Christopher L. Jones, Furui Wang, Christian Osswald, Xuan Kang, Nilanjan Sarkar, Derek G. Kamper

Abstract— Finger impairment following stroke results in

significant deficit in hand manipulation and the performance of

everyday tasks. Recent advances in rehabilitation robotics have

shown improvement in efficacy of rehabilitation. Current

devices, however, lack the capacity to accurately interface with

the human finger at levels of velocity and torque comparable to

the performance of everyday hand manipulation tasks. To fill

this need, we have developed the Actuated Finger Exoskeleton

(AFX), a three degree-of-freedom robotic exoskeleton for the

index finger. This paper outlines the implementation and initial

kinematic analysis of the AFX.

I. INTRODUCTION

Precise finger and thumb interactions are critical to hand manipulation in everyday tasks and are part of what separates humans from other animals. Neurological disorders such as stroke greatly impede this core function [1], [2], directly impairing quality of life [3].

In the United States, stroke is the leading cause of serious, long-term disability [4]. Out of an estimated 6.4 million stroke survivors in the U.S. [5], 30-50% will require ongoing care or experience chronic impairment [4]. The economic costs are high as well, with a total estimated price tag, including direct and indirect costs of stroke, of $73.7 billion for 2010 in the U.S. alone [5].

Thus, current research focuses on improving the efficacy of rehabilitation. Recent studies have shown that repetitive practice of desired movement leads to promising recovery. For example, promotion of use of the paretic upper limb through constraint-induced therapy has led to improved motor control [6], [7], [8]. Practice has been shown to improve plasticity and incite cortical functional reorganization leading to improved motor control following stroke [9], [10] .

Because of the complexity of the hand, with 21 mechanical degrees-of-freedom (DOF) and even more muscles, and the need for lengthy and consistent repetition of movement, researchers have begun to implement robotics for rehabilitation. Robot-assisted rehabilitation has been

This work was partially supported by NIH grants R24HD050821 and 5 R21HD055478-02.

Christopher L. Jones and Christian Osswald are with the Department of Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, 60616, USA (email: {jonechr, cosswald}@iit.edu).

Furui Wang and Nilanjan Sarkar are with the Department of Mechanical Engineering, Vanderbilt University, Nashville, TN, 37212, USA (email: {furui.wang, nilanjan.arkar}@Vanderbilt.edu).

Xuan Kang is from the Rehabilitation Institute of Chicago, Chicago, IL, 60611, USA (email: xuankang@gmail.com).

Derek G. Kamper is with the Department of Biomedical Engineering, Illinois Institute of Technology and the Rehabilitation Institute of Chicago, Chicago, IL, 60611, USA (e-mail: d-kamper@northwestern.edu).

demonstrated to enable longer training sessions while reducing the workload on therapists [11].

In recent years, a number of devices have been developed expressly for, or applied to, hand rehabilitation. These include both commercial products, such as CyberGrasp (Immersion Corporation, San Jose, CA) [12], the Hand Mentor (Kinetic Muscles Inc., Tempe, AZ), and the Amadeo System (Tyromotion GmbH, Graz, Austria) and experimental devices, including the Rutgers Master II-ND [13], HWARD [14], and HandCARE [15], among others [16]-[19].

A fundamental remaining question, however, is how to best use these devices for rehabilitation. The extent to which rehabilitation robots should assist, resist, or otherwise alter movement of the user is unclear. This is a key area for study. Unfortunately, most existing devices do not provide the speed, force, or independence of joint control to truly compare different training algorithms.

One exoskeleton which does allow independent control of finger joints has been designed for rehabilitation of occupational injuries [16]. Due to the nature of its transmission, however, response times are on the order of seconds which is too slow for our purposes. An 18 DOF device has recently been developed in Japan for hand and wrist rehabilitation following stroke [19]. As the joint torques are applied through motors positioned at the joints and thus limited in size, torques are limited to roughly 0.3 N-m in magnitude and the device platform prohibits arm movement.

The Actuated Finger Exoskeleton (AFX) will improve on current rehabilitation robotics solutions by providing a versatile framework with high performance, real-time control, and forces and speeds comparable to normal human function. The AFX will allow for normal task execution in a rehabilitation or motor study environment. This paper describes the current implementation of the AFX and analyzes preliminary kinematic performance.

II. FINGER EXOSKELETON DESIGN

To enable the exploration of rehabilitation and motor control strategies of the finger, the AFX must satisfy several design criteria. Primarily, the exoskeleton must be biomechanically compatible with an individual’s natural joint rotation and provide independent actuation of each of the joints of the finger. For application across users, the AFX must be versatile, capable of adjustment to accommodate finger segments of different lengths and thicknesses. To reduce impedance of normal movement, the exoskeleton must be lightweight, have low inertia and have a relatively small physical profile with respect to the index finger. For application to practical finger manipulation tasks, the device

Control and Kinematic Performance Analysis of an Actuated

Finger Exoskeleton for Hand Rehabilitation following Stroke

Proceedings of the 2010 3rd IEEE RAS & EMBSInternational Conference on Biomedical Robotics and Biomechatronics,The University of Tokyo, Tokyo, Japan, September 26-29, 2010

978-1-4244-7709-8/10/$26.00 ©2010 IEEE 282

must support angular velocities on the order of 1000°/s [20], representative of the speeds we have observed in normal movements. Similarly, the AFX must sustain torques of 2.5, 0.75 and 0.25 Nm at the MCP, PIP and DIP joints respectively, equaling half of the average maximal torque found in normal individuals (unpublished data). To control finger manipulation tasks, the AFX must support both position and torque control at each joint. To enable this control, joint angle and torque feedback must be collected. To meet these requirements, the exoskeleton was fabricated, actuated and instrumented as described in the following sections.

A. Mechanical structure

The complete mechanical design of the exoskeleton was discussed at length in a previous paper [21]. Briefly, the exoskeleton runs along the radial side of the index finger (Fig. 1). The three rotational joints of the exoskeleton are aligned with the flexion/extension axes of the metacarpophalangeal (MCP), proximal interphalangeal (PIP), and distal interphalangeal (DIP) joints. Pairs of parallel bars connect the structure to the proximal, middle, and distal segments of the finger. Thus, rotation of the exoskeleton produces equivalent rotation of the finger joint. The exoskeleton is attached to a plate on a fiberglass cast which encases the wrist and thus maintains its posture (Fig. 2).

Figure 1 – AFX located radial of the index finger with parallel posts

interfacing with each finger segment. Transmission pulleys are superior to the index finger. Guide pulleys direct cable over each joint.

All components are fabricated from aluminum or steel to withstand the relatively high torques required of the device. Thrust bearings at the MCP and PIP joints serve to accommodate potentially significant off-axis moments. The exoskeleton was designed to allow large ranges of motion: -15 to 75°, 0 to 90° and 0 to 90° for the MCP, PIP and DIP joints, respectively. Mechanical stops limit these ranges to prevent accidental injury to users. These stops can be adjusted to further limit joint ranges to match the passive range of motion of the user. The part of the exoskeleton that actually moves with the finger has a mass of 138 grams.

B. Actuation

In order to reduce the added mass on the hand, the exoskeleton is actuated by DC motors which can be located on the forearm. To maximize backdrivability, motors without any gearing were selected. Specifically, AKM motors (Kollmorgen, Munkekullen, Sweden) are being used, with

AKM13C, AKM12C and AKM11C for the MCP, PIP and DIP joints, respectively.

Cables (Spectra kite line) transmit motor torque to the exoskeleton joints. The cable drive design reduces friction and backlash in comparison with standard transmissions, thereby compensating for the cost of locating the motors a significant distance from the joints. Cable transmissions have been successfully implemented in commercial robots such as the Phantom (SensAble Technologies, Woburn, MA) and WAM (Barrett Technologies, Cambridge, MA). As these cables can only pull, similar to muscles, two cables and thus two motors are used for each joint for a total of 6 cables and motors (Fig. 2).

Figure 2 – AFX attached to mounting plate on forearm cast. AFX joints

(A-C), tension sensors (D), and motor pulleys (E) are indicated.

Primary gear reduction from motor cables to exoskeleton joints occurs directly at the joint. Namely, the cables are connected to pulleys which subsequently drive a section of a gear fixed to a rotating segment of the exoskeleton. This gear reduction at the joint reduces the tension in the cables from the motors to provide as much bandwidth for control as possible [22]. Total reduction is 11.8, 3.7, and 1.4 at the MCP, PIP, and DIP joints, respectively. A set of bearing and pulley cable guides leads each cable across joints to its proper pulley as necessary. The maximum width of the structure running along the finger is only 8 mm.

C. Sensing

Joint angles are computed from the motor shaft rotations, as measured from optical encoders integrated into each motor. The encoders provide 2000 counts per revolution of the shaft. Motor shaft rotation is converted to joint rotation through consideration of the pulley and gear reduction between the motor and the joint.

To compute joint torques, we measure cable tension from both the agonist and antagonist actuators. Custom load cells consist of three bearing-supported guide pulleys through which the cable runs (Fig. 3). Tension in the cable generates a downward force on the central pulley which is connected to the primary tension sensing element. This thin steel beam is equipped with dual strain gauges (KFG-02-120-C1- 11L1M2R, Omega, Stamford, CT), one superior and the other inferior on the beam. The gauges constitute one-half of a

283

Wheatstone bridge, whose output voltage is linearly related to cable force.

Figure 3 – Load cell deflects the tensioned cable (T), generating a central

downward force (F) resulting in deflection of the sensing element and strain gauges (red).

D. Control

The control system of the finger exoskeleton was

developed using MATLAB xPC target. The xPC target runs

on an independent computer which allows for real-time

control. Executable code is loaded onto this target PC from a

host PC running MATLAB Simulink software [23]. The

real-time data signals, acquired by target PC, are uploaded to

the host PC over a direct crossover Ethernet cable. The

experimenter can monitor the data signal in the host scope

and tune the parameters online in the Simulink model created

in the host PC (Fig. 4).

A PCI-6220 ADC board (National Instruments, Austin, TX) is installed on the target PC to perform analog-digital conversion of the signals from the tension sensors. The CNT32-8M encoder board (CONTEC, Sunnyvale, CA) inputs digital encoder signals, and the PCI-6703 DAC board (National Instruments) converts digital command signals into the analog signals which drive the S200 motor amplifiers (Kollmorgen). All signals are sampled at 10 kHz.

Target PCReal-time kernel

Sensor Acuator

Download Code

Finger exoskeleton

Data

acquisition

Control

signal

Position

ForceTorque

Simulink Model

Executable Code

VC++ Compiler

Host PC

Upload data

Host Scope

Figure 4 – Real-time Control System using xPC Target. The host PC

manages the control program, visual feedback and data storage; the target PC runs the real-time control and acquires sensor data.

Critical to the use of cable transmissions is maintaining taut cables. Thus, cables running to more distal joints must account for changes in more proximal joints. Additionally, the agonist-antagonist cables for a given joint must be adjusted simultaneously. To accomplish this, each motor receives a tension command to keep the cables taut, and all actuation commands are added to this baseline signal.

Command signals are generated by three PID control loops with friction compensation (Figure 11). The controlled variable for each loop is the angle measured by the quadrature encoders of the motors. Each motor in the pair for a given joint independently tracks joint angle and, when the joint is actuated, the motors in the pair pull the cables in opposite

directions. The driving motor winds its cable and the following motor unwinds its cable while maintaining tension. For the PIP and DIP joints, when a more proximal joint moves, the cables for both motors will be identically affected by the rotation. Thus, the motor pair rotates in opposite directions to actuate the joint and in the same direction to compensate for proximal joint movements. The actual joint angle is then the difference in the motor encoder values between motors of the same joint. In this way, the system passively accounts for changes in proximal joints; as the non-driving motor responds to maintain tension, the measured joint angle will change and the driving motor will counter the perceived movement accordingly.

Figure 5 – Block diagram of PID Control Loop with friction compensation.

The PID parameters of the AFX control loops were tuned to produce a stable motion when no external resistance is applied. The control loop runs at 10 kHz.

For safety consideration, an overdamped or critically

damped behavior in the joint angle control is preferred to an

underdamped system. The joint limits and torque limits are

also defined in the program and constantly checked.

Mechanically, guide slots restrict the range of motion of

each joint and can be adjusted to limit joint range as needed.

In our design, differently sized motors are used for the

different joints. In this manner, peak motor torque can be

better matched to peak voluntary subject torque and the

potential for excessive torque is minimized. An emergency

switch immediately terminates all power to the motors.

III. PERFORMANCE TESTING

A. Visual Tracking System

A two-camera setup employing high-resolution, monochrome CCD cameras (IPX-1M48, Imperx, Inc., Boca Raton, FL) was employed to examine performance of the AFX (Fig. 6). Namely, the cameras were used to measure exoskeleton joint angles to compare with the desired angles. Markers were attached to the exoskeleton to record movement. The markers were covered with ultraviolet- sensitive fluorescent paint (Wildfire, Modern Masters, Inc., N. Hollywood, CA) and illuminated with a UV light source.

Motion capture and analysis was carried out using a digital motion analysis suite (DMAS7, Spica Technology, Co., Maui, HI). The cameras were calibrated using the software provided in DMAS7 and a custom calibration form. An average calibration error of < 0.6 mm between cameras is

284

Figure 8 - AFX performance testing experimental setup. Black tape

marks the calibrated region.

Figure 6 - Target (green) and observed position over time of MCP (red), PIP (blue), and DIP (purple) joints. Positive angles for extension and

negative angles for flexion.

Figure 7 – Target position vs. observed position of MCP (red), PIP (blue) and DIP (purple) joints with linear regression (black).

achieved before proceeding. During motion capture, the 3-dimensional position of each marker was recorded and these positions were used to compute joint angles.

B. Experimental Procedures

For each experiment, the AFX was first rotated to the desired initial position (full extension, full flexion or in between) and slack was eliminated from the cables. During each trial, the camera system began recording before the control code was executed. The camera tracks actual positions while the control code tracks both desired and encoder-measured positions.

To test the exoskeleton’s ability to appropriately track a position, constant-velocity desired trajectories of 10°/s, 15°/s, and 15°/s for 5 seconds were created in either extension or flexion for the MCP, PIP or DIP joints, respectively. The camera system tracked the AFX as each joint tried to follow the path. The exoskeleton began each trial in the full extension or full flexion posture as appropriate.

To examine the ability of the AFX to achieve high rotation speed, a desired sinusoidal trajectory was generated in the MCP joint with an amplitude of 30º and an angular frequency

of 10π. This results in a theoretical maximum instantaneous velocity of 945°/s. The AFX began in a central posture to allow room for the tracking task.

To examine the ability of the AFX to simultaneously control each joint independently, desired sinusoidal trajectories were generated with amplitude 15° for 4 seconds, at distinct angular frequencies for each joint: π/4 (MCP), π/2 (PIP), and π (DIP). The AFX began in a central posture to allow room for oscillations.

C. Analysis

The achieved speed of the exoskeleton was computed from the derivative of the observed position. Unfortunately, the camera system is unable to track the exoskeleton at the high desired speeds, so we rely on encoder data from the motors. Accuracy of the encoder readings was first validated by comparing the encoder outputs to the camera data for slower movements for which camera data was valid.

The sinusoid data were processed by first aligning the camera data with the command signals for the desired trajectories. For each sinusoid, we calculated the sample correlation between observed and desired trajectories.

IV. RESULTS

For each experiment we record the desired position generated by the controller and the actual position as measured by the external camera.

A. Ramp

For each ramp, we first plotted the observed and desired angles for each joint (Fig. 7). For the most part, agreement was quite good. To quantify the accuracy of this tracking, the actual position regressed against the target position (Fig. 8) for every time step. For each joint, the slope of the resultant relationship and the associated R2 value are listed (Table 1).

285

Table 1 - Actual vs. observed ramp angle correlations.

Joint MCP PIP DIP

Slope 1.014 0.978 1.031

r2 0.999 0.999 0.992

Figure 9 – Encoder vs. camera recordings during MCP linear movement.

Figure 10 – MCP sinusoid at frequency 10π and amplitude 30°.

Table 2 - Correlation values for single trial MCP, PIP, and DIP sinusoids.

Joint MCP PIP DIP

r 0.91 0.95 0.83

B. Angular Velocity

Encoder readings are shown against the joint angles recorded from the camera for a ramp trajectory (Fig. 9). The sample correlation between encoder and camera observed position is 0.999.

The encoder-based joint angle measured at the MCP joint was then compared to the desired joint angle for the 10π sinusoid. The AFX was able to track this trajectory even as the rotational speed of 940°/s (Fig. 10).

C. Three-Joint Sinusoid

Desired and measured (camera system) trajectories for MCP, PIP and DIP for the three simultaneously applied sinusoids were then compared (Fig. 11). Sample correlation values ranged from 0.95 for PIP to 0.83 for DIP (Table 2).

V. DISCUSSION

While adding only 140 g of mass to the finger, the AFX can move at substantial speeds and provide considerable joint torque while retaining much of its backdrivability. With active control input, further compensation for the device can be provided to lessen its effects on voluntary movement. Thus, the AFX can be employed to examine the impact of a number of different control algorithms, from full assistance (position servo) to zero impedance (no assistance). This flexibility is valuable for efficacy testing of rehabilitation strategies and the study of motor control.

The high level of backdrivability is made possible by the chosen cable actuation system. The cable transmission permits the actuators to be placed proximal to the hand while minimizing the frictional losses that are inherent to other transmissions, such as Bowden cables. Of course, a trade-off must be made in terms of control complexity and additional hardware.

Namely, two motors are needed to control each joint, as each cable can only pull, not push. These two motors must act in concert to allow joint movement while maintaining tension in both cables. Additionally, the cables to more distal joints must respond to movement of more proximal joints; cable length is a function of all preceding joint angles as well as the joint the cable controls. Another benefit of this actuation mode that should be noted is the ability to set the joint stiffness as the tension in the agonist-antagonist cables can be modified in tandem.

Kinematic testing confirmed the ability of the AFX to track desired angle trajectories over time. Tracking of the desired ramps was quite good, with R2

values of 0.99 or greater. Admittedly, tracking of the sinusoids was less successful. During the sinusoid experiment, the MCP and PIP joints exhibited phase lag during flexion. This delay is likely caused by the cables running across joints to actuate the PIP and DIP. The tension in those cables generates an extension torque about more proximal joints, increasing the torque required for flexion. During extension, the cable tensions have the reverse effect, assisting the extension of the joints.

For the DIP, phase lag appears during extension, not flexion. Given that there are no cables running across the DIP joint, it fits that the result would differ from the proximal joints. While torque is not a problem, the motors must account for the differences in cable length caused by movement at both PIP and MCP segments in addition to the desired movement. Thus, DIP motor velocity must also accommodate PIP and MCP angular velocity. An additional

Figure 11 – Observed MCP (blue), PIP (red), and DIP (green) positions, plotted with target (grey) sinusoidal

position over time. Data shown for a single movement trial originating at each joint’s center of motion.

286

source of error is the tuning of the PID controller for DIP. As the segment is light weight and thus, has limited resistance, it is difficult to tune with stability. Control actually becomes easier when the device is loaded with a finger.

In regard to feedback control, the encoder appears to provide a reliable measure of joint angle. The angles measured during ramp trials were found to result in a highly linear relationship (r2 = 0.999) with angles observed by the camera system.. This validates the differential encoder method and indicates that the encoders will be more than sufficient for position feedback control of the AFX. It also suggests that cable winding has limited effect on joint angle.

Future implementations of the controller will include feed- forward angle compensation in addition to the currently implemented reactive tensioning. This will account for the posture of all joints and provide excursion for distal motor pairs to reduce surplus extension torque on proximal joints. Similarly, we will implement feed-forward torque and inertia compensation for the effect of distal joints.

Compared to other available exoskeletons, the AFX operates at speeds above and beyond what is currently available, while maintaining the controllability necessary to implement a wide range of rehabilitation and motor control study paradigms. The AFX is also contained in a form factor allowing for a wide finger workspace, low interference and arm mobility as necessary.

VI. CONCLUSION AND FUTURE WORK

The AFX successfully controlled position across all three joints. While there is room for improvement during rapid, multi-joint control, the analysis has demonstrated significant controllability with the selected sensors and proposed control strategies. Future improvements will include the addition of force feedback and feed-forward kinematics that will mitigate shortfalls discovered during this analysis and will include full bandwidth testing. Performance of the device with human subjects will be examined next.

Based on current findings, the AFX has the potential to satisfy all design requirements. Discoveries from this analysis will lead to significant improvements and enable implementation for rehabilitation and hand motor control studies.

ACKNOWLEDGMENT

We acknowledge Tom Worsnopp for his contributions in the initial development of the AFX, and Ed Colgate and Michael Peshkin from the Laboratory for Intelligent Mechanical Systems at Northwestern University for their insight and continued support of the project.

REFERENCES

[1] E.G. Cruz, H.C. Waldinger and D.G. Kamper "Kinetic and kinematic workspaces of the index finger following stroke," Brain vol. 128, no. Pt 5, pp. 1112-1121, 2005.

[2] N.J. Seo, W.Z. Rymer and D.G. Kamper "Altered digit force direction during pinch grip following stroke," Exp Brain Res 2010.

[3] G.S. Seale, I.M. Berges, K.J. Ottenbacher and G.V. Ostir "Change in positive emotion and recovery of functional status following stroke," Rehabil Psychol vol. 55, no. 1, pp. 33-39, 2010.

[4] C.f. Prevention "Outpatient rehabilitation among stroke survivors: 21 states and the District of Columbia," MMWR Morb

Mortal Wkly Rep. no. 56, pp. 504-507, 2007. [5] A.H. Association "Heart Disease and Stroke Statistics 2010

Update," Circulation no. 121, pp. 46-215, 2009. [6] S.B. S. L. Wolf, H. Baer, J. Breshears, and A. J. Butler

"Repetitive task practice: a critical review of constraint-induced movement therapy in stroke," Neurologist vol. 8, pp. 325-28 2002.

[7] D.K.R. C. J. Winstein, S. M. Tan, R. Lewthwaite, H. C. Chui, and S. P. Azen "A randomized controlled comparison of upper-extremity rehabilitation strategies in acute stroke: A pilot study of immediate and long-term outcomes," Arch Phys Med Rehabil vol. 85, pp. 620-8 2004.

[8] S.S. S. J. Page, P. Levine, and R. E. McGrath "Efficacy of modified constraint-induced movement therapy in chronic stroke: a single-blinded randomized controlled trial," Arch Phys Med Rehabil vol. 85, pp. 14-8, 2004.

[9] W.H.R.M. J. Liepert, H. Bauder, M. Sommer, C. Dettmers, E. Taub, and C. Weiller "Motor cortex plasticity during constraint-induced movement therapy in stroke patients," Neurosci

Letters vol. 250, pp. 5-8, 1998. [10] H.B. J. Liepert, W. H. R. Miltner, E. Taub, and C. Weiller

"Treatment-induced cortical reorganization after stroke in humans," Stroke vol. 31, pp. 1210-1216, 2000.

[11] L. Pignolo "Robotics in neuro-rehabilitation," J Rehabil Med vol. 41, no. 12, pp. 955-960, 2009.

[12] G.G.F. Sergei V Adamovich, Abraham Mathai, Qinyin Qiu, Jeffrey Lewis and Alma S Merians "Design of a complex virtual reality simulation to train finger motion for persons with hemiparesis: a proof of concept study," Journal of NeuroEngineering and Rehabilitation pp. 6-28, 2009.

[13] R.B. D. Jack, A. S. Merians, M. Tremaine, G. C. Burdea, S. V. Adamovich, M. Recce, and H. Poizner "Virtual reality enhanced stroke rehabilitation," IEEE Trans Neural Syst Rehabil Eng vol. 9, pp. 308-18, 2001.

[14] L.D. C. D. Takahashi, V. Le, R. R. Motiwala, and S. C. Cramer "Robot-based hand motor therapy after stroke," Brain vol. 131, pp. 425-37, 2008.

[15] O.L. L. Dovat, R. Gassert, T. Maeder, T. Milner, T. C. Leong, and E. Burdet "HandCARE: a cable-actuated rehabilitation system to train hand function after stroke," IEEE Trans Neural Syst Rehabil Eng vol. 16, pp. 582-91, 2008.

[16] A.W. Hommel "Development and control of a hand exoskeleton for rehabilitation of hand injuries," IEEE/RSJ

International Conference on Intelligent Robots and Systems 2005. [17] N.G.T. I. Sarakoglou, and D. G. Caldwell "Occupational and

physical therapy using a hand exoskeleton based exerciser," IEEE/RSJ

International Conference on Intelligent Robots and Systems, Sendai 2004.

[18] L.L. M. DiCicco, and Y. Matsuoka "Strategies for an EMG-controlled orthotic exoskeleton for the hand," IEEE International

Conference on Robotics and Automation 2004. [19] I.S. Kawasaki H, Ishigure Y, Nishimoto Y, Aoki T, Mouri H, et

al. "Development of a hand motion assist robot for rehabilitation therapy by patient self-motion control," IEEE 10th Intl Conf Rehab Rob pp. 234-240, 2007.

[20] B. Gutnik, Hudson, G., Ricacho, G., Skirius, J. "Power of Performance of the Thumb Adductor Muscles: Effect of Laterality and Gender," Medicina (Kaunas) vol. 42, no. 8, pp. 654-656, 2006.

[21] T. Worsnopp, M. Peshkin, J. Colgate and D. Kamper "An Actuated Finger Exoskeleton for Hand Rehabilitation Following Stroke," IEEE 10th International Conference on Rehabilitation

Robotics pp. 896-901, 2007. [22] W. Townsend and J. Salisbury "Mechanical bandwidth as a

guideline to high-performance manipulatordesign," IEEE International Conference on Robotics and Automation 1989.

[23] xPC target documents, Mathworks, http://www.mathworks.com/access/helpdesk/help/toolbox/xpc/

287