A

Sa

b

a

ARR2A

KCGTPR

1

tsspftlilr

ocr[(poAtios

1d

Medical Engineering & Physics 34 (2012) 1448– 1453

Contents lists available at SciVerse ScienceDirect

Medical Engineering & Physics

j o ur nal homep age : www.elsev ier .com/ locate /medengphy

n intrinsically compliant robotic orthosis for treadmill training

hahid Hussaina,∗, Sheng Quan Xiea, Prashant K. Jamwala, John Parsonsb

Department of Mechanical Engineering, The University of Auckland, Auckland, New ZealandSchool of Nursing, The University of Auckland, Auckland, New Zealand

r t i c l e i n f o

rticle history:eceived 3 November 2011eceived in revised form9 December 2011ccepted 7 February 2012

a b s t r a c t

A new intrinsically compliant robotic orthosis powered by pneumatic muscle actuators (PMA) was devel-oped for treadmill training of neurologically impaired subjects. The robotic orthosis has hip and kneesagittal plane rotations actuated by antagonistic configuration of PMA. The orthosis has passive mecha-nisms to allow vertical and lateral translations of the trunk and a passive hip abduction/adduction joint.A foot lifter having a passive spring mechanism was used to ensure sufficient foot clearance during swing

eywords:omplianceait rehabilitationreadmill trainingneumatic muscle actuators

phase. A trajectory tracking controller was implemented to evaluate the performance of the roboticorthosis on a healthy subject. The results show that the robotic orthosis is able to perform the treadmilltraining task by providing sufficient torques to achieve physiological gait patterns and a realistic steppingexperience. The orthosis is a new addition to the rapidly advancing field of robotic orthoses for treadmilltraining.

obotic orthosis

. Introduction

Robotic orthoses are gaining recognition among the rehabilita-ion engineering community for the treadmill training of subjectsuffering from neurologic impairments such as stroke [1,2] andpinal cord injuries (SCI) [3–5]. These robotic orthoses relieve thehysical therapist from the strenuous task of manual assistance andacilitates in delivering well-controlled repetitive and prolongedraining sessions at a reduced cost. The physical therapist’s role isimited to supervision. The subjectivity of manual training processs eliminated by providing measurement of interaction forces andimb movements to assess the quantitative level of motor functionecovery.

The first modern automated treadmill based gait trainingrthosis, LOKOMAT had been developed in late 1990s and isommercially available [6]. Several other research prototypes ofobotic gait training orthoses namely; active leg exoskeleton (ALEX)2], ambulation-assisting robotic tool for human rehabilitationARTHUR) [7], Lower extremity powered exoskeleton (LOPES) [8],elvic assist manipulator (PAM) and pneumatically operated gaitrthosis (POGO) [9] had also been developed during the last decade.ctuators hold a paramount importance in the design and func-

ioning of these robotic orthoses. Two approaches have been used

n actuator placement for powering the robotic orthoses [10]. Inne of the approaches, the actuators have been placed on a remotetation and the actuation has been transferred to the orthosis joints

∗ Corresponding author. Tel.: +64 9 3737599x87555; fax: +64 9 3737479.E-mail address: shus045@aucklanduni.ac.nz (S. Hussain).

350-4533/$ – see front matter © 2012 IPEM. Published by Elsevier Ltd. All rights reserveoi:10.1016/j.medengphy.2012.02.003

© 2012 IPEM. Published by Elsevier Ltd. All rights reserved.

via cables, rigid linkages [8,11] and pneumatic or hydraulic systems[9]. LOPES [8], PAM [9] and the cable-driven locomotor training sys-tem [11] are the rehabilitation devices using this approach. In thesecond approach the actuators have been directly mounted on theorthosis frame. LOKOMAT [6] and ALEX [2] use this approach ofactuator mounting.

The benefit of the first approach is that there are no limita-tions on actuator weight and hence the power capacity of theactuators. Lack of precise control, inefficient transfer of power andnon-durability of the actuation transfer mechanism (cables) are thedrawbacks associated with this approach [10]. The main advantageof the second approach is the efficient transfer of power and a bet-ter alignment of orthosis joints with patient joints [10]. But thedisadvantage is the use of geared electric motors which are eitherextremely heavy [10] or have high endpoint impedance (stiffness)[11]. The use of heavier electric motors and gear assembly increasesthe overall weight of the robotic orthosis which is not suitable forimplementing advance control strategies such as impedance con-trol [8]. If light weight electric motors are used, the force and torquegeneration capabilities of the robotic orthosis are seriously compro-mised [10]. Also these high endpoint impedance electric motors aremore suitable for industrial applications. Neurologically impairedpatients often suffer from severe spasms. These stiff actuators mayproduce large forces in response to the undesirable motions pro-duced by spasms (position errors) [12,13]. As a result the patientmay feel pain or discomfort. We believe that the second approach of

directly mounting the actuators on the orthosis frame may becomemore beneficial if the limitations on the weight and the endpointimpedance of the actuators could be overcome. Also adding com-pliance to the actuation mechanism would help in absorbing large

d.

ering & Physics 34 (2012) 1448– 1453 1449

pph

fitPhacnllorPnoiubPhwrra

2

2

itrw

1

2

3

4

5

6

2

rai

(

Fig. 1. Kinematic diagram of the robotic orthosis showing all the DOF. R1, rev-olute ankle joint; R2, revolute joint for knee flexion; R3, revolute joint for hipflexion/extension; R4, revolute joint for hip abduction/adduction; R5(a–d), revo-lute joints of the parallelogram mechanism for vertical translation; and T, sliders forlateral translation.

S. Hussain et al. / Medical Engine

osition errors and would insure safety of the patient. The com-liant actuation also provides impact resistance during events likeeel strike as well as a realistic stepping experience.

Pneumatic muscle actuators (PMA) are promising tools in theeld of rehabilitation robotics. Their actuation principle is similaro the actuation principle of skeletal muscles. Like skeletal musclesMA generate forces upon contraction. They are light weight andave high power/weight and power/volume ratios as compared toll the existing actuators [14]. PMA have intrinsic elasticity whichan be used in providing compliant actuation. PMA possess a highlyonlinear and time-varying behavior which present a control prob-

em but in the presence of above-mentioned advantages theseimitations become secondary. PMA have been used in the designf various upper extremity robotic orthoses [12,15] and single jointobotic orthoses for lower limb [16]. However, the potential ofMA in the design of robotic orthoses for treadmill training hasot been fully explored. In this work we developed a new roboticrthosis powered by PMA for treadmill training of neurologicallympaired subjects. The mechanism design and actuation conceptssed for the robotic orthosis are presented. According to authors’est knowledge, neither a multi-joint robotic orthosis powered byMA for treadmill training has been reported in the literature noras been controlled in a feedback control configuration. This workill add new dimensions to the compliant actuation of robotic gait

ehabilitation orthoses. Preliminary performance evaluation of theobotic orthosis was carried out with a healthy subject in order tossess its feasibility for the treadmill training.

. Robotic orthosis

.1. Design requirements

In order to address the functional, structural and cosmetic lim-tations of current robotic gait orthoses for treadmill training ando meet the clinical requirements of the treadmill training of neu-ologically impaired subjects, the new compliant robotic orthosisas designed based on the following criteria.

. The robotic orthosis should provide a realistic stepping experi-ence and orthosis joints should work in alignment with patient’sjoints.

. The actuation system should be powerful enough to guide thepatient’s limbs on reference trajectories and be able to producerequired joint torques.

. Actuators should be highly compliant with low mechanicalimpedance [10,17] in order to accommodate abrupt forces aris-ing from clonus. The relationship between moment applied bythe robotic orthosis and the joint angle is generally known as“mechanical impedance”.

. The robotic orthosis and the actuation mechanism should belight weight so that the advance control strategies such asimpedance control [8,18] could be implemented.

. The robotic orthosis should inhibit excessive knee extension andhip flexion/extension.

. Regarding the cosmetic requirements and ease of use the activeorthosis should be easy to wear and comfortable.

.2. Mechanism design and actuation

Based on the above-mentioned criterion the design of theobotic orthosis was carried out. The setup has following main parts

nd a simple kinematic diagram of the degrees of freedom (DOFs)s shown in Fig. 1.

1) A height adjustable frame to match subject’s height.

Fig. 2. Experimental setup of the robotic orthosis with a subject walking on a tread-mill.

1 ering

(

(

(

(

iddwrfkm

(otwAttApwou

rtlswcd

apssrntc

OtBlrkwis

2

l

450 S. Hussain et al. / Medical Engine

2) Trunk of the orthosis having two DOFs namely vertical andlateral translations.

3) Thigh segment of the orthosis having two DOFs with respect tothe trunk of the orthosis, one for flexion/extension and other forabduction/adduction motion. The thigh segment and hip braceare telescopic and can be adjusted to match anthropometricfeatures of a wider population range.

4) Shank segment has one DOF with respect to the thigh segmentfor knee flexion. Shank segment is also telescopic.

5) A foot lifter is attached to the shank section with a revoluteankle joint to ensure sufficient dorsiflexion during swing phase.

The complete design description of the robotic orthosis is shownn Figs. 2 and 3. Actuated and free DOF for the active orthosis wereecided based on the joint ranges of motion. The major rotationsuring gait cycle are in the sagittal plane [19]. The actuated DOFere hip and knee sagittal plane rotations. The maximum joint

anges of motion for hip and knee flexion/extension were selectedrom the studies reported in the literature [19]. All other DOFs wereept free. All the above-mentioned DOFs can help in achieving aore realistic stepping experience.PMA was used for providing actuation to the robotic orthosis

Fig. 4(a)). To provide the actuation at hip and knee joints, vari-us mechanisms were studied to transfer the actuation from PMAo the orthosis joints. Double groove aluminum discs (Fig. 4(b))ere employed at hip and knee joints for sagittal plane motions.n antagonistic disc-PMA mechanism (opposing pair configura-

ion) was selected for the actuation purpose (Fig. 4(c)). The verticalranslation was realized by a passive parallelogram mechanism.

spring was mounted on the parallelogram mechanism. The pur-ose of spring was to compensate the weight of the robotic orthosishich would help in reducing the relative slip between the robotic

rthosis and the subject’s joints. A passive slider mechanism wassed to allow the lateral translation.

The orthosis frame was made from light weight aluminumectangular tubing to meet the strength requirements for torqueransmissions. The orthosis frame was connected to the subjects’imbs with two braces, one at the thigh section and other at thehank section. Hip straps were used to secure the robotic orthosisith the subject’s hip. All the braces have soft straps to provide a

omfortable feel to the patient. The robotic orthosis is a unilateralevice and only attachment to the left leg was studied.

Ankle joint was not actuated. The rationale for eliminating anctuated ankle joint was that it is not necessary to provide an ankleush-off in the robotic orthosis in order to walk safely. For patientafety the only necessary ankle function is to assure enough dor-iflexion during swing phase for foot clearance [6]. This could beealized by simpler means such as elastic straps or spring mecha-isms. A foot lifter with a passive spring mechanism was designedo provide the necessary dorsiflexion during swing phase for footlearance.

A kinematic model of the lower extremity was developed usingpen SIMM [20]. The model was used to study the effects of inser-

ion/origin locations of the PMA on the joint ranges of motion.ased on the anthropometric data of an average man, the required

ength and braid diameter of each PMA to generate the desiredange of motion were 34 cm and 3 cm, respectively for the hip andnee joints. These PMA can provide peak joint torques of 50 Nm,hich was sufficient for the proposed application. If larger torque

s required larger diameter PMA can be used under the safety con-traints.

.3. Control

The purpose of the robotic orthosis is to guide the subject’simbs on reference trajectories. For the trajectory tracking task a

& Physics 34 (2012) 1448– 1453

proportional digital pole placement controller was implementedin joint space to control the antagonistic actuation of PMA pair(0–5 bar) [21]. Hip and knee sagittal plane physiological gait trajec-tories reported by Winter in [19] were used to define the desiredjoint angle trajectories. These joint angle trajectories are scalablein time, amplitude offset and range in order to adjust it to theindividual gait parameters of subjects. Matrix 820 solenoid 2/2valves were used for the control of pressure in PMA. Each PMA wasequipped with gauge pressure sensor and joint encoders were usedto measure the hip and knee joint angle trajectories. The joint anglemeasurements were used to calculate and control the lengths ofPMA. All the hardwares were controlled by using a DSPACE, Ds1103operating system.

Since the robotic orthosis was designed for treadmill trainingof neurologically impaired subjects and has mechanical actuators,certain safety features were incorporated to ensure the safety ofthe subject in the training environment as discussed below.

2.4. Safety

Safety is a key concern for the robotic devices working in closeproximity with human subjects. As the proposed robotic ortho-sis was designed to work in close proximity with neurologicallyimpaired subjects, several safety features were incorporated inthe robotic orthosis mechanism and control hardware. Mechanicalstops were placed on the hip and knee joint discs to avoid the ortho-sis to go beyond the physiological ranges of motion (maximumranges between 60◦/30◦ and 90◦/0◦ for the hip and knee flex-ion/extension, respectively). These mechanical stops can withstandthe maximum torques applied by the PMA. Also, an independentsafety circuit was created that can power the system down in caseof any danger or if the subject feels uncomfortable in the device.Two emergency switches were wired such that a single push canstop the whole system by exhausting the air from each PMA. Oneswitch button was held by the subject while the other was held bythe person invigilating the training process.

3. Experimental evaluation

In order to evaluate the performance of the design and to avoidextensive subject tests, several hardware tests were performed. Adummy leg was used in the robotic orthosis. One healthy subject(age 27 years, body mass 76 kg) with no history of neurological dis-orders gave written consent and walked within the robotic orthosisto evaluate its performance. The University of Auckland, HumanParticipants Ethics Committee approved this protocol. The roboticorthosis was attached to the left leg of the subject. The subject wasasked to walk within the passive (unpowered) robotic orthosis for15 min so that he should become familiar with the robotic ortho-sis and the training environment. Similar procedure was repeatedfor the robotic orthosis in active (powered) mode. The subject wasasked to walk with the robotic orthosis and allow it to guide the tra-jectory of his legs during the active mode. After the initial 15 minsession the data for 15 gait cycles during the active mode wasrecorded for analysis purpose.

Fig. 5 shows the desired and measured hip and knee sagittalplane joint angles obtained with the healthy subject walking in therobotic orthosis. Walking speed was set to 0.7 m/s. Control param-eters of the proportional controller were adjusted in such a waythat the deviations from the desired trajectory were minimum. Themaximum trajectory tracking errors were always less than 5◦ dur-

ing the experiments with the healthy subject. The robotic orthosisperformance during the trajectory tracking mode is in accordancewith the other gait rehabilitation orthoses such as LOKOMAT.For LOKOMAT the maximum trajectory tracking errors during the

S. Hussain et al. / Medical Engineering & Physics 34 (2012) 1448– 1453 1451

Fig. 3. The robotic orthosis, its major components and all the degrees of freedom labeled.

Fig. 4. (a) Pneumatic muscle actuator (PMA), (b) double groove aluminum disc, and (c) antagonistic disc-PMA (opposing pair configuration) mechanism.

1452 S. Hussain et al. / Medical Engineering & Physics 34 (2012) 1448– 1453

F percent he rob

tbdwc

4

tt[afSimprmoi

prtatdbt

ldw

ig. 5. Hip and knee joint sagittal plane angle trajectories with a healthy subject as a

aken as reference and plotted against the trajectories obtained during walking in t

rajectory tracking mode with minimum compliance must beelow 15◦ [5]. The passive foot lifter provided sufficient clearanceuring the swing phase and no cases of foot touching the treadmillere observed during the experiments. The subject also reported

omfortable trunk motions in the vertical and lateral planes.

. Discussion

Robot assisted treadmill training is establishing its worth inhe field of rehabilitation engineering. Several robotic orthoses forreadmill training have been developed during the last decade2,6–10]. These robotic orthoses use high endpoint impedancectuators such as electric motors [2,6,11] which may produce largeorces in response to the undesirable motions produced by spasms.everal attempts such as impedance control have been made tontroduce compliance in the robotic orthoses driven by electric

otors [5]. But these attempts have added an extra layer of com-lexity and extra cost [11]. PMA are intrinsically compliant and thuseduces the complexity of control system as compared to electricotors. We have designed a new intrinsically compliant robotic

rthosis powered by PMA for treadmill training of neurologicallympaired subjects.

There were two reasons for using PMA for the actuation pur-ose. The first reason was related to the geometrical design of theobotic orthosis. The design was made simple, light weight and easyo wear. The decrease in weight of the robotic orthosis resulted in

reduction of the relative slip between the robotic orthosis andhe subject’s joints. The second reason to use PMA was to intro-uce intrinsic compliance in the orthosis design. This compliance iseneficial for human–robot interaction and provides greater shockolerance on heel strike as well as low actuator impedance.

The robotic orthosis performance was evaluated with dummyeg and a healthy subject. The subject walked in the device unhin-ered and no complaints about any kind of pain and discomfortere reported. The robotic orthosis was able to generate the

tage of gait cycle obtained during trajectory tracking mode. Normal gait trajectoriesotic orthosis. (a) Hip joint and (b) knee joint.

required forces to guide the subject’s limbs on reference tra-jectories. No body weight support (BWS) was used during theexperiment as the test subject had no neurological impairmentsand did not require any external support. Also a proportional digitalpole placement controller was implemented for trajectory track-ing task as this study was aimed to evaluate the performanceof the robotic orthosis mechanism. The trajectory tracking errorsobtained with this controller were always less than 5◦. These tra-jectory tracking errors are due to the structured uncertainties inthe model of PMA [14,22]. Implementation of nonlinear controllerssuch as sliding mode control [23] may help in reducing these tra-jectory tracking errors. A proportional controller was sufficient forthe presented work, as the scope of this work was limited to thedesign and evaluation of the performance of the robotic orthosisin terms of achieving physiological gait trajectories. Also a fixedtreadmill speed was used during the presented experiments. Inorder to further enhance the voluntary participation of human sub-jects, the treadmill speed may be adapted according to individualpatient’s disability level and intention. A method for automatictreadmill speed adaptation has been proposed by Zitzewitz et al.[24]. The robotic orthosis was only evaluated on one healthy sub-ject. Rigorous clinical trials with neurologically intact as well asneurologically impaired subjects are required to establish the ther-apeutic efficacy of the intrinsically compliant robotic gait trainingorthosis. Another limitation of the present study is that only thesagittal plane kinematics was recorded. The kinematics of otherplanes with passive DOFs needs to be recorded in order to evaluatetheir significance during the treadmill training task.

According to authors’ best knowledge the prototype developedin this research is the first multi-joint robotic orthosis poweredby PMA for treadmill training of neurologically impaired subjects.

Also the experimental evaluation of multi-joint robotic gait trainingorthosis powered by PMA in a feedback control configuration hasnot been reported in the literature before. This work is an advanceon the current state of the art in the compliant actuation of robotic

ering

gmgrsijgca

C

mtf

A

uTlpo1

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

S. Hussain et al. / Medical Engine

ait rehabilitation orthoses. Future work shall involve the imple-entation of advanced assist-as-needed or patient-cooperative

ait training strategies [10] and their validation on healthy and neu-ologically impaired subjects. These assist-as-needed gait trainingtrategies will adapt the robotic assistance according to the disabil-ty level and rehabilitation stage of neurologically impaired sub-ects. This will also enhance patient’s voluntary participation in theait training process. Similarly, the energy storing properties of theompliant actuators will be an interesting research aspect and canlso be studied in relation to the robotic gait rehabilitation orthoses.

onflict of statement

The authors declare no conflict of interest including employ-ent, consultancies, stock ownership, honoraria, paid expert

estimony, patent applications/registrations, and grants or otherunding

cknowledgments

The authors would like to acknowledge the support of the Fac-lty Research Development Fund from the Faculty of Engineering,he University of Auckland, New Zealand. The authors would alsoike to thank Logan Stuart for his technical contribution to theroject. The authors would also like to acknowledge the supportf Auckland Medical Research Foundation under grant number111014.

eferences

[1] Husemann B, Müller F, Krewer C, Heller S, Koenig E. Effects of locomotion train-ing with assistance of a robot-driven gait orthosis in hemiparetic patients afterstroke: a randomized controlled pilot study. Stroke 2007;38:349–54.

[2] Banala SK, Kim SH, Agrawal SK, Scholz JP. Robot assisted gait training withactive leg exoskeleton (ALEX). IEEE Transactions on Neural Systems and Reha-bilitation Engineering 2009;17:2–8.

[3] Jezernik S, Schärer R, Colombo G, Morari M. Adaptive robotic rehabilitationof locomotion: a clinical study in spinally injured individuals. Spinal Cord2003;41:657–66.

[4] Cai LL, Fong AJ, Otoshi CK, Liang Y, Burdick JW, Roy RR, et al. Implica-tions of assist-as-needed robotic step training after a complete spinal cordinjury on intrinsic strategies of motor learning. Journal of NeuroScience2006;26:10564–8.

[5] Riener R, Lunenburger L, Jezernik S, Anderschitz M, Colombo G, Dietz V. Patient-cooperative strategies for robot-aided treadmill training: first experimental

results. IEEE Transactions on Neural Systems and Rehabilitation Engineering2005;13:380–94.

[6] Colombo G, Joerg M, Schreier R, Dietz V. Treadmill training of paraplegicpatients using a robotic orthosis. Journal of Rehabilitation Research and Devel-opment 2000;37:693–700.

[

& Physics 34 (2012) 1448– 1453 1453

[7] Emken JL, Wynne JH, Harkema SJ, Reinkensmeyer DJ. A robotic devicefor manipulating human stepping. IEEE Transactions on Robotics 2006;22:185–9.

[8] Veneman JF, Kruidhof R, Hekman EEG, Ekkelenkamp R, Van Asseldonk EHF, VanDer Kooij H. Design and evaluation of the LOPES exoskeleton robot for interac-tive gait rehabilitation. IEEE Transactions on Neural Systems and RehabilitationEngineering 2007;15:379–86.

[9] Daisuke Aoyagi WEI, Susan J, Harkema, David J, Reinkensmeyer, James E, et al.A robot and control algorithm that can synchronously assist in naturalis-tic motion during body-weight-supported gait training following neurologicinjury. IEEE Transactions on Neural Systems and Rehabilitation Engineering2007;15:387–400.

10] Hussain S, Xie SQ, Liu G. Robot assisted treadmill training: mechanisms andtraining strategies. Medical Engineering & Physics 2011;33:527–33.

11] Wu M, Hornby TG, Landry JM, Roth H, Schmit BD. A cable-driven locomo-tor training system for restoration of gait in human SCI. Gait and Posture2011;33:256–60.

12] Sugar TG, He J, Koeneman EJ, Koeneman JB, Herman R, Huang H, et al. Designand control of RUPERT: a device for robotic upper extremity repetitive ther-apy. IEEE Transactions on Neural Systems and Rehabilitation Engineering2007;15:336–46.

13] Vallery H, Veneman J, van Asseldonk E, Ekkelenkamp R, Buss M, van Der KooijH. Compliant actuation of rehabilitation robots. IEEE Robotics and AutomationMagazine 2008;15:60–9.

14] Chou CP, Hannaford B. Measurement and modeling of McKibben pneu-matic artificial muscles. IEEE Transactions on Robotics and Automation1996;12:90–102.

15] Ueda J, Ming D, Krishnamoorthy V, Shinohara M, Ogasawara T. Indi-vidual muscle control using an exoskeleton robot for muscle functiontesting. IEEE Transactions on Neural Systems and Rehabilitation Engineering2010;18:339–50.

16] Ferris DP, Gordon KE, Sawicki GS, Peethambaran A. An improved poweredankle–foot orthosis using proportional myoelectric control. Gait and Posture2006;23:425–8.

17] Campolo D, Accoto D, Formica D, Guglielmelli E. Intrinsic constraints of neuralorigin: assessment and application to rehabilitation robotics. IEEE Transactionson Robotics 2009;25:492–501.

18] Veneman JF, Ekkelenkamp R, Kruidhof R, Van Der Helm FCT, Van Der Kooij H.A series elastic- and bowden-cable-based actuation system for use as torqueactuator in exoskeleton-type robots. International Journal of Robotics Research2006;25:261–81.

19] Winter DA. The Biomechanics and Motor Control of Human Gait: Normal,Elderly and Pathological. 2nd ed. Waterloo: University of Waterloo Press;1991.

20] Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, et al. OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEETransactions on Biomedical Engineering 2007;54:1940–50.

21] Van Der Linde RQ. Design, analysis, and control of a low power joint for walk-ing robots, by phasic activation of McKibben muscles. IEEE Transactions onRobotics and Automation 1999;15:599–604.

22] Reynolds DB, Repperger DW, Phillips CA, Bandry G. Modeling the dynamiccharacteristics of pneumatic muscle. Annals of Biomedical Engineering2003;31:310–7.

23] Lilly JH, Yang L. Sliding mode tracking for pneumatic muscle actuators in

opposing pair configuration. IEEE Transactions on Control Systems Technology2005;13:550–8.

24] Von Zitzewitz J, Bernhardt M, Riener R. A novel method for automatic tread-mill speed adaptation. IEEE Transactions on Neural Systems and RehabilitationEngineering 2007;15:401–9.