A Small-Scale Robotic Manipulandumfor Motor Training in Stroke Rats

Bogdan Vigaru∗‡, Olivier Lambercy∗‡, Lina Graber∗, René Fluit∗, Pascal Wespe∗,Maximilian Schubring-Giese†‡, Andreas Luft†‡ and Roger Gassert∗‡

∗Rehabilitation Engineering Lab, ETH Zurich, Switzerland†Department of Neurology, University of Zurich, Switzerland

‡Rehabilitation Initiative and Technology Platform Zurich (RITZ)Emails: {vigarub, olambercy, gassertr}@ethz.ch, {andreas.luft, maximilian.schubring-giese}@usz.ch

Abstract—The investigation and characterization of sensori-motor learning and execution represents a key objective for thedesign of optimal rehabilitation therapies following stroke. Bysupplying new tools to investigate the learning process and objec-tively assess recovery, robot assisted techniques have opened newlines of research in neurorehabilitation aiming to complementcurrent clinical strategies. Human studies, however, are limited bythe complex logistics, heterogeneous patient populations and largedropout rates. Rat models may provide a substitute to explorethe mechanisms underlying these processes in humans withlarger and more homogeneous populations. This paper describesthe development and evaluation of a three-degrees-of-freedomrobotic manipulandum to train and assess precision forelimbmovement in rats before and after stroke. The mechanical designis presented based on the requirements defined by the interactionwith rat kinematics and kinetics. The characterization of therobot exhibits a compact, stiff and low friction device suitablefor motor training studies with rodents. The manipulandumwas integrated with an existing training environment for rodentexperiments and a first study is currently underway.

I. INTRODUCTION

Motor rehabilitation following stroke requires neural adap-tations that also occur during skill learning. These adaptationscomprise motor refinement, skill acquisition and decision mak-ing [16]. During motor learning, the central nervous system(CNS) builds a representation of the external dynamics withwhich we interact, commonly referred to as internal models[8], [19]. Gaining a detailed understanding of how thesemechanisms operate and how they are impaired after damageto the CNS is important to optimally support the recoveryof sensorimotor function following brain injury, as well asto design novel therapies to accelerate and further promoterehabilitation.

In recent years, robotic interfaces have provided novelinsights into motor learning, offering scientific evidence for theexistence of internal models. A typical experimental paradigmwidely used to study motor learning involves having subjectshold the output handle of a planar robotic manipulandumand perform reaching movements while the robotic interfacerenders virtual dynamics that perturb the movement [15].Robot-assisted rehabilitation represents a promising approachto complement current clinical strategies for rehabilitationafter injury to the CNS, and supplies new tools to investigateand objectively assess sensorimotor recovery [7]. Patient-

Fig. 1. Rat grasping the end effector of the robotic manipulandum. Thesystem has three active degrees of freedom, allowing controlled interactionduring planar reaching and pronosupination tasks.

tailored haptic training to restore reaching skills in patientswith poststroke hemiparesis demonstrated a great potential forrehabilitation tools that augment error to facilitate functionalrecovery [12]. Nevertheless, the mechanisms underlying robot-driven neural adaptations are not yet fully understood andthere is a lack of evidence regarding the most efficient us-age of robotic systems for sensorimotor recovery. Moreover,performing clinical investigations with stroke patients usingrobotic devices, such as the study reported by Lo et al. [9],is an enormous challenge due to large dropouts and highlyheterogeneous patient populations.

Rat models may provide a substitute to explore the mech-anisms underlying sensorimotor control and recovery afterstroke in humans. The similarity of fundamental brain regionsto those of the human model, the possibility to study large,homogeneous populations, and the fact that sensorimotorlearning and performance can already be assessed beforethe stroke, make rats the ideal model to investigate thesemechanisms. The ability to induce stroke in the rodent modelcould give many novel insights into stroke recovery and neuralplasticity. Further, investigating the behavioral deficits andtherapeutic treatments in animal models of stroke is essentialfor potential translational applications [13].

2011 IEEE International Conference on Rehabilitation Robotics Rehab Week Zurich, ETH Zurich Science City, Switzerland, June 29 - July 1, 2011

978-1-4244-9861-1/11/$26.00 ©2011 IEEE 198

Mechanisms of motor learning and re-learning (after stroke)can be investigated by training precision movements in rats,such as accurate reaching for food pellets. Schubring-Gieseand coworkers used this method to investigate the influence ofprior knowledge on re-learning of a precision reaching skillafter a cortical lesion in a rat [14]. However, the assessment ofmotor performance was limited because of the variability ofthe outcome measure, i.e. the number of successful reaches.Current experimental methodologies suffer from limited taskcomplexity and a lack of objective performance metrics of therat forelimb, such as kinematics and kinetics, thus restraininginsights into the motor learning process [18].

Using a robotic manipulandum that the rat is trained toactively manipulate in a specific way in order to obtain afood reward would not only increase the intricacy of theassignment, but also give more control over how this task isperformed. Similar to robotic devices used in human motorlearning studies, force fields could be implemented to perturbmotion of the rat forearm in a well-controlled and repeatablemanner, while providing a quantitative assessment of motorperformance and adaptation. Further, this system could beintegrated with current rat training environments, includingresponse buttons, pellet dispensers, automated doors, etc.,allowing to automate the training procedure. The robot canbe used to study skill learning as well as transient motoradaptations.

Several robotic systems have been developed for lowerand upper limb movement tasks with rodents. Nessler et al.designed a robotic device (the ”rat stepper”) to train and assesslocomotor function of spinal cord injured rodents [10]. Thedevice consists of a pair of lightweight, robotic arms thatattach to the rodent hindlimbs, a body weight support anda motorized treadmill. Devices that can be manipulated by therat forearm have also been reported. Francis et al. proposed aone degree-of-freedom (DOF) manipulandum that could beactively grasped and pulled or pushed to a specific targetposition by a water deprived rat against programmed forcefield perturbations, in an attempt to understand the feedforwardand feedback mechanisms of motor control [6]. This device islimited to the learning of skills of restrained complexity.

This paper presents the design and development of anactuated 3-DOF small-scale planar robotic manipulandum totrain and measure precision forelimb movements in rats beforeand after brain injury. The system has two translational DOFand one rotational DOF, hence allowing in-plane movementand pronation/supination of the rat forelimb. The latter DOFis particularly important for the acquisition of motor skills,in which movement sequences and movement complexity arerequired, and is also of high functional relevance in graspingand manipulation of objects. The design and characteristicdata of the device are reported, and the potential of this noveltechnology is discussed.

II. DESIGN AND IMPLEMENTATION

A. Requirements for the Robotic Manipulandum

To allow interaction of the robotic device with the forepawof a rat, the workspace of the device and the forepaw shouldoverlap as much as possible. No studies investigating the rangeof motion of rat forepaws were found, but a few studiesanalyzed the inverse dynamics of rat locomotion, which reportjoint angles, joint moment and joint power [1], [4], [17]. Usingthese, the range of motion of a rat forepaw can be estimatedfrom the averaged joint angle ranges, and thus approximationscan be made for the required workspace and the resultingdimensions of the robotic manipulandum.

Besides the range of motion, the maximum force which therat is able to exert at the end-effector of the manipulandumneeds to be estimated in order to select the appropriateactuators. By combining the maximum joint moments withthe segment lengths, a maximum possible force at the paw of2.33 N for a 300 g rat was estimated, a value that is well abovethe one reported by Fowler et al. (0.65 N) using a pressuresensor [5].

No scientific literature was found describing the maximumpronation or supination moment of the rat forelimb. Thesevalues were estimated by downscaling the torques reported ina study involving 24 healthy human male subjects [11] usingthe law of similitude based on the proportionality betweenweight and moment. A mean maximum value of 61 mNm and49 mNm for forepaw supination and pronation torques in therat can be estimated, respectively.

The requirements for the robotic device are summarized inTable I.

TABLE IRAT MANIPULANDUM REQUIREMENTS

force at the end-effector > 2.4 N [5]torque for pronation/supination > 61 mNm [11]force resolution < 3.2 mN [5]position error < 1.1 mm [10]inertia at the end-effector < 8 g [6]controller update rate > 100 Hz [6]workspace > 20x40 mm [1], [4], [17]

B. Mechanical Design

Both serial and parallel mechanisms were investigated forthe design of the robotic manipulandum. Despite the simplerforward kinematics and large workspace with respect to theirvolume, serial robots present several drawbacks such as lowstiffness, relatively low effective load, error summation fromlink to link and reduced dynamics [3]. Considering the rapidmovements of the rat forearm and the need for a transparentdevice, a parallel mechanism was selected since it can achievehigh dynamics with good precision while assuring a transpar-ent interaction.

The design of the robotic manipulandum is based onthe Pantograph, a 2-DOF, five-bar-linkage planar mechanismdeveloped at McGill University, Canada [2]. The structurewas redesigned so that it would meet the requirements for

199

training and measuring of planar movements of rats. Severalmechanical changes had to be performed in order to adapt theworkspace and downscale the output forces based to the ratkinematics. In addition, to incorporate training of pronationand supination movements, an extra DOF was added. Fig. 2presents a detailed CAD drawing of the rat manipulandumdesign.

The base of the robot is composed of four aluminum plates(lower base, upper base, and motor plate, which are connectedto a back plate), holding the system together. The movableparts of the robot consist of four aluminum rigid arm-linkages(two proximal and two distal) connected together at the end-effector, defining the planar workspace of the device. The end-effector, represented by a titanium sphere of diameter 6 mm,is attached to a telescopic brass shaft that transmits rotationfor the pronation/supination DOF. The respective actuator ismounted on the back plate and connected to the telescopicshaft through a universal joint. The two actuators driving thefour rigid links are mounted on the motor plate and connectedto the manipulandum arms through couplings.

The lower base plate is mounted on an aluminum mechan-ical support adjustable in the x and y directions, that servesthe purpose of both precisely positioning the manipulandum infront of the rat cage and ensuring its physical stability duringexperiments.

C. Robot Kinematics

The kinematic structure of the manipulandum is shown inFig. 3. The interaction with the rat takes place at the outputP6, which represents the position of the spherical end-effector,which can move in a horizontal plane and additionally rotatearound the P6P7 segment. The two actuated joints for planarmovement are located at points P1 and P5 respectively.

1) Forward Kinematics: the position of the end-effectorsphere P6(x6,y6) can be determined using the sensed angularpositions θ1 and θ2. The nominal lengths of the links ai inmm are:

anom = [50,45,45,50,26,15]T (1)

According to geometrical inspection, it follows that:

P2 =

(x2y2

)=

(a1cosθ1a1sinθ1

)(2)

P4 =

(x4y4

)=

(a4cosθ2−a5

a4sin(180◦−θ2)

)(3)

Fig. 2. Mechanical structure of the rat manipulandum: (1) DC torque motorsfor planar x-y movement; (2) high-resolution optical encoders; (3) DC torquemotor for pronosupination DOF with high-resolution optical encoder; (4) fourbase plates - three horizontal and one vertical; (5) two proximal and two distalarms interconnected with a telescopic shaft; (6) support system for alignmentand attachment to rat cage. All dimensions are in mm.

By performing a coordinate system transformation, thecoordinates of point P3 can be computed:

P3 =

(x3y3

)=

(a1cosθ1 +a2cos(θ1 + γ)a1sinθ1 +a2sin(θ1 + γ)

)(4)

where γ = 180◦− (α + β ) and α and β can be determinedfrom the triangles P1P2P4 and P2P3P4 respectively using thelaw of cosines:

α = cos−1(

a21 +D21−D22

2a1D1

)(5)

β = cos−1(

a22 +D21−a23

2a2D1

)(6)

The position P3(x3,y3) can be translated to the position ofthe sphere end-effector P6(x6,y6) using the following equa-tion:

P6 =

(x6y6

)=

((L+a6)sinφ(L+a6)cosφ

)(7)

where:

L =

√(x3 +

a52

)2+ y23 (8)

200

Fig. 3. Kinematics of the rat manipulandum.

φ = tan−1(

x3 +a52

y3

)(9)

2) Inverse Kinematics: the angles θ1 and θ2 can be calcu-lated given the position of the sphere end-effector P6(x6,y6),as follows from Fig. 3:

θ1 = α1−β1 (10)

θ2 = α2 +β2 (11)

where α1 and α2 are trivial, while β1 and β2 can be determinedfrom the triangles P1P2P3 and P3P4P5 using the law ofcosines:

α1 = sin−1(

y3L1

)(12)

α2 = sin−1(

y3L2

)(13)

β1 = cos−1(

a21 +L21−a22

2a1L1

)(14)

β2 = cos−1(

a24 +L22−a23

2a4L2

)(15)

3) Equations of Motion of the Telescopic Shaft: the angulardisplacement (rotational position) γo of the end-effector de-scribing the pronosupination movement is determined by twoangular positions, i.e. the rotational position of the input driveshaft γi and the angle of the universal joint β connecting theinput and output drive shafts:

γo = tan−1(

tanγicosβ

)(16)

The angular velocity of the output varies with the rotationalposition of the universal joint and the angle between the inputand output links. The relationship between the output (ωo) andinput (ωi) angular velocities is given by:

ωo =ωicosβ

1− sin2βcos2γi(17)

4) Singularities and Kinematic Constraints: informationabout the singular configurations of the rat manipulandum isgiven by the velocity Jacobian, which maps the joint velocitiesω onto the sphere end-effector velocities v:

v = [ẋ6, ẏ6]T = Jω = J[θ̇1, θ̇6]T (18)

The Jacobian is calculated by taking the partial derivatives ofthe forward kinematics map with respect to the actuated jointangles θ1 and θ2. The singularities of the system correspondto the zeros of the determinant of the Jacobian:

det[J] = a1a2sinγ (19)

This yields two singularities when α = β = 0◦ and α +β =180◦ respectively. If the former case is not reachable due torobot design, the latter must be prevented by implementingsoftware limitations. Another kinematic restriction that has tobe taken into account is represented by the configuration whenpoints P2 and P4 overlap, determining the ”crossing” of the leftand right arms. In order to avoid this, an additional constraintx4 < x2 is imposed at all times.

D. Actuation, Sensors, Control and Safety

The three DOF of the manipulandum are actuated usingthree motors as follows:• two brushed DC motors (Maxon Motor, Switzerland;

RE25, graphite brushes, 20 W) coupled with two highresolution rotary optical encoders (Gurley Precision In-struments, USA; R119B, 65k counts/rev) control andmonitor the movement of the end-effector in the x-yplane. They are capable of producing a force of at least2 N in any direction throughout the usable workspace.

• one brushed DC motor (Maxon Motor, Switzerland;RE-max 24, graphite brushes, 11 W) coupled with ahigh resolution rotary optical encoder (Gurley PrecisionInstruments, USA; R112, 32k counts/rev) provides therequired pronation and supination torque. The maximumcontinuous torque it can deliver is 66 mNm, which isabove the requirement defined previously.

Control of the motors is performed by means of three servodrive amplifiers (Maxon Motor, Switzerland; 4-Q-DC ServoControl LSC 30/2). Data acquisition is carried out via two

201

−150

−100

−50

0

50

100

0

20

40

60

80

100

120

0

1

2

3

x−coordinate

F(x,y)

y−coordinate

F [N

]

0.5

1

1.5

2

Fig. 4. The manipulability ellipsoid showing the maximum forces exertedat the sphere end-effector throughout the usable workspace.

data acquisition cards (National Instruments Corp., USA; PCI-6221, 68-pin). For safety, ease of debugging and transportationreasons, all the electronic components are comprised in anelectronic box.

The control program, implemented in LabVIEW 9.0 (Na-tional Instruments Corp., USA), runs on a personal computer(Dell, Optiplex 960, Intel Core(TM)2 Quad CPU 2.83 GHz,3.25 GB RAM, Windows 7 Enterprise) at a frequency of 1kHz. A Graphical User Interface (GUI) enables the user tovisualize in real time the preferred trajectories, choose betweenvarious modes of operation and select which data to record forsubsequent offline analysis.

Safety measures have been implemented in the form of bothhardware and software emergency units. Mechanical stopsserve the purpose of limiting the movement range of theproximal arms. The user can select a velocity limit for theend-effector, as well as a current limit for the actuators, thuslimiting the maximum force that can be produced. Moreover,an emergency push button is available to the operator, cuttingthe power to the entire system if pressed. A custom madeacrylic protection cover is placed over the robot to protect itduring transportation and experiments.

III. RESULTSA. General Characteristics

The rat manipulandum has a compact structure, with exter-nal dimensions of 212×160×150 mm3. Using the kinematicsdimensions and equations, the workspace of the manipulan-dum was calculated and overlapped with the rat forearmkinematics to determine an usable area of approximately40×20 mm2.

10−1

100

101

−40

−20

0

20Bode Plot

Frequency [Hz]

Mag

nitu

de [d

B]

10−1

100

101

−200

−100

0

Frequency [Hz]

Phas

e [d

eg]

Fig. 5. Bode plot of the system in closed loop PID control over a frequencyrange of 0.01 Hz to 30 Hz. The closed loop position control bandwidth isabout 15 Hz, with a structural resonance around 11 Hz.

The manipulability ellipsoid has been determined by inves-tigating the eigenvectors and eigenvalues of JJT . The largestforces can be applied in the direction where maximum velocityis the smallest. The results, depicted in Fig. 4, indicate thatthroughout the usable workspace a force of at least 2 N canbe applied at the sphere end-effector, except close to the edgeof the workspace. The values are in good accordance with theforces exerted by rats.

Due to friction and lubrication between various movingcomponents (e.g. motors, arms, couplings and ball-bearings),the device exhibits a static friction torque of 7.2 mNm, whichhas to be compensated for in the control program.

The control program capable of running at frequencies upto 1 kHz allows the user to drive the robot in different modesof operation, selected from a GUI (e.g. haptic tunnel, error-enganced guidance, etc).

B. Position BandwidthIn order to identify the dynamic behavior of the system,

a position bandwidth test was conducted. A sinusoidal os-cillation with an amplitude of 2 mm was commanded, withfrequencies ranging from 0.01 Hz to 30 Hz in steps of 0.5 Hz,for a time period of 5 sec for each frequency. Fig. 5 showsthe frequency response of the system when controlled using aPID controller (the gains were tuned to P = 0.11, I = 0.01 andD = 0.00017 according to the Ziegler-Nichols method). Theresonance frequency of the system is around 11 Hz, while theposition bandwidth is around 15 Hz.

C. System Integration for Experimental TrialsFig. 6 depicts the experimental setup and how the different

components are interconnected. The manipulandum interacts

202

Pellet Dispenser

IR S

en

sor

Operator Console

Support System

Rat Manipulandum

ComputerNI PCI-6221

Sound System

Forces

Motion

PositionInformation

TorqueCommands

RatCage

Fig. 6. System diagram (left) and setup (right). The robotic manipulandum is integrated and communicates with the previously used training environment,including an infrared response button, a pellet dispenser and sound system. An operator interface provides three buttons to trigger a success or failure(overriding the automated performance rating) or cut the power to the system over an emergency button.

directly with the rat using a conventional impedance-typehaptic architecture, measuring the position imposed by therat forelimb, based on which it displays the correspondingforces. In case of a successful trial, the rat is rewarded apellet supplied by a commercial pellet dispenser (LafayetteInstrument Comp., USA, Model 80208) integrated into thecontrol scheme. At the same time, a particular sound is played.If the trial has failed, no reward is given and a different soundis presented. After each trial, whether successful or failed,the arms of the manipulandum retract to a predefined initialposition. In order to reposition the end-effector in the startposition for the next trial, the rat has to touch an infrared(IR) sensor (response button) located at the back of thecage. Should unexpected events occur, an emergency buttonis available on the operator console to stop the system. Thisconsole further incorporates two push buttons, allowing tooverride the automated performance rating and to manuallyrate a task as being successful or failed.

IV. CONCLUSION AND FUTURE WORK

This paper presented the mechanical design and devel-opment of a three-degree-of-freedom robotic manipulandumto interact with the rat forelimb in motor learning experi-ments. The device can render virtual dynamics in a well-controlled and repeatable manner, offering the possibility ofimplementing various force fields to assist or perturb themovement in order to investigate motor learning, adaptationand recovery after stroke in animal models. The three degrees

of freedom (in-plane movement and pronosupination) will forthe first time give the possibility to study skill learning ofvarious complexities versus movement adaptation (e.g. forcefield adaptation) in rodents. Furthermore, the robotic deviceprovides objective assessments of motor performance, andallows to automate the time-consuming training period.

Preliminary training has shown that rats can be trained tograsp, pull and rotate the end-effector of the robotic system.Extensive studies are now planned with both healthy andstroke rats for a complete validation of the usefulness ofthe device. Several training stages are under development,through which the rats will be guided to accomplish variousmotor tasks of increasing complexities. We will investigatemechanisms responsible for acquisition of new motor skillsand recovery after stroke, in an attempt to gain a betterunderstanding of the mechanisms underlying sensorimotorcontrol and recovery after brain injury.

ACKNOWLEDGMENTS

The authors gratefully acknowledge the support of the Na-tional Center for Competence in Research in Neural Plasticityand Repair of the Swiss National Science Foundation.

REFERENCES

[1] S. W. Bennett, “Biomechanical assesment of locomotion in two rodentmodels,” Master’s thesis, University of Saskatchewan, 2009.

[2] G. Campion, Q. Wang, and V. Hayward, “The pantograph mk-ii: a hapticinstrument,” in Intelligent Robots and Systems, 2005. (IROS 2005). 2005IEEE/RSJ International Conference on, 2005, pp. 193 – 198.

203

[3] H. Choset, K. M. Lynch, S. Hutchinson, G. Kantor, W. Burgard, L. E.Kavraki, and S. Thrun, Principles of robot motion: theory, algorithms,and implementation. The MIT Press, Cambridge, MA, 2005.

[4] M. S. Fischer, N. Schilling, M. Schmidt, D. Haarhaus, and H. Witte,“Basic limb kinematics of small therian mammals.” J Exp Biol, vol.205, no. Pt 9, pp. 1315–1338, May 2002.

[5] S. C. Fowler, K. H. Davison, and J. A. Stanford, “Unlike haloperi-dol, clozapine slows and dampens rats’ forelimb force oscillationsand decreases force output in a press-while-licking behavioral task.”Psychopharmacology (Berl), vol. 116, no. 1, pp. 19–25, Sep 1994.

[6] J. T. Francis and J. K. Chapin, “Force field apparatus forinvestigating movement control in small animals.” IEEE Trans BiomedEng, vol. 51, no. 6, pp. 963–965, Jun 2004. [Online]. Available:http://dx.doi.org/10.1109/TBME.2004.827463

[7] V. S. Huang and J. W. Krakauer, “Robotic neurorehabilitation: acomputational motor learning perspective.” J Neuroeng Rehabil, vol. 6,p. 5, 2009. [Online]. Available: http://dx.doi.org/10.1186/1743-0003-6-5

[8] J. W. Krakauer, M. F. Ghilardi, and C. Ghez, “Independent learningof internal models for kinematic and dynamic control of reaching.”Nat Neurosci, vol. 2, no. 11, pp. 1026–1031, Nov 1999. [Online].Available: http://dx.doi.org/10.1038/14826

[9] A. C. Lo, P. D. Guarino, L. G. Richards, J. K. Haselkorn, G. F.Wittenberg, D. G. Federman, R. J. Ringer, T. H. Wagner, H. I.Krebs, B. T. Volpe, C. T. Bever, D. M. Bravata, P. W. Duncan,B. H. Corn, A. D. Maffucci, S. E. Nadeau, S. S. Conroy, J. M.Powell, G. D. Huang, and P. Peduzzi, “Robot-assisted therapy forlong-term upper-limb impairment after stroke.” N Engl J Med,vol. 362, no. 19, pp. 1772–1783, May 2010. [Online]. Available:http://dx.doi.org/10.1056/NEJMoa0911341

[10] J. A. Nessler, W. Timoszyk, M. Merlo, J. L. Emken, K. Minakata, R. R.Roy, R. D. de Leon, V. R. Edgerton, and D. J. Reinkensmeyer, “A roboticdevice for studying rodent locomotion after spinal cord injury.” IEEETrans Neural Syst Rehabil Eng, vol. 13, no. 4, pp. 497–506, Dec 2005.[Online]. Available: http://dx.doi.org/10.1109/TNSRE.2005.858432

[11] L. W. O’Sullivan and T. J. Gallwey, “Upper-limb surface electro-myography at maximum supination and pronation torques: the effectof elbow and forearm angle.” J Electromyogr Kinesiol, vol. 12, no. 4,pp. 275–285, Aug 2002.

[12] J. L. Patton, M. Kovic, and F. A. Mussa-Ivaldi, “Custom-designed haptictraining for restoring reaching ability to individuals with poststrokehemiparesis.” J Rehabil Res Dev, vol. 43, no. 5, pp. 643–656, 2006.

[13] K. L. Schaar, M. M. Brenneman, and S. I. Savitz, “Functionalassessments in the rodent stroke model.” Exp Transl StrokeMed, vol. 2, no. 1, p. 13, 2010. [Online]. Available: http://dx.doi.org/10.1186/2040-7378-2-13

[14] M. Schubring-Giese, K. Molina-Luna, B. Hertler, M. M. Buitrago,D. F. Hanley, and A. R. Luft, “Speed of motor re-learningafter experimental stroke depends on prior skill.” Exp Brain Res,vol. 181, no. 2, pp. 359–365, Aug 2007. [Online]. Available:http://dx.doi.org/10.1007/s00221-007-0930-3

[15] R. Shadmehr and F. A. Mussa-Ivaldi, “Adaptive representation ofdynamics during learning of a motor task.” J Neurosci, vol. 14, no.5 Pt 2, pp. 3208–3224, May 1994.

[16] R. Shadmehr and S. P. Wise, The computational neurobiology ofreaching and pointing: a foundation for motor learning. MIT Press,Cambridge, MA, 2005.

[17] A. K. Thota, S. C. Watson, E. Knapp, B. Thompson, and R. Jung,“Neuromechanical control of locomotion in the rat.” J Neurotrauma,vol. 22, no. 4, pp. 442–465, Apr 2005. [Online]. Available:http://dx.doi.org/10.1089/neu.2005.22.442

[18] I. Q. Whishaw, M. Alaverdashvili, and B. Kolb, “The problem ofrelating plasticity and skilled reaching after motor cortex stroke inthe rat.” Behav Brain Res, vol. 192, no. 1, pp. 124–136, Sep 2008.[Online]. Available: http://dx.doi.org/10.1016/j.bbr.2007.12.026

[19] D. M. Wolpert, Z. Ghahramani, and M. I. Jordan, “An internal model forsensorimotor integration.” Science, vol. 269, no. 5232, pp. 1880–1882,Sep 1995.

204