Autonomous Robots 15, 3551, 2003c 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.

Upper Limb Robot Mediated Stroke TherapyGENTLE/s Approach

RUI LOUREIRO AND FARSHID AMIRABDOLLAHIANtHRILthe Human Robot Interface Laboratory, Department of Cybernetics, School of Systems Engineering,

The University of Reading, Whiteknights, PO Box 225, Reading, RG6 6AY, UKr.c.v.Loureiro@reading.ac.uk

MICHAEL TOPPINGStaffordshire University, School of Design, College Road, Stoke on Trent, Staffordshire, ST4 2XN, UK

BART DRIESSENTNO-TPD Institute of Applied Physics, Control Engineering, Delft, The Netherlands

WILLIAM HARWINtHRILthe Human Robot Interface Laboratory, Department of Cybernetics, School of Systems Engineering,

The University of Reading, Whiteknights, PO Box 225, Reading, RG6 6AY, UK

Abstract. Stroke is a leading cause of disability in particular affecting older people. Although the causes of strokeare well known and it is possible to reduce these risks, there is still a need to improve rehabilitation techniques.Early studies in the literature suggest that early intensive therapies can enhance a patients recovery. According tophysiotherapy literature, attention and motivation are key factors for motor relearning following stroke. Machinemediated therapy offers the potential to improve the outcome of stroke patients engaged on rehabilitation forupper limb motor impairment. Haptic interfaces are a particular group of robots that are attractive due to theirability to safely interact with humans. They can enhance traditional therapy tools, provide therapy on demandand can present accurate objective measurements of a patients progression. Our recent studies suggest the useof tele-presence and VR-based systems can potentially motivate patients to exercise for longer periods of time.The creation of human-like trajectories is essential for retraining upper limb movements of people that have lostmanipulation functions following stroke. By coupling models for human arm movement with haptic interfaces andVR technology it is possible to create a new class of robot mediated neuro rehabilitation tools. This paper providesan overview on different approaches to robot mediated therapy and describes a system based on haptics and virtualreality visualisation techniques, where particular emphasis is given to different control strategies for interactionderived from minimum jerk theory and the aid of virtual and mixed reality based exercises.

Keywords: hemiplegia, motor relearning, robot mediated therapy, virtual environments, assistive technology

1. Introduction

Cerebral Vascular Accidents (CVAs), often-calledstrokes occur when the blood supply to the brainis interrupted either by a blood clot or internal bleed-ing. Due to this interruption some parts of the brain do

not receive fresh oxygenated blood and neurons in thatregion die. Depending on the region of neuronal death,control, sensory or cognitive functions may be lost orimpaired. The most common of all strokes account-ing for 7080% are cerebral thrombosis and cerebralembolism (Westcott, 2000). They are a leading cause

36 Loureiro et al.

of disability with estimates of the annual incidence ofstroke ranging from 180 per 100,000 in the USA, 1251per 100,000 in Europe (Stewart et al., 1999) to 200 per100,000 in England and 280 per 100,000 in Scotland(Isard and Forbes, 1992). England and Wales statis-tics show that every year around 100,000 people havetheir first stroke from which 10,000 are under retire-ment age.2 ,3 Approximately one-third of the peoplesurviving from a stroke are left with severe disabili-ties (Abrams and Berkow, 1997). One common con-sequence is movement and co-ordination of the UpperLimb (UL). Almost 85% of people with stroke show aninitial deficiency in the UL (Parker et al., 1986) fromwhom only 50% recover function of their affected UL(Parker et al., 1986; Broeks et al., 1999).

Stroke is the third most common cause of deathin England and Wales, after heart disease and cancer(Petersen et al., 2000). Mortality statistics for the UKshows that in 1999 a total of 64,971 deaths4 were dueto stroke (ONS, 2000). The cost of stroke to the Na-tional Health Service is estimated to be over 2.3 billion( 3.45 billion) per year. The total cost of stroke care isexpected to rise in real terms by around 30 per cent bythe year 2023 (Westcott, 2000). The UK governmentis determined to reduce the death rate from stroke andrelated diseases in people less than 75 years by at leasttwo fifths by 2010saving up to 200,000 lives per yearin total (Petersen et al., 2000).

Following the stroke the effects are almost immedi-ate and vary according to the level of damage done tothe brain. Some of the most important symptoms in-clude: unexpected insensibility, weakness or paralysison one side of the body, or in a higher level both sides.Signs of this might be a weak arm, leg or eyelid, ora dribbling mouth, difficulty finding words or under-standing speech, sudden blurring, disturbance or lossof vision, especially in one eye, dizziness, confusion,unsteadiness and/or severe headache (Westcott, 2000).

During the initial critical phase the aim is to stabilisethe condition, control blood pressure and prevent com-plications. Once the patient is stable, he/she is engagedin an individual rehabilitation programme designed tohelp regain independence and relearn the lost skills.

Several authors (Parker et al., 1986; Feys et al., 1998)have shown that the reason that Upper Limb (UL) re-covery of function is more difficult to achieve than withLower Limbs (LL) is due to the UL complexity. UL areused for gesture and non-purposeful movements suchas reflex perception and balance but their main func-tions are task oriented. These tasks involve UL func-

tions such as locating a target, reaching (transport ofarm and hand), grasping an object (grip formation andrelease) and postural control. The recovery of such taskoriented function is of extreme importance in every daymanipulative tasks (Ashburn, 1993).

1.1. Physiotherapy Practise

Physiotherapy is inconsistent, varying from one thera-pist to another and from hospital to hospital. Many dif-ferent approaches for treatment techniques have beenproposed (e.g. Bobath, 1978; Cotton and Kinsman,1983; Knott and Voss, 1968; Rood, 1954). The Carr &Shepherd method (Carr and Shepherd, 1987) focuseson motor relearning where relearned movements arestructured to be task specific. That is, motor controlis both anticipatory and ongoing and postural controllimb activities are interrelated. The practice of specificmotor skills leads to the ability to perform the task.The task should be practised in the appropriate environ-ment where sensory input modulates the performanceof motor tasks. The model is based on normal motorlearning, the elimination of abnormal movement, feed-back, practice and the relationship between posture andmovement.

Physiotherapy practice is partially based on theo-ries but is also heavily reliant on the therapists trainingand past experience. Based on this many authors de-scribe that there is insufficient evidence to prove thatone treatment is more effective than any other (Sackleyand Lincoln, 1996). Recent reviews of Upper Limbpost-stroke physiotherapy concluded that the best ther-apy has not yet been found (Ernst, 1990; Coote andStokes, 2001). Nonetheless, several studies (Kwakkelet al., 1997; Sunderland et al., 1992; Lincoln and Parry,1999) point out that UL motor re-learning and recov-ery levels tend to improve with intensive physiotherapydelivery. The need for conclusive evidence supportingone method over the other and the need to stimulate thestroke patient (Sackley and Lincoln, 1996) clearly sug-gests that traditional methods lack high motivationalcontent, as well as objective standardised analyticalmethods for evaluating a patients performance and as-sessment of therapy effectiveness.

1.2. Robot Mediated Therapy Approaches

Several authors have already proposed the use ofrobotics for the delivery of Upper Limb post-stroke

Upper Limb Robot Mediated Stroke Therapy 37

physiotherapy. The first far-reaching study on the ac-ceptance of robot technology in occupational therapyfor both patients and therapists was done by Dijkerset al. (1991) using a simple therapy robot. Dijkersstudy reported a wide acceptance from both groups,together with a large number of valuable suggestionsfor improvements. Advantages of Dijkers therapy in-cluded the availability of the robot to successively re-peat movements, as well as the ability to record move-ments. However, there was no measure of movementquality, and patient cooperation was not monitored.

At the VA Palo Alto Rehabilitation R&D center andStanford University, Johnson et al. (1999) have devel-oped the SEAT: simulation environment for arm ther-apy to test the principle of the mirrored-image bythe provision of a bimanual, patient controlled thera-peutic exercise based on a driving simulator. The devicecomprises a customised design of a car steering wheelequipped with sensors to measure the forces appliedby a patients limbs, and an electrical motor to providepre-programmed assistance and resistance torques tothe wheel. Visual cues were given to the patient viaa commercially available low cost PC-based drivingsimulator that provided graphical road scenes. The in-terface allowed the participation of the patient in thetask and the involvement of the paretic limbs in theexercise. The SEAT system implemented 3 differenttherapy modes: passive, active and normal mode. Inthe passive mode, the servomechanism compensatedfor the weight of the paretic limb which was guided bythe non-paretic limb. Active mode is used when the sub-jects demonstrated some level of voluntary control ofthe paretic limb. In this mode, the servomechanism onthe steering wheel encouraged the participation of theparetic limb when performing the steering task withthe paretic limb while relaxing the non-paretic limb.The normal mode was designed to assess the force dis-tribution and analyse the participation of the pareticlimb by the participation of both limbs in the steeringtask and assess the limbs co-ordination. Recent resultssuggest that SEAT system increased the interest of sub-jects in using the impaired limb in the steering task andthe use of the automated constraint discouraged com-pensatory use of the stronger limb (Johnson et al., 2001,2002).

Based on the same mirror image concept, Lum et al.(1999) at the VA Palo Alto Rehabilitation R&D centerproduced the MIME: Mirror-image motion enabler.A Puma 260 robot was used for the initial prototype,which was attached via a force-torque sensor to the

arm splint. In the current prototype a Puma-560 robotreplaces the original Puma-260 and its paretic limb mo-bile arm support while a 6DOF digitiser replaces thenon-paretic arm support. The MIME system can workin pre-programmed position and orientation trajecto-ries or in a slave configuration where it mirrors themotions of the non-paretic limb. A computer controlsmovement of the robot, with specific pre-programmedtasks tailored to the subjects level of recovery and ther-apeutic goals. Clinical trials with 27 chronic strokesubjects (>6 months post stroke) based on the Fugl-Meyer exam, have shown no negative effects. Resultsalso suggested that robot-aided therapies are safe andeffective for neuro development treatment (Shor et al.,2001).

Rao et al. (1999) have used the Puma-260 with apassive and active mode. In the passive mode the robotguided the patients arm through a specified path and inthe latter the patient lead the robot along a predefinedpath based on graphical interface resembling a tunnel.If the patient collided with the wall of the tunnel, therobot would take control and bring the patients armback to the normal path. One of the advantages of thisimplementation is that the tunnel constraints could bechanged according to each individuals need over therecovery time slot. The results from the test bed imple-mentation of the Puma-260 suggested that the subjectslearned to minimise deviations from the centre line inrepeated trials and the torque applied to the end-effectorbecame smoother over exercise time.

Work done at MIT by Krebs et al. (1999) on thedevelopment of a manipulandum that allowed the sub-jects to exercise against therapist nominated stiffnessand damping parameters uses a different approach fromthe systems described so far. Their project defines anew class of interactive, user-friendly clinical devicefor evaluating and delivering therapies via the use ofvideo games. They have designed and used the com-mercialised MIT-MANUS (Hogan et al., 1995) a 3DOF(2DOF active 1DOF passive) planar manipulator toperform a series of clinical trials. Reports on initialresults with 20 subjects with stroke, where 10 usedthe MIT-MANUS in addition to normal therapy foran additional 45 hours per week suggests that theyhad improved substantially compared to the ones un-dertaking normal therapy. Recent results were reported(Krebs et al., 1999) from a total of 76 subjects. It wasshown that the manipulation of the impaired limb influ-ences recovery, the improved outcome was sustainedafter 3 years, the neuro-recovery process continued far

38 Loureiro et al.

beyond the commonly accepted 3 months post-strokeinterval, and the neuro-recovery was dependent on thelesion location. The MIT-MANUS mechanism how-ever limits the range of possible therapies, has limiteddata collection facilities and does not allow bimanualtherapies as the SEAT and MIME systems.

Reinkensmeyer et al. (2001) introduce a differentapproach with their web-based force feedback telere-habilitator called Java Therapy. Java Therapy is aninexpensive robotic telerehabilitation system for armand hand therapy. It consists of a web site with alibrary of evaluation and therapy activities that canbe performed with a commercial force feedback joy-stick, which can physically assist or resist movementas the user performs therapy. It also allows for somelevel of quantitative feedback of movement perfor-mance, allowing users and their caregivers to assessrehabilitation progress via the web. The MicrosoftSidewinder Force Feedback Joystick was used to movethe patients arm while the subject interacted withsimple 2D games and perform speed, co-ordinationand strength tests. Initial comments on this new ther-apy indicate that while the subject gains concentra-tion, there is no significant improvement in motorcontrol of the upper arm mainly due to the verysmall workspace and force feedback provided by thejoystick.

The systems and research reviewed above focused onthe rehabilitation of the upper arm. Work in the con-text of haptic feedback in the rehabilitation of the handwas done at Dartmouth College (Brown et al., 1993)on an exoskeleton used as a prosthetic device by sub-jects who had lost muscular control of their hand. Thedevice consists of an instrumented aluminium struc-ture attached to the back of the hand via a Lycra glove.Position is measured by potentiometers and five cablesrouted to the palmar side are used to close the index andthumb fingers in a pinch grasp. DC motors located onthe forearm caused finger flexion, and restoring springsin the exoskeleton pull the fingers to a neutral positiononce the actuators are de-energised. Initial tests showedgood range of motion, good repeatability but calibra-tion was needed for every new patient, and cable staticfriction and exoskeleton weight were judged to be toolarge.

A different approach was used by Popescu et al.(2000) for orthopaedic rehabilitation. It consisted ofa PC based rehabilitation station with a Polhemustracker, and used the Rutgers Master II glove. Theyhave developed three different 3D graphical exercises

and two functional games (Pegboard and Ball game).Data is collected into an Oracle database and sent viathe Internet to a remote site for analysis. The system isat present undergoing clinical trials at Stanford MedicalSchool.

2. GENTLE/s Neuro-Rehabilitation Approach

The GENTLE/s project is financed by the EuropeanCommission under the Quality of Life initiative ofFramework 5, which aims to evaluate robot-mediatedtherapy in stroke rehabilitation. The project takes awide group of users to include patients, family mem-bers, physicians, physiotherapists, and healthcare man-agers. GENTLE/s is focused on neuro and physicalrehabilitation and particularly concentrates on devel-oping new, challenging and motivating therapies to aidthe increase of sensory input, relearning stimulation inthe brain, and achieve functional goals that improve in-dependence and coordination. A more detailed descrip-tion of the different project phases is given in Harwinet al. (2001).

The GENTLE/s approach utilises haptic and VirtualReality (VR) technologies. Haptics is the study of in-tegrating tactile, proprioceptors and other sensors intomeaningful information (Gibson, 1966) and a hapticinterface uses constrained motion devices to replicatesome of these sensory modalities. An initial user needsstudy along with brainstorming sessions and input frommembers of the Young Stroke Association at Stoke-on-Trent in the UK, encouraged the group to develop theidea of using haptic and VR technologies to delivertherapy. It was postulated from group discussions thatbetter functional and motor recovery outcomes couldbe achieved when patients receive a challenging andmotivational machine mediated therapy in a contextthat allows the stroke patient to feel comfortable andin control.

2.1. Assumptions

Some studies have shown that repetitive task-orientedmovements are of therapeutic benefit. With the use ofhaptics and VR technology, patient attention and moti-vation can be enhanced by means of Active Feedbackthat will further facilitate motor recovery through brainplasticity (Butefisch et al., 1995; Hesse et al., 1995).Four different levels for Active Feedback have been

Upper Limb Robot Mediated Stroke Therapy 39

identified: visual, haptic, auditory and performancecues. The creation of active agents and biofeedbackcan be a way of implementing and integrating activefeedback in a neuro-rehabilitation robotic system.

Visual cues: In some cases, following a stroke, hemi-plegic subjects tend to be confused about what theysee (Westcott, 2000). The brain needs to be re-educated to associate (for example) colours and ob-jects. As a result of the need of cognitive re-learningit is important that visual cues be simple, yet stim-ulating. Visual cues can be represented using realtasks based on the ones used in occupational ther-apy sessions, to realistic and accurate goal oriented3D computer environments. This can be anythingfrom a virtual room (for example a virtual kitchen ormuseum) to an interactive game.

Haptic cues: Kinaesthetic feedback can help to dis-criminate physical properties of virtual objects, suchas geometry. It can also be used to deliver physicaltherapy to a human subject using haptic interfaces.The force delivered in this way can be very ther-apeutic dependent on the way we apply this forceto human muscular and skeletal systems. It will un-doubtedly play an important role when manipulatingobjects, either virtual or real. In conjunction withinteractive virtual and augmented tasks, it can sim-ulate the shape of a virtual pen, bingo card or thefriction/drag when writing on the virtual bingo card.

Auditory cues: In some cases it may be appropriateto give encouraging words and sounds when the per-son is trying to perform a task and congratulatory orconsolatory words on task completion.

Performance cues: In a haptic stroke rehabilitationsystem, results of the previous tasks can be displayedindicating the errors committed and the level ofhelp given when completing the task. These perfor-mance cues should be designed to give constructivefeedback.

A robotic/haptic rehabilitation system should be er-gonomically comfortable. The therapy should be en-joyable and the system should be considered trust wor-thy by the patient. Such a concept can be achieved bythe introduction of a personality to the system, such asa character (wizard) that interacts with the patient byusing different identified cues. Different wizards canbe implemented for different personalities. These canbe defined and assigned to the patient by the therapist.

An example could be in the case where the patientis performing a simple exercise, such as reaching for

an object in a virtual world. In this case, with the aidof a good sensor system, analytical measures can beobtained in order to identify if the patient is strugglingin reaching the target. Taking this into account, thewizard could then offer encouragement to the patientto finish the movement. Another scenario is to presentthe score at the end of the session.

2.2. Current Prototype

The current prototype system (Fig. 1) consists of aframe, a chair, a shoulder support mechanism, a wristconnection mechanism, an elbow orthosis, two embed-ded computers, a large computer screen with speakers,an exercise table, a keypad and a 3DOF haptic inter-face (HapticMaster from FCS Robotics, see www.fcs-robotics.com). Current generations of haptic technolo-gies allow relative large reaching movements in threeactive degrees of freedom. This couples to three pas-sive degrees of freedom to allow arbitrary positioningof the persons hand. The patient is seated on the chairwith his/her arm positioned in an elbow orthosis sus-pended from the overhead frame (Fig. 2). This is toeliminate the effects of gravity and address the prob-lem of shoulder subluxation. The wrist is placed in awrist-orthosis connected to the haptic interface usinga quick release magnetic mechanism (Fig. 2). At thispoint, the physiotherapist can select the patient profilefrom the database, select an exercise or create a newexercise using a 3D graphical user interface (Fig. 3).The setup exercise (Figs. 4 and 5) allows the therapistto define the exercise path, amount of help needed foreach segment of the exercise, duration of the move-ment for each segment, the 3D context of the virtualexercise environment. When the exercise definition hasbeen saved and the exercise is selected, the system isready to perform that exercise with the patient.

2.3. Exercises & Movement Guidance

In the current prototype, three different virtual environ-ments can be used (Fig. 6):

1. Empty roomA simple environment that repre-sents the haptic interface workspace and intends toprovide early post-stroke subjects with awarenessof physical space and movement (Fig. 6(A)).

2. Real roomAn environment that resembles whatthe patient sees on the table in the real world.

40 Loureiro et al.

Figure 1. Initial concept design for GENTLE/s Robot mediated therapy.

The mat with 4 different shapes on the table(Figs. 2 and 3) is represented in the 3D graph-ical environment (Fig. 6(B)). This environmentwas developed to help discriminating the third di-mension that is represented on the Monitor 2Dscreen.

3. Detail roomA high detail 3D environment of aroom comprising of a table, several objects (a book,can of soft drink), portrait of a baby, window, cur-tains, etc (Fig. 6(C)).

In order to allow the user to navigate and interact witha virtual/real task, several mathematical models havebeen implemented (further explained in Section 3) asa control strategy capable of correcting the patientsmovement. An operation button on the keypad must bepressed continuously by the user (Fig. 3) for the du-

ration of the movement. Since movement control wasdefined to be in between two points, a new concept wasintroduced. The Bead Pathway concept assumes thatmovement takes place in between a start point and anend point. It is assumed that its behaviour is similar tothe behaviour of beads on a wire, they can only movealong their pathway. To achieve this behaviour the en-deffector is connected to a virtual spring and damper(Fig. 7) where the bead is constrained to move alonga wire-pathway that defines both the path and thevelocity profile of the movement.

Deviations from the movement profile are permittedbut constrained depending on the restoring force of thespring and associated damper. Different levels of guid-ance and correction can be programmed for differentsubjects with different recovery levels. For a patient inearly days after stroke, more help is needed to move

Upper Limb Robot Mediated Stroke Therapy 41

Figure 2. Subject using the GENTLE/s system. Arm is positionedin an elbow orthosis suspended from the overhead frame and con-nected to the robot using a wrist-orthosis that is secured using a quickrelease magnetic mechanism.

Figure 3. Therapist setting therapy tasks and instructing subject.

along the pathway, and this behaviour can be achievedby implementing a velocity profile for the bead on thepathway and proper spring-damper combination formore assistance. For a patient who has recovered moremotor function, we may need a different velocity pro-file along the pathway and a different setting definedfor the spring-damper behaviour.

Figure 8 shows the representation of the pathway,where in part (A) of the figure can be seen the yel-low start point (light shaded), the blue end point (darkshaded) and the desired trajectory between these twopoints (pathway). The choice these of colours addressesthe problem of colour blindness with some primarycolours, such as green and red that was noticed withsome subjects in earlier studies (Loureiro et al., 2001).In part (B) of Fig. 8, shadows and lines connectingshadow to object are used to help the user to perceive thedepth and height of the positioned points with respectto the table on the screen.

3. Implementation of Different Therapies

Several studies have agreed that the importance of aclear understanding of how the human arm moves isachieved to supplement interaction between a machineand a human subject. The urge to understand humandynamics with emphasis on explaining how humansbrain plans for reaching movements was first studiedby Bernstein (1967) and later by Bizzi et al. (1984).This lead to the study of several kinematic (Hogan,1984), dynamic (Uno et al., 1989), and neural features(Georgopoulos et al., 1981) of human arm reachingmovements. In most of these studies, one characteri-sation of multi-joint planar reaching movements wasfound to be a straight path with a bell shaped veloc-ity profile (Flash and Hogan, 1985; Uno et al., 1989;Abend et al., 1982; Atkenson and Hollerbach, 1985).These studies suggest different optimisation modelsbased on kinematic, dynamic or neural terms. Anoverview of these optimisation models and techniquescan be found in Wolpert et al. (1995) and Nakano et al.(1999).

The empirical minimum jerk approach is the sim-plest to implement in a real-time system and such amodel was first proposed for a single joint by Hogan(1984) and further developed for multi-joint move-ments by Flash and Hogan (1985). The later concludesthat humans by nature tend to minimise the jerk pa-rameter over the duration of the reaching movement ofthe arm. Jerk is the rate of the change of accelerationwith respect to time, (third time derivative of the posi-tion). Minimum jerk theory states that any movementwill have maximum smoothness when the magnitudeof the J parameter given by Eq. (1) is minimised overthe duration of the movement.

J = d

0| d3x/dt3 |2 dt (1)

42 Loureiro et al.

Figure 4. Setup exercise allows the therapist to define the exercise path. Here a fork exercise is shown (further explained in Section 3.1.4).

Figure 5. Users view of a reaching exercise.

Upper Limb Robot Mediated Stroke Therapy 43

Figure 6. (A) Empty room, (B) real room, and (C) detail room.

Figure 7. Spring and damper combinationBead pathway.

3.1. Theoretical Models

The models presented in this paper are based on theuse of polynomials to control position, velocity andacceleration parameters encountered in a human basedmovement profile. The use of polynomials has com-putational advantages for use in real-time applications,particularly in rehabilitation. Using this methodology,control of human trajectory is enhanced by the flexi-bility of being able to redefine polynomials or super-imposing a new trajectory over the previous one inreal-time.

Figure 8. Representation of a movement trajectory in the virtual environment.

Figure 9. Point-to-point movement definition.

3.1.1. Minimum Jerk Point-to-Point Model. Thefirst model is based on the assumption that every move-ment happens in between a start point and an endpoint, from which a straight-line path is generated(Fig. 9).

In this case it is advantageous to have accelerationsthat are zero at the start and end of the movement. Forthis a parameter is chosen such that:

1 1

44 Loureiro et al.

This parameter ( ) in a later stage can be scaledto the movement exact time. Symmetrical movementshave a mid range position, velocity and accelerationoccurring when = 0 which in turn simplifies the cal-culation of the polynomial coefficients at a later stage.To ensure that the acceleration at the start and end iszero, a polynomial with odd power is used. In the liter-ature (Flash and Hogan, 1985) a 5th order polynomialhas already been used. However, in order to allow fornon-symmetric, non-minimum jerk polynomials to co-exist with symmetric minimum jerk polynomials, a 7thorder polynomial is used:

p = a + b + c 2 + d 3 + e 4 + f 5 + g 6 + h 7(2)

The derivatives with respect to the parameter ( ) aredenoted as the more familiar p, p, p. The followingidentities are constraints applied to the start and end ofthe movement:

Start and end positions are defined:

p|=1 = pstart p|=1 = pend

Start and end velocities and accelerations are zero:

p|=1 = 0 p|=1 = 0p|=1 = 0 p|=1 = 0

Hence, the polynomial becomes:

p = a + b + d 3 + f 5 + h 7 (3)

We can then identify the coefficients of the polynomialin Eq. (3) as:

a = (pstart + pend)2

(4)b = p|=0 = vmid (5)d = 35

16p 3b (6)

f = 3b 218

p (7)

h = 1516

p b (8)

Where

p = pend pstart (9)

Mid velocity (vmid) needs to be determined in orderto minimise the integral given by Eq. (10) and achievea minimum jerk movement.

J = 1

1|p|2d (10)

Thus, to achieve minimum jerk, the mid velocityshould be:

b = 1516

p (11)

At which point h = 0 and the polynomial becomes5th order. The minimum jerk model and polynomialspresented in this section were used to implement thetherapy modes explained in Section 3.2 and to generateminimum jerk paths for the Bead-Pathway explainedin Section 2. Polynomial coefficients are calculated be-tween end-effectors current position (P1 in Fig. 9) andtarget position (P2 in Fig. 9).

3.1.2. Up and Over Model. Often rehabilitation ex-ercises involve the recovery of function for actionsinvolving lifting and transporting an object from one lo-cation to another. Such curvilinear non-minimum jerkmovement patterns require an even order polynomial(Eq. (12)) for the vertical axis of the movement:

p = a + c 2 + e 4 + g 6 (12)

In order to elevate the trajectory by the amount of Hand go back to the start point, these additional assump-tions are made:

Start and end positions are the same:

pstart = pend

Movement is symmetric:

a = p|=0 = pstart + H

Hence the polynomial coefficients become:

g = H (13)c = 3H (14)e = 3H (15)

Upper Limb Robot Mediated Stroke Therapy 45

Figure 10. Super-positioning model. Left: superimposed positions, and Right: superimposed velocities.

3.1.3. Super-Positioning Model. Certain move-ments, however, are not straight-line movements.These can be movements such as placing cubes on topof each other or placing a book on a shelf.

For this, a technique comparable to the one usedby Flash and Henis (1991) whereby movements thatare not straight-line or non-symmetrical by nature areachieved can be used. It consists of superimposing twodistinct trajectories to create a third one.

If two different polynomials are given:

p = a + b + d 3 + f 5 + h 7 (16)q = a + c 2 + e 4 + g 6 (17)

A different trajectory can be created by addingthe polynomials in Eqs. (16) and (17) vectorially(Fig. 10):

h = p + q (18)

The super-positioning model allows for smooth tra-jectories to be achieved when the minimum jerk param-eters for Eqs. (16) and (17) are minimised separately.In this case the parameter ( ) for the input polyno-mials can be mapped to time differently, therefore thesecond trajectory (Eq. (17)) does not need to begin at = 1.

3.1.4. Fork Model. The Fork model was created withthe intention of augmenting the existing therapy mod-els by allowing the user to have the freedom to decidewhich target to choose. Comparing this model to thepoint-to-point model, the only difference is that move-ments are not generated sequentially (i.e., from P1 toP2, P2 to P3, etc.) but instead the user is able to decide

Figure 11. Fork model. (A) Target selection and (B) vector dotproducts.

if it is more appropriate to move from P1 to P2 or fromP1 to P4 (Fig. 11(A)).

From a clinical point of view, apart from providingthe stroke patient with repetitive challenge therapies,the ability to choose can be motivational and be oftherapeutic benefit.

The fork model uses the force readings provided bythe force sensor mounted on the end-effector of thehaptic interface to identify the target that the patientwants to select. Once the user attempts to move in thedirection of his/her chosen target, the vector dot prod-ucts are used to detect the direction and the amount offorce exerted by the user in the direction of the targetpoint (Fig. 11(B)).

Vector dot product of the users force vector, ontovectors V1 and V2, can be used to detect which targetis selected by the user. We know that:

V1 = p2 p1 (19)V2 = p3 p1 (20)

46 Loureiro et al.

If the user is exerting a force F on the haptic inter-face, then 1 and 2 can be estimated as:

cos 1 =F V1| F || V1|

(21)

cos 2 =F V2| F || V2|

(22)

It is also known that, if both vectors are on the sameside of the plane normal to the force applied, then cos1 0, and 0 1 . Thus the algorithm fordetecting the target becomes:

Step 1. Calculate V1 and V2 using Eqs. (19) and (20)Step 2. Calculate cos 1 0 and cos 2 0 using

Eqs. (21) and (22)Step 3. IF cos 1 0 OR cos 2 0 AND F FActivation

(a defined threshold value) then:

Step 3.1. IF cos 1 cos 2 then assume 2 1which means target P2 is selectedELSE IF cos 2 cos 1 then assume 1 2which means target P3 is selected

Step 3.2. Initiate minimum jerk trajectory to appro-priate target.

In order to determine the intended target from morethan two choices, the algorithm becomes to find themaximum positive value for cos ( i ) where i is the in-dex number assigned to targets from the fork junction.

3.1.5. Time Mapping Model. The parameter ( ) pro-vides a useful scaling mechanism for the polynomialimplementation. This parameter can be scaled to time

Figure 12. Time mapping model. Left: time mapped positions and Right: time mapped velocities.

using a linear or quadratic scaling, which in turn al-lows for different movements to be attained dependingon the scaling factor. In this context, Eq. (23) can beused to scale time (t) to ( ) linearly with respect to tStart= 0 and tEnd = duration.

= 1 2(t tend)tstart tend (23)

Different time mappings are possible; for example aquadratic mapping can be used:

= at2 + bt + c (24)

Where:

0 t duration (25)

Hence the coefficients for this quadratic become:

c = 1 (26)a =

(2d2

bd

)(27)

Figure 12 shows a comparison in between linear andnon-linear mapping (b = 1) of parameter ( ) for bothpositions and velocities.

3.2. Different Therapy Modes

Using the minimum jerk polynomials three differenttherapy modes are implemented on the GENTLE/ssystem:

3.2.1. Patient Passive Mode. The Patient Passivemode was the first therapy mode implemented. As

Upper Limb Robot Mediated Stroke Therapy 47

the patient lacks the power to initiate the movementand remains passive, the haptic interface will move thearm along the pre-defined path. When the patients armreaches the target, depending on the exercise selected,the movement can be reversed back to the start positionor continued towards the next defined position.

3.2.2. Patient Active Assisted Mode. The secondmode is Patient Active-Assisted mode. In this mode,the haptic interface starts moving as soon as the pa-tient initiates a movement in the direction of the path-way. The haptic interface initiates the movement whenFUser U > FActivation, where U is the position vectorbetween the start point and the end point. After the ini-tiation is made, the haptic interface helps the user toreach to the end point.

3.2.3. Patient Active Mode. The third mode is Bead-Pathway (ratchet) mode or Active mode. The velocityprofile for this mode was set to zero to provide un-limited time for the patient to finish the correct task.This mode provides a unidirectional movement, wherethe amount of deviation can be controlled by chang-ing spring-damper coefficients. Similar to the previousmode, the user initiates the right movement. The hap-tic interface stays passive until the user deviates fromthe predefined path. In this case, the spring-dampercombination encourages the patient to return to thepathway. This operation can end by reaching the endpoint or releasing the operation button. Upon arrivalat the end point, it is up to the user to continue thesame movement back to the start point, a new pointor end the whole session in this mode. To implementthis mode, a ratchet or energy function was calculatedso that the user can only move towards the movementgoal. The ratchet function relies on the actual posi-tion of the robot pa and the position of the bead pto calculate an energy. Thus at a particular setting ofthe parameter t the energy would be E(t) = (p(t) =pa)2.

When user moves along the path, if the movement in-volves less energy, then that position is accepted at thecurrent position of the bead on the pathway, otherwise,the virtual spring damper will resist the users move-ment to higher energy states. This means, if t = t1,then for t2 > t1, E(t1) and E(t2) are calculated, ifE(t2) < E(t1) then t is adjusted to be the new value t2.This algorithm has the effect of only allowing one-waymovement along the pathway.

4. Clinical Trials

A pilot study was carried out in the spring of 2001 and aprinciple study conducted from autumn of 2001 to au-tumn of 2002. The choice of sites in both the UK andIreland gives the study access to a greater number ofsubjects (31 in total) for inclusion in the clinical trials.In Dublin the studies were conducted at the Adelaide& Meath Hospital, a teaching hospital of Trinity Col-lege, Dublin, and in the UK at the Battle Hospital inReading.

Some of the initial results of the pilot study are pub-lished (Loureiro et al., 2001) with more detailed resultsof the principal clinical trial to be published (Cooteet al., 2003; Amirabdollahian et al., 2003). Results fromthe initial pilot studies showed that the majority of thesubjects were enthusiastic about the use of visual andhaptic cues. The trial suggested that the system as awhole was able to motivate people with hemiplegia asa result of stroke and encourage them to participate andexercise more. The pilot study was used to evaluate thelevel of forces that should be imposed by the virtualsprings and dampers and the subjects difficulties in in-teracting with the forces. It also assessed the level ofinteraction with the three virtual rooms and the conse-quent motivation. An encouraging data was that 7 ofthe 8 patients in the pilot stopped exercising becauseof fatigue rather than boredom, confirming the designobjective of providing motivating therapies.

The principle study was conducted in the Adelaide& Meath Hospital (AMH) center (19 subjects) and inBattle Hospital (BH) center (11 subjects). In both cen-tres subjects were divided into two randomised groups,ABC (AMH = 10, BH = 6) and ACB (AMH = 9, BH =5) and each group involved in three phases, each phasehaving duration of 9 trial sessions in three weeks. Un-der phase one the subject was assessed using validatedoutcome measures in order to identify the underlyingbaseline. At the start of the next two phases, the sub-jects paretic limb was first assessed using selected out-come measures. The B phase was the Robot MediatedTherapy (RMT) customised for each individual. Dur-ing phase C the subjects paretic limb was suspendedusing sling suspension techniques. The results in bothcentres have shown a positive development towards theuse of RMT as a useful intervention late after stroke inthe assistance of UL rehabilitation. The data from theprincipal study is in accordance with findings of theMIT MANUS studies (Krebs et al., 1999). The salientconclusion to arise from this study is that this mode of

48 Loureiro et al.

therapy appears to be particularly appropriate for moreseverely disabled patients. Further studies are neededto establish the level, nature, onset and determinationof treatments that will result in the best recovery for anindividual patient.

5. Conclusion

This paper introduces a new approach to machinemediated neuro-rehabilitation, based on integrating ap-propriate haptic technologies to high quality virtual en-vironments, so as to deliver challenging and meaning-ful therapies to people with upper limb impairment inconsequence of a stroke.

A mathematical framework is given to computenatural paths for machine-assisted movements basedon polynomials that can be constrained to provide socalled minimum jerk movements that have been shownto characterise typical reaching movements. Oncethe polynomial pathway is established these can betraversed in a number of ways from patient passivewhere the haptic interface shapes the movement, toproviding an errorless constraint such that, the hapticdevice only corrects deviations from the ideal path.This framework is ideal for providing the challengingand motivating therapies to people with upper limbhemiplegia following a stroke, as it can be adaptedto provide a variety of levels of patient participationincluding the requirement for the patient to preplan anddecide which of two or more movements he/she shouldcomplete.

Two identical prototypes have undergone extendedclinical trials in the UK and Ireland with a cohort of31 stroke subjects. The clinical outcomes for these stud-ies are in the process of publication and are summarisedin this paper.

A second prototype is under development to improvethe current prototype at hardware and software levels.

Although 3D visual environments are relatively easyto implement the current generation requires wearingof glasses or a head-up display and the therapist in-put considered this too distracting for people who mayalso have visual and cognitive problems associated withtheir stroke. It is clear also that high quality therapydevices of this nature have a role in future deliveryof stroke rehabilitation, and machine mediated thera-pies should be available to patient and his/her clinicalteam from initial hospital admission, through to longterm placement in the patients home following hospitaldischarge.

Acknowledgments

The work presented in this paper has been car-ried out with financial support from the Commis-sion of the European Union, Framework 5, spe-cific RTD programme Quality of Life and Man-agement of Living Resources, QLK6-1999-02282,GENTLE/SRobotic assistance in neuro and motorrehabilitation. It does not necessarily reflect its viewsand in no way anticipates the Commissions futurepolicy in this area.

We are grateful to all our colleagues in theGENTLE/s consortium (University of Reading,UK; Rehab Robotics, UK; Zenon, Greece; Virgo,Greece; University of Stafordshire, UK; University ofLjubljana, Slovenia; Trinity College Dublin, Ireland;TNO-TPD, Netherlands; University of Newcastle, UK)for their ongoing commitment to this work.

Special thanks to Mr. Paul Hawkins for providingthe drawing used in Fig. 1.

GENTLE/s website: http://www.gentle.rdg.ac.uk

Notes

1. EASREuropean Age Standardised Rate.2. 55 years old.3. (2002). StrokeFacts and figures statistics, UK National

Stroke Association; web resource: http://www.stroke.org.uk/noticeboard/facts.htm (Accessed on 28-08-02).

4. Mortality rate: 109 per 100,000, 1999.

References

Abend, W., Bizzi, E., and Morasso, P. 1982. Human arm trajectoryformation. Brain, 105:331348.

Abrams, W. and Berkow, R. (Ed.). 1997. Merk Manual of Geriatrics,Merk Research Laboratories, Whithouse Station, NJ.

Amirabdollahian, F., Gradwell, E., Loureiro, R., Collin, C., andHarwin, W. 2003. Effects of the gentle/s robot mediated therapyon the outcome of upper limb rehabilitation post-stroke: Analysisof the battle hospital data. In Proc. 8th Int. Conf. Rehab. Robotics,(ICORR 2003), Published by HWRS-ERC Human-friendlyWelfare Robot System Engineering Center, KAIST, Republic ofKorea, pp. 5558, April 2325 (ISBN 89-88366-09-3 93560).

Ashburn, A. 1993. Physiotherapy in the rehabilitation of stroke: Areview. Clin Rehabil, 7:337345.

Atkenson, C.G. and Hollerbach, J.M. 1985. Kinematic features ofunrestrained vertical arm movements. Journal of Neuroscience,5:23182330.

Bernstein, N. 1967. The Coordination and Regualtion of Movements,Pergamon, London.

Bizzi, E., Accornero, N., Chapple, W., and Hogan, N. 1984. Posturecontrol and trajectory formation during arm movement. Journalof Neuroscience, 4:27382744.

Upper Limb Robot Mediated Stroke Therapy 49

Bobath, B. 1978. Adult Hemiplegia: Evaluation and Treatment,London, Heinemann.

Broeks, J.G., Lankhorst, G.J., Rumping, K., and Prevo, A.J. 1999.The long-term outcome of arm function after stroke: Results of afollow up study. Disabil & Rehabil, 21(8):357364.

Brown, P., Jones, D., Singh, S., and Rosen, J. 1993. The exoskeletonglove for control of paralysed hands. In Proc. IEEE Int. Conf. onRobotics and Automation, Atlanta, GA, pp. 642647.

Butefisch, C., Hummelsheim, H., Denzler, P., and Mauritz, K.H.1995. Repetitive training of isolated movements improves the out-come of motor rehabilitation in the centrally paretic hand. JournalNeurological Sciences, 13:5968.

Carr, J.H. and Shepherd, R.B. 1987. A Motor Relearning Programmefor Stroke, 2nd ed., Oxford, Butterworth Heinemann.

Coote, S. and Stokes, E. 2001. Physiotherapy for upper extremitydysfunction following stroke. Physical Therapy Reviews, 6:6369(W.S. Maney & Son Ltd).

Coote, S., Stokes, E., Murphy, B., and Harwin, W. 2003. The ef-fect of GENTLE/s robot mediated therapy on upper extremitydysfunction post stroke. In Proc. 8th Int. Conf. Rehab. Robotics,(ICORR 2003), Published by HWRS-ERC Human-friendlyWelfare Robot System Engineering Research Center, KAIST,Republic of Korea, pp. 5961, April 2325 (ISBN 89-88366-09-393560).

Cotton, E. and Kinsman, R. 1983. Conductive Education for AdultHemiplegia, Edinburgh, Churchill Livingstone.

Dijkers, M.P., deBear, P.C., Erlandson, R.F., Kristy, K., Geer, D.M.,and Nichols, A. 1991. Patient and staff acceptance of robot tech-nology in occupational therapy: A pilot study. Journal of Rehabil-itation Research and Development, 28(2):3344.

Ernst, E. 1990. A review of stroke rehabilitation and physiotherapy.Stroke, 21:10811085.

Feys, H.M., de Weert, W.J., Selz, B.E., Cox Steck, G.A., Spichiger,R., Vereeck, L.E., Putman, K.D., and van Hoydonk, G.A. 1998.Effect of a therapeutic intervention for the hemiplegic upper limbin the acute phase after stroke: A single-blind, randomised, con-trolled multicentre trial. Stroke, 29(4):785792.

Flash, T. and Henis, E. 1991. Arm trajectory modification duringreaching toward visual targets. Journal of Cognitive Neuroscience,3:220230.

Flash, T. and Hogan, N. 1985. The coordination of arm movements:An experimentally confirmed mathematical model. Journal ofNeuroscience, 5(7):16881703.

Georgopoulos, A.P., Kalaska, J.F., and Massey, J.T. 1981. Spatialtrajectories and reaction times of aimed movements: Effects ofpractice, uncertainity and change in target location. Journal ofNeurophysiology, 46:725743.

Gibson, J.J. 1966. The Senses Considered as Perceptual Systems,Houghton Mifflin Company.

Harwin, W., Loureiro, R., Amirabdollahian, F., Taylor, M., Johnson,G., Stokes, E., Coote, S., Topping, M., Collin, C., Tamparis, S.,Kontoulis, J., Munih, M., Hawkins, P., and Driessen, B. (2001).The GENTLES/S project: A new method of delivering neuro-rehabilitation. In Proc. Association for the Advancement of Assi-stive Technology in Europe (AAATE 2001), Assistive Technology-Added Value to the Quality of Life, Assistive Technology ResearchSeries, Crt Marincek, C. Buhler, H. Knops, and R. Andrich (Eds.),IOS Press, 10:3641.

Hesse, S., Bertelt, C., Jahnke, M.T., Schaffrin, A., Baake, P., Malezic,M., and Mauritz, K.H. 1995. Treadmill training with partial body

weight support compared with physiotherapy in non-ambulatoryhemiparetic patients. Arch Phys Med Rehabil, 75(10):10871093.

Hogan, N. 1984. An organizing principle for a class of voluntarymovements. Journal of Neuroscience, 4:27452754.

Hogan, N., Krebs, H.I., Sharon, A., and Charnnarong, J.1995. Interactive Robotic Therapist, U.S. Patent #5,466,213,MIT.

Isard, P.A. and Forbes, J.F. 1992. The cost of stroke to the National-Health-Service in Scotland. Cerebrovascular Diseases, 2:4750.

Johnson, M.J., Van der Loos, H.F.M., Burgar, C.G., and Leifer, L.J.1999. Drivers SEAT: Simulation environment for arm therapy. InProc. 6th Int. Conf. Rehab. Robotics, ICORR99, Stanford, CA,USA, pp. 227234.

Johnson, M.J., Van der Loos, H.F.M., Burgar, C.G., Shor, P., andLeifer, L.J. 2001. Designing a robotic stroke therapy deviceto motivate use of the impaired limb. In Proc. 7th Int. Conf.Rehab. Robotics, ICORR 2001, Integration of Assistive Technol-ogy in the Information Age, Evry, France, Vol. 9, M. Mokhtari(Ed.), IOS Press, pp. 123132.

Johnson, M.J., Van der Loos, H.F.M., Burgar, C.G., Shor, P., andLeifer, L.J. 2002. Design of drivers SEAT: A car steering sim-ulation environment for upper limb stroke therapy. In Robotica,to appear.

Knott, M. and Voss, D.E. 1968. Proprioceptive Neuromuscular Stim-ulation, New York, Harper & Row.

Krebs, H.I., Hogan, N., Volpe, B.T., Aisen, M.L., Edelstein, L., andDiels, C. 1999. Robot-aided neuro-rehabilitation in stroke: Three-year follow-up. In Proc. 6th Int. Conf. Rehab. Robotics, ICORR99,Stanford, CA, USA.

Kwakkel, G., Wagenaar, R.C., Koelman, T.W., Lankhorst, G.J., andKoetsier, J.C. 1997. Effects of intensity of rehabilitation afterstroke, a research synthesis. Stroke, 28:15501556.

Lincoln, N.B. and Parry, R.H. 1999. Randomised, controlled trialto evaluate increased intensity of physiotherapy treatment of armfunction after stroke. Stroke, 30(10):22422243.

Loureiro, R., Amirabdollahian, F., Coote, S., Stokes, E., and Harwin,W. 2001. Using haptics technology to deliver motivational ther-apies in stroke patients: Concepts and initial pilot studies. InProc. 1st European Conference on Haptics, EuroHaptics 2001,Educational Technology Research Paper Series, University ofBirmingham, UK, pp. 16, ISSN 14639394.

Lum, P.S., Burgar, C.G., Kenney, D.E., and Van der Loos, H.F.M.1999. Quantification of force abnormalities during passive andactive-assisted upper-limb reaching movements in post-strokehemiparesis. IEEE Transactions on Biomedical Engineering,46(6):652662.

Nakano, E., Imamizu, H., Osu, R., Uno, Y., Gomi, H., Yoshioka, T.,and Kawato, M. 1999. Quantitative examinations of internal rep-resentations for arm trajectory planning: Minimum commandedtorque change model. Journal of Neurophysiology, 81(5):21402155.

ONS 2000. Mortality Statistics: Cause, Office for National Statistics,276.

Parker, V.M., Wade, D.T., and Langton-Hewer, R. 1986. Loss of armfunction after stroke: Measurement, frequency, and recovery. IntRehabil Med, 8:6973.

Petersen, S., Rayner, M., and Press, V. 2000. Coronary Heart Dis-ease Statistics, 2000 ed., The British Heart Foundation StatisticsDatabase.

50 Loureiro et al.

Popescu, V.G., Burdea, G.C., Bouzit, M., and Hentz, V.R. 2000.A virtual-reality-based telerehabilitation system with force feed-back. IEEE Trans. on Information Technology in Biomedicine,4(1):4551.

Rao, R., Agrawal, S.K., and Scholz, J.P. 1999. A robot test bed forassistance and assessment in physical therapy. In Proc. 6th Int.Conf. Rehab. Robotics, ICORR99, Stanford, CA, USA, pp. 187200.

Reinkensmeyer, D.J., Pang, C.T., Nessler, J.A., and Painter, C.C.2001. Java therapy: Web-based robotic rehabilitation. In Proc.7th Int. Conf. Rehab. Robotics, ICORR 2001, Integration of As-sistive Technology in the Information Age, Evry, France, Vol. 9,M. Mokhtari (Ed.), IOS Press, pp. 6671.

Rood, M. 1954. Neurophysiological reactions as a basis for physicaltherapy. Phys Ther Rev., 34:444449.

Sackley, C.M. and Lincoln, N.B. 1996. Physiotherapy for strokepatients: A survey of current practice. Physiotherapy Theory &Practice, 12:8796.

Shor, P.C., Lum, P.S., Burgar, C.G., Van der Loos, H.F.M.,Majmundar, M., and Yap, R. 2001. The effect of robotic-aidedtherapy on upper extremity joint passive range of motion pain. InProc. 7th Int. Conf. Rehab. Robotics, ICORR 2001, Integration ofAssistive Technology in the Information Age, Evry, France, Vol. 9,M. Mokhtari (Ed.), IOS Press, pp. 7983.

Stewart, J.A., Dundas, R., Howard, R.S., Rudd, A.G., and Wolfe,C.D.A. 1999. Ethnic differences in incidence of stroke: Prospec-tive study with stroke register. British Medical Journal (BMJ),318:967971.

Sunderland, A., Tinson, D.J., Fletcher, D., Langton-Hewer, R., andWade, D.T. 1992. Enhanced physical therapy improves arm func-tion after stroke, a randomised controlled trial. J. Neurol NeurosurgPsychiatry, 55:530535.

Uno, Y., Kawato, M., and Suzui, R. 1989. Formation and control ofoptimal control trajectories in human multijoint arm movements:Minimum torque change model. Biological Cybernetics, 61:89101.

Westcott, P. 2000. StrokeQuestions and Answers, The StrokeAssociation, Stroke House, Whitecross Street, London.

Wolpert, D.M. et al. 1995. Are arm trajectories planned in kinematicor dynamic coordinatesAn adaptation study. Exp Brain Res,103(3):460470.

Rui Loureiro is a research officer working in the field of Med-ical Robotics in the department of Cybernetics at Reading Uni-versity, a Member of the Institution of Electrical Engineers (IEE)and a student Member of the Institution of the Electrical and Elec-tronics Engineers (IEEE). He is also pursuing a Ph.D. in Neuro-Haptics Rehabilitation Systems on a part-time basis. He received the

B.Eng. (HONS) degree in Electronic & Computer Engineering fromBolton Institute, Manchester in 1998 and a M.Sc. degree in AdvancedRobotics from Salford University, Manchester in 2000. From 1997to 1999 he worked in industry as a technical consultant in the fieldof Electronic Process Control. His research interests are in Neuro-Rehabilitation Robotics, Haptic Interfaces, Novel-Actuator Systemsand Real Time Simulation & Processing. He was involved in theGENTLE/s project since its start and is currently working on a Euro-pean project looking at Upper Limb Dynamically Responsive TremorSuppression.

Farshid Amirabdollahian is a research student and a research offi-cer in the department of Cybernetics, Reading University. He is alsoa student member of Institute of Electrical and Electronics Engineers(IEEE). He received his B.Sc. in Mathematics majoring ComputerScience in Azad University, Tehran, Iran in 1995. He is currentlyfinishing his Ph.D. in Upper Limb Stroke Rehabilitation Robotics.From 1995 to 1997 he worked as a software engineer, until 1999he worked as senior Engineer and database administrator for sev-eral companies. His current research interests include human armtrajectories for neuro-rehabilitation, arm movement quantification,control strategies and real-time control issues in man-machine in-teraction. He has been involved in the GENTLE/s project since itsstart.

Michael Topping received a BA, Cert Ed. From Keele Uni-versity in 1988 and is currently a professor at StaffordshireUniversity; Research Development Manager at the North StaffsHealth Authority, Stoke-on-Trent, and a Clinical Scientist at Re-hab Robotics, Birmingham. Professor Topping has extensive experi-ence with assistive technologies for people with motor impairments,where in 19881992 designed and developed the Handy 1 RoboticAid to Eating and Drinking when Research Manager at Keele Uni-versity. He his responsible for the assessment, placing and after careof over 120 Handy 1 eating and drinking systems in the UK todate. Project Leader for several university projects now underway todetermine the benefits gained from using eating and drinking aids toacquire independence at mealtimes. Participant and leader of severalpast and current European Union projects.

Upper Limb Robot Mediated Stroke Therapy 51

Ir. Bart J.F. Driessen is with TNO TPD in the Netherlands. He stud-ied electrical engineering in Delft from 19841990. His graduationsubject was the design of a control architecture for the MANUSwheelchair mounted manipulator. At TNO he is senior project man-ager with a focus on intelligent autonomous systems. Main applica-tion areas are i) service robotics (MANUS manipulator, GENTLE/ssystem), ii) transportation systems (autonomous container transport,obstacle detection), and iii) agricultural systems (automatic mush-room harvesting, automatic flower picking). Current research area ison visual servoing combined with force control.

William Harwin received a BA in engineering from CambridgeUniversity in 1982, and MSc in Bioengineering from Strathclyde

University in 1983. Following a period as a research assistant atthe Engineering Department of Cambridge University he began aPh.D. on computer recognition of the head gestures made by peo-ple with cerebral palsy, work he completed in 1991. His researchinterest is in machines and the human system. In 1983 he workedat the Cambridge University Engineering Department on rehabili-tation robotics research. Between 1990 and 1995 he directed reha-bilitation robotics research at the Applied Science and EngineeringLaboratories in Delaware, USA. He moved to the Department ofCybernetics at the University of Reading, England in 1996 wherehe directs work at the Human-Robot Interface Laboratory (tHRILwww.cyber.rdg.ac.uk/W.Harwin). Current research work includesrobot assistance of human movements, information content of hap-tic interfaces and lumped parameter modeling of human interactionsinvolving power exchange.