Research ArticleInteractive Compliance Control of a Wrist RehabilitationDevice (WReD) with Enhanced Training Safety

Dong Xu 1 Mingming Zhang 23 Han Xu 1 Jianming Fu4 Xiaolong Li1

and Sheng Q Xie 56

1Auckland Tongji Medical amp Rehabilitation Equipment Research Centre Tongji Zhejiang College Jiaxing China2Department of Biomedical Engineering Southern University of Science and Technology Shenzhen China3Key Lab of Digital Manufacturing Equipment amp Technology Huazhong University of Science and Technology Wuhan China4Rehabilitation Medical Centre 0e Second Hospital of Jiaxing Jiaxing China5School of Electronic and Electrical Engineering University of Leeds Leeds UK6Institute of Robotics Faculty of Engineering Qingdao University of Technology Qingdao China

Correspondence should be addressed to Mingming Zhang zhangmmsustceducn

Received 3 May 2018 Accepted 21 January 2019 Published 18 February 2019

Academic Editor Maria Linden

Copyright copy 2019 Dong Xu et al +is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Interaction control plays an important role in rehabilitation devices to ensure training safety and efficacy Compliance adaptationof interaction is vital for enabling robot movements to better suit the patientrsquos requirements as human joint characteristics vary+is paper proposes an interactive compliance control scheme on a wrist rehabilitation device (WReD) for enhanced trainingsafety and efficacy +is control system consists of a low-level trajectory tracking loop and a high-level admittance loop Ex-periments were conducted with zero load and human interaction respectively Satisfactory trajectory tracking responses wereobtained with the normalized root mean square deviation (NRMSD) values being 108with zero load and the NRMSD values nogreater than 14 with real-time disturbance and interaction from human users Results demonstrate that such an interactivecompliance control method can adaptively adjust the range of training motions and encourage active engagement from humanusers simultaneously +ese findings suggest that the proposed control method of the WReD has great potentials for clinicalapplications due to enhanced training safety and efficacy Future work will focus on evaluating its efficacy on a large sampleof participants

1 Introduction

According to a report from American Heart Association 33million stroke cases happened worldwide in 2010 with 169million people having a first stroke [1] In the United Statesmore than 700000 people suffer a stroke each year andapproximately two-thirds of these individuals survive andrequire rehabilitation [2] In New Zealand (NZ) there is anestimated 60000 stroke survivors and many of them havemobility impairments [3] Stroke is the third reason forhealth loss and takes the proportion of 39 percent especiallyfor the group starting on middle age suffering the stroke as anonfatal disease in NZ [4] Stroke is also a serious disease inEurope 200 to 300 of 100000 people in Europe suffer from a

stroke as new sufferers and approximately 30 survive withmotor deficits [4] In China approximately 16 millionpeople die of stroke annually in a total population of 14billion people [5] In general stroke has a great adverseimpact for many people worldwide regardless of ethnicgroups and regions

Professor Caplan who studies Neurology at HarvardMedical School describes stroke as a term which is a kind ofbrain impairment as a result of abnormal blood supply in aportion of the brain [5] +e brain injury is most likelyleading to dysfunctions and disabilities +ese survivorsnormally have difficulties in activities of daily living such aswalking speaking and understanding and paralysis ornumbness of the human limbs +e goals of rehabilitation

HindawiJournal of Healthcare EngineeringVolume 2019 Article ID 6537848 10 pageshttpsdoiorg10115520196537848

are to help survivors become as independent as possible andto attain the best possible quality of life

Physical therapy is conventionally delivered by thetherapist and requires a one-to-one physical therapist to domanual interactions with patients While this has beendemonstrated as an effective way for motor rehabilitation[6] it is time consuming and costly Treatments manuallyprovided by therapists require to take place in a specificenvironment (in a hospital or rehabilitation center) and maylast several months for enhanced rehabilitation efficacy [7]A study by Kleim et al [8] has shown that physical therapylike regular exercises can improve plasticity of a nervoussystem and then benefits motor enrichment procedures inpromoting rehabilitation of brain functional models It is atruth that physical therapy should be a preferable way to takepatients into regular exercises and guided by a physicaltherapist but Chang et al [9] showed that it is a money-consuming scheme While manual physical therapy plays acrucial role in the recovery from a stroke it is likely toprovide a lack of consistency for objective assessment fordetermining rehabilitative plans [10] For this reason theindustry has started to seek more solutions through themanufacture of robotic technologies integrated into therecovery process for supporting or substituting manualtherapy

Robot-assisted rehabilitation solutions as therapeuticadjuncts to facilitate clinical practice have been activelyresearched in the past few decades and provide an overduetransformation of the rehabilitation center from labor-intensive operations to technology-assisted operations[11] Engineers and people from medical fields are makingtremendous efforts to make the rehabilitation robots saferand more adaptable for the human body [12 13] +e robotcould also provide a rich stream of data from built-in sensorsto facilitate patient diagnosis customization of the therapyand maintenance of patient records As a popular neuro-rehabilitation technique Liao et al [14] indicated that robot-assisted therapy presents market potential due to quantifi-cation and individuation in the therapy session +equantification of robot-assisted therapy refers that a robotcan provide consistent training pattern without fatigue withthe given parameter +e characterization of individuationallows therapists to customize a specific training scheme foran individual

Upper extremity function is of paramount importance tocarryout various activities of daily living [15] in which thehuman wrist plays a vital role when orienting of an objectFor the rehabilitation of the human wrist a variety of roboticdevices have been developed in the past few decades [16]Some rehabilitation robots have been developed by com-bining the movement of both arm and wrist +e MIT-MANUS has been clinically demonstrated as an excellenttool for shoulder and elbow rehabilitation on stroke patients[17] Krebs et al [18] further developed a wrist rehabilitationrobot with three rotational degrees-of-freedom (DOFs)+iswrist device can be operated stand-alone or mounted at thetip of the MIT-MANUS Faghihi et al [19] constructed athree DOFs wrist robot by using a similar structure as thework by Krebs et al [18] +ey only introduced the design

and fabrication of the wrist robot without the introductionof a control system Toth et al [20] clinically validated thesafety and efficacy of the REHAROB in helping spastichemiparetic patients with passive physiotherapy +is robotwas built from two industrial robots which is not cost ef-fective Oblak et al [15] developed a universal haptic devicethat enables rehabilitation of either arm or wrist movementdepending on locking or unlocking of a passive universaljoint Some portable devices were developed with the focuson rehabilitation of the wrist joint Colombo et al [21]proposed a single-DOF rehabilitation robot for flexion andextension of the wrist joint+is device is actuated by a directcurrent motor with the integration of a torque transducerand a potentiometer in providing feedback signals whichallows the implementation of advanced interactive trainingstrategies Khor et al [22] developed a single-DOF devicethat can enable wrist and forearm training in differentconfigurations +is device has not been controlled withadvanced interaction modes while it has advantages of costeffectiveness and portability

+is paper proposes an interactive compliance controlscheme on a wrist rehabilitation device (WReD) for en-hanced training safety and efficacy +is control systemconsists of a low-level trajectory tracking loop and a high-level admittance loop Experiments were conducted withzero load and human interaction respectively to evaluatethe dynamic performance of the control system +ispaper is organized as follows Following the Introductiona description of theWReD development is given as well asits trajectory tracking control system withwithoutcompliance adaptation +e control stability is also ana-lyzed and presented +e section of Experimental Resultsis introduced next with the Discussion and Conclusion atlast

2 Methods

21 Wrist Rehabilitation Device (WReD) +e wrist jointanatomically possesses two DOFs flexion and extensionand abduction (radial deviation) and adduction (ulnardeviation) [23] +e WReD presented in this studyfunctions only for wrist flexion and extension re-habilitation training While this device can be reconfig-ured into wrist training of radialulnar deviation this doesnot affect the study design in evaluating the proposedcontrol scheme

+e prototype of WReD has been built and reported inour previous work [24] as in Figure 1 where Figure 1(a) isthe three-dimensional (3D) model of the device created inSolidWorks Figure 1(b) shows its use on a healthy subjectand Figure 1(c) is the control box +e WReD mainlyconsists of three components (the base module the ac-tuation module and the arm-hand module) +e basemodule acts as a foundation to support the actuation andthe arm-hand module It consists of the bottom base andthree vertical support bars +e motor and two-stagereduction gear box comprise the actuation module +earm-hand module consists of the arm holder the handleholder and the handle +e handle can be designed to be

2 Journal of Healthcare Engineering

subject-specific based on the requirement+e mechanicallimit part is set to avoid excessive rotation of the handlefor safety reasons Some bearings are also set to allow lowfriction motion transmission between the rotational axisand the base module Before training the human userswill be instructed to make appropriate adjustments fortheir hands including grabbing the handle though thehandle holder placing forearms on the arm holder andfixing forearms with Velcro straps

+e electrical components of the WReD consist of a DCmotor (EC 45 Maxon) a motor driver (ESCON 505Maxon) a static torque sensor (JNNT 25degNm Zhong-wan) a magnetic rotary encoder (AS5048A AMS) and anembedded controller (National Instruments myRIO-1900)+e ESCON driver myRIO controller and the transmitterof the torque sensor are set in the control box as shown inFigure 1(c) +e motor EC 45 nominally outputs834degmNm through the gear box with reduction ratio 1 300 and there is an estimated torque output of 2502degNmWith the consideration of the transmission efficiency being07614 the WReD can have a torque output of 1905degNm atthe end effector (also the handle) +e torque sensor isinstalled between the output shaft of the actuation module(through the coupling) and the handle holder for mea-suring the torque of human and robot interaction Amagnetic rotary sensor is installed on the shaft of thehandle holder for measuring the angular position of thehuman hand in real time An emergency stop is also set toensure training safety Predefined data and those from theelectrical components of the WReD communicate with a

computer through the embedded controller (myRIO-1900)

22 Control System +e trajectory tracking control of theWReD is the basis of a variety of robot-assisted rehabilitationexercises It can be directly used for passive training onpatients with weak active motor ability Tracking desiredtrajectories is not only a simple but also an effectivemethod for rehabilitation applications [25] Introducingcompliance to trajectory tracking control can lead toenhanced training comfort and safety and also allowsactive engagement for better rehabilitation efficacy [26ndash28] Compliance control is an important element duringthe interaction between patients and robots since theinteraction torque needs to be in a safe range Meanwhilethe position of the robot should also be controlled in thesame way to minimize the follow error during the tra-jectory tracking [29] Impedance control describes thedynamic relation between position and torque well and donot require accurate knowledge of the external environ-ment when compared to the other methods such ashybrid force and position control [17 29] +erefore thistechnique has been broadly implemented in rehabilitationrobots

Due to the large variability in human wrist character-istics and lack of accurate information of the environmentthis paper proposes a two-level control system for trajectorytracking control of the WReD with compliance adaption asin Figure 2 In the low level a closed loop system is used to

Bottom baseTorque sensor

Coupling

Motor andgearbox

Handle holderHandle

Magnetic rotary sensor Mechanical limit part

Arm holderSupport bar

(a)

Emergency stop

(b)

Control box

(c)

Figure 1 +e WReD system (a) +e 3D model (b) its use on a healthy subject and (c) the control box

Journal of Healthcare Engineering 3

achieve trajectory tracking by comparing desired trajectoryfrom the high-level controller and actual trajectory measuredfrom themagnetic rotary sensor To allow trajectory generationwith compliance adaptation an admittance law is adopted inthe high level In general the high-level controller dynamicallyadjusts the desired trajectory based on the feedback of themeasured interaction torque and sends the angular positioncommand to the low-level controller as an input signal

In the high level a reference trajectory can be definedby a physiotherapist or based on pilot tests denoted asθr(t) +e admittance control law is proposed as inequation (1) where θr(t) and θd(t) represent the referencetrajectory and the recalculated desired trajectory re-spectively and Ti(t) is the patient-robot interaction torqueB and K respectively represent the damping and stiffnesscoefficients and M is the inertia tensor In this studyparameter M and B are assumed to be negligible due tolow-velocity movement environment and low friction ofthe bearings +us equation (2) can be derived With theknowledge of the inertial property of the WReD the de-sired trajectory position θd(t) can be derived In this studyfurther in the middle level the measured trajectory positionθm(t) is compared with θd(t) for error in equation (3)Lastly the error e(t) is input to the PID controller and therequired speed of the motor v(t) can be calculatedaccording to equation (4) with well-tuned Kp Ki and Kd+e parameter K in equations (1) and (2) is an indication ofthe device compliance Specifically the training with thiscontroller can be considered to be completely passive whenK is infinite leading to θd(t) approximately equal to θr(t)With the variable K decreasing the participant gets to beable to change the compliance of the WReD with real-timehuman-robot interaction +e less the K is the morecompliance the device has thus the participant can affectthe predefined trajectory more easily

Ti(t) M( euroθd(t) minus euroθr(t)1113857 + B _θd(t) minus _θr(t)1113872 1113873

+ K θd(t)minus θr(t)( 1113857 (1)

θd(t) θr(t) +Ti(t)

K (2)

e(t) θd(t)minus θm(t) (3)

v(t) Kpe(t) + Ki 1113946t

0e(t) dt + Kd

de(t)

dt (4)

23 Stability Analysis To ensure the safety of the proposedcompliance control strategy it is essential to conduct sta-bility analysis Equation (1) can be transformed to equation(5) where θi(t) θr(t)minus θd(t) θi is the position disturbancewhen the human wrist interacts with the WReD With theinitial condition θi(0) 0 _θi 0 the transfer function (6) ofthe proposed compliance controller can be obtained throughLaplace transformation +us equation (6) can be written asequation (7) to describe the relationship between the outputand the input

Ti(t) minus1113858M euroθi(t) + B _θi(t) + Kθi(t)1113859 (5)

Ti(s) minus Ms2

+ Bs + K1113960 1113961X(s) (6)

G(s) X(s)

Ti(s) minus

1Ms2 + Bs + K

(7)

+e complex eigenvalue analysis method is a useful toolfor analyzing the stability of control systems which requiresthe calculation of the systemrsquos complex eigenvalues +ecorresponding characteristic equation of equation (7) can bewritten as follows

Ms2

+ Bs + K 0 (8)

An eigenvalue of equation (8) is in complex form ofα + jω where α is the real part of s which indicates thestability of the system and ω is the imaginary part of s whichindicates the modal frequency [30] +e eigenvalues ofequation (8) can be obtained as follows

s minusB plusmn

B2 minus 4MK

radic

2M (9)

With Bgt 0 and Kgt 0 it is obviously seen that the systemis stable with both eigenvalues in the left-half of the complexplane +e stability level depends on the real parts of ei-genvalues +erefore the controller parameter design iscritical to achieve a stable control system

To visualize the stability status the Bode diagram methodis adopted To conduct stability analysis with differentcompliance levels the variable K was set with different valuesPretests were conducted and it was found that when the robotcompliance has significant differences when K is 08 or 16 Asequence of pretests was also carried out with the differentvalues ofM and B From the Bode diagrams the phasemarginplots can be obviously presented when M 015 and B 02By using this method the controller parameters are assumedwith two groups according to the compliance difference

Maxon controller Wred

Mndash1

B

intint

K

PID

Handle

Ti Stactic torque sensor

Magnetic rotary encoderθm

Measured human-robot interaction torque

High-level trajectory adaptation Low-level trajectory trackingθr

θd

Figure 2 +e control diagram of the WReD

4 Journal of Healthcare Engineering

M 015 B 02 and K 08 and M 015 B 02 andK 16 +e Bode diagrams of the proposed compliancecontroller are presented in Figures 3(a) and 3(b) In the case ofK 08 it shows the phase margin c 402deg with the gaincrossover frequency ωC 42 rads While in the case ofK 16 the phase margin is c 505deg with the gain crossoverfrequency ωC 39 rads Since both of the phase marginplots do not cross with minus180deg the gain margins becomeinfinity According to the Bode stability criterion [31] theminimum phase system is stable with the phase margin cgt 0and the phase margin hgt 1 Back to the compliance controllerpart it is a minimum phase system since both eigenvalues arein the left-half of the complex plane +erefore the com-pliance controller part is stable

24 Participants and Training Protocol To preliminarilyevaluate the performance of the proposed control system ontheWReD a healthy subject volunteered to participate in thetest in a lab environment +e participant is male with theage of 27 years old 170 cm height and 65 kg body weightBefore the training the actual range of motion (ROM) of thewrist joint was measured for safety +e healthy subjectfollowed the training instructions well and had no confusionin using the WReD +e test was under the supervision ofexperienced clinical personnel and engineers

Passive training has been demonstrated as an effectiveway to induce quick recovery and expand joint ROM[32 33] Trajectory tracking is a conventional method usedto investigate the performance of passive motions In thisstudy the training trajectory is set as a sine wave piecewisefunction as in equation (5) which allows the training to beslow at wrist flexionextension limited position for comfortand safety Based on the measurements by using traditionalgoniometers the participant has the wrist flexion up to 80degand the wrist extension up to 65deg +is is consistent withpublished data of normal wrist range of motion 0ndash80deg forflexion and 0ndash70deg for extension [23] Here it is defined thewrist flexion for negative and extension for positive +usthe parameter Aext in equation (5) is set as 65deg and Afle asminus80deg +e frequency f is set to be 005Hz In equation (4)the parameters Kp Ki and Kd are tuned by using CohenCoon method [34] to be 03 06 and 0 for tracking desiredtrajectory θd(t)

θr(t) Aext sin(2lowast pilowastflowast t) when θm(t)ge 0

Afle sin(2lowast pilowastflowast t) when θm(t)lt 01113896

(10)

To validate the trajectory tracking responses and whetherthe proposed controller is capable of changing the complianceof theWReD four experiments were conducted with each for10 cycles Experiment 1 was conducted with zero load usingonly trajectory tracking control (without trajectory adapta-tion) By the same control method Experiment 2 was con-ducted with the hand of the participant on Experiments 3 and4 were conducted to check whether the proposed method iscapable of changing the compliance of the WReD with theparameter K being 08 and 16 respectively During these

experiments the healthy subject was verbally encouraged torelax his wrist during passive training

3 Experimental Results

+e trajectory tracking response of Experiment 1 is pre-sented in Figure 4 by using MATLAB R2016a where thedark line sine wave is the desired trajectory the dotted redline is the measured trajectory the blue line is the error as inequation (3) and the dark line in the bottom plot is themeasured torque from the torque sensor Without externalload the desired trajectory compares well with the desiredtrajectory To allow quantitative analysis of the trajectorytracking performance under this control method the sta-tistical results are given in Table 1 with the error rangingfrom minus2998deg to 3074deg the root mean square deviation(RMSD) being 15682deg and the normalized root meansquare deviation (NRMSD) being 108 Figure 4 also showsonly a slight variation of the interaction torque throughoutthe training +is is due to the lack of human-robot in-teraction and the small variation is caused by friction be-tween the arm-hand module and the actuation module Asstatistical results in Table 1 the mean of absolute torquevalue (MATV) is only 0152degNm

Figure 5 presents the trajectory tracking response ofExperiment 2 using the same control method as that inExperiment 1 During the whole process the participant wasencouraged to relax and did not apply intentional activeforce on the handle To allow quantitative comparison withExperiment 1 statistical results are also given in Table 1 +etrajectory tracking error ranges from minus3054deg to 3110deg theRMSD value is 15767deg and the NRMSD value is 108 Bycomparing with the interaction torque presented in Figure 4and Table 1 there is more obvious torque variation duringeach cycle of training with the MATV value being1825degNm +is is caused by the resistance when the wrist isat limited joint position Experiments 1 and 2 both showsatisfactory trajectory tracking performance with theNRMSD value being 108

To evaluate the proposed control method especially thehigh-level trajectory adaptation method to adjust thecompliance of the WReD Experiments 3 and 4 were con-ducted with the same participant +e parameter K inequation (2) represents the stiffness coefficient of theWReDExperiment 3 has the K value of 08 and trajectory trackingresponse is presented in Figure 6 In the top plot the darkline is the desired trajectory after adaptation from the ref-erence sine wave trajectory (plotted in green) the dotted redline is the measured trajectory and the blue line is the errorof the trajectory tracking As summarized in Table 1 thetrajectory tracking error ranges from minus12784deg to 7998deg theRMSD value is 20879deg and the NRMSD value is 14Experiment 4 has the K value of 16 and trajectory trackingresponse is shown in Figure 7 As summarized in Table 1 thetrajectory tracking error ranges from minus6081deg to 5211deg theRMSD is 16211deg and the NRMSD is 112 It was found thatthe desired trajectory was adapted based on the interactiontorque with respect to the reference trajectory

Journal of Healthcare Engineering 5

20

ndash20

ndash40

ndash60

ndash800

ndash45

ndash90

ndash135

ndash180

Mag

nitu

de (d

B)Ph

ase (

deg

)

10ndash1 100

Frequency (rads)101 102

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 42

System G (s)Phase margin (deg) 402Crossover freq (rads) 42

(a)

20

ndash20

ndash40

Mag

nitu

de (d

B)Ph

ase (

deg

)

ndash60

ndash80

ndash45

ndash90

ndash135

ndash180

Frequency (rads)10ndash1 100 101 102

0

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 39

System G (s)Phase margin (deg) 505Crossover freq (rads) 39

(b)

Figure 3 +e Bode diagrams of the proposed compliance controller with two different configurations (a) M 015 B 02 and K 08(b) M 015 B 02 and K 16

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

DesiredMeasuredError

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

Figure 4 +e trajectory tracking responses of Experiment 1 (zero load)

Table 1 Experimental results of the tracking responses of all experiments

MinError (deg) MaxError (deg) RMSD (deg) NRMSD () MATV (Nm) NRMSDlowast ()Zero load Experiment 1 minus2998 3074 15682 108 0152 NAPassive training Experiment 2 minus3054 3110 15767 108 1825 NA

Interaction training K 08 Experiment 3 minus12784 7998 20879 14 3669 413K 16 Experiment 4 minus6081 5211 16211 112 4446 224

Note MinError is the maximum tracking error in negative direction which is numerically minimum MaxError is the maximum tracking error in positivedirection which is numerically maximum RMSD root mean square deviation NRMSD normalized root mean square deviation MATV mean of absolutetorque value NRMSDlowast represents the NRMSD value of the adapted desired trajectory θd(t) with respect to the reference trajectory θr(t) NA not applicable

6 Journal of Healthcare Engineering

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

subject-specific based on the requirement+e mechanicallimit part is set to avoid excessive rotation of the handlefor safety reasons Some bearings are also set to allow lowfriction motion transmission between the rotational axisand the base module Before training the human userswill be instructed to make appropriate adjustments fortheir hands including grabbing the handle though thehandle holder placing forearms on the arm holder andfixing forearms with Velcro straps

+e electrical components of the WReD consist of a DCmotor (EC 45 Maxon) a motor driver (ESCON 505Maxon) a static torque sensor (JNNT 25degNm Zhong-wan) a magnetic rotary encoder (AS5048A AMS) and anembedded controller (National Instruments myRIO-1900)+e ESCON driver myRIO controller and the transmitterof the torque sensor are set in the control box as shown inFigure 1(c) +e motor EC 45 nominally outputs834degmNm through the gear box with reduction ratio 1 300 and there is an estimated torque output of 2502degNmWith the consideration of the transmission efficiency being07614 the WReD can have a torque output of 1905degNm atthe end effector (also the handle) +e torque sensor isinstalled between the output shaft of the actuation module(through the coupling) and the handle holder for mea-suring the torque of human and robot interaction Amagnetic rotary sensor is installed on the shaft of thehandle holder for measuring the angular position of thehuman hand in real time An emergency stop is also set toensure training safety Predefined data and those from theelectrical components of the WReD communicate with a

computer through the embedded controller (myRIO-1900)

22 Control System +e trajectory tracking control of theWReD is the basis of a variety of robot-assisted rehabilitationexercises It can be directly used for passive training onpatients with weak active motor ability Tracking desiredtrajectories is not only a simple but also an effectivemethod for rehabilitation applications [25] Introducingcompliance to trajectory tracking control can lead toenhanced training comfort and safety and also allowsactive engagement for better rehabilitation efficacy [26ndash28] Compliance control is an important element duringthe interaction between patients and robots since theinteraction torque needs to be in a safe range Meanwhilethe position of the robot should also be controlled in thesame way to minimize the follow error during the tra-jectory tracking [29] Impedance control describes thedynamic relation between position and torque well and donot require accurate knowledge of the external environ-ment when compared to the other methods such ashybrid force and position control [17 29] +erefore thistechnique has been broadly implemented in rehabilitationrobots

Due to the large variability in human wrist character-istics and lack of accurate information of the environmentthis paper proposes a two-level control system for trajectorytracking control of the WReD with compliance adaption asin Figure 2 In the low level a closed loop system is used to

Bottom baseTorque sensor

Coupling

Motor andgearbox

Handle holderHandle

Magnetic rotary sensor Mechanical limit part

Arm holderSupport bar

(a)

Emergency stop

(b)

Control box

(c)

Figure 1 +e WReD system (a) +e 3D model (b) its use on a healthy subject and (c) the control box

Journal of Healthcare Engineering 3

achieve trajectory tracking by comparing desired trajectoryfrom the high-level controller and actual trajectory measuredfrom themagnetic rotary sensor To allow trajectory generationwith compliance adaptation an admittance law is adopted inthe high level In general the high-level controller dynamicallyadjusts the desired trajectory based on the feedback of themeasured interaction torque and sends the angular positioncommand to the low-level controller as an input signal

In the high level a reference trajectory can be definedby a physiotherapist or based on pilot tests denoted asθr(t) +e admittance control law is proposed as inequation (1) where θr(t) and θd(t) represent the referencetrajectory and the recalculated desired trajectory re-spectively and Ti(t) is the patient-robot interaction torqueB and K respectively represent the damping and stiffnesscoefficients and M is the inertia tensor In this studyparameter M and B are assumed to be negligible due tolow-velocity movement environment and low friction ofthe bearings +us equation (2) can be derived With theknowledge of the inertial property of the WReD the de-sired trajectory position θd(t) can be derived In this studyfurther in the middle level the measured trajectory positionθm(t) is compared with θd(t) for error in equation (3)Lastly the error e(t) is input to the PID controller and therequired speed of the motor v(t) can be calculatedaccording to equation (4) with well-tuned Kp Ki and Kd+e parameter K in equations (1) and (2) is an indication ofthe device compliance Specifically the training with thiscontroller can be considered to be completely passive whenK is infinite leading to θd(t) approximately equal to θr(t)With the variable K decreasing the participant gets to beable to change the compliance of the WReD with real-timehuman-robot interaction +e less the K is the morecompliance the device has thus the participant can affectthe predefined trajectory more easily

Ti(t) M( euroθd(t) minus euroθr(t)1113857 + B _θd(t) minus _θr(t)1113872 1113873

+ K θd(t)minus θr(t)( 1113857 (1)

θd(t) θr(t) +Ti(t)

K (2)

e(t) θd(t)minus θm(t) (3)

v(t) Kpe(t) + Ki 1113946t

0e(t) dt + Kd

de(t)

dt (4)

23 Stability Analysis To ensure the safety of the proposedcompliance control strategy it is essential to conduct sta-bility analysis Equation (1) can be transformed to equation(5) where θi(t) θr(t)minus θd(t) θi is the position disturbancewhen the human wrist interacts with the WReD With theinitial condition θi(0) 0 _θi 0 the transfer function (6) ofthe proposed compliance controller can be obtained throughLaplace transformation +us equation (6) can be written asequation (7) to describe the relationship between the outputand the input

Ti(t) minus1113858M euroθi(t) + B _θi(t) + Kθi(t)1113859 (5)

Ti(s) minus Ms2

+ Bs + K1113960 1113961X(s) (6)

G(s) X(s)

Ti(s) minus

1Ms2 + Bs + K

(7)

+e complex eigenvalue analysis method is a useful toolfor analyzing the stability of control systems which requiresthe calculation of the systemrsquos complex eigenvalues +ecorresponding characteristic equation of equation (7) can bewritten as follows

Ms2

+ Bs + K 0 (8)

An eigenvalue of equation (8) is in complex form ofα + jω where α is the real part of s which indicates thestability of the system and ω is the imaginary part of s whichindicates the modal frequency [30] +e eigenvalues ofequation (8) can be obtained as follows

s minusB plusmn

B2 minus 4MK

radic

2M (9)

With Bgt 0 and Kgt 0 it is obviously seen that the systemis stable with both eigenvalues in the left-half of the complexplane +e stability level depends on the real parts of ei-genvalues +erefore the controller parameter design iscritical to achieve a stable control system

To visualize the stability status the Bode diagram methodis adopted To conduct stability analysis with differentcompliance levels the variable K was set with different valuesPretests were conducted and it was found that when the robotcompliance has significant differences when K is 08 or 16 Asequence of pretests was also carried out with the differentvalues ofM and B From the Bode diagrams the phasemarginplots can be obviously presented when M 015 and B 02By using this method the controller parameters are assumedwith two groups according to the compliance difference

Maxon controller Wred

Mndash1

B

intint

K

PID

Handle

Ti Stactic torque sensor

Magnetic rotary encoderθm

Measured human-robot interaction torque

High-level trajectory adaptation Low-level trajectory trackingθr

θd

Figure 2 +e control diagram of the WReD

4 Journal of Healthcare Engineering

M 015 B 02 and K 08 and M 015 B 02 andK 16 +e Bode diagrams of the proposed compliancecontroller are presented in Figures 3(a) and 3(b) In the case ofK 08 it shows the phase margin c 402deg with the gaincrossover frequency ωC 42 rads While in the case ofK 16 the phase margin is c 505deg with the gain crossoverfrequency ωC 39 rads Since both of the phase marginplots do not cross with minus180deg the gain margins becomeinfinity According to the Bode stability criterion [31] theminimum phase system is stable with the phase margin cgt 0and the phase margin hgt 1 Back to the compliance controllerpart it is a minimum phase system since both eigenvalues arein the left-half of the complex plane +erefore the com-pliance controller part is stable

24 Participants and Training Protocol To preliminarilyevaluate the performance of the proposed control system ontheWReD a healthy subject volunteered to participate in thetest in a lab environment +e participant is male with theage of 27 years old 170 cm height and 65 kg body weightBefore the training the actual range of motion (ROM) of thewrist joint was measured for safety +e healthy subjectfollowed the training instructions well and had no confusionin using the WReD +e test was under the supervision ofexperienced clinical personnel and engineers

Passive training has been demonstrated as an effectiveway to induce quick recovery and expand joint ROM[32 33] Trajectory tracking is a conventional method usedto investigate the performance of passive motions In thisstudy the training trajectory is set as a sine wave piecewisefunction as in equation (5) which allows the training to beslow at wrist flexionextension limited position for comfortand safety Based on the measurements by using traditionalgoniometers the participant has the wrist flexion up to 80degand the wrist extension up to 65deg +is is consistent withpublished data of normal wrist range of motion 0ndash80deg forflexion and 0ndash70deg for extension [23] Here it is defined thewrist flexion for negative and extension for positive +usthe parameter Aext in equation (5) is set as 65deg and Afle asminus80deg +e frequency f is set to be 005Hz In equation (4)the parameters Kp Ki and Kd are tuned by using CohenCoon method [34] to be 03 06 and 0 for tracking desiredtrajectory θd(t)

θr(t) Aext sin(2lowast pilowastflowast t) when θm(t)ge 0

Afle sin(2lowast pilowastflowast t) when θm(t)lt 01113896

(10)

To validate the trajectory tracking responses and whetherthe proposed controller is capable of changing the complianceof theWReD four experiments were conducted with each for10 cycles Experiment 1 was conducted with zero load usingonly trajectory tracking control (without trajectory adapta-tion) By the same control method Experiment 2 was con-ducted with the hand of the participant on Experiments 3 and4 were conducted to check whether the proposed method iscapable of changing the compliance of the WReD with theparameter K being 08 and 16 respectively During these

experiments the healthy subject was verbally encouraged torelax his wrist during passive training

3 Experimental Results

+e trajectory tracking response of Experiment 1 is pre-sented in Figure 4 by using MATLAB R2016a where thedark line sine wave is the desired trajectory the dotted redline is the measured trajectory the blue line is the error as inequation (3) and the dark line in the bottom plot is themeasured torque from the torque sensor Without externalload the desired trajectory compares well with the desiredtrajectory To allow quantitative analysis of the trajectorytracking performance under this control method the sta-tistical results are given in Table 1 with the error rangingfrom minus2998deg to 3074deg the root mean square deviation(RMSD) being 15682deg and the normalized root meansquare deviation (NRMSD) being 108 Figure 4 also showsonly a slight variation of the interaction torque throughoutthe training +is is due to the lack of human-robot in-teraction and the small variation is caused by friction be-tween the arm-hand module and the actuation module Asstatistical results in Table 1 the mean of absolute torquevalue (MATV) is only 0152degNm

Figure 5 presents the trajectory tracking response ofExperiment 2 using the same control method as that inExperiment 1 During the whole process the participant wasencouraged to relax and did not apply intentional activeforce on the handle To allow quantitative comparison withExperiment 1 statistical results are also given in Table 1 +etrajectory tracking error ranges from minus3054deg to 3110deg theRMSD value is 15767deg and the NRMSD value is 108 Bycomparing with the interaction torque presented in Figure 4and Table 1 there is more obvious torque variation duringeach cycle of training with the MATV value being1825degNm +is is caused by the resistance when the wrist isat limited joint position Experiments 1 and 2 both showsatisfactory trajectory tracking performance with theNRMSD value being 108

To evaluate the proposed control method especially thehigh-level trajectory adaptation method to adjust thecompliance of the WReD Experiments 3 and 4 were con-ducted with the same participant +e parameter K inequation (2) represents the stiffness coefficient of theWReDExperiment 3 has the K value of 08 and trajectory trackingresponse is presented in Figure 6 In the top plot the darkline is the desired trajectory after adaptation from the ref-erence sine wave trajectory (plotted in green) the dotted redline is the measured trajectory and the blue line is the errorof the trajectory tracking As summarized in Table 1 thetrajectory tracking error ranges from minus12784deg to 7998deg theRMSD value is 20879deg and the NRMSD value is 14Experiment 4 has the K value of 16 and trajectory trackingresponse is shown in Figure 7 As summarized in Table 1 thetrajectory tracking error ranges from minus6081deg to 5211deg theRMSD is 16211deg and the NRMSD is 112 It was found thatthe desired trajectory was adapted based on the interactiontorque with respect to the reference trajectory

Journal of Healthcare Engineering 5

20

ndash20

ndash40

ndash60

ndash800

ndash45

ndash90

ndash135

ndash180

Mag

nitu

de (d

B)Ph

ase (

deg

)

10ndash1 100

Frequency (rads)101 102

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 42

System G (s)Phase margin (deg) 402Crossover freq (rads) 42

(a)

20

ndash20

ndash40

Mag

nitu

de (d

B)Ph

ase (

deg

)

ndash60

ndash80

ndash45

ndash90

ndash135

ndash180

Frequency (rads)10ndash1 100 101 102

0

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 39

System G (s)Phase margin (deg) 505Crossover freq (rads) 39

(b)

Figure 3 +e Bode diagrams of the proposed compliance controller with two different configurations (a) M 015 B 02 and K 08(b) M 015 B 02 and K 16

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

DesiredMeasuredError

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

Figure 4 +e trajectory tracking responses of Experiment 1 (zero load)

Table 1 Experimental results of the tracking responses of all experiments

MinError (deg) MaxError (deg) RMSD (deg) NRMSD () MATV (Nm) NRMSDlowast ()Zero load Experiment 1 minus2998 3074 15682 108 0152 NAPassive training Experiment 2 minus3054 3110 15767 108 1825 NA

Interaction training K 08 Experiment 3 minus12784 7998 20879 14 3669 413K 16 Experiment 4 minus6081 5211 16211 112 4446 224

Note MinError is the maximum tracking error in negative direction which is numerically minimum MaxError is the maximum tracking error in positivedirection which is numerically maximum RMSD root mean square deviation NRMSD normalized root mean square deviation MATV mean of absolutetorque value NRMSDlowast represents the NRMSD value of the adapted desired trajectory θd(t) with respect to the reference trajectory θr(t) NA not applicable

6 Journal of Healthcare Engineering

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

achieve trajectory tracking by comparing desired trajectoryfrom the high-level controller and actual trajectory measuredfrom themagnetic rotary sensor To allow trajectory generationwith compliance adaptation an admittance law is adopted inthe high level In general the high-level controller dynamicallyadjusts the desired trajectory based on the feedback of themeasured interaction torque and sends the angular positioncommand to the low-level controller as an input signal

In the high level a reference trajectory can be definedby a physiotherapist or based on pilot tests denoted asθr(t) +e admittance control law is proposed as inequation (1) where θr(t) and θd(t) represent the referencetrajectory and the recalculated desired trajectory re-spectively and Ti(t) is the patient-robot interaction torqueB and K respectively represent the damping and stiffnesscoefficients and M is the inertia tensor In this studyparameter M and B are assumed to be negligible due tolow-velocity movement environment and low friction ofthe bearings +us equation (2) can be derived With theknowledge of the inertial property of the WReD the de-sired trajectory position θd(t) can be derived In this studyfurther in the middle level the measured trajectory positionθm(t) is compared with θd(t) for error in equation (3)Lastly the error e(t) is input to the PID controller and therequired speed of the motor v(t) can be calculatedaccording to equation (4) with well-tuned Kp Ki and Kd+e parameter K in equations (1) and (2) is an indication ofthe device compliance Specifically the training with thiscontroller can be considered to be completely passive whenK is infinite leading to θd(t) approximately equal to θr(t)With the variable K decreasing the participant gets to beable to change the compliance of the WReD with real-timehuman-robot interaction +e less the K is the morecompliance the device has thus the participant can affectthe predefined trajectory more easily

Ti(t) M( euroθd(t) minus euroθr(t)1113857 + B _θd(t) minus _θr(t)1113872 1113873

+ K θd(t)minus θr(t)( 1113857 (1)

θd(t) θr(t) +Ti(t)

K (2)

e(t) θd(t)minus θm(t) (3)

v(t) Kpe(t) + Ki 1113946t

0e(t) dt + Kd

de(t)

dt (4)

23 Stability Analysis To ensure the safety of the proposedcompliance control strategy it is essential to conduct sta-bility analysis Equation (1) can be transformed to equation(5) where θi(t) θr(t)minus θd(t) θi is the position disturbancewhen the human wrist interacts with the WReD With theinitial condition θi(0) 0 _θi 0 the transfer function (6) ofthe proposed compliance controller can be obtained throughLaplace transformation +us equation (6) can be written asequation (7) to describe the relationship between the outputand the input

Ti(t) minus1113858M euroθi(t) + B _θi(t) + Kθi(t)1113859 (5)

Ti(s) minus Ms2

+ Bs + K1113960 1113961X(s) (6)

G(s) X(s)

Ti(s) minus

1Ms2 + Bs + K

(7)

+e complex eigenvalue analysis method is a useful toolfor analyzing the stability of control systems which requiresthe calculation of the systemrsquos complex eigenvalues +ecorresponding characteristic equation of equation (7) can bewritten as follows

Ms2

+ Bs + K 0 (8)

An eigenvalue of equation (8) is in complex form ofα + jω where α is the real part of s which indicates thestability of the system and ω is the imaginary part of s whichindicates the modal frequency [30] +e eigenvalues ofequation (8) can be obtained as follows

s minusB plusmn

B2 minus 4MK

radic

2M (9)

With Bgt 0 and Kgt 0 it is obviously seen that the systemis stable with both eigenvalues in the left-half of the complexplane +e stability level depends on the real parts of ei-genvalues +erefore the controller parameter design iscritical to achieve a stable control system

To visualize the stability status the Bode diagram methodis adopted To conduct stability analysis with differentcompliance levels the variable K was set with different valuesPretests were conducted and it was found that when the robotcompliance has significant differences when K is 08 or 16 Asequence of pretests was also carried out with the differentvalues ofM and B From the Bode diagrams the phasemarginplots can be obviously presented when M 015 and B 02By using this method the controller parameters are assumedwith two groups according to the compliance difference

Maxon controller Wred

Mndash1

B

intint

K

PID

Handle

Ti Stactic torque sensor

Magnetic rotary encoderθm

Measured human-robot interaction torque

High-level trajectory adaptation Low-level trajectory trackingθr

θd

Figure 2 +e control diagram of the WReD

4 Journal of Healthcare Engineering

M 015 B 02 and K 08 and M 015 B 02 andK 16 +e Bode diagrams of the proposed compliancecontroller are presented in Figures 3(a) and 3(b) In the case ofK 08 it shows the phase margin c 402deg with the gaincrossover frequency ωC 42 rads While in the case ofK 16 the phase margin is c 505deg with the gain crossoverfrequency ωC 39 rads Since both of the phase marginplots do not cross with minus180deg the gain margins becomeinfinity According to the Bode stability criterion [31] theminimum phase system is stable with the phase margin cgt 0and the phase margin hgt 1 Back to the compliance controllerpart it is a minimum phase system since both eigenvalues arein the left-half of the complex plane +erefore the com-pliance controller part is stable

24 Participants and Training Protocol To preliminarilyevaluate the performance of the proposed control system ontheWReD a healthy subject volunteered to participate in thetest in a lab environment +e participant is male with theage of 27 years old 170 cm height and 65 kg body weightBefore the training the actual range of motion (ROM) of thewrist joint was measured for safety +e healthy subjectfollowed the training instructions well and had no confusionin using the WReD +e test was under the supervision ofexperienced clinical personnel and engineers

Passive training has been demonstrated as an effectiveway to induce quick recovery and expand joint ROM[32 33] Trajectory tracking is a conventional method usedto investigate the performance of passive motions In thisstudy the training trajectory is set as a sine wave piecewisefunction as in equation (5) which allows the training to beslow at wrist flexionextension limited position for comfortand safety Based on the measurements by using traditionalgoniometers the participant has the wrist flexion up to 80degand the wrist extension up to 65deg +is is consistent withpublished data of normal wrist range of motion 0ndash80deg forflexion and 0ndash70deg for extension [23] Here it is defined thewrist flexion for negative and extension for positive +usthe parameter Aext in equation (5) is set as 65deg and Afle asminus80deg +e frequency f is set to be 005Hz In equation (4)the parameters Kp Ki and Kd are tuned by using CohenCoon method [34] to be 03 06 and 0 for tracking desiredtrajectory θd(t)

θr(t) Aext sin(2lowast pilowastflowast t) when θm(t)ge 0

Afle sin(2lowast pilowastflowast t) when θm(t)lt 01113896

(10)

To validate the trajectory tracking responses and whetherthe proposed controller is capable of changing the complianceof theWReD four experiments were conducted with each for10 cycles Experiment 1 was conducted with zero load usingonly trajectory tracking control (without trajectory adapta-tion) By the same control method Experiment 2 was con-ducted with the hand of the participant on Experiments 3 and4 were conducted to check whether the proposed method iscapable of changing the compliance of the WReD with theparameter K being 08 and 16 respectively During these

experiments the healthy subject was verbally encouraged torelax his wrist during passive training

3 Experimental Results

+e trajectory tracking response of Experiment 1 is pre-sented in Figure 4 by using MATLAB R2016a where thedark line sine wave is the desired trajectory the dotted redline is the measured trajectory the blue line is the error as inequation (3) and the dark line in the bottom plot is themeasured torque from the torque sensor Without externalload the desired trajectory compares well with the desiredtrajectory To allow quantitative analysis of the trajectorytracking performance under this control method the sta-tistical results are given in Table 1 with the error rangingfrom minus2998deg to 3074deg the root mean square deviation(RMSD) being 15682deg and the normalized root meansquare deviation (NRMSD) being 108 Figure 4 also showsonly a slight variation of the interaction torque throughoutthe training +is is due to the lack of human-robot in-teraction and the small variation is caused by friction be-tween the arm-hand module and the actuation module Asstatistical results in Table 1 the mean of absolute torquevalue (MATV) is only 0152degNm

Figure 5 presents the trajectory tracking response ofExperiment 2 using the same control method as that inExperiment 1 During the whole process the participant wasencouraged to relax and did not apply intentional activeforce on the handle To allow quantitative comparison withExperiment 1 statistical results are also given in Table 1 +etrajectory tracking error ranges from minus3054deg to 3110deg theRMSD value is 15767deg and the NRMSD value is 108 Bycomparing with the interaction torque presented in Figure 4and Table 1 there is more obvious torque variation duringeach cycle of training with the MATV value being1825degNm +is is caused by the resistance when the wrist isat limited joint position Experiments 1 and 2 both showsatisfactory trajectory tracking performance with theNRMSD value being 108

To evaluate the proposed control method especially thehigh-level trajectory adaptation method to adjust thecompliance of the WReD Experiments 3 and 4 were con-ducted with the same participant +e parameter K inequation (2) represents the stiffness coefficient of theWReDExperiment 3 has the K value of 08 and trajectory trackingresponse is presented in Figure 6 In the top plot the darkline is the desired trajectory after adaptation from the ref-erence sine wave trajectory (plotted in green) the dotted redline is the measured trajectory and the blue line is the errorof the trajectory tracking As summarized in Table 1 thetrajectory tracking error ranges from minus12784deg to 7998deg theRMSD value is 20879deg and the NRMSD value is 14Experiment 4 has the K value of 16 and trajectory trackingresponse is shown in Figure 7 As summarized in Table 1 thetrajectory tracking error ranges from minus6081deg to 5211deg theRMSD is 16211deg and the NRMSD is 112 It was found thatthe desired trajectory was adapted based on the interactiontorque with respect to the reference trajectory

Journal of Healthcare Engineering 5

20

ndash20

ndash40

ndash60

ndash800

ndash45

ndash90

ndash135

ndash180

Mag

nitu

de (d

B)Ph

ase (

deg

)

10ndash1 100

Frequency (rads)101 102

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 42

System G (s)Phase margin (deg) 402Crossover freq (rads) 42

(a)

20

ndash20

ndash40

Mag

nitu

de (d

B)Ph

ase (

deg

)

ndash60

ndash80

ndash45

ndash90

ndash135

ndash180

Frequency (rads)10ndash1 100 101 102

0

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 39

System G (s)Phase margin (deg) 505Crossover freq (rads) 39

(b)

Figure 3 +e Bode diagrams of the proposed compliance controller with two different configurations (a) M 015 B 02 and K 08(b) M 015 B 02 and K 16

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

DesiredMeasuredError

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

Figure 4 +e trajectory tracking responses of Experiment 1 (zero load)

Table 1 Experimental results of the tracking responses of all experiments

MinError (deg) MaxError (deg) RMSD (deg) NRMSD () MATV (Nm) NRMSDlowast ()Zero load Experiment 1 minus2998 3074 15682 108 0152 NAPassive training Experiment 2 minus3054 3110 15767 108 1825 NA

Interaction training K 08 Experiment 3 minus12784 7998 20879 14 3669 413K 16 Experiment 4 minus6081 5211 16211 112 4446 224

Note MinError is the maximum tracking error in negative direction which is numerically minimum MaxError is the maximum tracking error in positivedirection which is numerically maximum RMSD root mean square deviation NRMSD normalized root mean square deviation MATV mean of absolutetorque value NRMSDlowast represents the NRMSD value of the adapted desired trajectory θd(t) with respect to the reference trajectory θr(t) NA not applicable

6 Journal of Healthcare Engineering

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

M 015 B 02 and K 08 and M 015 B 02 andK 16 +e Bode diagrams of the proposed compliancecontroller are presented in Figures 3(a) and 3(b) In the case ofK 08 it shows the phase margin c 402deg with the gaincrossover frequency ωC 42 rads While in the case ofK 16 the phase margin is c 505deg with the gain crossoverfrequency ωC 39 rads Since both of the phase marginplots do not cross with minus180deg the gain margins becomeinfinity According to the Bode stability criterion [31] theminimum phase system is stable with the phase margin cgt 0and the phase margin hgt 1 Back to the compliance controllerpart it is a minimum phase system since both eigenvalues arein the left-half of the complex plane +erefore the com-pliance controller part is stable

24 Participants and Training Protocol To preliminarilyevaluate the performance of the proposed control system ontheWReD a healthy subject volunteered to participate in thetest in a lab environment +e participant is male with theage of 27 years old 170 cm height and 65 kg body weightBefore the training the actual range of motion (ROM) of thewrist joint was measured for safety +e healthy subjectfollowed the training instructions well and had no confusionin using the WReD +e test was under the supervision ofexperienced clinical personnel and engineers

Passive training has been demonstrated as an effectiveway to induce quick recovery and expand joint ROM[32 33] Trajectory tracking is a conventional method usedto investigate the performance of passive motions In thisstudy the training trajectory is set as a sine wave piecewisefunction as in equation (5) which allows the training to beslow at wrist flexionextension limited position for comfortand safety Based on the measurements by using traditionalgoniometers the participant has the wrist flexion up to 80degand the wrist extension up to 65deg +is is consistent withpublished data of normal wrist range of motion 0ndash80deg forflexion and 0ndash70deg for extension [23] Here it is defined thewrist flexion for negative and extension for positive +usthe parameter Aext in equation (5) is set as 65deg and Afle asminus80deg +e frequency f is set to be 005Hz In equation (4)the parameters Kp Ki and Kd are tuned by using CohenCoon method [34] to be 03 06 and 0 for tracking desiredtrajectory θd(t)

θr(t) Aext sin(2lowast pilowastflowast t) when θm(t)ge 0

Afle sin(2lowast pilowastflowast t) when θm(t)lt 01113896

(10)

To validate the trajectory tracking responses and whetherthe proposed controller is capable of changing the complianceof theWReD four experiments were conducted with each for10 cycles Experiment 1 was conducted with zero load usingonly trajectory tracking control (without trajectory adapta-tion) By the same control method Experiment 2 was con-ducted with the hand of the participant on Experiments 3 and4 were conducted to check whether the proposed method iscapable of changing the compliance of the WReD with theparameter K being 08 and 16 respectively During these

experiments the healthy subject was verbally encouraged torelax his wrist during passive training

3 Experimental Results

+e trajectory tracking response of Experiment 1 is pre-sented in Figure 4 by using MATLAB R2016a where thedark line sine wave is the desired trajectory the dotted redline is the measured trajectory the blue line is the error as inequation (3) and the dark line in the bottom plot is themeasured torque from the torque sensor Without externalload the desired trajectory compares well with the desiredtrajectory To allow quantitative analysis of the trajectorytracking performance under this control method the sta-tistical results are given in Table 1 with the error rangingfrom minus2998deg to 3074deg the root mean square deviation(RMSD) being 15682deg and the normalized root meansquare deviation (NRMSD) being 108 Figure 4 also showsonly a slight variation of the interaction torque throughoutthe training +is is due to the lack of human-robot in-teraction and the small variation is caused by friction be-tween the arm-hand module and the actuation module Asstatistical results in Table 1 the mean of absolute torquevalue (MATV) is only 0152degNm

Figure 5 presents the trajectory tracking response ofExperiment 2 using the same control method as that inExperiment 1 During the whole process the participant wasencouraged to relax and did not apply intentional activeforce on the handle To allow quantitative comparison withExperiment 1 statistical results are also given in Table 1 +etrajectory tracking error ranges from minus3054deg to 3110deg theRMSD value is 15767deg and the NRMSD value is 108 Bycomparing with the interaction torque presented in Figure 4and Table 1 there is more obvious torque variation duringeach cycle of training with the MATV value being1825degNm +is is caused by the resistance when the wrist isat limited joint position Experiments 1 and 2 both showsatisfactory trajectory tracking performance with theNRMSD value being 108

To evaluate the proposed control method especially thehigh-level trajectory adaptation method to adjust thecompliance of the WReD Experiments 3 and 4 were con-ducted with the same participant +e parameter K inequation (2) represents the stiffness coefficient of theWReDExperiment 3 has the K value of 08 and trajectory trackingresponse is presented in Figure 6 In the top plot the darkline is the desired trajectory after adaptation from the ref-erence sine wave trajectory (plotted in green) the dotted redline is the measured trajectory and the blue line is the errorof the trajectory tracking As summarized in Table 1 thetrajectory tracking error ranges from minus12784deg to 7998deg theRMSD value is 20879deg and the NRMSD value is 14Experiment 4 has the K value of 16 and trajectory trackingresponse is shown in Figure 7 As summarized in Table 1 thetrajectory tracking error ranges from minus6081deg to 5211deg theRMSD is 16211deg and the NRMSD is 112 It was found thatthe desired trajectory was adapted based on the interactiontorque with respect to the reference trajectory

Journal of Healthcare Engineering 5

20

ndash20

ndash40

ndash60

ndash800

ndash45

ndash90

ndash135

ndash180

Mag

nitu

de (d

B)Ph

ase (

deg

)

10ndash1 100

Frequency (rads)101 102

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 42

System G (s)Phase margin (deg) 402Crossover freq (rads) 42

(a)

20

ndash20

ndash40

Mag

nitu

de (d

B)Ph

ase (

deg

)

ndash60

ndash80

ndash45

ndash90

ndash135

ndash180

Frequency (rads)10ndash1 100 101 102

0

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 39

System G (s)Phase margin (deg) 505Crossover freq (rads) 39

(b)

Figure 3 +e Bode diagrams of the proposed compliance controller with two different configurations (a) M 015 B 02 and K 08(b) M 015 B 02 and K 16

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

DesiredMeasuredError

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

Figure 4 +e trajectory tracking responses of Experiment 1 (zero load)

Table 1 Experimental results of the tracking responses of all experiments

MinError (deg) MaxError (deg) RMSD (deg) NRMSD () MATV (Nm) NRMSDlowast ()Zero load Experiment 1 minus2998 3074 15682 108 0152 NAPassive training Experiment 2 minus3054 3110 15767 108 1825 NA

Interaction training K 08 Experiment 3 minus12784 7998 20879 14 3669 413K 16 Experiment 4 minus6081 5211 16211 112 4446 224

Note MinError is the maximum tracking error in negative direction which is numerically minimum MaxError is the maximum tracking error in positivedirection which is numerically maximum RMSD root mean square deviation NRMSD normalized root mean square deviation MATV mean of absolutetorque value NRMSDlowast represents the NRMSD value of the adapted desired trajectory θd(t) with respect to the reference trajectory θr(t) NA not applicable

6 Journal of Healthcare Engineering

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

20

ndash20

ndash40

ndash60

ndash800

ndash45

ndash90

ndash135

ndash180

Mag

nitu

de (d

B)Ph

ase (

deg

)

10ndash1 100

Frequency (rads)101 102

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 42

System G (s)Phase margin (deg) 402Crossover freq (rads) 42

(a)

20

ndash20

ndash40

Mag

nitu

de (d

B)Ph

ase (

deg

)

ndash60

ndash80

ndash45

ndash90

ndash135

ndash180

Frequency (rads)10ndash1 100 101 102

0

0

System G (s)Magnitude (dB) 0Crossover freq (rads) 39

System G (s)Phase margin (deg) 505Crossover freq (rads) 39

(b)

Figure 3 +e Bode diagrams of the proposed compliance controller with two different configurations (a) M 015 B 02 and K 08(b) M 015 B 02 and K 16

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

DesiredMeasuredError

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

Figure 4 +e trajectory tracking responses of Experiment 1 (zero load)

Table 1 Experimental results of the tracking responses of all experiments

MinError (deg) MaxError (deg) RMSD (deg) NRMSD () MATV (Nm) NRMSDlowast ()Zero load Experiment 1 minus2998 3074 15682 108 0152 NAPassive training Experiment 2 minus3054 3110 15767 108 1825 NA

Interaction training K 08 Experiment 3 minus12784 7998 20879 14 3669 413K 16 Experiment 4 minus6081 5211 16211 112 4446 224

Note MinError is the maximum tracking error in negative direction which is numerically minimum MaxError is the maximum tracking error in positivedirection which is numerically maximum RMSD root mean square deviation NRMSD normalized root mean square deviation MATV mean of absolutetorque value NRMSDlowast represents the NRMSD value of the adapted desired trajectory θd(t) with respect to the reference trajectory θr(t) NA not applicable

6 Journal of Healthcare Engineering

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

Furthermore to analyze the experimental results pre-sented in Figures 6 and 7 and Table 1 while Experiment 3has a smaller interaction torque with the MATV of3669degNm it presents a larger NRMSDlowast value (413) of thetrajectory adaptation +is demonstrates that the controlmethod with K value being 08 has more compliance withrespect to Experiment 4 (K 16) As the stiffness coefficientof the WReD increases the training can be considered toapproach to passive mode similar to Experiment 2 In termsof the trajectory tracking accuracy of Experiments 3 and 4the latter has a better tracking performance than the formerwith NRMSD values being 112 and 14 respectively

+is finding is reasonable since the WReD system with morecompliance is more susceptible to external disturbancewhich is also reflected in Table 1 with the MinError ofExperiment 3 up to 12784deg +e data for Experiment 1 2 3and 4 are included within the supplementary informationfiles to support the findings of this study

4 Discussion and Conclusion

+is paper presents the development of the WReD and acompliance control method for comfortable and safehuman-robot interaction By contrast with existing wrist

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 6 +e trajectory tracking responses of Experiment 3 (K 08)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash100

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasured Error

Reference

Figure 7 +e trajectory tracking responses of Experiment 4 (K 16)

0 20 40 60 80 100 120 140 160 180 200

ndash50

0

50

Ang

le (d

egre

es)

0 20 40 60 80 100 120 140 160 180 200Time (seconds)

ndash10

0

10

Inte

ract

ion

torq

ue (N

m)

DesiredMeasuredError

Figure 5 +e trajectory tracking responses of Experiment 2 (passive training)

Journal of Healthcare Engineering 7

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

rehabilitation robots [15 18 22] the WReD is portable foruse in hospital or home environment due to its reconfig-urable structure design In this study the WReD is con-figured for the rehabilitation training of wrist flexion andextension as shown in Figure 1 since wrist flexion andextension are worth 70 of total wrist function [35] andtraining therapy along this direction is more commonly usedthan that of radial and ulnar deviation It is worth men-tioning that the WReD can be easily reconfigured fortraining of wrist radial and ulnar deviation by rotating thehandle to a vertical posture along the handle holderHowever this is not the focus of this study More impor-tantly this has no effect on the experiment design as well asthe feasibility evaluation of the proposed control method+is study presents experimental validation of the proposedinteractive compliance control with satisfactory perfor-mance summarized in Table 1 While tests were conductedonly with wrist flexion and extension it is reasonable topredict that the proposed method will also apply to thetraining of wrist radial and ulnar deviation

Human-robot interactive tasks cannot be handled bypure motion control that generally rejects forces exerted byhuman users as disturbances +e impedance controlscheme is mostly considered as the basis of interactive ro-botic training It has been widely adopted in rehabilitationrobots [29 36] to obtain user comfort and enhance thetraining safety +ere are two ways of implementing im-pedance control based on the controller causality imped-ance control (force- or torque-based method) andadmittance control (position-based method) [37ndash39] Whileboth implementations were referred to as impedance controlby Hogan [40] we make a distinction since this is highlyrelevant to the work presented in this study +e impedancecontroller takes a displacement as input and reacts with aforce while in an admittance control mode the controller isan admittance and the manipulator is an impedance [28]+e interactive compliance control of the WReD wasachieved based on an admittance law due to the availabilityof direct measurement of the interaction torque Underadmittance mode the WReD deviates from the referencetrajectory in the presence of patient-robot interaction but isotherwise following the reference trajectory

Experiments were conducted by implementing the tra-jectory tracking combined with compliance control As thestatistical results in Table 1 the proposed compliance controlmethod has a satisfactory trajectory tracking accuracy withall NRMSD values no greater than 14 +e stability of thecompliance controller was analyzed by calculating thecomplex eigenvalues and also has been validated with thesufficient phase margin and gain margin according to theBode stability criteria Different stiffness coefficients weremanually set to evaluate the effect on the compliance of thepresented control method +e introduction of the com-pliance control method can make robot-assisted wrist re-habilitation training more comfortable and safer by avoidingexcessive interaction torque on human wrists

While experiments have been conducted with satisfac-tory compliance and trajectory tracking performance thisstudy suffers from some limitations First the developed

WReD can be further improved for clinical applicationsSome measures should be taken to address the issue ofmisalignment between the device and the human wrist andto comfortably fix human arms preventing relative move-ments +is may have affected the measurement of theposition and interaction torque of human wrists Second toachieve better control performance the inertial property ofthe WReD should be considered in equation (1) as well asthe inertial property of human hands +ird this study hasonly one participant for such a preliminary test moresubjects with hand disabilities should be encouraged andrecruited in future tests

Future work will focus on the improvement of the WReDin terms of its functionality and clinical evaluation Mechan-ically we will consider designing different handles for use ondifferent patients addressing the misalignment issue alsoadding another actuation module for multiple dimensiontraining For applications this device can be used for not onlyrehabilitation training but also assessment purpose in mea-suring wrist stiffness and ranges of motion Its efficacy forassessment will be investigated next Directly following thisstudymore research is needed to improve the robustness of theproposed interactive compliance control scheme to externaldisturbances and involuntary human-robot interaction Futurework will also consider proposing a high-level algorithm toenable automatic K value tuning

To summarize this paper proposes an interactive com-pliance control scheme on the WReD for enhanced trainingsafety and efficacy +is control system consists of a low-leveltrajectory tracking loop and a high-level admittance loopExperiments were conducted with zero load and humaninteraction respectively Satisfactory trajectory tracking re-sponses were obtained with the NRMSD values being 108with zero load and the NRMSD values no greater than 14with real-time disturbance and interaction from the humanuser +e interactive compliance can be adjusted in subjectspecific by setting different stiffness coefficients of the controlsystem Results demonstrate that such an interactive com-pliance control method can adaptively adjust the range oftraining motions and encourage human usersrsquo active en-gagement +ese findings suggest that the proposed controlmethod of the WReD has great potentials for clinical ap-plications due to enhanced training safety and efficacy

Data Availability

+edata for Experiment 1 2 3 and 4 are included within thesupplementary information files to support the findings ofthis study Also we share the original data as the supple-mentary information files

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is research project is supported by Jiaxing Science andTechnology Bureau (2017AY13023) China and Key Lab of

8 Journal of Healthcare Engineering

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

Digital Manufacturing Equipment amp Technology(DMETKF2018005) Huazhong University of Science andTechnology China In addition the authors would like togive special thanks to Ya Sun Xu Zhang and Yibin Li whohave made contributions in the development of the wristrehabilitation device in the previous work

Supplementary Materials

+e original data of Experiment 1 to Experiment 4 are in-cluded +e data file of each experiment contains fourcolumns named Time (second) Desired Angle (degree)Interaction Torque (Nm) and Measured Angle (degree)Figures 4ndash7 were derived from the original data of Exper-iment 1 to Experiment 4 by using MATLAB Below are thedescriptions of the four experiments Experiment 1 thetrajectory tracking responses with zero load Experiment 2the trajectory tracking responses with passive trainingExperiment 3 the trajectory tracking responses with K 08Experiment 4 the trajectory tracking responses with K 16(Supplementary Materials)

References

[1] S C Johnston S Mendis and C D Mathers ldquoGlobal vari-ation in stroke burden and mortality estimates from moni-toring surveillance and modellingrdquo 0e Lancet Neurologyvol 8 no 4 pp 345ndash354 2009

[2] D Mozaffafian E J Benjamin A S Go et al ldquoHeart diseaseand stroke statisticsmdash2015 Updaterdquo Circulation vol 131no 4 pp 29ndash322 2015

[3] ldquoFacts about stroke in New Zealandrdquo httpswwwstrokeorgnzfacts-and-faqs

[4] M Turley and Ministry of Health Health Loss in New Zea-land a Report from the New Zealand Burden of DiseasesInjuries and Risk Factors Study 2006ndash2016 Ministry of MTurley Health Ministry of Health Wellington New Zealand2013

[5] L R Caplan Stroke AAN Press American Academy ofNeurology New York NY USA 2006

[6] P S Lum C G Burgar P C Shor M Majmundar andM Van der Loos ldquoRobot-assisted movement training com-pared with conventional therapy techniques for the re-habilitation of upper-limb motor function after strokerdquoArchives of Physical Medicine and Rehabilitation vol 83no 7 pp 952ndash959 2002

[7] R Loureiro F Amirabdollahian M Topping B Driessenand W Harwin ldquoUpper limb robot mediated stroke ther-apymdashGENTLEs approachrdquo Autonomous Robots vol 15no 1 pp 35ndash51 2003

[8] J A Kleim T A Jones and T Schallert ldquoMotor enrichmentand the induction of plasticity before or after brain injuryrdquoNeurochemical Research vol 28 no 11 pp 1757ndash1769 2003

[9] Y-J Chang W-Y Han and Y-C Tsai ldquoA Kinect-basedupper limb rehabilitation system to assist people with cerebralpalsyrdquo Research in Developmental Disabilities vol 34 no 11pp 3654ndash3659 2013

[10] A U Pehlivan S Lee and M K OrsquoMalley ldquoMechanicaldesign of RiceWrist-S a forearm-wrist exoskeleton for strokeand spinal cord injury rehabilitationrdquo in Proceedings of IEEERas amp Embs International Conference on Biomedical Roboticsand Biomechatronics pp 1573ndash1578 Rome Italy June 2012

[11] H I Krebs J J Palazzolo L Dipietro et al ldquoRehabilitationrobotics performance-based progressive robot-assistedtherapyrdquo Autonomous Robots vol 15 no 1 pp 7ndash20 2003

[12] G Kwakkel B J Kollen and H I Krebs ldquoEffects of robot-assisted therapy on upper limb recovery after stroke a sys-tematic reviewrdquo Neurorehabilitation and Neural Repairvol 22 no 2 pp 111ndash121 2008

[13] J Stein H I Krebs W R Frontera S E Fasoli R Hughesand N Hogan ldquoComparison of two techniques of robot-aidedupper limb exercise training after strokerdquoAmerican Journal ofPhysical Medicine amp Rehabilitation vol 83 no 9 pp 720ndash728 2004

[14] W-W Liao C-Y Wu Y-W Hsieh K-C Lin andW-Y Chang ldquoEffects of robot-assisted upper limb re-habilitation on daily function and real-world arm activity inpatients with chronic stroke a randomized controlled trialrdquoClinical Rehabilitation vol 26 no 2 pp 111ndash120 2012

[15] J Oblak I Cikajlo and Z Matjacic ldquoUniversal haptic drive arobot for arm and wrist rehabilitationrdquo IEEE Transactions onNeural Systems amp Rehabilitation Engineering vol 18 no 3pp 293ndash302 2010

[16] S Mazzoleni M C Carrozza P Sale M FranceschiniF Posteraro and M Tiboni ldquoEffects of upper limb robot-assisted therapy on motor recovery of subacute stroke pa-tients a kinematic approachrdquo in Proceedings of IEEE 13thInternational Conference on Rehabilitation Robotics (ICORR)pp 1ndash5 Seattle WA USA June 2013

[17] B T Volpe H I Krebs N Hogan L Edelsteinn C M Dielsand M L Aisen ldquoRobot training enhanced motor outcome inpatients with stroke maintained over 3 yearsrdquo Neurologyvol 53 no 8 pp 1874ndash1876 1999

[18] H I Krebs B T Volpe D Williams et al ldquoRobot-aidedneurorehabilitation a robot for wrist rehabilitationrdquo IEEETransactions on Neural Systems and Rehabilitation Engi-neering vol 15 no 3 pp 327ndash335 2007

[19] A Faghihi S A Haghpanah F Farahmand and M JafarildquoDesign and fabrication of a robot for neurorehabilitationsmart robo wristrdquo in Proceedings of 2015 2nd InternationalConference on Knowledge-Based Engineering and InnovationKBEI pp 447ndash450 Tehran Iran November 2015

[20] A Toth G Fazekas G Arz M Jurak and M HorvathldquoPassive robotic movement therapy of the spastic hemipareticarm with REHAROB report of the first clinical test and thefollow-up system improvementrdquo in Proceedings of 2005 IEEE9th International Conference on Rehabilitation Roboticsvol 2005 pp 127ndash130 Chicago IL USA July 2005

[21] R Colombo F Pisano S Micera et al ldquoRobotic techniquesfor upper limb evaluation and rehabilitation of stroke pa-tientsrdquo IEEE Transactions on Neural Systems and Re-habilitation Engineering vol 13 no 3 pp 311ndash324 2005

[22] K X Khor P J H Chin A R Hisyam C F YeongA L T Narayanan and E L M Su ldquoDevelopment of CR2-Haptic a compact and portable rehabilitation robot for wristand forearm trainingrdquo in Proceedings of IEEE Conference onBiomedical Engineering and Sciences (IECBES) pp 424ndash429Miri Malaysia December 2014

[23] Y Youm and A E Flatt ldquoDesign of a total wrist prosthesisrdquoAnnals of Biomedical Engineering vol 12 no 3 pp 247ndash2621984

[24] M Zhang D Xu Y Sun et al ldquoDevelopment of a recon-figurable wrist rehabilitation device with an adaptive forearmholderrdquo in Proceedings of IEEEASME International Confer-ence on Advanced Intelligent Mechatronics (AIM) pp 454ndash459 Auckland New Zealand July 2018

Journal of Healthcare Engineering 9

[25] T Proietti V Crocher A Roby-Brami and N JarrasseldquoUpper-limb robotic exoskeletons for neurorehabilitation areview on control strategiesrdquo IEEE Reviews in BiomedicalEngineering vol 9 pp 4ndash14 2016

[26] M Zhang T C Davies and S Xie ldquoEffectiveness of robot-assisted therapy on ankle rehabilitationndasha systematic reviewrdquoJournal of Neuro Engineering and Rehabilitation vol 10 no 1p 30 2013

[27] H Yu S Huang G Chen Y Pan and Z Guo ldquoHumanndashrobotinteraction control of rehabilitation robots with series elasticactuatorsrdquo IEEE Transactions on Robotics vol 31 no 5pp 1089ndash1100 2015

[28] M Zhang S Q Xie X Li et al ldquoAdaptive patient-cooperativecontrol of a compliant ankle rehabilitation robot (CARR) withenhanced training safetyrdquo IEEE Transactions on IndustrialElectronics vol 65 no 2 pp 1398ndash1407 2017

[29] Y H Tsoi and S Q Xie ldquoImpedance control of ankle re-habilitation robotrdquo in Proceedings of IEEE InternationalConference on Robotics and Biomimetics pp 840ndash845Bangkok +ailand February 2009

[30] X Wang J L Mo H Ouyang et al ldquoSqueal noise of frictionmaterial with groove-textured surface an experimental andnumerical analysisrdquo Journal of Tribology vol 138 no 2 article021401 2015

[31] D R Coughanowr and L B Koppel Process Systems Analysisand Control McGraw-Hill New York NY USA 1991

[32] B R Salter and P Field ldquo+e effects of continuous com-pression on living articular cartilage an experimental in-vestigationrdquo Journal of Bone amp Joint Surgery vol 42 no 1pp 31ndash90 1960

[33] S W OrsquoDriscoll and N J Giori ldquoContinuous passive motion(CPM) theory and principles of clinical applicationrdquo Journalof Rehabilitation Research amp Development vol 37 no 2pp 179ndash188 2000

[34] M Z F B M Zawawi I Elamvazuthi A B A Aziz andS A Daud ldquoComparison of PID and fuzzy logic controller forDC servo motor in the development of lower extremityexoskeleton for rehabilitationrdquo in Proceedings of IEEE In-ternational Symposium in Robotics and Manufacturing Au-tomation pp 1ndash6 Rome Italy September 2017

[35] A B Swanson C G Hagert and G D Swanson ldquoEvaluationof impairment of hand functionrdquo Journal of Hand Surgeryvol 8 no 5 pp 709ndash722 1983

[36] Y Yang L Wang J Tong and L Zhang ldquoArm rehabilitationrobot impedance control and experimentationrdquo in Pro-ceedings of IEEE International Conference on Robotics andBiomimetics pp 914ndash918 Roma Italy April 2007

[37] D A Lawrence ldquoImpedance control stability properties incommon implementationsrdquo in Proceedings of IEEE In-ternational Conference on Robotics and Automation vol 2pp 1185ndash1190 Philadelphia PA USA April 1988

[38] C Ott R Mukherjee and Y Nakamura ldquoUnified impedanceand admittance controlrdquo in Proceedings of IEEE InternationalConference on Robotics and Automation pp 554ndash561 An-chorage AK USA May 2010

[39] C Ott and Y Nakamura ldquoBase forcetorque sensing forposition based Cartesian impedance controlrdquo in Proceedingsof IEEERSJ International Conference on Intelligent Robots andSystems pp 3244ndash3250 St Louis MO USA October 2009

[40] N Hogan ldquoImpedance control an approach to manipula-tionrdquo in Proceedings of American Control Conferencepp 481ndash489 St Louis MO USA June 2009

10 Journal of Healthcare Engineering

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom