ADDITIONAL DESIGN FEATURES OF A MASTER–SLAVE … · ADDITIONAL DESIGN FEATURES OF A MASTER–SLAVE CONTROL SYSTEM WITH FORCE SENSING AND ENERGY RECYCLING FOR UPPER LIMB REHABILITATION

ADDITIONAL DESIGN FEATURES OF A MASTER–SLAVE CONTROLSYSTEM WITH FORCE SENSING AND ENERGY RECYCLING FORUPPER LIMB REHABILITATION ROBOTS

Chunguang Li, Yoshio Inoue, Tao Liu, Kyoko Shibata, and Koichi Oka

Department of Intelligent Mechanical Systems Engineering, Kochi University ofTechnology, Kami-City, Kochi, Japan

& Traditional upper-limb rehabilitation robots usually realize force feedback with force sensors orimpedance controllers. Otherwise, assistant or resistant force required in different training modes isgiven by the robot, which does not motivate the initiative of patients sufficiently. This article intro-duces a self-controlled upper-limb rehabilitation robot to implement force sensing without a forcesensor or an impedance controller. The system supports bimanual exercises in different trainingmodes with one limb providing a proper force for the contralateral limb. The above characteristicsand the capability of master–slave motion tracking with a kind of energy recycling were verifiedwith preliminary experiments.

Keywords bimanual training, energy recycling, force sensing, master–slave motiontracking, self-controlled operation

INTRODUCTION

Stroke or brain injury is a common disease, which results in anincreasing number of hemiplegic patients with unilateral limb impairment.Meanwhile, the percentage of aged persons is continuously increasing inmany countries. In the elderly, the prevalence of physical deterioration isvery high, and their physical deterioration generally leads to degenerationof motor function. Therefore, motor function recovery and strengthenhancement are necessary in our aging society. This emerging require-ment has stimulated considerable interest in the development of upperlimb rehabilitation robots, which can act as a therapeutic aid for physicaltherapists, especially under existing conditions in many countries wherethe physical therapy resources are quite limited.

Address correspondence to Chunguang Li, Department of Intelligent Mechanical Systems Engin-eering, Kochi University of Technology, 185 Miyanokuchi, Tosayamada-Cho, Kami-City, Kochi 782-8502,Japan. E-mail: [email protected]

Instrumentation Science and Technology, 38:385–410, 2010Copyright # Taylor & Francis Group, LLCISSN: 1073-9149 print/1525-6030 onlineDOI: 10.1080/10739149.2010.508322

Instrumentation Science and Technology, 38:385–410, 2010Copyright # Taylor & Francis Group, LLCISSN: 1073-9149 print/1525-6030 onlineDOI: 10.1080/10739149.2010.508322

Downloaded By: [Chunguang, Li] At: 12:00 28 October 2010

Numerous robots have been developed to deliver arm therapy, such asMIT-MANUS,[1,2] ARM-GUIDE,[3,4] MIME,[5,6] and ARMin,[7] which are allrepresentative robotic devices that have been tested extensively on hemiple-gic patients and have achieved good results in improving arm function.MIT-MANUS, which is a 2-degrees-of-freedom assisted robot, can supportpatients in executing reaching movements in a horizontal plane. In orderto achieve a better improvement in shoulder strength and function,ARM-GUIDE, MIME, and ARMin were developed to give training in athree-dimensionalworkspace. ARM-GUIDEallows subjects to performactive-assisted reaching movements against gravity. It can be used as both a diag-nostic tool and a treatment tool for addressing arm impairment inhemiparetics. MIME supports unilateral training in passive, active-assisted,and active-constrained modes. Furthermore, it can assist the affected limbto move with the same manner of motion as the contralateral limb, with thetwo limbs performing bilateral mirror image movements. ARMin is roboticdevice that can deliver patient-cooperative arm therapy. It allows patients toplay ball games or perform activities of daily living (ADL)-related tasks bycombining an audiovisual display, which can motivate the activity ofpatients in exercises. However, these robots are relatively complex anddifficult to set up by patients themselves. In addition, the supervision froma therapist is always required during the training process, and so the dura-tions that patients spend in rehabilitation activities are limited, and theeconomic burden to patients is increased.

Home-based rehabilitation makes possible treatments with relativelyhigh intensity and frequency that are favorable for improving motor recov-ery.[8] Therefore, the development of telerehabilitation robots that can beused in patients’ homes is a new trend. A portable telerehabilitationsystem[9–11] with haptic feeling was developed for the treatment and assess-ment of elbow deformity of stroke patients. A real-time control strategy anda teach-and-replay control method are achieved for tasks that involve slowmovements and fast movements, respectively. Also, the torque and positioncontrol modes for the master and slave devices can be exchanged for passiveand active movements. Thus the system supports both the passive and theactive movements including slow and fast tasks. However, the teach-and-replay control does not actually accomplish transparent haptic feeling.Other-wise, the control strategy is complex, since a hybrid force-position controlleris employed and a separate control method is required for the master andslave devices. Furthermore, the operator is a therapist rather than patientsthemselves. Patients are trained passively or with insufficient initiative(visual=video feedback). This is unfavorable for acquiring a good recoveryeffect. Even though therapists can optimize the therapy scheme accordingto the feedback force, the degree of comfort of patients cannot be sensed,so it is possible to make patients feel pain in the process of training.

386 C. Li et al.


Therefore, many self-controlled upper limb rehabilitation robots havebeen developed for use at home.[12–15] Popescu et al. introduced a PC-based rehabilitation system to support home-based VR (virtual-reality) exer-cises.[16] The included networking allows clinic to telemonitor the trainingprocess and to change the difficulty level of exercises. Also, the rehabili-tation component can apply resistive forces to patients. However, no experi-mental results were presented. Colombo et al. presented two robots forhome-based upper limb rehabilitation training.[17] In particular, a newevaluation metrics was proposed for observing the improvement rate andselecting the targeted rehabilitative strategies. During exercise, patientsthemselves control the handle of the device to track a target position repeat-edly. The experimental results demonstrate the validity of this rehabili-tation technique both in recent and in chronic post-stroke patients.

Meanwhile, in order to motivate much more activation of patients inexercises and facilitate motor recovery, some robots that support bimanualtraining were introduced. BATRAC[18] is a robotic device that can deliverbilateral arm therapy for stroke patients. Clinical tests on BATRAC verify thatbilateral movements improved arm function by inducing reorganization ofbrain regions involved in motor control. Hesse et al.[19] proposed a portablerobot for training bilateral forearms and wrists of hemiparetic patients. Thedevice supported passive practice and active bimanual practice with or with-out a resistance to the impaired arm. Preliminary experiments performed on12 chronic hemiparetic patients confirmed that this bilateral arm exercisecan achieve a reduction of wrist and finger spasticity. The results also indi-cated that the device has potential for training severely affected strokepatients. A subsequent trial[20] on 44 acute stroke patients revealed thatthe robot-trained group produced a superior enhancement in upper limbmotor control and power compared with the group that practiced with elec-trical simulation of the paretic wrist extensor. This is probably explained bythe bilateral practice and extensive repetitions of the robot therapy.

However, the robotic systems mentioned above realize force feedback=sensing with torque=force sensors or impedance controllers, which increasethe difficulty of hardware mounting or control complexity of the system. Asfor our research, a self-controlled master-slave device has been developedfor supporting rehabilitation training in a home environment.[21] Thedevice realized force sensing without any force sensor, master-slave motiontracking, and a kind of energy recycling with a compact structure. Previouswork verified the feasibility of the proposed master-slave control scheme.However, the device’s output torque was too small to drive a human limbin rehabilitation exercise. Therefore, we redesigned the device to increaseits driving power and make it capable of supporting bilateral arm rehabili-tation training. In the new system, master and slave units are controlled by apatient with two limbs. This can avoid the time-delay caused by distance

Master–Slave Control System 387


between the master and slave devices (tele-rehabilitation). The system sup-ports bimanual training in passive, active-assisted, and active-resisted modesfor patients with different residual motor function. Different modes arerealized with the healthy limb providing a suitable force (a control forcein passive mode, an assistant force in active-assisted mode, or a resistantforce in active-resisted mode) for the impaired limb.

In this article, except for a further verification of force sensing andenergy recycling, the influence of different gearbox combinations, thecapability of supporting different training modes, and the frequencyresponse range of the system were also confirmed.

MATERIALS AND METHODS

The master–slave system contains two identical DC motors with adirectly wired connection. The master motor works in generating stateand powers the slave motor, which works in electromotive state and outputsan assistant or resistant force for the attached limb, which reproduces themotion of the contralateral limb in the master site. Equivalent closed-loopcircuit of the system is given in Figure 1, in which Mm and Ms represent thetwo motors. In this article, the subscripts m and s mean the master and theslave, respectively, and the variables with subscripts in and out denote theterminal values in the master and slave sites, respectively.

Theoretical Analysis

Based on the dynamics mechanism, the motion equations are written asfollows:

TingmNm

¼ TM þ T0 m þ JmNmdxin

dtTout

Nsgs¼ TM � T0 s � JsNs

dxout

dtTM ¼ CT i

8<: ð1Þ

FIGURE 1 Equivalent closed-loop circuit of the master–slave system.

388 C. Li et al.


where Tin and Tout stand for terminal torques; TM denotes electromagnetictorque; the two motors possess identical electromagnetic torque, since thesame torque constant CT and the shared closed-loop current i; T0_m andT0_s are motor no-load torques caused by no-load losses, including mechan-ical loss, magnetic core loss, and additional loss in the motors; Jm and Jsare motor inertial moments; xin and xout are terminal velocities; and Nm,Ns, and gm, gs represent the gear ratios and working efficiencies of the gear-boxes. According to Equation (1), the relationship between the torques inthe two sites can be written as follows:

Tin ¼ kTTout þ Nm

gmT0 m þ T0 s þ JmNm

dxin

dt þ JsNsdxout

dt

� �kT ¼ Nm

Ns

1gmgs

(ð2Þ

where kT is defined as the force sensing coefficient, which expresses forcesensing capability towards the impedance variation in the slave site. Equation(2) indicates that the torques in the two terminals correspond to each other.This is realized due to the closed-loop current. When the impedance in theslave site increases=reduces, the current as well as the electromagnetic tor-ques of the two motors increase=reduce, and then operators can sense thisvariation in the master site and regulate the input force accordingly in orderto achieve a new balance between the torques in the two sites. This meansthat the system is capable of mirroring the force in the slave site to the mastersite without a force sensor. However, the no-load torques and inertial torquesof the motors increase the requirement for the input torque in the mastersite. But this can not affect the accurate sensation on the variation of externalforce in the slave site for a rehabilitation application, because the no-load tor-ques and inertial torques of the motors are very small compared to externalimpedances, even though they are enlarged by the gearboxes.

Under the control of operators, the master motor generates electricenergy and transmits it to the slave motor, and the slave motor is drivento move with the motion trend of the master motor. However, the energylosses in the circuit make it impossible to realize accurate motion tracking.This is especially the case when large impedance is attached to the slavesite; a large current and therefore large energy losses in the circuit willresult, and furthermore, the two motors will have a big difference in velo-cities and positions. In order to realize high motion tracking performance,a certain amount of energy is compensated for the circuit to offset theenergy losses in the circuit. Based on the electrical mechanism, the voltagebalance equation of the circuit can be written as follows:

2Ri þ 2L didt ¼ em þ esup � es

em ¼ CTxm ; es ¼ CTxs

�ð3Þ



where R and L denote the motor resistance and inductance (since the twomotors are connected in series, a factor of 2 is needed here); and em and esare motor armature voltages, which depend on the motor torque constantand the velocities of the motors. es is referred to as counter-electromotiveforce (EMF), since it has an opposite direction with the current; and esupis the supplementary voltage. The energy generated by the master motor,together with the supplementary energy, enables the slave=gear unit toreproduce the movement of the master=gear unit. That is, the systemachieves a kind of energy recycling. When accurate motion tacking is rea-lized (xin¼xout), the supplementary voltage can be expressed as follows:

esup ¼ 2 Ri þ Ldi

dt

� �ð4Þ

If the master and slave motors have no connection, and an independentsupply power is used to drive the slave=gear unit to accomplish motionimitation, the corresponding driving voltage (edri) in the slave site will bewritten as follows:

edri ¼ es þ Ri þ Ldi

dtð5Þ

in which the current is the same as that in Equation (4) because of theidentical external force and velocity. The armature voltage is usually largerthan the energy losses in the resistance and inductance for a rehabilitationapplication, thus the supplementary voltage in Equation (4) is always lessthan the driving voltage in Equation (5). That is, the required energy isreduced due to the recycled energy from the master motor.

According to the above analysis, the system is equivalent to a dampingsystem, through which operators can adjust the input force properlyaccording to the impedance variation in the other side, even though thereis frictional loss in the power transmission process, whereas a certainamount of energy is required for achieving the same motion manner ofthe two motors.

Energy Supplement Design

An H-bridge driver is employed to compensate energy for the master–slave circuit. The hardware connection diagram is given in Figure 2. Thecontrol inputs of the H-bridge driver are a pulse-width modulation (PWM)signal and a direction control signal, which control the magnitude anddirection of the supplementary voltage, respectively. The supplementaryvoltage can be calculated by the following equation:

esup ¼ aUs ð6Þ

390 C. Li et al.


where Us is the supply voltage of the H-bridge driver; and the magnitudeand the sign of a represent the duty cycle of the PWM signal and the direc-tion of the supplementary voltage, respectively.

A motion tracking controller is used to regulate the control signals ofthe H-bridge driver based on the velocity difference and the position differ-ence between the master and the slave terminals. A revised proportional–integral–differential (PID) control method is adopted for both the positionand the velocity controls, and the control output summation is used toadjust the duty cycle of the PWM signal and the direction signal. Sincethe movements are controlled by patients themselves, a sudden changemay happen to the input velocity=position in the master side. If the differ-ential operation is applied to the velocity=position difference directly, thevariation of the input velocity=position may lead the system to overshootand fluctuation of the whole system. Therefore, the differential operationis applied only to the output velocity=position in the slave side. The motioncontrol equation is given as follows:

a ¼ KxP ðxin � xoutÞ þ Kx

I

Zðxin � xoutÞ � Kx

D

dxout

dt

� �

þ K hP ðhin � houtÞ þ K h

I

Zðhin � houtÞ � K h

D

dhoutdt

� �ð7Þ

where KP, KI, and KD denote the proportional, integral, and differentialcoefficients, respectively. The parameters with the superscripts x and h

FIGURE 2 Connection of the H-bridge driver and the master–slave circuit.



denote the coefficients corresponding to the velocity and position, respect-ively. Since the compensated energy is used to offset the energy losses in thecircuit and to increase the equivalent energy provided for the slave motor,the compensatory voltage should have the same direction as the armaturevoltage of the master motor. Therefore, the velocity=position difference isobtained by subtracting xout=hout from xin=hin. Based on Equations (1),(3), and (6), the block diagram of the master–slave system with the motiontracking controller is shown in Figure 3, where um and us represent theworking efficiency of the gearboxes. The energy generated by the mastermotor and the supplementary energy drives the slave unit to reproducethe movement of the master unit.

Power Transmission Flow

The power transmission flowchart of the master–slave control system isshown in Figure 4, in which Pin and Pout denote the input and output powerin the terminals; Pin_m, PM_m, and Pout_m represent input mechanical power,electromagnetic power, and output electrical power of the master motor,respectively; Pin_s, PM_s, and Pout_s are input electrical power, electromag-netic power, and output mechanical power of the slave motor, respectively;and Psup is the compensatory energy power. The various energy losses in the

FIGURE 3 Block diagram of the master–slave system with the motion tracking controller. (a) Overalldiagram; (b) block diagram of the slave motor.

392 C. Li et al.


system are listed in Table 1. Gear loss is caused by coulomb friction ingearboxes. Mechanical loss, magnetic core loss, and excitation loss, whichare mainly caused by mechanical friction and the alternative magnetic fieldtowards the armature core, are called the no-load loss in general and areprimarily related to the velocity. Resistance loss and contact loss, whichare caused by the armature current, are called load loss. The load lossand excitation loss are energy losses in the circuit. As shown in Figure 4,the power balance equation in the circuit is given by the following:

PM m þ Psup ¼ PM s þ 2ðpa þ pb þ pf Þ ð8Þ

When the system achieves motion tracking accurately, xout equals xin, andthe compensatory energy power can be expressed as

Psup ¼ CT ixoutðNs � NmÞ þ 2ðpa þ pb þ pf Þ ð9Þ

where the resistance loss accounts for the main part of the energy losses.This suggests that the compensatory energy relies on the gear ratios ofthe two gearboxes and the energy losses in the circuit.

Based on Equations (2) and (9), the gearboxes with different gearratios affect the requirements for the input power from the healthy limband the supplementary energy. There are two cases as follows:

1. Nm>Ns: As indicted in Equation (2), a larger input torque=power isrequired to drive the same load compared to the system with identicalgearboxes, and the force sensing coefficient is increased. In contrast,the demand for the supplementary energy is reduced because the mas-ter can generate more electromagnetic power than that required in theslave site [refer to Equations (8) and (9)]. Actually, the two motors do

FIGURE 4 Power transmission flowchart of the master–slave system.

TABLE 1 Various Losses in the Control System

Gear Loss Mechanical Loss Core Loss Added Loss Resistance Loss Contact Loss Excitation Loss

pG pm pFe pD pa pb pf



not achieve motion tracking, whereas the two terminals can realizemotion tracking due to the function of different gearboxes.

2. Nm<Ns: This is a reverse case of Nm>Ns. That is, a smaller input torque=power is required, while the demand for the compensatory energy isincreased.

Different Operating Modes

The force sensing mechanism enables the system to support passive,active-assisted, and active-resisted operating modes. The working states ofthe two motors depend on the magnitudes of the produced torques in themotor mechanical axes. The motor attached with a larger torque works ingenerating state and acts as the master motor, while the other one works inelectromotive state and acts as the slave motor. In passive mode, the healthylimb provides a control force to overcome the impedance caused by theimpaired limb, which is driven to reproduce the movements of the healthyone symmetrically. In active-assisted mode, the impaired limb starts move-ments actively, whereas the force is insufficient to accomplish movements.Since the operator is the patient himself=herself, according to themovement consciousness, the patient provides an auxiliary force with thehealthy limb to help the impaired one to complete movements. In active-resisted mode, the impaired limb starts movements actively, while thehealthy one provides a proper resistant force according to the sensed forceof the contralateral limb. This mode can be used for performing strengthenhancement training. No matter which operating mode the system is in,the required control force, assistant force, or resistant force is providedby the healthy limb. Therefore, the system can support real bimanual train-ing. In the process of self-controlled bimanual training, the magnitudes ofimposed forces from the two limbs are determined by trainees themselves.In assisted mode, the trainees can control the assistant force from thehealthy limb as small as possibly with the consciousness of achieving func-tion recovery quickly. In addition, the assistant force can be increased prop-erly when the impaired limb is too tired, to avoid further producing tiredfeeling in exercises.

EXPERIMENTAL STUDY

Experimental Platform

A preliminary test platform, as shown in Figure 5, was built for verifyingthe viability of the proposed system. The main part was composed of masterand slave motor=gear units (motor 3863012C, combined with PlanetaryGearhead 38=2 A, and Encoder IE2–512, Faulhaber Group, Germany),

394 C. Li et al.


an H-bridge driver (LMD18200, National Semiconductor Co., USA), and adSPACE control platform (CLP1104, dSPACE Inc., Germany). The H-bridge driver was connected with the two motors to construct a closed-loop circuit. It supplied compensatory energy for the circuit under thecontrol of the motion tracking controller, which was realized in theCLP1104.

In addition, two torque transducers (TP-20KCE, Kyowa Co., Japan) anda torque signal amplifier were applied to measure the torques attached tothe two terminals for verifing force sensing capability. However, they will benot required in future applications. Additionally, in order to simplify systemperformance analysis, a DC driving motor was used to drive the master=gear unit instead of a human operator. It was coaxially connected to themaster and was driven by another H-bridge driver, herein referred to asdriver 2, while the driver connected with the closed-loop circuit is referredto as driver 1. A corresponding driving motor controller was implementedin the CLP1104 to regulate the input voltage of the driving motor based onthe difference between a predefined reference velocity and the actualvelocity in the master terminal.

Gearboxes with gear ratios of 43 and 66 were selected in different test-ing experiments, and the corresponding maximum output torques of themotor=gear units were 3.311Nm and 5.082Nm, respectively, which willbe sufficient to drive a forearm to perform elbow flexion=extensionmovement. The torque caused by gravity of a forearm was estimated for ahuman with a weight of 65 kg and height of 175 cm,[21] and the resultwas 1.519Nm. During operation, the unhealthy limb may produce impe-dance except for the gravity due to its limited residual motor capability

FIGURE 5 Experimental setup of the master–slave system.



and movement positions. Therefore, gearboxes with larger drivingtorque were selected.

In order to ensure safety throughout the training process, position limitcan be regulated by setting parameters according to the motor capacity ofpatients. If the slave terminal is moved beyond the position limit, theH-bridge driver stops compensating energy for the circuit and disconnectsthe master and the slave motors. As a result, the slave unit stops movementimmediately. In addition, if the actual current of the motors is larger thanthe allowable maximum value, the H-bridge driver also stops working toensure the normal operation of the system. This can aslo prevent misopera-tion caused by the sudden and large forces that result from the spasticity ofthe impired limb. Since the driving force in passive mode or the assistant=resistant force in active-assisted=resisted mode is exerted by the healthylimb, patients can regulate the force according to force sensation andthe feel of the impaired limb. Therefore, there is no need to set a torquelimit based on the residual motor function of patients. In addition, a push-button can be manipulated by the trainee to switch off the power of thesystem in case of emergency.

Information flow of the system is shown in Figure 6 (the dashed linesrepresent that the devices or the information are not needed in futureapplications). The CLP1104 collected the velocity and position informationthrough the incremental encoder interface. The driving motor controllerworked out control signals (a PWM signal and a directional signal) forthe H-bridge driver 2, and regulated the input voltage of the driving motor,which further drove the master to rotate with a predefined velocity. Mean-while, the motion tracking controller calculated the control signals for theH-bridge driver 1, and directed the driver to supply a proper amount ofenergy for the circuit. The compensated energy, together with the energygenerated by the master motor, drove the slave motor to mimic the motionof the master. In addition, the torque information and closed-loop currentwere collected through AD modules of the CLP1104 in order to verify the

FIGURE 6 Information flow of the system.

396 C. Li et al.


force sensing capability and the effect of the gearboxes with different gearratios. The outputs of the motion tracking controller were also recordedfor calculating the amount of the compensated energy. All the informationmentioned above is sampled every 1ms.

Energy Recycling Test

Two identical gearboxes with the gear ratio of 66 were employed and aconstant load of 500 gram (the corresponding resistant torque was0.0492Nm) was attached to the slave site. During the test, the referencevelocity increased with an angular acceleration of 1 deg=s2. In order to con-firm the capability of energy recycling and the function of supplementaryenergy, two experiments were carried out. In the first experiment, thetwo motors were connected directly and there was no supplementaryenergy; in the second experiment, the H-bridge driver was connected withthe two motors and supplied supplementary energy for the circuit. Thearmature voltages of the two motors and the supplementary voltage werecalculated with the following:

em ¼ CTNmxin

es ¼ CTNsxout

esup ¼ aUs

8<: ð10Þ

where Us¼ 12V. The comparison of the two motors’ armature voltages,which also represent velocities according to Equation (10), was carriedout for each experiment. In addition, the two motors’ armature voltage dif-ference in the first experiment, which actually was the voltage drop in theresistance and inductance and indirectly reflected the amount of energylosses in the circuit, was calculated and compared with the compensatedvoltage in the second experiment.

Force Sensing Test

The regulation of the input torque following the variation of resistanttorque was used to verify the force sensing capability. In this experiment,the gearboxes with the same gear ratios as the energy recycling test wereemployed. The DC driving motor provided a control force for themaster–slave device and rotated the master terminal with a referencevelocity of 100 deg=s. The constant velocity was used to minimize the vari-ation of the no-load=inertial torques of motors and thus to acquire theforce sensing coefficient accurately. With the supplementary energy fromthe H-bridge driver 1, the slave motor was rotated in the same velocity ofthe master. In the process of rotation, an operator exerted an increasing



resistant force on the slave site with one hand for a moment. In this period,the corresponding force information was used to analyze force sensingperformance.

Tests for Verifying the Effect of Different Gearboxes

As for the gearboxes in the master and slave sites, four sets of gear ratiocombinations were employed, as listed in Table 2. With each combination,two experiments were carried out. The testing objectives and the experi-mental conditions are also listed in Table 2, in which gsys represents systemefficiency. It was defined as

gsys ¼Pout

Pin þ Psup� 100% ð11Þ

where Pin, Pout, and Psup were calculated by

Pin ¼ Tinxin

Pout ¼ Toutxout

Psup ¼ aUsi

8<: ð12Þ

The first experiment was aimed at testing gearboxes’ influence on forcesensing capability, and also at confirming the relationship of the twomotors’electromagnetic powers and the supplementary energy power when gear-boxes had different gear ratios. The velocity was fixed, and the resistantforce attached to the slave site was increased, which was the same as thatin the force sensing test. The second experiment was used to test gear-boxes’ effect on system efficiency as well as the requirements for inputpower and compensatory energy. In order to carry out a comparisonbetween the different gear ratio combinations easily, the output powerwas controlled with a constant value by attaching a load of 500 g to the slavesite and setting a reference velocity of 100 deg=s. The corresponding out-put power was 0.085W.

TABLE 2 Testing Objects and Experimental Conditions

Gear RatioCombination Experiment

TestingObjectives Testing Conditions

Nm Ns 1 kT . Constant velocity: 100deg=s. The impedance in the slave site was increased.66 66

66 4343 66 2 Pin, Psup, gsys . Constant velocity: 100deg=s

. Constant load: 500 g43 43

398 C. Li et al.


Different Operating Modes Test

In this test, the DC driving motor connected with the master unit wasremoved from the experimental platform, and the same handles were att-ached to the terminal axes of the two torque transducers. The two handleswere controlled by the two limbs of an operator (Figure 7). In order to con-firm that the system can support training in different operating modes, thetest was performed in two steps. First, the left limb provided a small forceactively, and the right limb gave an assistant force to assist the left one so itcould accomplish movements. Second, the left limb started movements witha larger force, while the right limb exerted a resistant force to increase the bur-den on the left one. The two cases were used to imitate active-assisted andactive-resisted modes, respectively. The second case can also be used to imitatepassivemode if the resistant force of the right limb is considered as impedancecaused by an impaired limb with nomotor function, and the active force of theleft limb is considered as a driving force from the healthy limb. In order tocompare the torque relationship, the same upward and downwardmovements(elbow flexion and extension, respectively) were performed for the two cases.

System Stability Test

The test was aimed at investigating the stability of the system when themotion velocity was increased and at verifying the same working perform-ance of the system for both control directions. This experiment was per-formed with the same platform as the one used in the different operating

FIGURE 7 Experimental schematic for bimanual coordinated control.



modes test. First, upward and downward movements were carried out byattaching an increased force to the right site, while no external load=forcewas attached to the left site. The left motor=gear unit tracked the move-ments of the right motor=gear unit with an increased velocity. The attachedforce was increased until there was a sudden change in the sensed force, thatis, the master and slave terminals could not mirror each other in motionbehavior, and then stopped the movement immediately. Secondly, the aboveprocess was repeated for the reverse control direction: The control force wasattached to the left site, and no external load=force was attached to the rightsite. After the experiment, in order to confirm the stable frequency responserange of velocity, a fast Fourier transform (FFT) analysis was carried out withthe velocity information collected from the two terminals during the periodthat the master and slave made mirror symmetric movements.

EXPERIMENTAL RESULTS AND ANALYSIS

Energy Recycling Test

The example results for the two experiments are given in Figures 8aand 8b, respectively. From Figure 8a, we can conclude that the mastermotor was able to drive the slave one, even though there was no supple-mentary energy. This confirms the energy recycling capability of the system.However, there was a big difference in the armature voltages (velocities);this was caused by the energy losses in the circuit. As for Figure 8b, whenthe energy was compensated for the circuit, the motor armature voltages(velocities) were basically identical. The compensated voltage shows agree-ment with the two motors’ armature voltage difference given in theFigure 8a. This demonstrates that the supplementary energy was used tooffset the energy losses in the circuit, and, as a result, accurate motiontracking was achieved. In addition, if an independent supply power isapplied to drive the slave unit for implementing symmetrical movements,the driving voltage will equal the summation of the es and half of the twomotors’ armature voltage difference [the voltage drop in the resistanceand inductance of one motor; refer to Equation (5)]. The amount willbe about 2.35V at the beginning and increase following the increment ofthe velocity. This will be larger than the compensated voltage (around1.1 V). Therefore, it can be concluded that a certain amount of energycan be saved with the function of energy recycling.

Force Sensing Test

The corresponding results are shown in Figure 9. As can be seen fromFigure 9a, the input torque increased linearly following the increment of

400 C. Li et al.


the resistant torque. This demonstrates that the system realized force sens-ing without a force sensor. The actual force sensing coefficient was calcu-lated with the following:

kT ¼ DTin

DTout¼ Tk

in � T 0in

T kout � T 0

out

ð13Þ

where Tout equaled the resistant torque exerted by the operator based onNewton’s theorem (the action and reaction force). T 0

out and T 0in represent

FIGURE 8 Results of the energy recycling test. (a) Two motors’ armature voltages and their difference:without supplementary energy; (b) two motors’ armature voltages and the supplementary voltage: withsupplementary energy.



the initial torques when there was no external impedance in the slave site,and the system was rotated with the velocity of 100deg=s. T 0

out (0.014Nm)was mainly cuased by the worm gear in the slave site. T 0

in (0.665Nm) wasused to overcome the frictional torques in the gearboxes and the no-loadtorques in the two motors, in order to drive the two motors and the wormgear in the slave site to rotate; the variables with the superscript k denotethe sampling values in the k time. The force sensing coefficient, as shownin Figure 9b, was approximately constant, even though a varying resistanttorque was attached to the slave terminal. The average force sensing coef-ficient equaled 1.626, which was larger than unit one due to the frictionalloss in the gearboxes. This result shows agreement with Equation (2).

FIGURE 9 Results of the force sensing test. (a) Relationship between the input and output torque; (b)force sensing coefficient versus resistant torque; (c) velocity tracking curve; (d) position tracking curve;(e) two motors’ electromagnetic powers and the supplementary energy power.

402 C. Li et al.


Combining Equations (2) and (13), the force sensing coefficient can beexpressed as

kT ¼ Nm

Ns

1

gmgsþ Nm

gm

DT0 m þ DT0 s þ DJmNmdxin

dt þ DJsNsdxout

dt

� �DTout

ð14Þ

In reference to Figure 9c, the terminal velocities in the master and theslave sites had a small fluctuation; as a result, the no-load and inertial tor-ques were not constant. That is, the second item in Equation (14) variedslightly. Also, the gearbox efficiency was not unchanged for the increasedresistant force. Therefore, the calculated force sensing coefficient was nota constant. In this test, the maximum change rate of the force sensingcoefficient was 0.04 within the velocity varying range of 1.625 deg=s. Thisfluctation can be ignored for a human-controlled operation, and the forcesensing coefficient of 1.626 is enough for human operators to sense thevariation of the acting force in the slave site.

Figures 9c and 9d give the results of velocity tracking and position track-ing. The maximum velocity error was 0.5987 deg=s, and the maximumangular position error was 0.0719 degrees. This demonstrates that highmotion tacking performance was achieved.

The electromagnetic powers of the two motors and the supplementaryenergy power were given in Figure 9e. In the experiment, the system rea-lized velocity tracking and the gearboxes in both sites had a symmetricstructure (identical gear ratio), thus the electromagnetic powers of thetwo motors were basically identical. This confirms that the energy losses inthe circuit were compensated completely [refer to Equations (8) and (9)].Otherwise, the supplementary energy power was less than the electromag-netic power of the slave motor, which further confirms that the mastermotor provided energy for the slave motor.

TESTS FOR VERIFYING THE EFFECT OF DIFFERENTGEARBOXES

For the gearbox combination with different gear ratios, the relationshipof the two motors’ electromagnetic powers and the compensated energypower obtained in the first experiment are shown in Figure 10. It can beseen that when Nm was larger than Ns, the electomagnetic power of the mas-ter motor was more than that of the slave motor when the two terminalsachieved motion tracking. That is, the master generated more energy thanthat required in the slave site. Therefore, the amount of the compensatedenergy decreased even though there were energy losses in the circuit. Incontrast, when Nm was smaller than Ns, the electomagnetic power of the



master motor was less than that of the slave motor. Thus, the supplemen-tary energy was also used to provide extra energy for the slave motor exceptfor offsetting the energy losses in the circuit. In looking at the initial phaseof Figure 10b, one can observe that the compensated energy power was lessthan the electromagnetic power of the slave motor, and this verifies that theenergy generated by the master motor was recycled even though it was lessthan that reqired in the slave site.

In addition, for each gear ratio combination, the average values of thetesting results were calculated and depicted in Table 3. By comparing the

FIGURE 10 Two motors’ electromagnetic powers and the supplementary energy power when thesystem employed gearboxes with different gear ratios. (a) Nm¼ 66, Ns¼ 43; (b) Nm¼ 43, Ns¼ 66.

404 C. Li et al.


case of Nm¼Ns¼ 43 with that of Nm¼ 66 and Ns¼ 43, we can see thatthe force sensing coefficient and the demand for the input powerincreased distinctly when the gear ratio of the master=gear unit wasincreased; in contrast, the compensated energy decreased significantly.By comparing the cases of Nm¼Ns¼ 66 with that of Nm¼ 43 and Ns¼ 66,it can be conclued that when the gear ratio of the master=gear unit wasreduced, the force sensing coefficient and the demand for input powerdeclined, while the supplemetary energy increased. The results fullyappraise the theoretical analysis introduced in the Power TransmissionFlow section of this article. As for the cases of Nm¼Ns¼ 43 andNm¼Ns¼ 66, the compensated energy had no obvious change, while theneed for the input power increased with the increment of the gear ratios.This was because the actual working efficiency of the gearboxes decreasedwith the increment of gear ratios. Furthermore, it can be concluded thatthe system efficiency mainly depended on the gear ratio of the mastermotor: the larger the gear ratio of the master, the lower the resulting systemefficiency.

Different Operating Modes Test

The torque relations for the two cases are presented in Figure 11,where TL denotes the produced torque in the left site, and TR representsthe produced torque in the right site. In active-assisted mode, the torquesin the two sites had the same direction, while in the active-resisted mode,the torques in the two sites had a reverse direction. In addition, the activetorque exerted by the left limb in the former case was very small com-pared to that in the latter case. This corresponds to the theoretical analy-sis introduced in the Different Operating Modes section if this article.Additonally, high-motion tracking performance was achieved, and the cor-responding maximum velocity and position errors were 0.5926 deg=s and0.0613 degrees in the active-assisted mode and were 0.2373 deg=s and0.024 degrees in the active-resisted mode. The results demonstrate thatthe system can support different operating modes by coordinating theforces of the two limbs, and no matter in which mode, high motion track-ing performance can be achieved.

TABLE 3 Results of the Comparison Experiments

Nm Ns kT Pin Psup gsys

66 66 1.534 1.265 0.330 5.48166 43 2.442 1.302 0.196 5.46843 66 0.862 0.771 0.443 7.33843 43 1.498 0.788 0.344 7.412



System Stability Test

Figure 12 shows the frequency response result of velocity when the con-trol force was exerted by the right limb. When the velocity was varied abovethe frequency of around 30Hz, the slave could not mirror the movement ofthe master any more, and there was a sudden change in the sensed force.Because of that, in a sampling period, the velocity=position differencebetween the two terminals became larger when the rotational velocityhad a higher frequency, which resulted in a larger control output of themaster–slave motion tracking controller. Furthermore, the two terminalshad larger differences in motion velocities and positions, which easily ledto a movement fluctuation of the slave terminal and a great change inthe current and the sensed force. Almost the same result was obtainedwhen the control force was exerted by the left limb. This confirmed the

FIGURE 11 Torque relations for different operating modes. (a) Left limb: active force, right limb:assistant force; (b) left limb: active force, right limb: assistant force.

FIGURE 12 Frequency response result of velocity force: control force was exerted by the right limb.

406 C. Li et al.


same operational performance in the two control directions and gave thesystem potential in training patients no matter which limb is impaired.Overall, the system can maintain stability if the rotational velocity is keptwithin the frequency range of 30Hz. This frequency range of velocity is suf-ficient for patients to perform rehabilitation training.

CONCLUSION

This article introduces a new master–slave control system with the twomotors constructing a closed-loop circuit. The master motor works in gen-erating state and supplies energy for the slave motor, which is driven toreproduce the movement of the master motor and to support the connec-ted limb in motion imitation.

The system is capable of mirroring the force from the slave to themaster without a force sensor, thus the hardware mounting difficulty canbe reduced greatly. Accurate motion tracking was achieved within the velo-city’s frequency responce range of around 30Hz. The tracking precision iscomparative with the system introduced in Nef et al.[7] It satisfies the appli-cation requirement for a rehabilitation robot. Force sensing and motiontracking are realized simultaneously, with only a master–slave motion track-ing controller. This reduces the control complexity significantly. Also, thecharacteristic of energy recycling makes it so that a lightweight batterycan supply enough power for the system, thus, reducing the design burdenof the power supply unit. Additionally, self-controlled operation, togetherwith the force sensing capability, will enable patients to adjust the forceof the healthy limb timely according to the feel of the impaired limb andforce sensation, further avoiding pain as well as unpredictable reactions.This makes it possible to carry out rehabilitation in the home environment.

Moreover, the force sensing coefficient mainly depends on the gearratios of the gearboxes. Based on the effects of different gear ratio combi-nations, the system hardware configuration can be considered according tothe emphases in applications. For example, when the healthy limb pos-sesses sufficient power, and the patient wants to exercise the healthy limbsimultaneously while carrying out rehabilitation training, the master=gearunit can select a gearbox with a larger gear ratio. Then, the energy ofthe power supply unit, such as a battery, can be saved greatly. However,the gear ratio is not allowed to be very large, since the gearbox in themaster site works in the back-drivable state and it is easy to destroy itwhen the gear ratio is too large. In contrast, if patients are older subjectsand the healthy limb has not enough power due to the weak motorfunction, the gearbox in the master site with a smaller gear ratio will bepreferable.



Furthermore, different operating modes can be achieved by coordinat-ing the forces of the two limbs. The clinical trial with MIME[22,23] has pro-ven that the treatments with patient’s active participation (robot-assistedtreatment) can produce larger improvements on a motor impairmentscale, and active-constraint training can achieve a greater strength gains.However, the bilateral training has no advantage over the unilateral train-ing except that hypertonia and abnormal synergies can be reduced. Finally,the training mode that combines the unilateral and bilateral exercises wasconsidered as a preferred selection. In this system, the required controlforce, assistant force, or resistant force comes from the healthy limb ratherthan the robot. This is different from MIME, in which the healthy limbprovides a reference movement while the robot gives an assistant forcefor bilateral exercises. This means that the healthy limb not only providesa reference but also produces a suitable force for the impaired one. There-fore, much more cognitive processing will be involved in training tasks.This will make the proposed system has a great potential in improvingmotor function in a larger scale. However, this presumption should beconfirmed in future study.

Since the master and the slave units have a wired connection, the rela-tive position of the two units can be regulated if they are designed withindependent and position-adjustable base plates. Based on this character-istic, the system will be able to support multiple movements, such aswrist=elbow=shoulder flexion=extension, shoulder abduction=adduction,forearm pronation=supination and, so on, by adjusting the positions ofthe two units and repalcing the handles in the two terminals. This will bean advantage over some conventional systems that achieve a mechanicalcoupling between the unimpaired and impaired arms, such as the systemintroduced in Hesse et al.[24]

However, the no-load torques of motors and the efficiency of gearboxesaffect back-drivability and have a negative influence on force sensing,especially when the gear ratio is large and the velocity is not constant.Therefore, motors and gearboxes with higher efficiency should be emplo-yed to enhance force sensing accuracy, whereas the system cost will beincreased relatively. Otherwise, on the condition that the output torqueis enough to drive an impaired limb, gearboxes with a smaller gear ratiocan be selected to weaken the above influence and to reduce the require-ment for the input power from operators. In addition, motors with asmaller armature resistance are favorable to reduce the requirement forsupplementary energy.

In a future study, a visual display will be added to the system in order tooffer desired motion trajectory tracking for the two limbs, and will givescores after each training task to motivate the active participation of pati-ents in training and evaluate the improvement of movement performance.

408 C. Li et al.


Tests on numbers of subjects will be conducted for appraising theadvantages of bimanual training. Additionally, in order to make the systemmore suitable for the application in rehabilitation robots, the master andslave units with independent and position-adjustable base plates will bedesigned. Also, except for the current limitation, an acceleration limitationshould be considered to prevent abnormal operations caused by thespasticity of patients and to increase the stability of the system. As well, thisone DOF (degree of freedom) system will be expanded to a multiple DOFmechanism with multiple master–slave motor combinations. It may bepossible to control each DOF with independent master–slave motor combi-nation and an independent motion tracking controller.

REFERENCES

1. Hogan, N.; Krebs, H. I. Interactive robots for neuro-rehabilitation. J. Restorative Neurology andNeuroscience 2004, 22(3), 349–358.

2. Krebs, H. I.; Volpe, B. T.; Aisen, M. L.; Hogan, N. Increasing productivity and quality of care:Robot-aided neuro-rehabilitation. J. Rehabil. Res. Develop. 2000, 37(6), 639–652.

3. Reinkensmeyer, D. J.; Kahn, L. E.; Averbuch, M.; McKenna-Cole, A.; Schmit, B. D.; Zev Rymer, W.Understanding and treating arm movement impairment after chronic brain injury: Progress withthe ARM Guide. J. Rehabil. Res. Develop. 2000, 37(6), 653–662.

4. Reinkensmeyer, D. J.; Dewald, J. P. A.; Rymer, W. Z. Guidance-based quantification of arm impair-ment following brain injury: A pilot study. IEEE Trans. Rehabil. Eng. 1999, 7(1), 1–11.

5. Lum, P. S.; Burgar, C. G.; Shor, P. C.; Majmundar, M.; Van der Loos, M. Robot-assisted movementtraining compared with conventional therapy techniques for the rehabilitation of upper-limb motorfunction after stroke. Arch. Phys. Med. Rehabil. 2002, 83(7), 952–959.

6. Lum, P. S.; Burgar, C. G.; Shor, P. C. Evidence for improved muscle activation patterns after retrain-ing of reaching movements with the MIME robotic system in subjects with post-stroke hemiparesis.IEEE Trans. Neural Syst. Rehabil. Eng. 2004, 12(2), 186–194.

7. Nef, T.; Mihelj, M.; Riener, R. ARMin: A robot for patient-cooperative arm therapy. Med. Biol. Eng.Comput. 2007, 45(9), 887–900.

8. van Exel, N. J. A.; Koopmanschap, M. A.; Scholte op Reimer, W.; Niessen, L. W.; Huijsman, R.Cost-effectiveness of integrated stroke services. QJM-J. Assoc. Physicians 2005, 98(6), 415–425.

9. Park, H.-S.; Peng, Q.; Zhang, L.-Q. A portable telerehabilitation system for remote evaluationsof impaired elbows in neurological disorders. IEEE Trans. Neural Syst. Rehabil. Eng. 2008, 16(3),245–254.

10. Peng, Q.; Park, H.-S.; Zhang, L.-Q. A low-cost portable tele-rehabilitation system for the treatmentand assessment of the elbow deformity of stroke patients. In Proceedings of the 2005 IEEE 9th Inter-national Conference on Rehabilitation Robotics, Chicago, Illinois, June 28–July 1, 2005; Patton, J. L.et al., Eds.; IEEE: USA, pp. 149–151.

11. Park, H.-S.; Peng, Q.; Zhang, L.-Q. Causality-based portable control system design fortele-assessment of elbow joint spasticity. In Proceedings of the 2005 IEEE 9th International Conferenceon Rehabilitation Robotics, Chicago, Illinois, June 28–July 1, 2005; Patton, J. L. et al., Eds.; IEEE:USA, pp. 303–306.

12. Reinkensmeyer, D. J.; Pang, C. T.; Nessler, J. A.; Painter, C. C. Web-based telerehabilitation for theupper extremity after stroke. IEEE Trans. Neural Syst. Rehabil. Eng. 2002, 10(2), 102–108.

13. Jadhav, C.; Nair, P.; Krovi, V. Individualized interactive home-based haptic telerehabilitation. IEEEMultimedia 2006, 13(3), 32–39.

14. Perry, J. C.; Rosen, J.; Burns, S. Upper-limb powered exoskeleton design. IEEE=ASME Trans.Mechatronics 2007, 12(4), 408–417.



15. Feng, X.; Winters, J. M. An interactive framework for personalized computer-assisted neurorehabil-itation. IEEE Trans. Inf. Technol. Biomed. 2007, 11(5), 518–526.

16. Popescu, V. G.; Burdea, G. C.; Bouzit, M.; Hentz, V. R. A virtual-reality-based telerehabilitationsystem with force feedback. IEEE Trans. Inf. Technol. Biomed. 2000, 4(1), 45–51.

17. Colombo, R.; Pisano, F.; Micera, S.; Mazzone, A.; Delconte, C.; Chiara Carrozza, M.; Dario, P.;Minuco, G. Robotic techniques for upper limb evaluation and rehabilitation of stroke patients. IEEETrans. Neural Syst. Rehabil. Eng. 2005, 13(3), 311–324.

18. Whitall, J.; Waller, S. M.; Silver, K. H. C.; Macko, R. F. Repetitive bilateral arm training with rhythmicauditory cueing improves motor function in chronic hemiparetic stroke. Stroke 2000, 31(10),2390–2395.

19. Hesse, S.; Schulte-Tigges, G.; Konrad, M.; Bardeleben, A.; Werner, C. Robot-assisted arm trainer forthe passive and active practice of bilateral forearm and wrist movements in hemiparetic subjects.Arch. Phys. Med. Rehab. 2003, 84(6), 915–920.

20. Hesse, S.; Werner, C.; Pohl, M.; Rueckriem, S.; Mehrholz, J.; Lingnau, M. L. Computerized armtraining improves the motor control of the severely affected arm after stroke: A single-blindedrandomized trial in two centers. Stroke 2005, 36(9), 1960–1966.

21. Li, C.; Liu, T.; Shibata, K.; Inoue, Y. A master-slave control system with energy recycling and forcesensing for upper limb rehabilitation robots. IEEE=ASME International Conference on AdvancedIntelligent Mechatronics, AIM, pp. 36–41, 2009.

22. Zatsiorsky, Vladimir M.; Seluyanov, V. N. Mass and inertia characteristics of the main segments of thehuman body. International Series on Biomechanics 1983, 4B, 1152–1153. http://www.dh.aist.go.jp/bodyDB/m/k-05.html(accessed 12 Oct 2010).

23. Lum, P. S.; Burgar, C. G.; Van Der Loos, M.; Shor, P. C.; Majmundar, M.; Yap, R. The MIME roboticsystem for upper-limb neuro-rehabilitation: Results from a clinical trial in subacute stroke. InProceedings of the 2005 IEEE 9th International Conference on Rehabilitation Robotics, Chicago, Illinois, June28–July 1, 2005; Patton, J. L. et al., Eds.; IEEE: USA, pp. 511–514.

24. Lum, P. S.; Burgar, C. G.; Van Der Loos, M.; Shor, P. C.; Majmundar, M.; Yap, R. MIME robotic devicefor upper-limb neurorehabilitation in subacute stroke subjects: A follow-up study. J. Rehabil. Res.Develop. 2006, 43(5), 631–642.

25. Hesse, S.; Schmidt, H.; Werner, C. Machines to support motor rehabilitation after stroke: 10 years ofexperience in Berlin. J. Rehabil. Res. Develop. 2006, 43(5), 671–678.

410 C. Li et al.


Documents

ADDITIONAL DESIGN FEATURES OF A MASTER–SLAVE … · ADDITIONAL DESIGN FEATURES OF A MASTER–SLAVE CONTROL SYSTEM WITH FORCE SENSING AND ENERGY RECYCLING FOR UPPER LIMB REHABILITATION