buttons is pressed. These buttons, which are available to the subject and experimenters during the experiment, are connected to another two safety relays and a master switch. In total, four safety relays (PSR-MS35, Phoenix Contact, Finland) control the power supply of the controller. If any safety threshold is reached the power motor is set of. 2.4. System characterisation 33

Laser location Receptor box Laser emitter box Inspection cap Lens Switch Battery

Laser Photodiode Laser diode

Figure 2.3: Laser safety system composed of a laser emitter box (with a focusing lens) and a receptor box. Any obstacle blocking the laser transmission will immediately shut down the power supply to the motor controller.

Beside the safety measures described above, various factors are included to provide a comfortable environment for diferent subjects. The dimensions of NED, that includes rail lengths and chair size, are designed based on subjects of height between 1.5-1.8m. The rigid chair is cov- ered with memory-foamed cushions to increase experiment contentment, and handrail location is adjustable to optimised body weight support.

2.4 System characterisation

This section frst analyses the kinematics of the developed system. It then examines sensor measurement errors and solutions to ensure an accurate recording. Lastly, a series of system identifcation tests are performed to characterise the system in diferent dynamical conditions.

2.4.1 Kinematics and sensitivity analysis

NED is designed to measure the lower limb joints’ fexion-extension biomechanics assuming that the cable and leg motions are restricted to the sagittal plane (with high enough cable pretension). For small angular displacements with the knee stretched and locked to maintain the leg straight, the kinematics is:

ρ ρ θ˙ =x ˙ = L θ˙ → θ˙ = m θ˙ (2.2) m m L m 34 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a Side-way motion induced error

ZD θ1 XD xu L1 x θ2 L2

b Leg rotation induced off-plane motion

yd L3' θ3 θ3

θ4 L4 θ4

L4'

Figure 2.4: Of-plane motion induced error. Panel (a) illustrates the measurement error resulting from side-way motions. θ1 and θ2 are the misalignments between the load-cell and the desired leg motion, L1 and L2 the distances between the foot and both pulleys. The maximum permissible side- way motion xu can be calculated by trigonometry. Panel (b) depicts the error resulting from a large leg rotation. Since both pulleys are fxed for each experiment, large leg motions will cause an angle between the load-cell and line of motion, which is described as angles θ3 and θ4. L3 and L4 are the ′ ′ distances between the foot and the pulleys. On the other hand, L3 and L4 are the distances between the foot and the ideal pulley location. xd and yd are the distances between the ideal pulley location and actual pulley location. 2.4. System characterisation 35

˙ where ρm is the motor pulley diameter, θm is the motor speed,x ˙ is the cable linear motion speed, L the leg length and θ˙ the hip joint rotation speed.

To ensure that the aforementioned planar movement assumption is valid we investigated how lateral leg movements can potentially infuence the accuracy of the biomechanical measurements. We assume that during an experiment the leg-cable attachment point can displace sideways by an undesired distance xu measured from the normal plane of movement as shown in Fig.2.4a. Then, the cable force measurements are afected by the of-plane confguration described by the angles θ1 = arctan(xu/L1) and θ2 = arctan(xu/L2) as shown in Fig.2.4a. By considering diferent leg lengths {80-95cm}, hip angles {5◦-60◦} and diferent pulley locations {80-165cm horizontal and 45-110cm vertical}, it is shown that the side-way motion should be limited within 14.3cm to result in a force measurement error below 5%. A side-way motion test (with 200N cable pretension) shows that a 14cm side-way motion requires an external force of 225N. Experiments should therefore be limited within such force limitations.

As large angular displacement cannot be considered as linear motion, we further investigated the infuence of limb rotation upon measurement accuracy. For each experiment, the pulley locations are relocated and fxed to yield a perpendicular cable connection minimising the measurement errors (shown as the gray dashed line in Fig.2.4b). Moving far away from the

2 ′2 2 ′ initial position will cause measurement error due to angles θ3 = (L3 + L3 − Yd )/(2 ∗ L3 ∗ L3)

2 ′2 2 ′ and θ4 = (L4 + L4 − Xd )/(2 ∗ L4 ∗ L4) (the black lines). By considering diferent leg lengths {80-95cm}, diferent experiment hip angles {5◦-60◦} and diferent sizes of leg motions, the largest acceptable leg motion before reaching an 5% error in measurement is 21cm. This can be considered as maximum acceptable position displacement to design experiments.

2.4.2 Spatial and temporal dependency of cable tension

The mass of the load-cells, connecting the elements and the cable itself (0.5kg) will slightly bend the cable and create cable sagging as shown in the left panel of Fig.2.5. As the cable is not perfectly straight, we observe that the tension does not change monotonically (as shown in Fig.2.5b with cable tension measured from a single load-cell during a back and forth motion) 36 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a Cable sagging

mgx mgy

b Measurement change due to sagging

50N pretension 5 100N pretension

0 Cable tension (N)

-5 5 10 15 20 Time (sec)

Figure 2.5: Cable sagging. The weight of the cable system (i.e. cable, harness, load-cells and turnbuckles) deforms the cable as illustrated in (a). (b) During a back and forth motion, the measured force will not change monotonically, which is marked by a red circle. Cable sagging can be minimised by increasing the cable tension. 2.4. System characterisation 37 and this discontinuity in force measurements is caused by a misalignment between the load- cells’ axis and the cable’s motion. Trigonometric calculations showed that a pretension of 200N limits the relative error between the measured and actual tensions below 2.5%. All further experiments were then performed with 200N pretension.

As in most cable-based systems, NED has mechanical characteristics that can slightly change over time. During the validation tests, it was observed that the measured cable tension will drop for 1N every 33.6s (during a cyclic movement test of speed 750mm/s with a pretension values of 200N while holding a 18kg dummy leg, which will be described in Section 2.5.1). This tension drop is negligible since most movements for neuromechanics evaluation require short perturbations with duration <1s (as will be developed in Section 2.5.2).

2.4.3 Cable temporal dependency

To characterise the drop in cable tension over time we measured the tension force while during slow periodic cable displacements of ± 6 cm with the average cable speed was 0.2 mm/s. The initial pretension was 200 N. In total fve cyclic motions were performed for diferent initial positions of the cable and the results are shown in Fig. 2.6). A relatively small decrease in cable tension, f(t), was observed which could be modelled with an exponential function:

f(t) = −202.4 e1.5 10−8 t, (2.3) where t is the time in seconds with compensated result shown in Fig.2.6 right panel. As described in Section 2.4.2, this tension drop is negligible for fast perturbations. However, it might be required to consider cable tension drop for timely tests, depending on the experimental tasks. 38 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

Temporalychangeyinycableytension 1stycycle beforeycompensation 2ndycycle afterycompensation 3thycycle 180 4thycycle 180 5thycycle 170 170 Forcey(N)

160 Forcey(N) 160 -60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60 Positiony(mm) Positiony(mm)

Figure 2.6: Measured temporal dependency of the cable tension for diferent cable (load-cell) positions.

2.4.4 Cable system modelling

We performed system identifcation tests to demonstrate that the behaviour of the designed interface can be characterised with the second order linear dynamical system under diferent speed. Hence, the transfer function describing the cable’s dynamics with cable tension, ∆F (s), as input, and cable displacement, ∆X(s), as output can be expressed as:

∆X(s) 1 = 2 (2.4) ∆F (s) Mxs + Bxs + Kx

with Mx the mass of the moving components on the cable, Bx and Kx the cable viscosity and stifness, respectively.

For system identifcation tests, we pre-programmed NED’s controller to perform 10 saw-shape displacement patterns with ±60mm amplitude and diferent speeds of 20-750mm/s as shown in Fig.2.7a. The force acting on the cable was measured with a load-cell and recorded at 1kHz. All ten trials at a given speed conditions were used to estimate the transfer function (2.4) using a least square method tfest of Mathworks Matlab. Additionally the mass of all moving mechanical components was known and was used as an initial estimate for mass component

(Mx) of function (2.4).

The parameters identifed at diferent speeds are shown in Fig.2.7b, demonstrating that our interface is characterised as high stifness and low viscosity (which reduces with the speed). This indicates that (despite the inherent cable compliance) NED is a rigid device that can 2.4. System characterisation 39

a Cable0movement0profiles0for0dynamics0identification 0.02m/s 0.06 0.04m/s 0.12m/s 0.04 0.2m/s 0.4m/s 0.6m/s 0.02 0.75m/s Position0Cmb 0 1 3 5 7 Time0Csb

b Estimated0impedance0value 600 600

400 400

200 200 StiffnessCN/mb Viscosity0CNs/mb 0 0 20 40 120 200 400 600 750 Speed0Cmm/sb

Bode0plot c From: Force To: Position -50 Natural frequency -100

Magnitude0CdBb -150 -45 -90 -135

Phase0Cdegb -180 10-2 10-1 100 101 102 103 Frequency0CHzb

Figure 2.7: Identifcation results of NED as a linear second order system. Panel (a) depicts the perturbation profles at diferent speed with estimated impedance values shown in panel (b). Panel (c) is the Bode plot of the system (with average values) that contains two poles at 1.6 and 17.2Hz (marked with dashed line). 40 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a NonlinearxHammerstein-Wienerxmodel F(t) Input U(t) Linear W(t) Output x(t) nonlinearity system nonlinearity

b Transferxfunctionxestimationxresults 1

0.8

0.6 MWxmodel 0.4 linearxmodel 0.2 Fitxresulxtx(NRMSE)

20 40 120 200 400 600 750 Speedx(mm/s)

Figure 2.8: The scatter plots of panel (a) exhibits the ftness of the identifcation, with circles representing the linear model and triangles the (piecewise linear) Hammerstein-Wiener model (HW model). Panel (b) shows a block diagram of the nonlinear HW model. F (t) is the interaction force exerted on the subject while x(t) is the resulting position movement. U(t) and W (t) describe nonlinear internal variables interconnecting the linear system with input and output. be used to identify the lower limb mechanics. The performance of the ftting were confrmed by normalised root mean square error value (NRMSE) with values higher than 70%. Due to both low variance in estimated impedance and relatively high NRMSE value, the mechanical characteristic of NED can be described by the average transfer function parameter’s values, with the averaged Bode plot shown in Fig.2.7c which has two poles, at 1.6 and 17.2Hz respectively.

2.4.5 Cable Nonlinearities

To further assess the nonlinear characteristics (such as Coulomb friction) within the system, an additional identifcation was carried out by adding a piecewise linear Hammerstein-Wiener (HW) model as shown in Fig.2.8a. The average value identifed in Section 2.4.4 was used as the linear system described in the centre of Fig.2.8a. The light blue circles and black triangles in Fig.2.8b shows how using the HW model slightly improves the identifcation, which is probably due to consideration of motor Coulomb friction. 2.5. Validation 41

2.5 Validation

In this section, we demonstrate how NED can be used for lower limb neuromechanics characterisation. First, two experiments were conducted to identify the dynamic parameters of a dummy leg and a pair of springs seperately, and then the experiment identifcation results were compared to know mechanic properties of the components. We then determined an optimal position perturbation for estimating hip joint stifness of healthy subjects.

2.5.1 Dummy leg mechanics

To validate the functionality of the developed interface, experiments are carried to identify the mechanical properties dummy leg and compare with the values obtained from CAD calculations (Fig.2.9a). The design parameters of the leg were: mass 18kg (resembles the leg mass of a 90kg subject), length of 70cm and moment of inertia of 1.84kg·m2 with respect to the hip joint. This mechanical dummy leg was fxed in NED for neuromechanics experiments. During the experiment, the leg was rotated about the hip joint (fexion/extension sequences) with the amplitude of 5◦ (6cm endpoint displacement) and with the speed range of 20-750mm/s as shown in Fig.2.9d. The prior mentioned fexion/extension movements were tested at fve diferent hip angles between 15◦ and 55◦ to test the the infuence of gravity to the identifcation results. In total, 20 repetitions were performed at each combined condition of speed {20-750mm/s} and hip angle {15◦-55◦}.

As described in Section 2.4.4, the developed robot exhibits high stifness and low viscosity which resembles a rigid device. Therefore, the cable dynamics in series with the leg could be neglected, and the recorded displacements and cable-leg interaction forces were used to estimate a linear second-order model of the mechanical dummy leg (with the least square method tfest of Mathworks Matlab):

¨ ˙ ∆τ = I∆θ + B∆θ + K∆θ , ∆τ = (F1 − F2)L (2.5) 42 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a b CADMdrawingM EstimatedMimpedanceMvalue ofMtheMdummyMleg 4 revolute 25 25 joint 3 20 20

2 15 15

70cm 10 10 1

5 5 0 0 0 Inertiaykg0m2b DampingMyNms/radb StiffnessyNm/radb EstimatedMinertia yCADMestimationbM

c FitMresultsMofMdifferentMangleMunderMdifferentMspeed 1 20mm/s 0.9 40mm/s 0.8 100mm/s 200mm/s 0.7 350mm/s 0.6 550mm/s 750mm/s 0.5 0.4 0.3

Fit Results (NRMSE) 0.2 0.1 0 55 45 35 25 15 Hip angle (deg)

d 6 MeasuredMdataMwMmodelMoutput cpredictio 4 Model n Measurement 2

0 Position(deg) -2 0.5 1 1.5 2 2.5 3 3.5 40

0 Torque(Nm)

-20 0.5 1 1.5 2 2.5 3 3.5 Timec(seconds)

Figure 2.9: Identifcation of the mechanics of a 18kg dummy leg at joint angles between 15◦-550 and speed 20-750mm/s, with 20 trials at each condition. Panel (a) shows the CAD drawing of the dummy leg, Panel (b) the boxplot of the estimated inertia, damping and stifness value of the dummy leg, with a green line indicating the estimated inertia from CAD (1.84kg·m2). Panel (c) shows the NRMSE value (which defnes the estimation’s quality) at fve diferent hip angles and seven diferent speeds. Panel (d) shows the measurement results (dash line) and estimated model output (solid line) of the dummy leg identifcation. 2.5. Validation 43 where ∆τ is the change of interactive torque, I the leg inertia, B the viscous parameter of the joint, K the hip joint stifness, F1 and F2 are the two load-cells’ signals. It is important to note that the change in interaction torque (∆(F1 − F2)L) was used here rather than change in one load-cell measurement (∆F1), as the purpose of this section was to estimate dummy leg impedance through position displacement and resulting force changes.

The estimated impedance values are shown in Fig.2.9b with the NRMSE value depicted in Fig.2.9c which suggests that the dummy leg’s dynamic parameters were successfully identifed for velocities larger than 40mm/s (with NRMSE > 80%). The inertia estimated in this dynamic identifcation is close to the value predicted from the CAD parameters (1.84kg·m2) while the viscosity and stifness values are both low. These results indicate that NED can be used to identify the hip mechanical impedance.

2.5.2 Stifness estimation

To evaluate whether NED can be used to identify stifness, we used two parallel springs attached to the cable as shown in Fig.2.10a. The stifness of these two springs was then identifed using a smooth position displacement as described in (Burdet et al., 2000). In this method, stifness K can be simply computed from: ∆τ = K∆θ (2.6) on the constant position plateau where inertia and viscosity have little infuence, as ∆θ˙ = ∆θ¨ = 0, see Fig.2.10b. Twenty tests were carried out for each perturbation with the amplitude of 2-8mm and duration 50-150ms. Stifness was then evaluated from (Equ.2.6) with mean displacement ∆θ and mean measured torque ∆τ during the plateau region. The results of Fig.2.10d demonstrate that this method can identify stifness accurately, with estimations improved with a larger perturbation amplitude and no observable diference in diferent duration or perturbation direction. Since the load-cells’ measurement sufers from noise with standard deviation of 0.29N and maximum 1N, a large amplitude with stronger spring force will increase the signal to noise ratio and improve the estimation. In the mean time, the angle measurement 44 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a Spring estimation setup

Springs b Perturbation of various amplitudes c Perturbation of various durations 8mm 10 10 6mm 150ms 4mm 100ms 2mm 50ms 0 0 Amplitude/(mm) Amplitude/(mm)

-10 -10 10 10

5 5

0 0

-5 -5 Force/Differece/(N) Force/Differece/(N) -10 -10 0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500 Time/(ms) Time/(ms) d Estimated stiffness

0.6 0.6

0.4 0.4

0.2 0.2 Stiffness/(N/mm) Stiffness/(N/mm) Amplitude/=/2/mm 4/mm 6/mm 8/mm Duration/=50ms 100ms 150ms

Figure 2.10: Stifness identifcation of a spring with known elasticity using NED. Panel (a) shows where the spring is attached. Panel (b) and (c) shows the perturbations of diferent amplitude and duration together with the resulting force diference. Panel (d) shows the values obtained with these perturbations, which are similar to the spring constant with less deviation using a larger perturbation. of the motor shaft has a resolution of 0.019o corresponding to a 0.35mm cable displacement and angle of 0.46◦ for a 90cm long leg. This also implies a small perturbation amplitude (e.g. 2mm amplitude) will sufer from measurement errors.

2.5.3 Optimal position perturbation to identify stifness

A pilot study with one healthy subject (male, weight: 54kg, height: 172cm) was carried out to evaluate the feasibility of using NED to identify hip stifness with the technique described in the previous section. The experimental protocol was approved by the Imperial College 2.5. Validation 45

Research Ethics Committee (Appendix B), and all procedures were performed according to the principles described in the Declaration of Helsinki. The subject was informed on the device and experiment, and signed an informed consent form prior to the experiment. The participant’s weight and leg length (from the anterior superior iliac spine to the lateral malleolus) were measured to estimate the leg inertia. A lockable knee brace was used to fx the knee joint at 0◦ angle.

The subject was half seated on the chair with one leg suspended, was asked to support his body weight on the handle and relax the lower limb. A harness was attached to the ankle of the tested leg which was connected to the cable and the motor (Fig.2.1). The participant was given an emergency stop and can stop the experiment whenever needed. The laser safety system was initiated and adjusted to defne the range of motion of the tested leg at 15◦. The motor performed a slow motion to move the participant’s leg to defne the comfortable range of motion for additional system adjustments. A single pulse perturbation was also given to provide the participant with an experience of a perturbation and adjust the system’s safety.

As described in Section 2.3.3, the motor controller implements the fastest displacement given the safety limits in speed, acceleration and jerk. The position perturbation with a constant displacement plateau was thus determined by these limits as well as by the plateau duration and displacement amplitude. Our goal was to use a perturbation of minimum amplitude and duration, which would disturb the subject minimally and thus avoid any voluntary reaction. However, this comes in trade-of with the perturbation amplitude, which should be large enough to maximise the signal to noise ratio (SNR) of the stifness estimation. On the other hand, very large accelerations to yield a fast perturbation could cause cable oscillations which would disturb the subject and afect the stifness estimation quality. To achieve fast displacement with minimal cable oscillations and stable force measurements, we increased the perturbation amplitude iteratively (starting from 6cm) and reduced the controller’s dynamic limits (speed, acceleration and jerk) while recording the magnitudes of cable oscillation. In total, 50 combinations of the speed, acceleration and jerk limits were tested for position perturbation command with the hip fexed at 15◦. 46 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED)

a Minimum oscillations profile

0.03 0.015m 0.0175m 0.02 0.02m 0.01 0

Positionv(m) -0.01 -0.02 -0.03 10

Forcev(N) -5

-10 0 0.2 0.4 0.6 0.8 Time (sec)

b Estimatedvparameters

5 150 15 ) 2 4

100 10 3

2 50 5 Estimatedvinertiav(kgm 1 Estimatedvstiffness(Nm/rad) Estimatedvviscosityv(Nm*s/rad)

0 0 0 Estimated inertia (Winter, 1995)

Figure 2.11: Estimation result of a pilot study with one subject. Panel (a) is an example of the profle with the position at the top and the force feedback at the bottom. Panel (b) shows the estimated joint impedance. The estimated inertia is close to the estimation from the anatomical model (Winter, 1995) shown is a green line. 2.6. Discussion 47

The resulting optimal perturbation is shown in Fig.2.11a. The amplitude of the perturbation is largely above the position resolution (0.35mm cable displacement) and from the test with the spring of Fig.2.10, should have a large signal to noise ratio. Considering the large motor variability in human movements, we selected a larger displacement than required to maintain a high signal to noise ratio. The collected kinematics and interaction forces were again least square ftted to estimate the hip joint impedance using Equ.2.5. As shown in Fig.2.11a, the optimised perturbation resulted in consistent and reproducible motions with negligible force oscillations.

Fig.2.11b shows the estimated joint impedance (I,B,K) of three diferent perturbation amplitudes {15,17.5,20}mm with a 150ms long plateau. The green line shown in Fig.2.11b is the inertia calculated from the anatomical model (Winter, 2009) using the subject’s weight and leg length. We can see that the variance of estimation is small (11% for inertia, 10% for stifness and 19% for viscosity). The inertia estimate is close to the anatomical model, and the values of viscoelasticity are in the same order of magnitude as reported in (Koopman et al., 2016).

2.6 Discussion

Investigating the lower-limb neuromechanics is critical to understanding the control of standing and walking in healthy and neurologically afected individuals, as well as to efciently control assistive and rehabilitation devices for performance augmentations. However, so far only few studies could use a single device to investigate lower-limb neuromechanics of diferent joints and specifc to the hip. Importantly, hip joint viscoelasticity investigation was only performed in multi-joint torque perturbation (Koopman et al., 2016), which is usually limited in accuracy, and never with precise single joint position displacement. In this context, we have developed and validated a novel robotic interface named NED (Neuromechanics Evaluation Device) to investigate the lower-limb neuromechanics.

NED can apply a large range of dynamic interactions to a subject’s leg at static posture or during movement. This enables the neuromechanics identifcation of hip and knee joints in 48 Chapter 2. A lower limb Neuromechanics Evaluation Device (NED) fexion/extension. Importantly, NED allows the experimenter to estimate a subject lower limb neuromechanics in a natural upright posture under controlled environment, which also makes the device well suited for carrying out investigations on patients’ neuromechanics. The device can be quickly adapted to a subject’s specifc anatomy and to carry out various measurements. The use of a closed mechanical cable loop with powerful actuator fxed outside the rigid supporting structure enables us to implement highly dynamic environments with little vibrations.

In this chapter, NED’s mechanics was characterised, and its performance to estimate a lower limb neuromechanics was demonstrated through the identifcation of a dummy leg and a spring’s mechanical impedance. As a result of a powerful actuator and stif mechanical frame of NED, it was possible to achieve accurate and repeatable position perturbation which enabled more efcient dynamics identifcation of individual leg joint compare to the mechanisms with rigid links (Hahn et al., 2011; Fujii et al., 2007). The good match of the identifed and measured parameters as well as the range of protocols that can be implemented on NED makes it an efective tool to identify the hip, knee and ankle joint biomechanics. The techniques developed in this chapter could be used to systematically investigate hip joint viscoelasticity as described in Chapter 3 and (Huang et al., 2019a). Chapter 3

Hip joint neuromechanics evaluation with cable driven robot NED

The previous chapter described the development of NED and its mechanical evaluation. In this chapter, we will develop neuromechanics estimation experiments to illustrate the functionality of NED with healthy subjects. A maximal voluntary isometric contractions experiment is frst carried out to estimate the hip maximal voluntary contraction. Then, the impedance measurement method is developed and tested with a parallel spring. An experiment with 10 healthy subjects is then conducted to estimate the hip joint impedance and study how it depends on the posture and applied force.

3.1 Maximal voluntary isometric contractions (MVIC)

One key neuromechanical characteristic of motor control is the maximal voluntary isometric contraction (MVIC) and how it depends on the joint angle, which often refers to as the torque-angle relation. The maximum voluntary contraction (MVC) of the limb during isometric experiments, which is sometimes also named maximum voluntary joint torque (MVJT), defnes motor capability under a specifc joint orientation (Hahn et al., 2011). It can be used to examine the motor capabilities in individuals afected by neurological disease such as SCI (Sisto

49 50 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED and Dyson-Hudson, 2007). The EMG measurements obtained during maximum contractions is useful to determine the muscle activation levels during various phases of movements such as locomotion (Buchanan et al., 2006). The activation pattern can illustrate how humans control motion under various scenarios, including rough terrains or external perturbations. Addition- ally, it was demonstrated that EMG normalisation with data collected from MVIC experiments was more reliable than other normalisation methods (Bolgla and Uhl, 2007). To further test the capabilities of NED, I performed a hip joint MVIC experiment with one healthy subject in both fexion and extension motion.

3.1.1 Experiment protocol

The experimental protocol is composed of three steps: preparation, data collection and neuromuscular parameter estimation. The preparation and data collection processes are similar to the pilot test listed in Section 2 but with additional EMG recording.

1. Preparation. The participant was given an introduction to both the device and the research prior to the experiment and was asked to sign an informed consent form. Both his weight (M, unit: kg) and leg length (Lleg, unit: m) were measured in order to estimate the leg inertia. The mass was collected by a scale and the leg length was estimated by a tailor tape from the anterior superior iliac spine to the lateral malleolus. Afterwards, bipolar electrodes were positioned at the belly of three muscles, which includes Rectus Femoris (RF), Biceps Femoris (BF) and Tibialis Anterior (TA) muscles of the measured leg, and connected to a signal amplifer to monitor muscle activation. The locations of the electrodes were defned with anatomical landmarks, and the skin surface area was cleaned with alcohol before placing the electrodes. After fxing the EMG electrodes, a lockable knee brace was used to fx the knee joint at a specifc experiment angle of zero degrees.

The participant was half seated within the robot with the experiment leg suspended. The subject was asked to support his body weight on the handle and relax until given command. A harness, which connects to the cable system of the robot, was afxed to the ankle joint. The participant was instructed to stop the experiment using one of the emergency buttons (with 3.1. Maximal voluntary isometric contractions (MVIC) 51

Table 3.1: Biographical information of the subject in MVIC experiment Subject Weight Height Leg length Age Sex Participated ex- No. [kg] [cm] [cm] periment 1 69 171 85 25 M MVIC experiment

additional buttons next to the master computer and the side of the workspace) whenever he would like. After setting the laser safety system to defne a safety range of the leg, the laser system was initiated.

2. Data collection stage. To start the experiment, the subject was verbally encouraged to pull NED’s cable with maximum leg strength for approximately two seconds before a ten-second break was given. This sequence was repeated for four times for both fexion and extension of the hip joint at fve diferent hip angular orientations in the sagittal plane. A ten-minute break was given between each hip angular orientation to prevent muscle fatigue. Both the force and the EMG were recorded to estimate the hip joint torque-angle relation. The angle was recorded and checked to prevent biasing the measurements. The experiment was repeated at fve diferent hip angles {15◦, 25◦, 35◦, 45◦, 55◦}.

3. Neuromuscular parameter estimation stage. EMG signal processing involves recti- fcation and bandpass fltering (10-500Hz, Butterworth) as well as notch fltering with 50Hz basis frequency to cancel the efect of the power supply of the UK. The EMG was further Butterworth lowpass fltered with a 5Hz cut-of frequency to create an envelope of the signal. The maximum value of the fltered EMG signal was used to normalise the signals to enable comparisons of the muscle activation. Fig.3.1a shows the procedure of the EMG processing. Panels (b) and (c) are examples of muscle activity during MVIC in both fexion and extension.

One healthy subject with no history of lower-limb injury or medical conditions was recruited for the MVIC experiment, with biographical information in Table 3.1. All procedures were performed according to the principles in the Declaration of Helsinki and approved by the Imperial College Research Ethics Committee (Appendix B). 52 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED

3.1.2 Results

Fig.3.2 shows the measured maximal voluntary isometric contractions at diferent hip angles with colours indicating separate trials. We see that the variation within the same hip angle was smaller in extension direction and larger in fexion, with a maximum diference of 30Nm and 75Nm respectively. However, by converting the variation into the percentage of maximum value, the variations were 38% and 37% respectively.

On the other hand, the mean value at diferent hip angles shows an angle dependency of measured torque at hip fexion direction, with a maximum change of 64Nm. In the extension direction, the angle dependency is less obvious with the largest change of 21Nm. By converting the observed change into the percentage of maximum measured torque, it was found to be 34% and 25% respectively.

(a).BEMGBprocessBflowBchart RawFEMG FilteredFEMG signal signal Band-passFfilter NotchFfilter Low-passFfilter Rectify 10-500Hz 50Hz 5Hz

(b).BExtension (c).BFlexion Trial 1 100 200 Trial 2 150 Trial 3 50 100 Trial 4

TorqueB(Nm) 50 0 0 1 1 0.75 0.5 0.5 0.25 (BicepsBFemoris) NormalizedBEMGB 0 (RectusBFemoris) 0 0 1 2 3 4 5 6 0 1 2 3 4 5 6 TimeB(sec) TimeB(sec)

Figure 3.1: EMG process and sample measurement of the maximal voluntary isometric contractions experiment. Panel (a) the block diagram describing the ofine EMG signal processing. Panel (b) and (c) examples of the measured data, with torque plotted at the top while EMG at the bottom. 3.1. Maximal voluntary isometric contractions (MVIC) 53

Torque-angle relation (Extension) 200 Trial 1 Trial 2 Trial 3 160 Trial 4 Mean

120

Torque (Nm) 80

0 15 25 35 45 55 Hip angle (degree) Torque-angle relation (Flexion) 200

160

120

Torque (Nm) 80 Trial 1 Trial 2 Trial 3 40 Trial 4 Mean

0 15 25 35 45 55 Hip angle (degree)

Figure 3.2: Torque-angle relation of hip joint fexion and extension motion. The maximal voluntary isometric contractions experiment was repeated for four times at fve diferent hip angles and results represented in diferent coloured circles. The mean value of each hip angle measurement depicts as black stars. 54 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED

3.1.3 Discussion

To my knowledge, there is only one research involving hip joint MVIC measurements in both fexion and extension direction (Anderson et al., 2007) and one with extension direction only (Worrell et al., 2001), perhaps due to limitations in the measurement systems. In particular, the devices used in these works for hip fexion and extension MVIC experiments are not suitable for SCI participants as it required standing postures. In this context, as expected the high standstill torque of NED helped to perform the experiment. No position change was observed even under such large muscle torque with our a healthy subject. Additionally, the structure design of NED potentially allows SCI subjects to carry out MVIC experiments.

As shown in Fig.3.2, there are large variations among diferent trials, which may be due to inconsistencies in producing identical experiments. As mentioned in the sensitivity analysis of Section 2.4.1, NED uses single-axis loadcells for measuring the interaction forces between the subject and the robot. Any of-plane motion could produce misalignment between the leg force and loadcells, and further resulting in geometrical error. It was estimated in Chapter 2 that a force of 225N is required to result in a measurement error of 5% at the sensor feedback. Since the MVIC experiment involves active contractions up to 205N in fexion motion, a measurement error of 10.25N value was expected and the residual variation should be considered as inconsistent among trials.

By comparing the measurement results with lower-limb torque-angle relations found in (An- derson et al., 2007) and (Worrell et al., 2001), both the amplitude and shape in my experiment were similar. It is shown by (Anderson et al., 2007) that the hip extension torque will decrease slightly with the decrease of hip angle, while (Worrell et al., 2001) illustrated a change up to 20% of maximum torque value. Both of the results correspond to above experimental results.

On the other hand, (Anderson et al., 2007) demonstrated an increase of maximum joint torque when the hip angle decreases in a fexion motion, while our measurement suggested a valley shape of torque-angle relation with a minimum point at 25◦. At this moment, it is unclear whether our fndings largely difers from (Anderson et al., 2007) due to subject posture, potential 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity55 measurement error listed in the previous paragraph and the experimental hip angles.

Additionally, the above results show the Tibialis Anterior muscle was always active while performing the MVIC experiments regardless of force direction. Therefore, the torque-angle relations collected in this experiment were multi-joint contractions instead of single-joint, which involves muscle forces from shank muscles and biarticular muscles. Using an ankle module, the identifcation experiment could start from the distal joint and further upwards to diferentiate the contribution respectively.

3.2 The infuence of posture, applied force and pertur-

bation direction on hip joint viscoelasticity

Limb viscoelasticity is a critical factor used to regulate the interaction with the environment. It plays a key role in modelling human sensorimotor control, and can be used to assess the condition of healthy and neurologically afected individuals. This section reports the estimation of hip joint viscoelasticity during voluntary force control using a novel device that applies a leg displacement without constraining the hip joint. The infuence of hip angle, applied limb force and perturbation direction on the stifness values was studied in ten subjects. No diference was detected in the hip joint stifness between the dominant and non-dominant legs, but a small dependency was observed on the perturbation direction. Both hip stifness and viscosity increased monotonically with the applied force magnitude, with posture to being observed to have a slight infuence. These results are in line with previous measurements carried out on upper limbs, and can be used as a baseline for lower limb movement simulation and further neuromechanical investigations.

3.2.1 Literature

Muscles are characterised by their viscoelasticity, where stifness and viscosity increase with activation. By co-activating the muscles acting on limbs, the human nervous system can control 56 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED its stifness or viscosity in magnitude, shape and orientation (Burdet et al., 2013). Critically, this enables humans to regulate their interaction with the environment (Hogan, 1985) e.g. during object manipulation, or for running optimally on diferent grounds.

In order to understand how humans control the limb viscoelasticity, a large body of experiments have estimated stifness and viscosity in the upper limb, in particular at the wrist and arm (Burdet et al., 2013). Stifness and viscosity can be measured indirectly by applying a mechanical disturbance on the limb and regressing the resulting changes of position and force. Measurements carried out using this method showed that stifness generally increases linearly with the applied force: in one deaferented muscle, in a single joint (thus including refexes), and in the arm (Burdet et al., 2013).

Much less is known on the viscoelasticity in the lower limbs, in part due to the difculty to carry out suitable experiments involving heavy leg mass. For instance, existing robotic interfaces to estimate viscoelasticity in the lower limb either require a sitting or lying position (Amankwah et al., 2004; Sinclair et al., 2006; Perell et al., 1996; Akman et al., 1999), or are not sufciently rigid to apply fast perturbations without causing non-negligible oscillations e.g. (L¨unenburger et al., 2005; Koopman et al., 2016; Meuleman et al., 2016; Farkhatdinov et al., 2019). In addition, all of these interfaces are afxed to the body and thus determine the joints around which the limb can move, while anatomical joints generally vary with the posture (e.g. the knee joint rotates and translates during locomotion). An alternative method consists of applying perturbations directly on the foot, which can be used to estimate (only) ankle viscoelasticity (Rouse et al., 2014).

In view of the limitations of previous devices to investigate the lower limb viscoelasticity, we have developed a dedicated robotic interface (Huang et al., 2019b). This rigid interface can be used to investigate the hip, knee or ankle neuromechanics in a natural upright posture. It uses an endpoint-based approach to apply dynamic environments on the leg, thus does not need to impose joint movement.

Due to the difculty to apply a mechanical disturbance on the leg for estimating viscoelasticity, experiments reported in the literature have been mainly restricted to a single joint disturbance 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity57 at the knee (Zhang et al., 1998; Ludvig et al., 2017) and ankle (Mirbagheri et al., 2000) joints. In (Koopman et al., 2016) the LOPES exoskeleton has been used to estimate viscoelasticity at the whole leg (including the hip joint), using a multi-joint random torque as perturbation and an indirect measurement of the resulting displacement from its series elastic actuators. While random torque perturbations enable experiments to identify both the stifness and viscosity simultaneously (Perreault et al., 2001, 2002), we preferred using a single position displacement to focus on accurately determining the joint stifness (Burdet et al., 2000). This allowed us to examine the efect of individual factors such as posture or force level separately.

3.2.2 Methods

Measurement system

The Neuromechanics Evaluation Device (NED) is a powerful cable-driven robotic interface to yield computer-controlled dynamic testing on one leg of subjects supported in a seated or upright posture ((Huang et al., 2019b), Fig.3.3a). NED’s open stand support allows for conducting biomechanics identifcation experiments on various subjects including subjects with impaired motor function. Used in diferent confgurations, this cable-based system can control the motion of the whole leg, foreleg, or foot in order to estimate the hip, knee or ankle neuromechanics. The pulley system can be adjusted to keep the cable orientation approximately normal to the limbs movement in diferent orientations for subjects of various size (Huang et al., 2019b).

Experiment

The experimental protocol was approved by the Imperial College Research Ethics Committee (Appendix B), and all procedures were performed according to the principles described in the Declaration of Helsinki. Safety measures with NED include software limits on the velocity, acceleration and jerk, an optical system to check perturbation limits, and emergency buttons for the subject and experimenters (Huang et al., 2019b). 58 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED

a Neuromechanics Evaluation Device visual5feedback of5force screen

θ adjustable pulley

F1 applied5force loadcell

τm,θm θ F2

x motor

b 0.03 hip flexion 0.02 0.01 0 -0.01 Position (m) -0.02 forward perturbation -0.03 hip extension backward perturbation

0 Force (N) -5

-10 0 200 400 600 800 Time (ms)

Figure 3.3: Sketch of Neuromechanics Evaluation Device (NED) and perturbation profle used to estimate the hip viscoelasticity. Panel (a) depicts an the experimental setup. The subject’s leg was moved by the motor via a cable closed loop. The interactive force was recorded by the loadcells in both the front and back of the ankle. The pulleys can be displaced to yield a force perpendicular to the subject’s leg. θ˙m and τm are the speed and torque at the motor, θ˙ the hip joint angle, X˙ the cable linear motion, F1 and F2 the force recorded at the loadcell in both front and the back. Visual feedback of the applied force enabled the subjects to control a desired force level while a perturbation was provided by the interface. Panel (b) shows the measured position and interaction force δ(F1 −F2).

Ten subjects (Age 21-27, with 6 females) without any known lower-limb injury or medical condition were recruited, they were informed on the device and experiment and signed a consent form prior to participation. Subjects’ weight and leg length (from the anterior superior iliac 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity59

Table 3.2: Biographical information of the subjects in hip joint impedance measurements no weight [kg] height [m] leg length [m] age sex 1 67 1.70 0.89 25 M 2 47 1.55 0.82 24 F 3 100 1.79 0.85 27 M 4 47 1.55 0.82 26 F 5 61 1.72 0.93 23 F 6 54 1.68 0.88 27 F 7 54 1.72 0.94 21 F 8 69 1.71 0.85 25 M 9 85 1.79 0.87 23 M 10 50 1.50 0.81 24 F spine to the lateral malleolus) were then measured to estimate leg inertia. These subjects’ parameters are reported in Table 3.2.

Bipolar electromyography (EMG) electrodes placed on the rectus femoris, biceps femoris and tibialis anterior muscles were used to check when subjects are relaxed. A locking knee brace was used to keep the knee joint fxed during the perturbations, and thus ensure that the leg is straight during the whole procedure.

Each participant was asked to relax while supporting their body weight using the handle. A harness was used to connect the ankle of the leg under test to the cable system (Fig.3.3a). The subject could familiarise with the device by experiencing several perturbations, after which the system workspace safety limits were set.

A position perturbation was used to estimate hip joint impedance. The perturbation shown in Fig.3.3b was used. It consists of a 150ms long plateau with 20mm amplitude (corresponding e.g. to an angle of 1.15◦ for a 90cm long leg) with smooth ramps up and down. This perturbation profle was determined by trial and error to ensure a force measurement profle with negligible oscillations (Huang et al., 2019b). All data was measured with a sampling rate of 1000 Hz.

For both legs, measurement was carried at diferent initial postures with the hip angle (relative to vertical) at {15◦, 25◦, 35◦, 45◦, 55◦}. At every posture, subjects were frst asked to relax (which was checked using EMG) while a perturbation (with profle as in Fig.3.3b) was applied by the system randomly in the forward or backward direction, with fve trials in each direction. The 60 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED

Subject7number Left7leg Right7leg Force

Hip7angle level 0 0 200 400 forward7perturbation7 400 200 backward7perturbation Stiffness7RNm/radh

Subject71 Subject72 Subject73 Subject74 Subject75 -20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N

5545352515 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N

-20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N

-20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N Hip7angle -20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N

-20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N 0 0 0 0 0 0 0 0 0 0 200 400 200 400 200 400 200 400 200 400 400 200 400 200 400 200 400 200 400 200

Subject76 Subject77 Subject78 Subject79 Subject710

-20N -20N -20N -20N -20N -10N -10N -10N -10N -10N

5545352515 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N

-20N -20N -20N -20N -20N -10N -10N -10N -10N -10N 0N 0N 0N 0N 0N 10N 10N 10N 10N 10N 20N 20N 20N 20N 20N

Figure 3.4: Hip stifness results for all subjects and conditions. time of a perturbation was also random so that the subject could not prepare for a perturbation.

Sequentially, each subject was asked to pull or push the leg to exert a force of {-20, -10, 10, 20}N (with positive value for backward kick) as was controlled by the subject using real-time feedback of the applied force displayed on a computer screen placed in front of them. The force level was taken relative to the relaxed condition of each subject, so that the efect of gravity was compensated by the interface. To prevent a subject from volitionally reacting to a perturbation, visual feedback was not updated during the perturbation.

The subjects carried out two such measurement cycles (5 minutes each), with a ten minute rest during which they were detached from NED. For the two legs of the ten subjects, there 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity61 were thus ten trials at each of the fve postures and fve force levels, using two perturbation directions (see Fig.3.4). The total experiment time was 100 minutes excluding the breaks.

Data analysis

The hip joint dynamics can be described as:

¨ ˙ τm ≡ τg + τe + Iδθ + Bδθ + Kδθ , (3.1)

where τm is a torque produced by muscle tension to counteract the gravitational torque τg and ¨ ˙ τe corresponding to the external forces. Iδθ is the inertia component and Bδθ + Kδθ the hip viscoelasticity component corresponding to a displacement angle δθ, where B is the viscosity and K stifness.

As two force sensors are used at the extremities of the ankle fxture which record signals F1 and F2, the dynamics can be simplifed to:

¨ ˙ δτ = δ(F1 − F2)L = Iδθ + Bδθ + Kδθ , (3.2) where δτ is the torque response to δθ and L the limb length. Both the static muscular torque

(τm due to static applied limb force) and gravity torque τg is eliminated as we removed the ofsets during data analysis (i.e. using δ(F1 − F2)). Additionally, the gravitational torque τg was found to have less than 1% efect on the overall joint torque with the 20mm perturbation amplitude, and is therefore negligible. Similar to the method described in (Burdet et al., 2000), a constant displacement (as shown in Fig.3.3b) was used to determine stifness K using:

δτ ≡ K δθ . (3.3)

For each participant, leg, posture, force level, and perturbation direction condition, the perturbation displacement δθ and resulting change of torque δτ in the last 100ms of the perturbation plateau of all 10 trials formed 1x1000 vectors, which were used to estimate K as the least-square 62 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED solution of Equ.3.3.

Viscosity was determined (using Matlab tfest command with search method set ’auto’ for best ft) as the least-square solution of the transfer function:

∆Θ(s) 1 = , (3.4) ∆T (s) Is2 + Bs + K where ∆Θ and ∆T are the Laplace transforms of δθ and δτ respectively. In this equation, inertia was computed from the biomechanical model of (Winter, 1995) and stifness was estimated from Equ.3.3. The weight of the leg was estimated as 16.1% of total weight, and the radius of gyration of the whole leg at the distal end is 0.56L, thus

I = 0.161M(L 0.56)2, (3.5) with the mass M and length L parameters from Table 3.2.

3.2.3 Results

Fig.3.4 summarizes the stifness estimation results of all ten subjects. These results were obtained with the two perturbation directions, for their two legs, at the selected fve postures and the fve force levels. Hip joint overall stifness changes with the perturbation direction, applied limb force level and hip angle (as was tested by separate Friedman’s tests with p<0.05). No diference was detected between stifness values in the dominant and non-dominant legs (as was tested using both Friedman’s test and paired t-test). In the following, we will thus, for each subject, use the stifness value of the two legs together, and investigate how stifness and viscosity depend on the perturbation direction, force level and hip posture.

Perturbation direction dependency. Fig.3.5 shows how the stifness values of all subjects, at all postures and force levels, depend on the perturbation direction. We see that a larger por- tion of the stifness values is below the identity line, suggesting that the backward perturbation results in larger stifness values than the forward perturbation. This was confrmed by a paired 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity63

500

400

300

200

100 Forward perturbation Stiffness (Nm/rad)

0 100 200 300 400 500 Backward perturbation Stiffness (Nm/rad)

Figure 3.5: Hip stifness measurement depends on the perturbation direction. Each dot represents the stifness at a specifc subject leg, posture and force level, with stifness measured with backward perturbation in the abscissa and with forward perturbation in the ordinate. The linear regression (green solid line) below the (dashed red) diagonal indicates larger values with perturbations in backward as in forward directions. t-test indicating that the diference between the estimation was diferent with the two diferent directions (p<0.05). The linear regression result (green solid line, with R2=0.72) described in Table 3.3 exhibits a diference of 26% between the estimation in the two directions. On the other hand, the estimated viscosity values showed no clear perturbation direction dependency, with regression close to identity line but R2 <0.1 for the best linear regression model.

Force-level dependency. To investigate the interrelationship between measured viscoelasticity, applied limb force level and hip angle, we performed three steps of mixed efect modelling to examine the stifness change due to the selected parameter. Firstly, stifness was assumed to vary linearly with applied limb force while posture may infuence this linear relation, modeled as:

Kik = A0Fik + A1 + bi0Fik + bi1, (3.6)

where Kik is the k-th stifness estimated at i-th hip angle, Fik the k-th applied limb force at i-th hip angle, A0 and A1 the fxed efects describing how stifness changes with applied limb force, bi0 and bi1 correspond to the random efects representing the infuence of hip angle upon 64 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED the identifed force-stifness relation.

By estimating mixed efect models for each subject’s leg, it was found that stifness increases monotonically with applied force amplitude in all subjects (presented in Fig.3.6a). The estimated force-level dependency weight (A0) has a mean value of 5.15Nm/rad per applied Newton force and a standard deviation of 0.98Nm/rad. This fnding indicates a positive relationship between applied limb forces and hip joint stifness, which is further confrmed by F-tests (p<0.05 for all subjects’ legs).

Furthermore, Friedman tests showed that the hip angle would change both fxed-efect parameters, namely the relaxed stifness (A1) and force-level dependency (A0) (with p=0.0006 and p<0.0001, respectively). To further emphasize stifness change due to hip angle, random ef- fects are presented as the relative percentage of fxed efects (bi0/A0 and bi1/A1). Furthermore, the acquired percentages were further subtracted by random efect percentages estimated at 55◦ hip angle in order to present stifness change with respect to 55◦ hip angle. As shown in Fig.3.6d, stifness changes with posture and reached statistically signifcance at 15◦ degree hip angle (tested with two tailed Wilcoxon rank sum test with Bonferroni correction). On the other hand, Fig.3.6c shows that the force-level dependency (A0) changed inconsistently due to posture and does not reach statistical signifcance at any specifc hip angle.

The same investigation was carried out on the estimated viscosity. All subjects had an increased viscosity with applied force (with a mean slope of 0.19Nm s/rad, presented in Fig.3.6b). How- ever, only 42% cases passed the F-test, indicating that the viscosity change due to the applied limb force may be insignifcant. Additionally, the identifed mixed efect models showed low prediction accuracy and a limited data variance explained by the model (with mean R2=0.35 lower than the stifness model prediction with R2=0.79) despite the inclusion of random efects. It is, therefore, unclear whether hip joint viscosity exhibits similar force-level dependency or whether the identifed trend was merely due to noise.

Posture dependency. A second investigation used a model assuming that stifness changes 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity65

a b 600 40 500 400 30 300 20 200 10 StiffnessF(Nm/rad) 100 ViscosityF(Nm*s/Frad) 0 0 -20 -10 0 10 20 -20 -10 0 10 20 ForceFlevelF(N) ForceFlevelF(N) c d Δb0/A0 Δb1/A1 <0.001 80 80 ** 40 40 0 0 40 40 PercentageF(%) ** 80 80 ** 15253545 15253545 AngleF(degree) AngleF(degree)

Figure 3.6: Violin plots showing the probability density of force-level dependency and how it changes due to hip angle. The dashed lines indicate the least square ftted force dependency. Panels (a) and (b) show how hip stifness and viscosity changes with the applied force. Panel (c) and panel (d) shows the infuence of hip angle upon force-level dependency. The infuence is presented as random efects (bi0 and bi1) and specifcally in the percentage of fxed efects (A0 and A1). Additionally, it is presented as changes with respect to hip angle 55◦ in order to examine changes from a specifc hip angle. Within each violin plots, a cross indicates the median value of the respective violin plot and a square the mean value. Random efects are found to change the identifed force-level dependency (A0) inconsistently and does not reach statistically signifcant at any hip angle. On the other hand, relaxed stifness (A1) is found to change with hip angle and confrmed to be statistical signifcant by two tailed Wilcoxon rank sum test and corrected by Bonferroni correction. 66 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED quadratically with hip angle, and the applied limb force may infuence this quadratic relation:

2 2 Kjk = A2θjk + A3θjk + A4 + bj2θjk + bj3θjk + bj4 (3.7)

where Kjk is the k-th stifness estimated at j-th limb force, θjk the k-th hip angle at j-th limb force (where 0 degree refers to the angle while standing straight), A2, A3 and A4 the fxed efects describing how stifness changes with posture, bj2, bj3 and bj4 the random efects relating the infuence of applied limb force upon the identifed posture-stifness relation. Note that a quadratic function of the position was used to best catch the larger stifness at 25◦ and 15◦.

The identifed fxed efect parameters indicated that most legs exhibit an inverse relationship between measured stifness and hip angle, as presented in Table 3.3 in combination with F- test results. In other words, it was found within our experiment range {15-55◦} that hip joint stifness would increase with the decrease of hip angle. However, the identifed posture dependency was less infuential compared to the previously identifed force-level dependency, as the linear regression model without random efects showed a low estimation accuracy (mean R2 = 0.16) and required random efects that consider applied limb forces (mean R2 = 0.65). The importance of force-level dependency was consolidated by theoretical likelihood tests (where 95% cases passed with p<0.05), and suggested that applied limb force is a stronger infuencing factor in comparison with hip angle.

The same process was repeated on estimated viscosity with Equ.3.7. The identifed models showed poor prediction accuracy and explained limited variance of data (mean R2=0.28) with 65% of the models failed the F-tests (indicating no posture dependency).

Force and posture dependency. Based on the aforementioned test results, we further hypothesised that stifness changes according to both applied limb force and hip angle, with each factor possibly afecting the other one:

′ ′ 2 ′ Kijk = A0Fik + A2θjk + A3θjk + bi0Fik (3.8) 2 ′ ′ +bj2θjk + bj3θjk + A5 + bi1 + bj4 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity67

p < 0.001 p = 0.018 1

0.8

0.6 2 R

0.4

0.2

0 force-level posture combined dependency dependency dependency

Figure 3.7: Model prediction accuracy comparison. Prediction accuracy is presented as R2 and compared between all three models. It is shown that both models that considers force-level dependency performed a better prediction (tested with two tailed Wilcoxon rank sum test). On the other hand, the combined model improves estimation accuracy, however, did not reach a statistical signifcant (with p = 0.3579).

where Kijk is the k-th stifness estimated at i-th hip angle and j-th limb force, Fik the k-th applied limb force at i-th hip angle, θjk the k-th hip angle at j-th limb force (0 degree refers to standing straight), A0’, A2’, A3’ and A5 are the fxed efects describing how stifness changes with applied force and posture; bi0’ and bi1’ the random efects relating the infuence of hip angle upon identifed parameters; bj2’, bj3’ and bj4’ the random efects relating the infuence of applied limb force upon the identifed parameters. The proposed model is the combination of previous models, and similar notations were used to allow comparison with previously identifed parameters.

Interestingly, the newly identifed fxed efects exhibited values similar to previous fndings.

Stifness was again found to increase with applied limb force, with slopes (mean A0’=4.98) close to previous values (mean A0=5.15). By calculating the diferences between both values,

′ 83% cases showed diferences less than 10% (calculated by A0 − A0/A0). Meanwhile, most subjects were again found to exhibit a negative relation between stifness and hip angle, and are presented in Table 3.3 along with F-test results. These fndings imply that the identifed 68 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED force-level and posture dependencies coexist.

The estimated generalised linear models, which refers to models without random efects, were shown to predict hip joint stifness of all subjects’ legs with acceptable variance being explained (mean R2=0.68, with standard deviation of 0.16). The model can be further improved by including random efects (mean R2=0.84, with standard deviation 0.09, 92.5% cases passed F-tests). This fnding demonstrates the importance of correlation among parameters (e.g. hip angle changing force-level dependency). On the other hand, random efects (bi1’ and bi4’) which afect the constant value (A5) are shown to decrease since both posture and force-level dependencies are considered in this model.

The model prediction accuracy of all three models is presented in Fig.3.7.

3.2.4 Discussion

We performed a systematic experimental investigation of hip viscoelasticity using NED, a novel rigid robotic interface dedicated to lower limb neuromechanics studies. A position displacement was used as a mechanical perturbation, that enabled us to obtain an accurate estimation of hip stifness. Viscosity was computed in a second step using a least-square minimization of the linear second order model. The relatively large perturbation amplitude ensured a reliable estimation despite large force measurement noise. We also analysed the infuence of the leg, posture, force level and perturbation direction on stifness and viscosity estimates. The dominant and non- dominant legs exhibited similar values of viscoelasticity, which may not be surprising as the legs are mostly used for the symmetric walking. Sports activities such as playing football might induce some asymmetry, although this could not be studied with the available population.

Stifness was found to be slightly larger when estimated from displacement applied in the posterior direction than in the anterior direction. This is probably due to stronger or larger muscles since stifness is known to vary proportionally to the cross-sectional area of a stretched muscle (Gonzalez et al., 1997), and the quadriceps femoris may be larger than the biceps femoris (Wickiewicz et al., 1983). The study (Koopman et al., 2016) estimated hip and knee 3.2. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity69

Table 3.3: Statistics of linear regression and mixed efect models Estimate Standard deviation Stifness: Perturbation direction dependency Y = 0.74X + 45.29,R2 = 0.717 Intercept 45.29 5.26 Slope 0.74 0.02 Stifness: Force level dependency 2 Kik = A0Fik + A1 + bi0Fik + bi1, mean R = 0.79 A0 [m/rad] 5.15 0.98 A1 [Nm/rad] 169.39 39.61 bi0/A0 0 19.93% bi1/A1 0 14.24% Identifed dependencies: 100% cases found force-level dependency Stifness: Posture dependency 2 2 Kjk = A2θjk + A3θjk + A4 + bj2θjk 2 +bj3θjk + bj4, mean R = 0.65 2 A2 [Nm/(rad degree )] 0.042 0.073 A3 [Nm/(rad degree)] -3.72 5.30 A4 [Nm/rad] 302.75 104.06 bj2/A2 0 21.41% bj3/A3 0 6.1% bj4/A4 0 14.19% Identifed dependencies: 20% cases failed F-tests, showing no posture dependency 5% cases showed positive posture dependency 75% cases showed negative posture dependency Stifness: Posture and force-level dependency ′ ′ 2 ′ 2 Kijk = A0Fik + A2θjk + A3θjk + bi0Fik + bj2θjk ′ ′ 2 +bj3θjk + A5 + bi1 + bj4, mean R = 0.84

A0’ [m/rad] 4.98 1.34 2 A2’ [Nm/(rad degree )] 0.036 0.070 A3’ [Nm/(rad degree)] -3.26 5.11 A5 [Nm/rad] 234.00 102.24 bi0’/A0’ 0 15.97% bi1’/A5 0 13.11% bj2’/A2’ 0 31.46% bj3’/A3’ 0 5.02% bj4’/A5’ 0 0.74% Identifed dependencies: 100% cases found force-level dependency 7.5% cases failed F-tests, showing no posture dependency 20% cases showed positive posture dependency 72.5% cases showed negative posture dependency 70 Chapter 3. Hip joint neuromechanics evaluation with cable driven robot NED multi-joint viscoelasticity using an exoskeleton, but could not study the infuence of applied force systematically. Using the dedicated NED interface, we could systematically analyse the infuence of posture and applied force on the single-joint viscoelastic parameters in a controlled manner. We found that stifness increases monotonically with the applied limb force, with a relation consistent with previous measurements in the upper limb (Burdet et al., 2013). The stifness value was found to be slightly infuenced by the hip angle, as was previously found in the ankle (Mirbagheri et al., 2000). The viscosity exhibited no clear dependency upon perturbation direction or hip angle, and slightly increases with the applied limb force. The difculty in identifying viscosity dependencies may originate from its low value relative to stifness.

The obtained viscoelasticity values we have observed with our subjects population are in the same order as reported in previous studies, although such comparison is limited by the fact that viscoelasticity depends on the individuals. In (Zhang et al., 1998), it was found that knee joint stifness in the relaxed condition is around 75Nm/rad and viscosity is about 2Nm s/rad, and both of these factors increase with muscle contraction. The values we obtained for the hip joint are larger (with stifness values between 75-318Nm/rad and 2-21Nm s/rad under relaxed condition), as expected as larger muscles are involved. Using the LOPES exoskeleton perturb- ing the whole leg and indirect position measurement from the serial elastic actuators used in LOPES, (Koopman et al., 2016) found stifness values between 50-220Nm/rad and viscosity between 0.5-10Nm s/rad. While being in the same order of magnitude, the diference with the values we have obtained may be in part due to the older population of that study with ages between 67-72 while our young adults were between 21-27. Chapter 4

Subject-specifc modelling and evaluation

After a spinal cord injury (SCI), patients’ active motor functions would degrade and passive resistive force of the limbs would increase thus afecting locomotions. Knowing the neurophysiological condition of an individual afected by a neurological disease can help designing a control for assistive exoskeletons increasing safety and considering the subject-specifc characteristics. Using rough modelling could lead to over-supported walking without using the patients’ residual motor functions, and potentially accelerate muscle atrophy. It is therefore demanded to develop a subject-specifc model to describe SCI subjects’ dynamics for exoskeleton application purpose. This promoted me to develop such a model capturing the typical neuromechanical characteristics of SCI.

This chapter introduces this subject-specifc model incorporating typical biomechanical characteristics often found to change after SCI. This proposed model contains only measurable parameters and is expressed in joint space thus allowing its use for the control of exoskeletons. A series of single-joint inverted pendulum simulations are developed to test the subject-specifc modelling under diferent scenarios and validated it, including balance recovery from both force perturbation or body tilt. The model is further tested with human measurement data in particular regarding its ability to provide balance. All work related to subject-specifc model was

71 72 Chapter 4. Subject-specifc modelling and evaluation written in a work package of the EU-FP7 grant ICT-611626 SYMBITRON.

4.1 Literature review

The spinal cord carries out various functions including transmitting ascending and descending neural signals between the motor/sensory cortex and body, implementing motor refexes and coordinating central pattern generators (Dietz and Harkema, 2004; Silva et al., 2014). A spinal cord injury can cause a temporary or permanent loss in muscle strength and body sensa- tions, and cause difculties in daily activities. In this section, neuromusculoskeletal alterations caused by a SCI are frst introduced along with their relevance to neuromuscular modelling (in Section 4.1.1). Afterwards, existing human models and their limitations are discussed in Section 4.1.2. A Hill’s type muscle model will serve as a basis for the proposed subject-specifc model described in the upcoming Section 4.2. Additionally, an inverted pendulum model presented in Section 4.1.3 will be used for validations and simulations described in Section 4.3.

4.1.1 Physiological alterations following spinal cord injury

Skeletal muscle fbres within humans are responsible for conducting muscle contractions to form active motions. When a descending signal from the central nervous system is delivered via the motor neurons to innervate muscle contractions, the protein flaments which locates inside the skeletal muscle fbre slide alongside each other to produce a muscle contraction. The type of skeletal muscle fbre defnes the motor capability of a muscle group and is generally separated into Type I and II muscle fbres. Type I is known to generate a slow but lasting muscle contraction force (commonly used during standing) while Type II can produce strong and fast-twitch but may fatigue quickly (commonly used in short sprinting).

Various research into histochemistry had revealed that a high percentage of paralysed muscle fbres would experience transformation from Type I and become predominantly composed of Type II muscle fbres a few months after SCI (Burnham et al., 1997; Lotta et al., 1991; Mc- 4.1. Literature review 73

Donald et al., 2005; Shields, 2002). This observation was also confrmed by other physiological assessments including fatigue index and contractile speed variance check (Shields, 2002). Due to this muscle type modulation, patients’ muscle fatigues easily and are less likely to hold weight for a longer period, as needed during level walking. Furthermore, research has revealed that such muscle transformation is irreversible after reaching a new steady-state of the body condition, in which the period varies from weeks to months (Biering-Sørensen et al., 2009; Shields, 2002; Dudley-Javoroski and Shields, 2008). Therefore, it is important to perform related physical therapy soon after an injury to increase the usage of muscle fbres and lower the percentage of transformation. Another musculoskeletal deterioration commonly happens after SCI is muscle atrophy, in which breakdowns and re-absorbs the muscle tissues a few weeks after the injury and thus causing a decrease in the mass and force production of the afected limb (Shields, 2002; Dudley-Javoroski and Shields, 2008). The impact of muscle type modulation and muscle atrophy infuences the motor performance and can be divided into two categories in neuromuscular modelling, one relates to the active muscle strength and the other to the resistance to stretch.

The active muscle strength is often depicted by the force-length and force-velocity relations, which relate the maximum tension to the muscle length and the change in muscle length, respectively. When referring to limb motions or joint rotations, it is commonly described as torque-angle and torque-angular velocity relation that considers the contributions of all muscles towards the motion. Both of these relationships were reported to change after a SCI. Through electrical stimulation experiments, it has been shown the force-length relation of ankle plantar fexor muscles was diferent between SCI and healthy subjects (Pelletier and Hicks, 2010). However, the dorsifexor muscle showed no signifcant diference between SCI and healthy subjects. A similar observation was reported with a torque-angle relation experiment measured using a motor-driven dynamometer (McDonald et al., 2005), with results suggesting that plantar fexor muscles torque-angle relation tended to shift towards plantarfexion direction while no diference was found in dorsifexor muscles. Through electrical stimulation experiments, it has been shown that the force-velocity relations among SCI subjects were diferent and could be easily afected by muscle fatigue (Sinclair et al., 2006; Fornusek et al., 2007). Due to the 74 Chapter 4. Subject-specifc modelling and evaluation evidence of the changes in muscle performance, it is expected that further measurements should be conducted with SCI subjects while developing neuromuscular models for them. Common modelling approaches using healthy subjects’ data are no longer applicable.

The restoring force in response to muscle stretch was also found to be afected by SCI, i.e. passive joint resistive moments were found to increase after SCI (Amankwah et al., 2004; Riener and Edrich, 1999). This restoring force is commonly defned as joint viscoelasticity that includes both joint stifness and viscosity, where joint stifness can be further categorised into three contributions, i.e. passive muscle stifness, active muscle stifness, and neurally mediated

Figure 4.1: Central nervous system. The central nervous system is composed of the brain and the spinal cord. The main function of the spinal cord in the central nervous system is to transmit the descending neural signal from the motor cortex to the body and the sensory feedback from the body to the sensory cortex. It is also responsible for many motor refex actions and coordination central pattern generator. Photo adapted from: https://www.christopherreeve.org/living-with-paralysis/health/how- the-spinal-cord-works 4.1. Literature review 75 refex stifness. Passive muscle stifness results from the elasticity of the tendon, connective tissues and muscle cross-bridges to resist bending or stretching. Refex stifness describes the increase in muscle stifness due to neural feedback with a delay of a few tens or hundreds of milliseconds after an external perturbation. Active muscle stifness, on the other hand, relates to the increase of stifness from muscle contraction. When a muscle contracts, the number of cross-bridges would increase and become more resilient to muscle stretches. Separating these three stifness types may enable us to characterise individual neurophysiological changes (Lieber et al., 2004) and to reach an optimised exoskeleton motion. For instance, Mirbagheri et al. (2001) and Lorentzen et al. (2010) found that SCI individuals have higher overall joint stifness in comparison to healthy subjects, and such abnormality strongly depends on the joint angle. Additionally, the contribution proportion between passive, active and refex stifness to overall joint stifness was found diferent in SCI subjects, where refex stifness has a larger contribution. Therefore, understanding the individual contribution would allow a better prediction in subject dynamics, and thus a better exoskeleton control.

Another important neurological factor is spasticity, which afects muscle behaviour and prompts an increase in mechanical resistance to stretch. Spasticity, which can be characterised by a velocity-dependent increase in tonic stretch refexes combined with intense tendon jerks, could occur after SCI and result in a resistive force opposing to the limb movement. Spasticity originates from the loss of motor neuron inhibition after SCI and could alter the muscle fbre composition and distribution (Lotta et al., 1991). It was shown that a spastic muscle has larger passive stifness compared to a healthy muscle (Sinkjaer and Magnussen, 1994; Sinkj´eret al., 1993; Lieber et al., 2004). Additionally, the co-activation and reciprocal activation patterns of the residual muscle were also modifed by the lesion (Levin et al., 2000). A series of theoretical explanations of how spasticity may relate to an alteration of the refex threshold was also developed (Mullick et al., 2013). By examining the literature on spasticity, it is suggested that an additional model is required in order to recognise and prevent damaging the subject if spasticity appears. However, despite the importance to model spasticity, it is not included in our subject-specifc model due to time constraint. 76 Chapter 4. Subject-specifc modelling and evaluation

4.1.2 Hill’s type muscle model

A forward dynamic model depicts how humans control their muscles based on physiological observations. To move the body, a descending command signal propagates from the central nervous system (CNS) to the muscles, resulting in muscle activation and a contraction force. A simple phenomenological muscle-tendon model often used to describe the complex activation process is Hill’s type model (Hill, 1938). Hill’s type muscle model is based on representing a muscle as a combination of mechanical springs and active contractile elements, as shown in Fig.4.2. In general, a force produced by muscles results in a limb fexion and a joint torque for the limb can be calculated by multiplying the force with the corresponding moment arm. The following equation specifes a Hill’s type muscle-tendon model:

Fm lm tendon Ft

t tendon F m φ muscle Fp fibre Fm lt/2

mt l M FA lm,vm

Figure 4.2: Hill’s muscle-tendon model. Hill’s muscle-tendon model describes the physical behaviour of the muscle-tendon unit as a series of spring and contractile unit (Hill, 1938).

Activation Force Torque Central Musculo-skeletal Equation Muscleb Nervous Geometry of Model System (momentbarms) Motion

Sensory θ,bθ,bθ Feedback

Figure 4.3: Simplifed forward dynamics for human motion. This is a simplifed forward dynamic to describe a motion generated in a human body described by Buchanan et al. (2006). A muscle command generates from the central nervous system to activate the muscle contraction in the model. The muscle then creates a muscle contractile force, and torque is calculated by including muscle-skeletal geometry. Eventually, the motion is generated and the sensory organs within the body provide the feedback to the central nervous system. 4.1. Literature review 77

t m m m F = Fm = [FmaxFA−L(l )FA−V (v )A(t) + FmaxFP −L(l )]cos(ϕ), (4.1)

t where F is the force exerted from the tendon to the joint, Fm is the force generated from

m the muscle fbre, Fmax is the maximum force that a muscle can generate, FA−L(l ) is the force-length relation which defnes the normalised active force capability depending on the

m muscle length, FA−V (v ) is the force-velocity relation which defnes the normalised active force capability depending on the rate of change of muscle length (muscle velocity), A(t) is the muscle

m activation and FP −L(l ) is the normalised passive resistive force which would depend on the muscle length, ϕ is the pennation angle which defnes the angle between the line of muscle fbre and the muscle-tendon unit.

Many models make use of Hill’s type muscle mechanics due to its simplicity (Buchanan et al., 2006; Lloyd and Besier, 2003; Anderson et al., 2007). Interestingly, the model complexity can be decided depending on the modelling needs. By restricting the modelling under a specifc workspace, the number of muscles can be reduced while remaining satisfactory result. On the other hand, more muscles could be incorporated to increase the variability of the model. For example, in Sartori et al. (2015), an EMG-driven musculoskeletal model with 13 Hill’s type muscles was constructed to predict the joint stifness under dynamic motion, although a Hill’s type model is not equipped to model the active stifness. Another example used six muscles to predict both knee and ankle torque from representative EMG data (Olney and Winter, 1985). An extended model such as the refex-based neuromuscular gait model driven by oscillatory systems known as central pattern generators (CPG) was developed to simulate complex full- body motion (Geyer and Herr, 2010).

A disadvantage of Hill’s type muscle modelling is that most of the parameters are only available from invasive measurements done on cadavers (Manal et al., 2006; Hawkins and Hull, 1990; McConville et al., 1980) while parameters from non-invasive measurements with subjects are usually limited (Manal et al., 2013; Maganaris, 2000). Therefore, it is sometimes hard to unfold the actual physiological meaning of each parameter. Muscle morphometric alteration (moving from Type I muscle to Type II) would be a signifcant barrier to the creation of a 78 Chapter 4. Subject-specifc modelling and evaluation

SCI forward dynamic model based on the neuromuscular model (Buchanan et al., 2006). For example, Sinclair et al. (2006) attempted to build a muscular-skeletal model from cadaver data (or numerical functions, which was also estimated from cadaver data) to match the motion recording of SCI subjects. It is shown that the ftted model can only generate ftted motion, and cannot be generalised to diferent dynamics. This fnding suggested that cadaver data is not suitable for SCI subject modelling.

Similarly, it is a complicated task to ft experimental data to the model simulation due to its redundancy. It was shown that a musculoskeletal model of a human lower extremity contains essential interdependency among parameters and it would be difcult to match the resulting motion by varying only a few parameters (Hoy et al., 1990). By matching the physical measurement with one specifc parameter, it would then lose track of the others. Eventually, the endpoint trajectory of the simulation could be matched the measurement result; however, this doesn’t imply that individual physiological parameters are still valid. Therefore, Hill’s type based modelling would need to be modulated before applying to a subject-specifc modelling for SCI patients.

Considering the above factors, I developed a subject-specifc model by frst reducing the redundancy in Hill’s type muscle modelling approach using reasonable assumptions such that all parameters can be identifed during well-defned experiments. This modifcation can consider the force-length and force-velocity alteration described above while preserving essential muscle features. Additionally, stifness estimation experiments can be conducted such that the modulation of passive properties can be incorporated into the model for better understanding of human-robot interaction. The detailed model is described in Section 4.2.

4.1.3 Inverted pendulum model

Another tool that I used for my investigations is an inverted pendulum model, which is widely used to model human upright standing (Loram and Lakie, 2002; Milton et al., 2009; Kuo, 1995; Jiang and Kimura, 2007). An inverted pendulum model refers to a pendulum system with its centre of mass above its revolute joint. The model contains similar unstable behaviour 4.1. Literature review 79 to humans since two-thirds of the human body weight is located at approximately two-thirds of their height. Therefore, by considering an ankle as the revolute joint, which includes a joint stifness and viscosity, and simplifes the body into a point mass located at two-thirds of their height, it is possible to simulate balancing behaviour of a human subject (Winter, 1995; Winter et al., 1998; Morasso and Schieppati, 1999; Winter et al., 2001; Morasso and Sanguineti, 2002). For example, Engelhart et al. (2016) conducted investigations on human balancing capability with an inverted pendulum model under external perturbation to distinguish between the control strategy of diferently aged persons.

As illustrated in these works, the unstable dynamics of the inverted pendulum model provided the opportunity to investigate the balance performance of a SCI subject without asking the subject to perform actual perturbation experiments, such as the stepping experiment conducted by Mille et al. (2003). Therefore, an inverted pendulum model was incorporated to simulate a human standing task with the subject-specifc modelling, which is detailed in this chapter. A series of simulations including various unstable conditions were tested to demonstrate the potential of subject-specifc modelling in examining stability performance of human standing.

Inverted pendulum y Fext Fext

τ Fg x

Figure 4.4: Inverted pendulum model. A man tilting can be described as an inverted pendulum model with the ankle joint described as a pivot joint while the mass of the body is at a distance above the ground. 80 Chapter 4. Subject-specifc modelling and evaluation

4.2 Subject-specifc modelling

As described in Section 4.1, Hill’s type model is one of the most commonly used muscles biomechanic model. However, the identifcation of its parameters is challenging, as it requires invasive measurements from cadavers. In particular, it is difcult to measure musculoskeletal characteristics such as the muscle force-length relationship. Therefore the model presented in this section considers the efects of the Hill’s model but in joint space, where they can be identifed in well-designed experiments.

4.2.1 Model description

Basic components of the proposed model consist of groups of agonist and antagonist muscles acting on a joint. For example, the fexion and extension of the ankle joint in the sagittal plane can be simplifed into a pair of muscle groups, as shown in Fig.4.5. The forces from all extensor muscles are grouped and modelled as a single unit, and the fexor muscles are grouped as another muscle group. In order to capture the length and velocity dependency of the Hill’s muscle-tendon model in joint space, the force produced by the muscle groups in the model is

a Inverted pendulum b Block diagram torque-angle torque-angular velocity Fenv dependency y active τmax_flex θ flexion τflex Fg extensor τ +τ - joint dynamics env g + muscle + I, B, K, muscle extension τext activation

τ ,τ flexor muscle flex ext τ active max_ext x

Figure 4.5: Sketch and block diagram of the single-joint subject-specifc modelling. Panel (a) is the sketch of the inverted pendulum model which is adapted with the subject-specifc model parameter. Panel (b) is the respective block diagram describing the subject-specifc modelling of the ankle joint. The active torque is generated separately by both fexor and extensor muscle and further combined with all external torques. The resulting net torque would result in the ankle dynamics and move the pendulum. 4.2. Subject-specifc modelling 81 characterised with torque-angle and torque-angular velocity dependencies. The active torque generated by each muscle group is modelled by the following neuromechanical characteristics:

˙ τmax τact = ⌈ τl(θ)τv(θ)A(t) ⌋ (4.2)

where active torque component τact (unit: Nm), which can also be described as an active torque ˙ constraint, is limited by the torque-angle τl(θ) and torque-angular velocity τv(θ) dependencies. ˙ Both τl(θ) and τv(θ) are nomalised parameters that depend on the musculoskeletal geometry (i.e. the moment arms) and the force-length and force-velocity relationship of the muscle. A(t) is the normalised muscle activation measured at the major representative muscle (i.e. Tibialis Anterior refers to the major dorsifexor of the ankle joint) ranging between zero and one. ⌈a⌋b denotes that a is limited to a fxed range [0; b] representing the capacity of muscle force generation τmax (units: Nm). The torque produced by both the fexor and extensor muscle are modelled similarly.

Joint impedance, which refers to the passive resistive torque due to joint displacement, is the second component of the model. It is composed of inertia(I), viscosity (B) and stifness (K), and originates from the passive resistive force generated due to muscle and tendon elongation and the number of the cross-bridges. τpass is defned as follows:

¨ ˙ τpass = Iδθ + Bδθ + Kδθ, (4.3) with I is the inertia, B is the joint viscosity and K as joint stifness.

After including two muscle groups and the joint impedance, the basic format of modelling a single joint is completed. It remains to include any external force. Note that it would be possible to incorporate additional joint refex components not considered in previous modellings.

A human upright standing case with the proposed subject-specifc model is illustrated in Fig.4.5a. The body dynamics is modelled as an inverted pendulum and the ankle joint as a revolute joint. The stabilising torque results from both the plantar fexor muscle (Soleus and 82 Chapter 4. Subject-specifc modelling and evaluation

Gastrocnemius) and the dorsifexor muscle (Tibialis Anterior) of the ankle. The equation of motion of the system is:

τ ≡ −τext + τflex + τpass + τg + τenv, (4.4)

where τext and τflex are active extension and fexion torque generated by the corresponding muscle groups and τg, τenv are the gravitational torque and the torque accounting for external forces, such as a push perturbation. The body dynamics are characterised by inertia I, viscosity B and stifness K. A block diagram of the model is depicted in Fig.4.5b.

It should be noted that the modelling approach proposed here contains two important features, which are measurability and subject-specifc, and requires few assumptions. First, the measured active torque component τact represents the collective contribution of all muscle-tendon components acting on each joint. This assumption yields a simplifed structure with fewer parameters compared to traditional Hill’s type muscle-based modelling. Furthermore, physiological parameters are incorporated within existing parameters, for example the pennation angle of the muscle fbres (usually described as a function of muscle fbre length and muscle optimal fbre length in Hill model), was not included in the proposed subject-specifc model. I assumed that the length/angle related components within the subject-specifc modelling can fully embody the feature of pennation angle under a normal dynamic condition. This assumption allows further simplifcations and results in measurable parameters only. Only by applying the assumption listed in this paragraph could the subject-specifc modelling to be valid.

4.2.2 Method to identify the model’s parameters

One important feature for the proposed subject-specifc modelling is that all parameters are measurable. The following section describes how the parameters can be estimated in a few measurement experiments.

¨ ˙ (1) Identifcation of the passive joint component. τpass(θ) = Iδθ + Bδθ + Kδθ, 4.2. Subject-specifc modelling 83

The structure of the model enabled us to separate the identifcation process into both passive and active conditions. To begin with, the joint dynamics (I, B and K) in the equation should frst be identifed during a relaxed condition(τact(A(t) = 0) = 0). The joint impedance can be identifed by performing a controlled position or force perturbation and recording the interaction between the actuated interface and the subject. For example, the endpoint stifness value can be estimated during the plateau period of a perturbation Burdet et al. (2000), after which the viscosity and inertia can also be determined, as was described in the previous chapter. It is important to notice that the proposed subject-specifc modelling is not restricted to any specifc impedance identifcation methods.

(2) Isometric experiment for τA−L(θ) and τmax identifcation

The torque-angle dependency τA−L(θ) and the maximum torque capacity τmax can be identifed by a maximum voluntary contraction experiment (MVC) under static condition (θ˙ = 0). The subject’s leg can be fxed at a specifc joint orientation while the subject produces maximum possible torque with EMG signal recorded in parallel. In this isometric condition, the torque- angle dependency can be identifed and normalised according to the maximum torque τmax on diferent joint angles.

˙ (3) Isokinetic experiment for τv(θ)

After the joint impedance identifcation and isometric experiment, the torque-angular velocity ˙ dependency τv(θ) can be identifed by an isokinetic experiment. The subject would perform constant joint angular-velocity motion with the assistance of metronome or music. A set of diferent angular velocities should be tested within the subject’s ROM to map out the decrease of muscle torque through the increase of velocity. During the motion, resistive torque would be increased until the subject could not withstand the velocity command.

After the experiments listed above, the primary form subject-specifc model is created. One advantage of the subject-specifc modelling approach is that it can be easily scaled up to a multi-DoF system by adding additional revolute joints. However, it should be noted that identifcation experiments should be done sequentially from the distal joint to the proximal 84 Chapter 4. Subject-specifc modelling and evaluation joints since it would be impossible to perform identifcation without the interference of other joints.

4.3 Model evaluation using a single-joint inverted pen-

dulum simulation

To validate the modelling approach, I developed a human standing simulation model with an inverted pendulum model which has a single revolute joint at the ankle, with the sketch as described in Fig.4.5. This section aims to determine whether the proposed modelling approach can capture the balancing behaviour of both healthy and SCI subjects. In order to do so, a simulation was frst conducted to perturb both a healthy and a SCI subject model away from the natural posture and observe the dynamic during recovery. The model parameters were collected from relevant literature sources (detailed in Section 4.3.1), and the model was simulated under a collective of scenarios of force perturbations and posture tiltings with diferent evaluation methods. Furthermore, the model was tested with one healthy subject perturbation measurement.

4.3.1 Model parameter

The inverted pendulum model with Equation (4.4) was used in simulations with the following parameters: M=70 kg, both healthy and SCI joint impedance B and K were collected from (Mirbagheri et al., 2001) and (Mirbagheri et al., 2000). Constant values of B=1.1Nms/rad, K=325Nm/rad were used for both healthy and SCI at the beginning of the simulations for simplicity, which is specifcally referring to the result of Section 4.3.3. Afterwards, the joint angle dependent joint impedance equations collected from Mirbagheri et al. (2001) were used in further experiments including sensitivity and robustness analysis. Torque-angle τl(θ) and ˙ torque-angular velocity τv(θ) dependency of dorsifexor and plantar fexor muscle were adopted from McDonald et al. (2005) and Sinclair et al. (2006), respectively. I used the healthy subject 4.3. Model evaluation using a single-joint inverted pendulum simulation 85 torque-angular velocity data reported in Sinclair et al. (2006) for both healthy and SCI subject since measurement for SCI subjects was not available. Both the torque-angle dependency and the torque-angular velocity dependency were plotted in Fig.4.6a-c which includes both healthy and SCI subjects measurements. Maximum torque capacity τmax were adopted from Pelletier and Hicks (2010). To model a muscle activation A(t), which is the active control provided by the central nervous system for upright body stabilisation, a PD controller (proportional and derivative) was used with the P and D value of 1400 and 400 respectively. The controller gains were identifed using Matlab from the balance recovery time series reported in Kuo (1995).

As described in Section 4.1.1, spinal cord injury subjects sufer from neuromuscular changes and muscle force-length relations would shift away from natural. In order to mimic diferent severities of spinal cord injury, an interpolation between healthy torque-angle dependency and a b Torque-angle dependency (Plantarflexor) Torque-angle dependency (Dorsiflexor) 100 100 Able Able 80 SCI 80 SCI 60 60 40 40

Normalized3TorqueC3f 20 Normalized3TorqueC3f 20 Dorsiflexion Dorsiflexion I40 I20 0 20 40 I40 I20 0 20 40 Joint3angle3bdeg5 Joint3angle3bdeg5 c Torque-angular velocity dependency d Muscle Torque-angle dependenct shifing 1 10f3SCI 100 Dorsiflexor 50f3SCI 100f3SCI 80 0.5

60 0 I50 0 50 40 1 Normalized3TorqueC3f 20 0.5 Normalized3relation Plantarflexor 0 100 200 300 0 Velocity3bdeg/s5 I50 0 50 Joint3angle3bdeg5

Figure 4.6: Normalised torque-angular velocity and torque-angle dependency of both able-bodied and SCI subjects. Panel (a), (b) and (c) are the normalised dependencies of both able-bodied and SCI subjects. Data is collected from McDonald et al. (2005) and Sinclair et al. (2006) which describes the muscle capability at diferent joint angle and joint angular velocity. Panel (d) shows the torque-angle dependency shifting which simulates diferent severity levels of SCI subjects. An interpretation of 10%, 50% and 100% SCI subject’s torque-length dependency was carried by interpolation and linear combination. The lower the injury level was, the closer the curve would be with the healthy subject’s data, as in this case dorsifexor and plantar fexor muscle would appear to the right of the graph. 86 Chapter 4. Subject-specifc modelling and evaluation

SCI torque-angle dependency was performed. Fig.4.6d demonstrates diferent torque-length dependency under various simulated level of injury. For example, a low-level SCI condition was imitated by combining 90% of healthy torque-angle dependency and 10% SCI torque-angle dependency. Since both relations were a normalised curve, no extra weight would be needed. A severe condition might have more infuence from the SCI torque-angle dependency, while a mild condition can be mimicked by having more infuence from healthy data. This mimicking procedure enables the understanding of torque-angle dependency shifting after SCI and the infuence upon standing stability.

4.3.2 Scenario description and evaluation methods

The goal of this simulation was to investigate whether the body can be stabilised in an upright position for diferent destabilisation factors and neuromechanical parameters in both healthy and SCI cases. Two simulation scenarios were considered: balance recovery after a force per-

Table 4.1: Values of the model’s parameters and relations Name Unit mass 70 kg height 1.80 m moment of inertia 12.64 kg*mˆ2 τmax (dorsi, able) 43.15 Nm τmax (plantar, able) 118 Nm τmax (dorsi, SCI) 14.17 Nm τmax (plantar, SCI) 22 Nm K (SCI) 754.51x2 + 472.13x + 116.44 Nm/deg K (able) 436.52x3 + 678.51x2 + 307.70x + 76.20 Nm/deg B (SCI) 55.34x5 + 38.37x4 − 5.54x3 − 4.23x2 + 1.96x + 1.08 Nm*s/deg B (able) −3.36x3 − 0.62x2 + 1.21x + 0.64 Nm*s/deg −7 4 −6 3 −4 2 τl (dorsi, able) −3.36 × 10 x + 2.90 × 10 x + 5.07 × 10 x − 1.86 × 10−2x + 0.36 −7 4 −6 3 −5 2 τl (plantar, able) −1.44 × 10 x − 9.47 × 10 x − 4.37 × 10 x + 2.06 × 10−2x + 0.68 −7 4 −5 3 −4 2 τl (dorsi, SCI) −8.35 × 10 x − 2.79 × 10 x + 6.18 × 10 x − 5.74 × 10−3x + 0.08 −7 4 −5 3 −3 2 τl (plantar, SCI) −4.83 × 10 x − 5.56 × 10 x − 1.52 × 10 x + 1.94 × 10−2x + 0.94 −6 2 −3 τv (able) 3 × 10 x˙ − 3.00 × 10 x˙ + 1.00 −6 2 −3 τv (SCI) 3 × 10 x˙ − 3.00 × 10 x˙ + 1.00 4.3. Model evaluation using a single-joint inverted pendulum simulation 87 turbation and balance recovery from a non-zero initial condition (tilted body). Lateral force perturbations of 200ms ranging from 100 to 600N, were applied to the centre of mass in the frst scenario. In non-zero tilted body scenario, the initial tilts were in a range of -60◦ to 60◦.

The stabilisation was defned successful if the simulated body was able to recover balance within 3sec from the time the perturbation was applied (lateral force impulse) or from the time the simulation starts (non-zero initial body tilt). Furthermore, the system was defned to be stable if the body was within the 5◦ of upright orientation and its absolute angular velocity was less than 5◦/s. The maximum sway angle during the simulation was also recorded to compare the behaviour of each simulation.

To understand the infuence of each biomechanical component of the model to upright stabilisation performance, I conducted simulations with step by step inclusion of the parameters in Section 4.3.3. First, the model was tested without any active muscle constraints. In other

Stability margin

F Unstable

Stable Force Perturbation Initial

θ Position -1 (rad) -0.5 0.5

-200

-300

-100

100

200 300

Force Perturbation (N)

Initial tilt

Figure 4.7: Stability margin description. The stability margin defnes the conditions which the simulated human could return to upright position. One of the axes refers to the size of force perturbation and the other axis refers to the amplitude of the initial tilting angle of the body. The maximum deviation angle during the sway is also recorded and coloured in the fgure. Darker colour refers to smaller sway and is defned to be more stable. 88 Chapter 4. Subject-specifc modelling and evaluation

˙ words, the parameter of τl(θ), τv(θ) and τmax were excluded. Then, muscle mechanics were sequentially included to investigate the infuence of coefcients. The torque-angle and torque- ˙ angular velocity dependencies (τl(θ), τv(θ)) were inserted in the second part of the simulation, while the full model was tested in the third part of the simulation. Since the torque-angular velocity dependency shown a small infuence on the result, it was included in the same time with torque-angle dependency. Full results were described in Section 4.3.3.

In order to defne the balancing performance of the human model under diferent scenarios, a criterion named ”stability margin” was defned and plotted in Fig.4.7. The stability margin defnes the conditions in which the simulated body can return to an upright position. One of the axes refers to the size of force perturbation and the other axis refers to the initial tilting angle of the body. The maximum sway angle was recorded in each simulation and coloured in the fgure. The colour is darker if the sway is smaller. This criterion was used in all simulations.

Additionally, to simulate the change of stability margin in SCI subject after wearing a commercial exoskeleton, an additional simulation was carried (shown in Fig.4.10c.2) under a few assumptions. First, the exoskeleton was assumed to support its own weight without transfer- ring any additional burden to the subject. Second, the only control strategy on the exoskeleton was to support the weight of the subject, which was set to reduce 60% body weight. This weight removing control strategy was set as an example of robotic-assisted motion, without over-complicating the simulation scenarios with advanced controller design such as trajectory prediction or oscillator designs.

To identify the most infuential parameter in model stability, the stability margin was investigated when the parameters independently varied within 0-200% of their nominal values. The results were shown in Section 4.3.4 and demonstrated that the simulation enabled us to identify the infuencing parameter under each condition. The efect of torque-angle dependency shifting was also described by sequentially moving the torque-angle dependency to diferent SCI severity level, where the residual parameters remain identical (healthy data). A random noise (of 10% of signal magnitude) was incorporated at the muscle activation of both fexor and extensor muscle groups with the aim to evaluate model consistency and robustness due to 4.3. Model evaluation using a single-joint inverted pendulum simulation 89 muscle command noise.

To further test the validity of the subject-specifc model, I conducted a balance experiment with a gait rehabilitation exoskeleton robot named LOPES III (Meuleman et al., 2016). The aim ˙ of this test was to examine whether the active muscle constraints (τl(θ), τv(θ) and τmax) could allow a better estimation of human dynamics (with the measured human balancing dynamics) in comparison to models without muscle constraints.

The experiment set-up is described in Fig.4.8. One subject was recruited to be connected to LOPES III while various forces are given at the pelvis of the subject. The subject was asked to maintain a relaxed upright posture with his hand crossed in front of his chest. As shown in Fig.4.8, he was asked to stand naturally without extensively co-contracting lower limb muscles to prevent falling. The subject could hold a handle or step if he were unable to maintain upright standing, and the specifc data would be marked and discarded. Also, an emergency button was always within his reach in case of need. The amplitude of the perturbation force varied between 25 to 175N in both directions with an interval of 25N. Both the force amplitude and the time interval between each perturbation were randomised. In total, each perturbation size was repeated for fve times in one cycle and three cycles were performed. Both the position and

F LOPES

Figure 4.8: Balance experiment with LOPES. A balance experiment was conducted in a gait rehabilitation robot named LOPES III which was developed by Meuleman et al. (2016). The subject was asked to maintain upright standing while a force was given by the robot to perturb the subject. 90 Chapter 4. Subject-specifc modelling and evaluation the force data were recorded and further converted into a rotation of the ankle (with respect to the centre of mass) and the respective force perturbation at the centre of mass.

After measuring the force and angle data, two single-joint inverted pendulum models with healthy data (also constructed with Equation (4.4)) were used to simulate the motion. All parameters were similar to previous sections and listed in Table 4.1 where the constant values of B=1.1Nms/rad, K=325Nm/rad were used. The diference between the two models was that one contains the active muscle constraints, which includes the torque-angle, torque-angular ˙ velocity dependencies (τl(θ) and τv(θ) ) and the maximum torque capacity (τmax), while the other did not. In other words, the model which contains no active muscle constraints was close to a simple inverted pendulum with no torque limit.

All measured data were grouped by the amplitude and direction. In comparison to previous simulations, the PD controller which models the muscle activation was varied in this experiment to fnd the best trajectory ftting under each force perturbation condition. For example, Fig.4.9 is a plot with all perturbation with the size of 125N and one PD value was ftted according to this specifc condition, which is presented as the solid line. By ftting a PD controller at each force condition allows us to compare between two models base on their best performance. Furthermore, each ftted model was further tested with diferent conditions in order to show if the model characteristics were transferable. The comparison function goodnessOfFit and the cost function NRMSE was selected in Matlab to compare between simulated trajectory and measured data. A value of 1 refers to a perfect ft suggested by NRMSE, where -1 refers to a bad ft.

4.3.3 Human balance simulation with subject-specifc modelling

Fig.4.10a presents the body sway when a lateral force perturbation is applied (at t=0.5 s, between ±400N, duration 200ms). Each curve in Fig.4.10a describes a simulated body sway to a given condition. The upright body corresponds to 0◦, while positive tilt values correspond to leaning forward. The left panel of Fig.4.10a shows the body tilt when Equation (4.4) was used ˙ with no neuromuscular constraints (without τl(θ), τv(θ), and τmax). In this case, the balance 4.3. Model evaluation using a single-joint inverted pendulum simulation 91 recovers in less than 1 second regardless of the perturbation magnitude. The central plot of Fig.4.10a describes the body deviation when the torque-angle and torque-angular velocity ˙ dependency (τl(θ) and τv(θ) ) were introduced to the model. As in the frst case, the system was able to recover from diferent magnitudes of perturbation, however, it took almost twice of the time and the tilt deviations were larger due to the introduction of position and velocity relations. Lastly, the right plot in Fig.4.10a shows the balancing results when all model constraints were introduced. This includes an additional infuence by torque capability. In this case, the model was stable only to a limited amplitude of force perturbations compare to the previous cases. As demonstrated, the gradual addition of the muscle biomechanics restrains the balancing capability.

The results of balance recovery from diferent non-zero initial body tilt (-40◦ to 40◦) are shown in Fig.4.10b. The only diference between Fig.4.10a and Fig.4.10b was that instead of a force perturbation, an initial tilting angle was given to the system to test balance recovery. The balance was recovered within 1 second regardless of the initial tilting angle in the left panel of Fig.4.10. After including the torque-angle and torque-angular velocity relations to the model, the required time to regain balance was longer as it is shown in the central plot of Fig.4.10b.

0.08 NeuromuscularpModelling-p125pN modelpoutput measuredpdata 0.06

0.04

0.02 Deviationp(rad) 0

-0.02 0 0.5 1 1.5 Timep(sec)

Figure 4.9: Time series data of both the perturbation experiment conducted with LOPES and model simulated output. The dotted line refers to the measurement data of the perturbation experiments while the solid line refers to simulation results. All data of diferent force perturbations (125 N forward force in this specifc fgure) are grouped and simulation is matched to have the best ft base with these data. 92 Chapter 4. Subject-specifc modelling and evaluation

Lastly, the right plot of Fig.4.10b is the result while maximum torque capacity was introduced. Only in a few conditions could the model maintain balance. Both force perturbation and

a Body deviation from vertical (force perturbation) full6model -400N no6constraints torque-position/velocity -300N dependency -200N increasing -100N perturbation 100N 200N 16 300N leaning6forward 400N

perturbation applied 16s

b Body deviation from vertical (initial tilt angle) -40 -30 no6constraints torque-position/velocity full6model dependency -20 40 increasing 40 40 -10 tilt6angle 10 20 20 20 20 30 40 leaning6forward 0 0 0

-20 -20 -20 Initial6tilting -40 -40 -40 16s 16s 16s

c Ability to maintain balance for different angles and perturbations c.1 Healthy6S6SCI c.2 SCI6with660&6weight6support forward stable 200

SCI6stability -200 region Perturbation,6N max.6deviation

leaning6forward leaning6forward

-60 -30 030 60 -30 030 60 unstable Initial6angle Initial6angle

Figure 4.10: Simulation results of balance maintaining task. Panels (a) and (b) are the time series of the body sway angle when neuromechanical constraints were added sequentially under force perturbation and initial non-zero tilt angle conditions, respectively. Panel (c) demonstrates the stability margin of diferent subjects, including healthy, SCI and SCI with exoskeleton weight supported, under various condition. 4.3. Model evaluation using a single-joint inverted pendulum simulation 93 initial angle tilting task demonstrated similar behaviour on muscle parameters limiting balance recovery.

Fig.4.10c summarised the balance recovery obtained from both aforementioned tests. The colour maps are the testing result of diferent combinations of initial angular deviations (horizontal axis) and force perturbations (vertical axis).

The simulation showed that a healthy subject could recover the balance from -10◦ (leaning backwards) to 20◦ (leaning forward), and for perturbation levels of -100N (push back) to 300N (push forward). The asymmetric performance could originate from the diferent size in both represented muscle groups as Gastrocnemius is usually larger than the Tibialis Anterior. The condition for a SCI subject was shown as a white point in the same fgure without having the capability to withstand diferent force perturbations or angular deviations when all neuromechanical constraints were included, suggesting that the shifting of torque-angle relation or the decrease in maximum torque restricted the dynamics of the simulated SCI subject. In addition, Fig.4.10c.2 showed the simulation result of a SCI subject with the exoskeleton providing 60% weight support. As shown in the plot, larger stability margin along force disturbance direction was spotted, implying that weaker motor functions of SCI subjects could still be manifested if properly supported by actuated exoskeletons.

4.3.4 Parameter sensitivity analysis and robustness to muscle noise

Following the experiments described in the previous section, I investigated how the overall stability is afected by the variation of parameters. The parameters were independently varied within 0-200% of their nominal values and Fig.4.11 describes a representative stability margin obtained in this study. A subset of points and ranges of the stability margin were scrutinised to compare the contribution of individual parameters. As shown in Fig.4.11a, points along both the x and y-axis were selected in order to demonstrate the maximum permissible perturbation force and maximum initial tilt from which balance recovery was possible. The results are shown in Fig.4.12a. The range of the stability margin in both angle and perturbation were collected to compare the stability contribution in the multi-parameter sensitivity analysis (shown in 94 Chapter 4. Subject-specifc modelling and evaluation

Fig.4.13). The total area of the stability margin, which is shown in Fig.4.11b, was used to further compare the infuence of parameters upon all simulated conditions with results shown in Fig.4.13.

The propagation of the purple square point, which refers to the maximum perceivable backward force perturbation, and the blue circle point, which refers to the maximum backward tilting,

a Stability margin

Rangeyofyinitial Rangeyofyforce angleybefore perturbation forward falling beforeyfalling 200

-200

Perturbation,yN Angleyofyleaning backwards Backwardyforce beforeyfalling perturbationybeforeyfalling

leaningyforward -30-60 0 6030 Initialyangle

b Area of stability margin forward 200

Areayofystability 0 margin

-200 Perturbation,yN

leaningyforward -30-60 0 6030 Initialyangle

Figure 4.11: Specifcation used for stability evaluation. Panel (a) shows the edges of the stability margin in both initial angle and force perturbation conditions, which were collected to examine the infuence of single parameter upon balancing performance (shown in Fig.4.12a). Furthermore, the range of the stability margin in both angle and perturbation were collected to compare the stability contribution in multi-parameter sensitivity analysis (shown in Fig.4.13). Panel (b) shows the area of the stability margin that was used to expressed as a percentage of the entire simulation region, which was also used in Fig.4.13 to imply the global efect of individual parameters. 4.3. Model evaluation using a single-joint inverted pendulum simulation 95 are shown in Fig.4.12a as examples. As shown in Fig.4.12a.1, among the tested parameters, stifness most afects the balance recovery capability under the tilting test. While the parameter was at 200% of its nominal value, it increases the stability margin by 8.6◦, in other words, 150% extra from the initial condition. On the other hand, increasing maximum torque capacity (τmax ) increases the stability margin, as an increase of 124% was observed when the max joint torque was lifted up to 200% of the normal value. Fig.4.12a.2 shows the result of the backward force perturbation experiment and the maximum torque capacity was the most afecting parameter. When the maximum torque capacity was at 200% of the nominal value, the model can withstand perturbation up to 200% of the original amplitudes. The stifness can only increase the stability of the system by 33%. Joint viscosity demonstrates a limited efect on both scenarios.

a Changes of stability margin a.1 Angle of leaning backwards before falling MaximumTorque Stiffness 5 Damping 0

-5 Changesb,bdeg

a.2 Backwards force perturbation before falling 100

-100 Changesb,bN 0 50 100 150 200 Percentagebofbnormal,b%

b Stability margin due to torque-angle dependency

10 200

0 0

-200 -10 Initialbangleb,bdeg 100 75 50 25 0 perturbationb,bN 100 75 50 25 0 PercentagebofbSCIbcondition,b% PercentagebofbSCIbcondition,b%

Figure 4.12: Single parameter sensitivity analysis of the modelling approach. Panel (a) describes the changes in the stability margin in both angle and force perturbations under the efect of each parameter. Panel (b) demonstrates the efect of torque-angle dependency on the stability margin. 96 Chapter 4. Subject-specifc modelling and evaluation

The infuence of diferent levels of SCI conditions as described in the methods are plotted in Fig.4.12b. When the model carries 100% of SCI subject torque-angle dependency, it could not withstand any perturbation or initial tilt despite other parameters being taken from healthy subject cases. In order to maintain balance under diferent initial tilt angles, more than 50% of healthy torque-angle relation required. A 75% healthy torque-angle was essential when the force is pushing the subject backwards.

Next, I investigated the change in stability margin due to the infuence of diferent parameters simultaneously, in order to determine the importance of individual parameters. The range of the stability margin in both x- and the y-axis was recorded as described in Fig.4.11. In comparison to previous simulations, this evaluation incorporated a joint-angle-dependent joint impedance model to better describe the joint impedance variation upon diferent postures. The results are presented as colourmaps in Fig.4.13 where the darker the colour, the larger the stability margin (increased initial tilt or higher force perturbation). The horizontal axis defnes the level of SCI with 0% corresponding to healthy case and 100% corresponding to SCI case.

As demonstrated in Fig.4.13a, the balance recovery from an initial tilted angle was more difcult than a force perturbation under upright posture. The increase of the SCI level results in a decrease of stability margin, such that the balance recovery became impossible when the SCI level was higher than 50% even if the maximum joint torque capacity was increased by two folds. The highest stability margin in Fig.4.13a was observed around 140% of maximum torque capacity. This demonstrates that the human balancing task under diferent initial angles does not only depend on muscle strength. Fig.4.13b, on the other hand, shows a clear correlation between robustness to force perturbation and maximum torque generation capability. For both healthy and lower severity level of SCI cases, higher torque capacity enables resistance to higher perturbation forces.

In Fig.4.13c, the colour corresponds to the normalised area circumscribed by the stability margin that indicates the overall stability to diferent force perturbations and non-zero initial tilts. Examining the gradient of the fgure indicates that the SCI condition has no remarkable efect on the model stability if the maximum torque was less than 60% of normal. On the 4.3. Model evaluation using a single-joint inverted pendulum simulation 97 other hand, the SCI condition would become the dominant factor for stability maintenance when the maximum torque was larger than 160% of its original value. The example of the area calculation is shown in Fig.4.11b.

To ensure that the obtained modelling results are consistent, I performed a set of simulations with white noise added to the muscle activation of the model. For diferent conditions, I

a Range of initial angle before falling 200 25

100 Percentage of

maximum torque, % 0 0

b Range of force perturbation before falling 200 450N

100 Percentage of

maximum torque, % 0 0N

c Stability margin in compare to total area 200 10%

100 Percentage of

maximum torque, % 0 0% 0 25 50 75 100 Percentage of SCI condition, %

Figure 4.13: Multi-parameter sensitivity analysis. The changes in stability margin caused by diferent parameters are demonstrated through the specifcation listed in Fig.4.11. Panel (a) describes the range of initial angle before falling. Panel (b) plots the tolerable force perturbation before falling. Panel (c) plots the stability margin in comparison to the total area. It is presented in percentage of total area. 98 Chapter 4. Subject-specifc modelling and evaluation computed the maximum force perturbation and maximum initial tilt from which the balance recovery was possible. Each simulation condition with added noise was performed 10 times, and the mean and maximum deviation of the maximum perturbation force and initial tilt was shown in Fig.4.14a and Fig.4.14b for maximum perturbation force and maximum initial tilt respectively. As shown, the addition of noise to the model did not change the results signifcantly, and the infuence of SCI conditions and torque capability on balance recovery performance remained the same. The average relative diference in maximum force was 2.5% and for the maximum initial tilt was 1.8%.

a Range of force perturbation before falling 500 0% SCI 20% SCI 400 40% SCI 300 60% SCI 80% SCI 200 100% SCI Perturbation,CN 100

0 50 100 150 200 PercentageCofCmaximumCtorque,C%

b Range of initial angle before falling

25 0% SCI 20% SCI 20 40% SCI 60% SCI 15 80% SCI 100% SCI 10

InitialCangle,Cdeg 5

0 25 50 75 100 PercentageCofCSCICcondition,C%

Figure 4.14: Noise resulted in fuctuation on multi-parameter sensitivity analysis. These are the crosscutting of Fig.4.13 which displays the noise resulted variance which is presented with error bar. Panel (a) is the crosscutting of Fig.4.13a and b are the crosscutting of Fig.4.13b. 4.3. Model evaluation using a single-joint inverted pendulum simulation 99

a Subject-specific model 175N

0 Comparison data

-175N b Model without muscle constraint 175N

0 Comparison data

-175N -175N 0 175N Fitted condition c Best fit of individual condition 0.7 Subjec-specific model Model without muscle constraint

0 Fit result (NRMSE)

-0.7 -175N 0 175N

Figure 4.15: Comparison between subject-specifc modelling and simple inverted pendulum model. Perturbation experiment data was collected and used as a reference for both the subject-specifc model and a simple inverted pendulum model. Both models were best ftted at a specifc force perturbation, which refers to the x-axis, and then compared to diferent perturbation conditions, which refers to the y-axis. Panel (a) refers to the result of the subject-specifc modelling while panel (b) refers to the pure inverted pendulum model. The results are described in NRMSE. Panel (c) describes specifcally the results at best-ftted condition of both models, in other words, the diagonal of both plots. 100 Chapter 4. Subject-specifc modelling and evaluation

4.3.5 Balance experiment with LOPES

As shown previously, Fig.4.8 presented measurements with LOPES III under 125N perturbation force and the simulated subject-specifc model output. Despite the variance of measured data, the simulated model could capture the body sway trajectories and simulate accordingly. Fig.4.15a and Fig.4.15b are the collected NRMSE under the mixed test with x-axis referring to the ftted condition and y-axis referring the comparison condition. As shown in Fig.4.15a, the subject-specifc model contains a distinct dependence on the force perturbation size, which was shown by the higher fttings (lighter colour) among both diagonal lines and drastically dropped in other conditions. This may imply that the subject-specifc model was sensitive in respective ftting conditions or incapable of adapting to diferent conditions. On the other hand, the inverted pendulum model (shown in Fig.4.15b) has a uniform performance with a minor improvement at both diagonal lines, which was not surprising since none of the parameters contains angle or angular-velocity dependent characteristics.

Fig.4.15c demonstrates the best ft result of each representative force. In other words, the diagonal line of both above fgures from the lower left to the upper right. Despite the fact that both models presented the best ft for each individual case, the subject-specifc model presented a better performance. This suggests that the subject-specifc modelling approach captured natural muscle features and could display human dynamics during a balancing task.

4.3.6 Discussion

A subject-specifc model to investigate spinal cord injury subjects’ balance capability was presented and evaluated in this chapter, with potential application toward tailored exoskeleton controller design thanks to the model structure. In contrast to the Hill’s type muscle based modelling, the subject-specifc model allows parameters to be measured through simple non-invasive experiments (as described in Section 4.2.2) with the prospective to accommodate neurophysiological modulations happened after SCI. Through the evaluation of the simulated subject- specifc model (with both healthy and SCI subject data collected from literature), various 4.3. Model evaluation using a single-joint inverted pendulum simulation 101 important features that correspond to human balance recovery behaviours. For example, the asymmetric balance recovery performance depicted in Fig.4.10 could originate from the higher force capacity of the plantar fexor muscle group with respect to dorsifexor muscles, potentially indicating physiological relativeness of the model in describing diferent motor capacity; A higher maximum torque capacity value allows a stronger resistance to rapid force perturbation (shown in Section 4.3.4), which could be extendedly discussed with similar fndings in balance measurements among both young and healthy elderly subjects (Engelhart et al., 2016) (where elderly were found to have longer neuro-delays and weaker muscles to resist perturbations).

The subject-specifc model combined with the stability margin could be used to examine human balance performance under diferent scenarios without asking the patient to carry out perturbation experiments. As found in Section 4.3.4, a stronger joint stifness may allow the subject to further counteract unbalanced terrains, which is equivalent to body tilting, while higher torque capacity may prevent falling due to external perturbations. Both fndings could support the development of a tailored exoskeleton control base on individuals need, and the model robustness was shown in the noise analysis test. The perturbation experiments conducted with LOPES III further demonstrated the advantage of subject-specifc modelling to predict balance recovery movement in comparison to the model without muscular constraints. Chapter 5

Conclusion

Understanding how the human central nervous system controls the body requires accurate neuromechanics investigations carried out using robotic interfaces to apply controlled interactions. In particular, knowing the lower-limb neuromechanics is critical to identify the diferences between healthy and pathologically afected individuals, and to develop efcient control strategies for gait assistive devices. In this context, this thesis has 1) developed and validated a novel robotic interface dedicated to investigate the lower-limb neuromechanics; 2) it has used it to identify for the frst time hip joint neuromechanics while investigating applied force and posture dependencies; 3) and developed a computational model to study balance integrating parameters due to neurological afections such as spinal cord injury. Let us now discuss each of these contributions below.

The neuromechanics evaluation device (NED) that was designed and evaluated in Chapter 2 can apply a variety of dynamic interactions to the participant’s leg in both static condition and during movement, and enables the systematic investigation of the hip and knee joints’ neuromechanics. NED’s design based on a closed mechanical mechanism using a cable transmission allows voiding to move the heavy actuator, and to place it outside the workspace. In comparison to exoskeletal robotic interfaces that have been used to investigate the lower-limb neuromechanics, NED enables subjects to carry out natural movements without joint constraints. Additionally, NED can be used to perform both hip and knee measurements with the

102 103 same mechanism, thus enabling a homogeneous characterisation of these joints. Importantly, NED’s supports natural upright postures makes it suitable for carrying out investigations on patients with weak motor functions. These features were illustrated through the investigations of Chapter 2.

Rigidity and controllability are key characteristics requested from neuromechanics measurement devices, which were indeed addressed by the NED mechanism. NED showed to be capable of achieving a 2cm displacement at the foot within 230ms with negligible vibration. This smooth and rapid motion allows the measurement of restoring force during limb displacement, which further yields an impedance estimation. The perturbation amplitude thus its duration could be lowered if the position measurement is improved. While the force measurement was carried out at the fxture, the position measurement stemmed from the motor encoder, thus is sufering from the cable deformation. An improved position measurement could however be provided using an external system e.g. based on optical motion capture. By upgrading the position feedback hardware, NED may perform other lower-limb neuromechanics investigations such as multi-sine torque perturbation for joint neuromechanics under various speed (Koopman et al., 2016), and rapid limb displacement that allows the separation of muscle refexes contributed stifness change (Mirbagheri et al., 2000, 2001).

The versatility of the novel device to carry out various neuromechanics investigation experiments was illustrated in two experiments reported in Chapter 3: The estimation of maximum voluntary hip torque, and the measurement of hip joint mechanical impedance. The frst experiment showed that human subjects were capable of producing torque up to 250Nm in hip extension, with an angle dependency consistent with the simulation of Anderson et al. (2007). On the other hand, hip joint stifness values are found to vary from 75 to 200Nm/rad under a relaxed condition, which is similar to fndings from Koopman et al. (2016). Additionally, stifness was found to vary with perturbation direction, limb force and joint angle, which is consistent with previous measurements carried out on other joints (Burdet et al., 2013; Mirbagheri et al., 2000). These results demonstrate that NED is an efective tool for hip and knee joint biomechanical identifcation, ready for single-joint experiments including impedance at static posture or during movement, as well as to investigate multijoint refexes. 104 Chapter 5. Conclusion

Many other neuromechanics investigations could be conducted with the developed NED, such as to study: the infuence of biarticular muscles on the viscoelasticity in multiple joints. Using NED together with a knee brace, we can keep the knee joint at diferent angles and measure the resulting hip viscoelastic values. This may allow a better understanding of limb viscoelasticity during diferent gait phases; The size of perturbation may infuence the measured viscoelasticity and are originated by the number of detached cross-bridges. Therefore, investigating the infuence displacement amplitude may enable the understanding of musculoskeletal properties, and can serve as an index for evaluating the pathological conditions of neurologically afected individuals.

On the other hand, the systematic investigation of hip joint viscoelasticity modulation may be used to develop the control of assistive robots. For example, it was shown that the joint stifness monotonically increases with applied limb force. A dynamic joint stifness prediction model may be created by correlating muscle activation with resulting limb force, and further with resulting joint stifness. Since muscle activation occurs a few tens of milliseconds before generating contraction force, we could therefore predict joint stifness before actually experiencing it. This may allow efcient robotic control and prevent interference in body movement.

Another contribution of the present thesis consists of the development of a subject-specifc modelling for lower limb joints’ biomechanics. Compared to other Hill’s type muscle-based modelling approaches (Sartori et al., 2015; Olney and Winter, 1985; Lloyd and Besier, 2003), the new model used a reduced number of parameters providing only key physiological characteristics: the torque-angle and angular-velocity dependency, the joint impedance and any external forces exerted on the model. Importantly, all of these parameters can be identifed with NED following simple identifcation protocols. A set of simulation studies were conducted with the developed model. These simulations demonstrate how neuromechanical constraints infuence the overall postural stability of both healthy and SCI human subjects. It was also shown that changes in the torque-angle dependency of the joint biomechanics afect the capability to perform successful balance functions. The results obtained through modelling can be used in the development of subject-specifc controllers for postural assistance robotic exoskeletons by adjusting the robotic assisting forces to a subject-specifc biomechanical profle. Bibliography

M N Akman, R Bengi, M Karatas, S Kilin¸c,S S¨ozay, and R Ozker. Assessment of spasticity using isokinetic dynamometry in patients with spinal cord injury. Spinal Cord, 37(9):638–43, 1999.

Kof Amankwah, Ronald J Triolo, and Robert F Kirsch. Efects of spinal cord injury on lower-limb passive joint moments revealed through a nonlinear viscoelastic model. Journal of Rehabilitation Research and Development, 41(1):15–32, 2004.

Dennis E. Anderson, Michael L. Madigan, and Maury A. Nussbaum. Maximum voluntary joint torque as a function of joint angle and angular velocity: Model development and application to the lower limb. Journal of Biomechanics, 40(14):3105–13, 2007.

T P Andriacchi, G B Andersson, R W Fermier, D Stern, and J O Galante. A study of lower-limb mechanics during stair-climbing. The Journal of Bone and Joint Surgery. American volume, 62(5):749–57, 1980.

D. Aoyagi, W.E. Ichinose, S.J. Harkema, D.J. Reinkensmeyer, and J.E. Bobrow. A robot and control algorithm that can synchronously assist in naturalistic motion during body-weight- supported gait training following neurologic injury. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 15(3):387–400, sep 2007.

Bo Biering-Sørensen, Ida Bruun Kristensen, Michael Kjaer, and Fin Biering-Sørensen. Muscle after spinal cord injury. Muscle and Nerve, 40(4):499–519, 2009.

Kathleen A. Bieryla, Dennis E. Anderson, and Michael L. Madigan. Estimations of relative efort during sit-to-stand increase when accounting for variations in maximum voluntary

105 106 BIBLIOGRAPHY

torque with joint angle and angular velocity. Journal of Electromyography and Kinesiology, 19(1):139–44, 2009.

Lori A. Bolgla and Timothy L. Uhl. Reliability of electromyographic normalization methods for evaluating the hip musculature. Journal of Electromyography and Kinesiology, 17(1):102–11, 2007.

Thomas S Buchanan, David G Lloyd, Kurt Manal, and Thor F Besier. Neuromusculoskeletal modeling: estimation of forces and joint moments and movements from mrasuerments of nueral command. Journal of Applied Biomechanics, 20(4):367–95, 2006.

E Burdet, R Osu, D W Franklin, T Yoshioka, T E Milner, and M Kawato. A method for measuring endpoint stifness during multi-joint arm movements. Journal of Biomechanics, 33(12):1705–9, 2000.

Etienne Burdet, David W. Franklin, and Theodore E. Milner. Human robotics : neuromechanics and motor control. MIT Press, 2013.

R Burnham, T Martin, R Stein, G Bell, I MacLean, and R Steadward. Skeletal muscle fbre type transformation following spinal cord injury. Spinal Cord, 35(2):86–91, 1997.

Tina L. Claiborne, Mark K. Timmons, and Danny M. Pincivero. Test - retest reliability of cardinal plane isokinetic hip torque and EMG. Journal of Electromyography and Kinesiology, 19(5):e345–52, 2009.

Jo˜aoBreno de Araujo Ribeiro Alvares, Rodrigo Rodrigues, Rodrigo de Azevedo Franke, Bruna Gon¸calves Cordeiro da Silva, Ronei Silveira Pinto, Marco Aur´elioVaz, and Bruno Manfredini Baroni. Inter-machine reliability of the Biodex and Cybex isokinetic dynamometers for knee fexor/extensor isometric, concentric and eccentric tests. Physical Therapy in Sport, 16(1): 59–65, 2015.

Karin L. De Gooijer-Van De Groep, Erwin De Vlugt, Jurriaan H. De Groot, H´el`eneCm Van Der Heijden-Maessen, Dennis Hm Wielheesen, Rietje S. Van Wijlen-Hempel, J Hans Arendzen, and Carel Gm Meskers. Diferentiation between non-neural and neural contributors to ankle BIBLIOGRAPHY 107

joint stifness in cerebral palsy. Journal of NeuroEngineering and Rehabilitation, 10(1):81, 2013.

V. Dietz and Susan J. Harkema. Locomotor activity in spinal cord-injured persons. Journal of Applied Physiology, 96(5):1954–60, 2004.

Johannes Dirnberger, Alexander K¨osters, and Erich M¨uller.Concentric and eccentric isokinetic knee extension: A reproducibility study using the IsoMed 2000-dynamometer. Isokinetics and Exercise Science, 20(1):31–5, 2012a.

Johannes Dirnberger, Hans-Peter Wiesinger, Alexander K¨osters,and Erich M¨uller. Repro- ducibility for isometric and isokinetic maximum knee fexion and extension measurements using the IsoMed 2000-dynamometer. Isokinetics and Exercise Science, 20(3):149–53, 2012b.

Shauna Dudley-Javoroski and Richard K Shields. Muscle and bone plasticity after spinal cord injury: review of adaptations to disuse and to electrical muscle stimulation. Journal of Rehabilitation Research and Development, 45(2):283–96, 2008.

D. Engelhart, J. H. Pasma, A. C. Schouten, R. G. K. M. Aarts, C. G. M. Meskers, A. B. Maier, and H. van der Kooij. Adaptation of multijoint coordination during standing balance in healthy young and healthy old individuals. Journal of Neurophysiology, 115(3):1422–35, 2016.

Ildar Farkhatdinov, Julia Ebert, Gijs Van Oort, Mark Vlutters, Edwin Van Asseldonk, and Etienne Burdet. Assisting human balance in standing with a robotic exoskeleton. IEEE Robotics and Automation Letters, 4(2), 2019.

Ch´eFornusek, Peter J. Sinclair, and Glen M. Davis. The force-velocity relationship of paralyzed quadriceps muscles during functional electrical stimulation cycling. Neuromodulation, 10(1): 68–75, 2007.

Atsushi Fujii, Shohei Oda, Satoshi Komada, and Junji Hirai. A muscle tension estimation method by using mechanical impedance of human knee joint during training by manipulator. In Mechatronics,IEEE International Conference on Mechatronics, pages 1–6. IEEE, 2007. 108 BIBLIOGRAPHY

Hartmut Geyer and Hugh Herr. A muscle-refex model that encodes principles of legged mechanics produces human walking dynamics and muscle activities. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 18(3):263–73, 2010.

Roger V. Gonzalez, Thomas S. Buchanan, and Scott L. Delp. How muscle architecture and moment arms afect wrist fexion-extension moments. Journal of Biomechanics, 30(7):705–12, 1997.

Daniel Hahn, Matthias Olvermann, Jan Richtberg, Wolfgang Seiberl, and Ansgar Schwirtz. Knee and ankle joint torque-angle relationships of multi-joint leg extension. Journal of Biomechanics, 44(11):2059–65, 2011.

D Hawkins and M L Hull. A method for determining lower extremity muscle-tendon lengths during fexion/extension movements. Journal of biomechanics, 23(5):487–94, 1990.

A. V. Hill. The heat of shortening and the dynamic constants of muscle. Proceedings of the Royal Society B: Biological Sciences, 126(843):136–95, 1938.

N Hogan. The mechanics of multi-joint posture and movement control. Biological Cybernetics, 52(5):315–31, 1985.

M G Hoy, F E Zajac, and M E Gordon. A musculoskeletal model of the human lower extremity: the efect of muscle, tendon, and moment arm on the moment-angle relationship of musculotendon acutators at the hip, knee and ankle, 1990.

Hsien-Yung Huang, Arash Arami, Ildar Farkhatdinov, Domenico Formica, and Etienne Burdet. The infuence of posture, applied force and perturbation direction on hip joint viscoelasticity.

2019a. URL http://arxiv.org/abs/1908.07800.

Hsien-Yung Huang, Ildar Farkhatdinov, Arash Arami, Mohamed Bouri, and Etienne Burdet. Cable-driven robotic interface for lower limb neuromechanics identifcation. 2019b. URL

http://arxiv.org/abs/1908.02689.

Yifa Jiang and Hidenori Kimura. Balance-keeping control of upright standing in biped human BIBLIOGRAPHY 109

beings and its application for stability assessment. In Humanoid Robots: New Developments. I-Tech Education and Publishing, jun 2007.

H Kawamoto, S Taal, H Niniss, T Hayashi, K Kamibayashi, K Eguchi, and Y Sankai. Voluntary motion support control of Robot Suit HAL triggered by bioelectrical signal for hemiplegia. In Annual International Conference of the IEEE Engineering in Medicine and Biology, pages 462–466. IEEE, 2010.

R E Kearney, R B Stein, and L Parameswaran. Identifcation of intrinsic and refex contributions to human ankle stifness dynamics. IEEE Transactions on Bio-Medical Engineering, 44(6): 493–504, 1997.

Bram Koopman, Edwin H. F. van Asseldonk, and Herman van der Kooij. Estimation of human hip and knee multi-joint dynamics using the LOPES gait trainer. IEEE Transactions on Robotics, 32(4):920–32, 2016.

A D Kuo. An optimal control model for analyzing human postural balance. IEEE Transactions on Bio-Medical Engineering, 42(1):87–101, 1995.

Mindy F. Levin, Ruud W. Selles, Martine H G Verheul, and Onno G. Meijer. Defcits in the coordination of agonist and antagonist muscles in stroke patients: Implications for normal motor control. Brain Research, 853(2):352–69, 2000.

Richard L. Lieber, Suzanne Steinman, Ilona A. Barash, and Hank Chambers. Structural and functional changes in spastic skeletal muscle, 2004.

David G Lloyd and Thor F Besier. An EMG-driven musculoskeletal model to estimate muscle forces and knee joint moments in vivo. Journal of Biomechanics, 36(6):765–76, 2003.

Ian D Loram and Martin Lakie. Human balancing of an inverted pendulum: position control by small, ballistic-like, throw and catch movements. The Journal of Physiology, 540(Pt 3): 1111–24, 2002.

Jakob Lorentzen, Michael J. Grey, Clarissa Crone, Dominique Mazevet, Fin Biering-Sørensen, and Jens B. Nielsen. Distinguishing active from passive components of ankle plantar fexor 110 BIBLIOGRAPHY

stifness in stroke, spinal cord injury and multiple sclerosis. Clinical Neurophysiology, 121 (11):1939–51, 2010.

S Lotta, R Scelsi, E Alfonsi, A Saitta, D Nicolotti, P Epifani, and U Carraro. Morphometric and neurophysiological analysis of skeletal muscle in paraplegic patients with traumatic cord lesion. Paraplegia, 29(4):247–52, 1991.

Daniel Ludvig, Maciej Plocharski, Piotr Plocharski, and Eric J. Perreault. Mechanisms con- tributing to reduced knee stifness during movement. Experimental Brain Research, 235(10): 2959–70, 2017.

Lars L¨unenburger, Gery Colombo, Robert Riener, and Volker Dietz. Clinical assessments performed during robotic rehabilitation by the gait training robot Lokomat. In Proceedings of the IEEE International Conference on Rehabilitation Robotics, pages 345–8, 2005.

Constantinos N Maganaris. In vivo measurement-based estimations of the moment arm in the human tibialis anterior muscle-tendon unit. Journal of Biomechanics, 33(3):375–9, 2000.

Kurt Manal, Dustyn P Roberts, and Thomas S Buchanan. Optimal pennation angle of the primary ankle plantar and dorsifexors: variations with sex, contraction intensity, and limb. Journal of Applied Biomechanics, 22(4):255–63, 2006.

Kurt Manal, Justin D. Cowder, and Thomas S. Buchanan. Subject-specifc measures of Achilles tendon moment arm using ultrasound and video-based motion capture. Physiological Reports, 1(6):e00139, 2013.

Laura Marchal-Crespo and David J Reinkensmeyer. Review of control strategies for robotic movement training after neurologic injury. Journal of NeuroEngineering and Rehabilitation, 6(1):20, 2009.

John T McConville, Charles E Clauser, Thomas D Churchill, Jaime Cuzzi, and Ints Kaleps. Anthropometric relationships of body and body segment moments of inertia, 1980.

Michael F. McDonald, M. Kevin Garrison, and Brian D. Schmit. Length-tension properties of BIBLIOGRAPHY 111

ankle muscles in chronic human spinal cord injury. Journal of Biomechanics, 38(12):2344–53, 2005.

Jos Meuleman, Edwin Van Asseldonk, Gijs Van Oort, Hans Rietman, and Herman Van Der Kooij. LOPES II - Design and evaluation of an admittance controlled gait training robot with shadow-leg approach. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 24(3):352–63, 2016.

M.-L. Mille, M. W. Rogers, K. Martinez, L. D. Hedman, M. E. Johnson, S. R. Lord, and R. C. Fitzpatrick. Thresholds for inducing protective stepping responses to external perturbations of human standing. Journal of Neurophysiology, 90(2):666–74, 2003.

John Milton, Juan Luis Cabrera, Toru Ohira, Shigeru Tajima, Yukinori Tonosaki, Christian W. Eurich, and Sue Ann Campbell. The time-delayed inverted pendulum: Implications for human balance control. Chaos: An Interdisciplinary Journal of Nonlinear Science, 19(2): 026110, 2009.

M. Mirbagheri, H. Barbeau, M. Ladouceur, and R. Kearney. Intrinsic and refex stifness in normal and spastic, spinal cord injured subjects. Experimental Brain Research, 141(4):446– 59, 2001.

M.M. Mirbagheri, H. Barbeau, and R.E. Kearney. Intrinsic and refex contributions to human ankle stifness: variation with activation level and position. Experimental Brain Research, 135(4):423–36, 2000.

Pietro G. Morasso and Vittorio Sanguineti. Ankle muscle stifness alone cannot stabilize balance during quiet standing. Journal of Neurophysiology, 88(4):2157–62, 2002.

Pietro G. Morasso and Marco Schieppati. Can muscle stifness alone stabilize upright standing ? Journal of Neurophysiology, 82(3):1622–6, 1999.

Aditi A. Mullick, Nadine K. Musampa, Anatol G. Feldman, and Mindy F. Levin. Stretch refex spatial threshold measure discriminates between spasticity and rigidity. Clinical Neurophys- iology, 124(4):740–51, 2013. 112 BIBLIOGRAPHY

S J Olney and D A Winter. Predictions of knee and ankle moments of force in walking from EMG and kinematic data. Journal of Biomechanics, 18(1):9–20, 1985.

C A Pelletier and a L Hicks. The length-tension relationship of human dorsifexor and plan- tarfexor muscles after spinal cord injury. Spinal cord, 48(3):202–6, 2010.

K Perell, A Scremin, O Scremin, and C Kunkel. Quantifying muscle tone in spinal cord injury patients using isokinetic dynamometric techniques. Paraplegia, 34(1):46–53, 1996.

Eric Perreault, Robert Kirsch, and Patrick Crago. Efects of voluntary force generation on the elastic components of endpoint stifness. Experimental Brain Research, 141(3):312–23, 2001.

Eric J. Perreault, Robert F. Kirsch, and Patrick E. Crago. Voluntary control of static endpoint stifness during force regulation tasks. Journal of Neurophysiology, 87(6):2808–16, 2002.

R Riener and T Edrich. Identifcation of passive elastic joint moments in the lower extremities. Journal of biomechanics, 32(5):539–44, 1999.

Elliott J. Rouse, Levi J. Hargrove, Eric J. Perreault, and Todd A. Kuiken. Estimation of human ankle impedance during the stance phase of walking. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 22(4):870–8, 2014.

Massimo Sartori, Marco Maculan, Claudio Pizzolato, Monica Reggiani, and Dario Farina. Mod- eling and simulating the neuromuscular mechanisms regulating ankle and knee joint stifness during human locomotion. Journal of Neurophysiology, 114(4):2509–27, 2015.

R K Shields. Muscular, skeletal, and neural adaptations following spinal cord injury. The Journal of orthopaedic and sports physical therapy, 32(2):65–74, 2002.

Nuno A. Silva, Nuno Sousa, Rui L. Reis, and Ant´onioJ. Salgado. From basics to clinical: A comprehensive review on spinal cord injury. Progress in Neurobiology, 114:25–57, 2014.

Peter J. Sinclair, Glen M. Davis, and Richard M. Smith. Musculo-skeletal modelling of NMES- evoked knee extension in spinal cord injury. Journal of Biomechanics, 39(3):483–92, 2006. BIBLIOGRAPHY 113

T Sinkjaer and I Magnussen. Passive, intrinsic and refex-mediated stifness in the ankle exten- sors of hemiparetic patients. Brain : a Journal of Neurology, 117:355–63, 1994.

Thomas Sinkj´er,Egon Toft, Knud Larsen, Steen Andreassen, and Hans J. Hansen. Non-refex and refex mediated ankle joint stifness in multiple sclerosis patients with spasticity. Muscle and Nerve, 16(1):69–76, 1993.

Sue Ann Sisto and Trevor Dyson-Hudson. Dynamometry testing in spinal cord injury. Journal of Rehabilitation Research and Development, 44(1):123–36, 2007.

Lizeth H. Sloot, Marjolein M. van der Krogt, Karin L. de Gooijer-van de Groep, Stijn van Eesbeek, Jurriaan de Groot, Annemieke I. Buizer, Carel Meskers, Jules G. Becher, Erwin de Vlugt, and Jaap Harlaar. The validity and reliability of modelled neural and tissue properties of the ankle muscles in children with cerebral palsy. Gait and Posture, 42(1):7–15, 2015.

Mukul Talaty, Alberto Esquenazi, and Jorge E. Briceno. Diferentiating ability in users of the ReWalk powered exoskeleton: An analysis of walking kinematics. In IEEE International Conference on Rehabilitation Robotics (ICORR), pages 1–5. IEEE, 2013.

William T; Townsend. The efect of transmission design on force-controlled manipulator performance. PhD thesis, Massachusetts Institute of Technology, 1988.

Tamara C. Valovich-mcLeod, Sandra J. Shultz, Bruce M. Gansneder, David H. Perrin, and Joshua M. Drouin. Reliability and validity of the Biodex system 3 pro isokinetic dynamometer velocity, torque and position measurements. European Journal of Applied Physiology, 91(1): 22–9, 2004.

T L Wickiewicz, R R Roy, P L Powell, and V R Edgerton. Muscle architecture of the human lower limb. Clinical Orthopaedics and Related Research, (179):275–83, 1983.

DA Winter. Human balance and posture control during standing and walking. Gait and Posture, 3(4):193–214, 1995.

David A. Winter. Biomechanics and motor control of human movement. Wiley, 2009. 114 BIBLIOGRAPHY

David A. Winter, Aftab E. Patla, Francois Prince, Milad Ishac, and Krystyna Gielo-Perczak. Stifness control of balance in quiet standing. Journal of Neurophysiology, 80(3):1211–21, 1998.

David A. Winter, Aftab E. Patla, Shirley Rietdyk, and Milad G. Ishac. Ankle muscle stifness in the control of balance during quiet standing. Journal of Neurophysiology, 85(6):2630–3, 2001.

Teddy W. Worrell, Gregory Karst, David Adamczyk, Randy Moore, Chris Stanley, Blaine Steimel, and Shane Steimel. Infuence of joint position on electromyographic and torque generation during maximal voluntary isometric contractions of the hamstrings and gluteus maximus muscles. Journal of Orthopaedic & Sports Physical Therapy, 31(12):730–40, 2001.

Tingfang Yan, Marco Cempini, Calogero Maria Oddo, and Nicola Vitiello. Review of assistive strategies in powered lower-limb orthoses and exoskeletons. Robotics and Autonomous Systems, 64:120–36, 2015.

L Q Zhang, G Nuber, J Butler, M Bowen, and W Z Rymer. In vivo human knee joint dynamic properties as functions of muscle contraction and joint position. Journal of Biomechanics, 31(1):71–6, 1998. Appendix A

Motor selection

The motor was selected in four steps to identify an actuator fulflling the design criteria of Section 2.2.2. Actuators from three manufacturers were considered: ETEL, Beckhof and In- franor, their specifcations were used in simple simulations to determine the most appropriate actuator.

First, actuators with unsuitable features (such as extreme power rated torque <1Nm or >200Nm or excessive dimensions such as length >1m) were removed from the list. Then simulation of impedance estimation was carried out to defne the minimum motor peak torque required for our robot. A typical perturbation profle for impedance identifcation Burdet et al. (2000) with 100ms rise time and 0.03◦ amplitude was selected to move a simulated leg. The simulated model represented a 90kg subject’s leg (with inertia calculated as in Winter (2009)). Motors with too low dynamics were removed from the list. Diferent gearbox ratio and pulley radius were considered in the same time to achieve an optimal solution for each motor.

Signifcant standstill torque is required for isometric experiments when the subject is performing motions involving maximal voluntary isometric contractions (MVIC). Therefore, a third selection was conducted by comparing the MVIC data available in literature Andriacchi et al. (1980); Anderson et al. (2007) with the selected motors capabilities, and motors with insufcient standstill torque were excluded.

115 116 Appendix A. Motor selection

In general, human movements are slower than a typical actuator’s speed. Therefore, reducing the actuator’s speed enables us to increase the output torque and thus select smaller motors. In our setup the motor pulley amplifes the motor torque by a factor of L/ρ, where L is the leg length and ρ the pulley diameter. While a pulley of small size would provide a large torque amplifcation and thus enable us selecting a smaller motor, a small pulley will be fragile. To have a sufciently rigid pulley, we impose a radius larger at least 1cm larger than the motor shaft. Considering all these factors, the motor Beckhof AM8061 with a gearbox of reduction ratio 5:1 was selected, with a pulley of diameter of 5.5cm. Appendix B

Ethical approval

The following documents are the ethical approvals which were submitted to the Imperial Col- lege Research Ethics Committee (ICREC). All procedures were performed according to the principles in the Declaration of Helsinki and approved by the Imperial College Research Ethics Committee.

117

Imperial College Research Ethics Committee (ICREC) Application Form

Section One - Details of Principal Investigator If the application is a student project, please name the student’s supervisor as the Principal Investigator. Please add the student’s name and details of all other co-investigators or collaborators in Section Six. Name: Etienne Burdet

Position: Professor of Human Robotics

Address: RSM 4.05, Department of Bioengineering, Royal School of Mines, Imperial College London South Kensington Campus, London SW7 2AZ, UK

Telephone: 02075945179

Email: [email protected]

Fax:

Summary of skills and experience relevant to this study: Human physiology and biomechanics, modelling, robotics.

Previous experience in any procedures to be used.(Including involvement with The PI has over 20 years of experience in psychophysical and sensorimotor vulnerable participants if relevant): control experiments with healthy and neurologically impaired participants.

Section Two - Project Summary Full title of study: Single joint neuromechanics of the lower limb

Proposed start date: 02/01/2018

Proposed end date: 31/12/2023

Does the study involve any of the following

Drugs/medication  Yes  No If yes, please attach the SmPc

Ionising radioactive substances/x-rays  Yes  No If yes, please also complete appendix one

Genetically modified materials  Yes  No If yes, please also complete appendix two and attach the GM Safety Committee letter

ICREC application form, version 4, January 2015

2.a Project Summary Please provide a summary of the project , including the expected outcomes, in LAY terms (max 200 words)

This project will use a computer-controlled device to estimate the single joint neuromechanics of the lower limb in the healthy participants. This knowledge will be acquired by applying computer-controlled forces/displacements on the lower limb isometrically and during single joint movement and measuring the resulting movements/forces. The acquired knowledge is critical to understanding standing and walking in humans, and will provide the knowledge basis for future studies with neurologically impaired individuals. However the studies here will involve only healthy individuals.

2.b Ethical Summary Please provide details of what you consider to be the ethical issues surrounding this project in LAY terms (max 200 words)

The experiment population will be formed of healthy individuals from 18 to 40 years of age. The position and interaction force will be recorded while forces will be applied by the interface. The experiment involves no serious ethical or safety issues. Redundant safety measures includes mechanical stops, software limits on the position, velocity and acceleration, multiple electrical circuit breakers, isolated transformer, residual current detector, multiple emergency buttons for the participants and researchers. Although unlikely, participants might experience temporary muscle fatigue due to repetitive experiments. However, participants will be able to rest between trials, and will be free to withdraw from the experiment at any time.

2.c Peer Review Has your research project been independently Peer Reviewed  Yes  No This can be organised by the Peer Review Office (within the JRCO)

Please give details below

Section Three - Project Details

3.a Protocol Please provide details of the scientific justification for the study, and the research methodology to be used in LAY terms (max 200 words). Please attach the protocol and any questionnaires that will be used. Human participants with no neuromuscular disorders or injury at the lower limbs will be selected. The participants will be full introduced to the protocol and the robotic system, then will be given enough time to read the participant information sheet before giving their consent to participate. No study will be commence until full consent has been given to participate. Each participant will sit on a chair while the targeted leg will be fixed with a harness to the robotic interface. This interface will provide a series of force perturbations at static position or during movement. To prevent muscle fatigue, resting time will be provided between trials according to the individual’s need. The recorded force and position will enable us to estimate the lower limb mechanics and how the brain controls it.

Surface EMG and lower limb kinematics and kinetics will be recorded using certified medical devices and load cells attached to a self-developed harness. After the experiment, the data would be analysed using statistical tools and mathematical models to verify model characteristics and behaviours.

ICREC application form, version 4, January 2015

3.b Value Please state in LAY terms the intended value of the project (in terms of the expected outcomes) (max 200 words).

The data acquired from this study can be used to model how the brain controls the lower limb. A series of well-documented experiments had demonstrated the potential of understanding motor control and learning on the upper limb motion (Human Robotics: Neuromechanics and Motor Control, MIT Press). We believe by applying it onto lower limb experiments, we can create model which are adaptive to individual neuromuscular conditions. This is important to develop individual therapies and dedicated devices for neurologically impaired individuals.

3.c Location Please state where the study will take place. If part or all of the study will take place abroad, please state any steps that have been taken to ensure compliance with research and ethical rules in those countries.

The study will take place at the laboratory room B120 in the Bessemer building, Department of Bioengineering Imperial College London.

3.d Dissemination Please state how and to whom the results of this study will be disseminated, including the communication of results to the research participants.

The results of the study will be submitted to relevant journals (in the fields of human neurophysiology and human motion science). Participants will be anonymised in the data treatment and they will be able to consult the published results.

3.e Previous ethical approval

Has any part of this proposal received prior ethical approval?  Yes  No

Has any part of this proposal previously been rejected by an ethics  Yes  No committee?

If yes to either of the above, please give details and attach relevant documentation:

3.f Funding

Has this project received funding, or is in the process of receiving  Yes  No funding?

ICREC application form, version 4, January 2015

If yes, please provide details:

EU FP7 project SYMBITRON. FP7-ICT-2013-10, ID 661626

3. g Insurance and Indemnity Cover

Have you applied to the Joint Research Compliance Office (JRCO)  Yes  No for indemnity cover?

If no, please state when you intend to apply:

If you do not intend to apply, please give reasons: The experiment does not involve healthcare studies.

Section Four - Participants

4.a Study population Please state the number of research participants to be recruited, including the inclusion and exclusion criteria for recruitment.

Up to 100 healthy individuals aged 18-40 years old will be recruited to this study. Inclusion criteria: Volunteers will be healthy participants without known neuromuscular disorder or injury on the lower limbs. They must understand and be able to fully express themselves in English. Participants outside of these criteria will be excluded.

4. b Vulnerable groups Please give details of any vulnerable groups to be used in the study (e.g. those under 18, prisoners, those in a dependent relationship, the mentally ill), and give reasons for their inclusion.

No vulnerable groups will be used in this study.

4.c Recruitment process

Please explain how you will recruit participants to the study. Please include any incentives (such as financial reimbursement). If advertising is to be used, please attach a copy of the advert. If you are planning to recruit via email, please include a copy of the email.

Voluntary participants will be mainly recruited through flyers placed around Imperial College South Kensington campus and through email distributed by the Department of Bioengineering.

The email will be in the following form:

Recruitment e-mail Study: "Single joint neuromechanics of the lower limb"

ICREC application form, version 4, January 2015

The Human Robotics group of Bioengineering Department at Imperial College London will investigate the neuromechanics of lower limb using a novel robotic interface. This can be used to understand the control of balance and how this control is affected by neurological impairments.

The present study will measure the neuromechanics of healthy adults. The participants should have no neuromuscular disorder or recent injury to the lower limbs. The study is composed of trials where participants will both actively and passively interact with the robotic interface while sitting on a chair.

The experimental protocol had been designed in accordance with the rules established by Imperial College Research Ethics Committee and approved by it.

The experiment will take place at Imperial College London’s South Kensington campus. It requires typically 45 minutes, with sufficient time for breaks between trials. Each participant will receive a lunch voucher as a compensation for your time spent.

If you are interested in participation of this study, please contact the following person:

Hsien Yung, Huang e-mail: [email protected] PhD Student Human Robotics group Department of Bioengineering, Imperial College London, London, SW7 2AZ, UK

4.d Informed Consent i. Please detail your process for ensuring the informed consent of all research participants. ii. If vulnerable persons are to be used in the study, please give separate specific information on how you will ensure informed consent. iii. If participants whose first language is not English are to be recruited, please state clearly how the details of the study will be explained and the informed consent process will be handled. Please attach a copy of the consent form and participant information sheet and any additional forms/information if appropriate.

The study will be explained to each participants by one of the researcher, who will be given an information sheet detailing the study aims and methodology. Each participant will then be given one day to read and process the information, and will be able to ask questions to the researcher. They will then be asked to fill in and sign a written consent form (of which they will receive a copy).

Participants presenting difficulties to understand or expressing themselves in English, will be excluded from the study as part of the inclusion criteria.

The participant information sheet and consent form are attached.

4.e Withdrawal Please state procedures for the withdrawal of participants from the study.

Participants will be free to refuse participation and to withdraw from the study at any time (e.g. at the recruitment, during data collection or within 1 year from data acquisition). Each participant will be provided with the contact details (both email and telephone) of the principal investigator, Professor Etienne Burdet, on the participant information sheet. If the publication of the study results is expected or happens to fall prior to the expiration of the 1 year deadline for withdrawal, the principal investigator will contact the participants, informing them about their rights to withdraw before publication of results.

ICREC application form, version 4, January 2015

4.f Confidentiality Please state the measures you have taken to ensure the confidentiality of the research participants and the data collected, including who will have access to the data. Please include information on any process to anonymise participants and their data.

All collected data will be anonymised to ensure that participants cannot be identified even if the data are compromised. The data will be stored digitally in a cloud storage which is managed by the Department of Bioengineering. The cloud had restricted access connected to the Imperial College account database such that only members of Imperial College with approval may access the data. Moving of data to off-site storage (such as personal computers at home or abroad) is strictly prohibited. The records and data are kept for 10 years after the completion of the project to allow for publication, revision and further analyses.

All information which is collected about participants during the course of the research will be kept strictly confidential. Any information which leaves Imperial College London will have the participant’s name and address removed for confidentiality. All procedures for handling, processing, storage and destruction of their data are compliant with the Data Protection Act 1998.

Section Five - Risks and Benefits

5.a Risks Has Imperial Colleges Risk Assessment procedures been followed.  Yes  No

BioSOP57

If No, please explain:

Does Imperial Colleges’ Insurer need to be notified about your project  Yes  No before insurance can be provided

Insurance for all Imperial studies is provided by a commercial insurer, (please refer to website) http://www3.imperial.ac.uk/clinicalresearchgovernanceoffice/projectplanning/clinicalresearchofficeapproval/insurance . For the majority of studies the cover is automatic. However for a minority of studies, in certain categories, the insurer requires prior notification of the project before cover can be provided.

If Yes, please provide confirmation that the appropriate insurance cover has been agreed. Please attach your Imperial College registration form and any related correspondence

Please state briefly the precautions being taken to protect the health and safety of researchers and others associated with the project (as distinct from the research participants).

The experimental device is regularly checked for safety and functionality. All devices are grounded and tested with certified equipment for continuity of the ground. Circuit breakers are installed and would close the system without delay. All electrical components are placed in enclosure boxes with the box grounded. No direct contact of any active potential is possible with finger contact.

ICREC application form, version 4, January 2015

Please state briefly the precautions being taken to protect the health and safety of research participants.

Redundant safety measures implemented in the experimental system include mechanical stops, software limits on the position, velocity and acceleration, multiple electrical circuit breakers, isolated transformer, residual current detector, multiple emergency buttons for the participant and the researcher. All electronic equipment will be checked before every experiment for proper functioning and safety. Two researchers will be present during the whole experiment. To ensure the health and safety of the research participants, we will only use additional systems such as electromyography amplifier that are fully safety-checked and approved for human recording, with documented limits for the potential current that may flow into the participants’ body. To eliminate hazards relate to power line interference, an electric isolator will be used to connect all instruments.

Will these participants participate in any activities that may be   potentially stressful or harmful in connection with this research Yes No

Please specify whether the following procedures are involved?

Any invasive procedure (s)  Yes  No

Physical contact  Yes  No

Any procedure (s) that may cause mental distress  Yes  No

Are you using any medical device in the UK that is CE marked and is it  Yes  No being used within its product limitation?

ICREC application form, version 4, January 2015

5.b Benefits Please describe any expected benefits to the participant, researchers or others (e.g. Imperial College, Industry funders)

Participants would be told about the impact of the study and the potential applications after the participation. They will also be given a lay summary of published journal papers if they are interested in the results.

For the researchers, the experiment result will help to learn more about neuromechanics and exoskeleton control strategies to be developed. It will also benefit us for further understanding the possible participant- specific control strategy for SCI participants. Robotic industries may consider implementing those strategy in their commercially-available exoskeleton or human gait trainers.

In the context of the EU-FP7 SYMBITRON, this would support us further understand of SCI participant’s behaviour and it would be implemented into the control algorithm of the developing exoskeleton.

Section 6 – Co-Investigators/Collaborators If there are more than 3 co-investigators, please use a separate sheet and follow the format below. Hsien Yung, Huang Name:

PhD student Position: RSM 4.28, Department of Bioengineering, Royal School of Address: Mines, Imperial College London South Kensington Campus, London, SW7 2AZ, UK

07846696903 Telephone: [email protected] Email:

Fax:

Summary of skills and experience relevant to this study: Application of electrodes for electromyography recording, control system, exoskeleton control strategy development, data analysis.

Previous experience in any procedures to be Application of electrodes for electromyography recording, used. (Including involvement with vulnerable participants if control system, exoskeleton control strategy development, data relevant): analysis. Arash Arami, PhD Name:

Assistant Professor Position: Department of Mechanical and Mechatronics Engineering Address: Faculty of Engineering University of Waterloo 200 University Avenue West Waterloo, ON, Canada N2L 3G1

Telephone: [email protected] Email:

ICREC application form, Version 4.0, January 2015

Fax: Arash Arami has 8 years of experience in gait analysis, Summary of skills and experience relevant to physiological and movement data collection from participants this study: ranged from senior citizens and individuals with orthopaedic disease to participants with spinal cord injury and Parkinson’s disease. Dr Arami has expertise both in using in-lab instruments and wearable systems and developing algorithms to improve the estimation of clinically relevant measures. He pioneered an instrumented knee prosthesis enabling the early detection of loosening in an objective way, estimation of 3D kinematics without soft tissue artifacts in addition to providing the knee contact forces and moments. He also built and controlled a robotic knee systems which has been used in several studies at EPFL. Dr Arami’s current research is more on design of instruments and techniques to identify the neuromechanics underlying the impaired movements, as well as objective modelling of motor deficits such as spasticity. Having a background in control engineering and machine learning, Dr Arami has designed several advanced controllers for electrometrical systems including wearable exoskeletons which have been already tested on healthy and spinal cord injury participants. Arash Arami has 8 years of experience in gait analysis, physiological and movement data collection from participants ranged from senior citizens and individuals with orthopaedic disease to participants with spinal cord injury and Parkinson’s disease. Dr. Arami has expertise both in using in-lab instruments and wearable systems and developing algorithms to improve the estimation of clinically relevant measures. He pioneered an instrumented knee prosthesis enabling the early detection of loosening in an objective way, estimation of 3D Previous experience in any procedures to be kinematics without soft tissue artifacts in addition to providing used. (Including involvement with vulnerable participants if the knee contact forces and moments. He also built and relevant): controlled a robotic knee systems which has been used in several studies at EPFL. Dr Arami’s current research is more on design of instruments and techniques to identify the neuromechanics underlying the impaired movements, as well as objective modelling of motor deficits such as spasticity. Having a background in control engineering and machine learning, Dr Arami has designed several advanced controllers for electrometrical systems including wearable exoskeletons which have been already tested on healthy and spinal cord injury participants.

Name:

Position:

Address:

Telephone:

Email:

ICREC application form, version 4, January 2015

Fax:

Summary of skills and experience relevant to this study:

Previous experience in any procedures to be used. (Including involvement with vulnerable participants if relevant):

ICREC application form, version 4, January 2015