Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

ARMin – Exoskeleton for Arm Therapy in Stroke Patients Tobias Nef, Matjaz Mihelj, Gabriela Kiefer, Christina Perndl, Roland Müller, Robert Riener, Member, IEEE consequence is that rehabilitation, i.e. the training sessions, are often shorter than required to gain an optimal therapeutic outcome. Finally, manually-assisted training lacks repeatability and objective measures of patient performance and progress. By the implementation of robotic devices, in contrast, training sessions can be automated and duration and number of training sessions can be increased. By the use of patientcooperative controllers that support the patient only when necessary and by the use of appropriate audiovisual displays to motivate the patient, we hypothesize that the training will be more intensive than traditional manual arm therapy [1417]. Many groups have therefore developed and evaluated arm therapy robots (see [18] and [19] for review), for example the MIT-Manus [20], Assisted Rehabilitation and Measurement (ARM) Guide [21], Mirror Image Motion Enabler (MIME) [22], Bi-Manu-Track [23], GENTLE/s [24], Neurorehabilitation Robot (NeReBot) [25], REHAROB [26], and Arm Coordination Training 3-D (ACT3D) [27]. These devices are so called endeffector-based robots, meaning that the endeffector of the robot is connected to a distal part of human limb, i.e. the hand. Depending on the available degrees of freedom of the robot, the human hand can be positioned and oriented in space. In contrast, the structure of exoskeleton robots resembles the human arm anatomy. The rotation axes of the robot must correspond to the rotation axes of the human skeleton and, therefore, the arm can be connected to the exoskeleton at several points. As the arm segments are individually actuated, single-joint torques can be applied. Rosen et al. [28] is using an exoskeleton robot with seven degrees of freedom (DoF) (shoulder: 3DoF, elbow: 1DoF, lower arm: 1DoF and wrist: 2DoF) allowing to control the most important joints of the human arm. With this design, the shoulder complex is reduced to a spherical joint, composed of 3 individual axes, intersecting at the center of the glenohumeral joint. The exoskeleton robot developed by Craig et al. [29] uses five motors to actuate the shoulder complex in order to get a more ergonomic shoulder movement. While many clinical studies (see [19] for review) have been conducted with endeffector-based robots with limited possibilities to control position and orientation of the human arm in the three dimensional space, not much clinical evidence has been reported from work with actuated armexoskeleton robots.

Abstract—Task-oriented repetitive movement can improve movement performance in patients with neurological lesions. The application of robotics can serve to assist, enhance, evaluate and document rehabilitation of movements. ARMin is a robot for arm therapy applicable to the arm training in clinics. It has an exoskeleton structure and is equipped with position and force sensors. Our latest version ARMin II has six degrees of freedom. The mechanical structure, the actuators, and the sensors of the robot are optimized for patientcooperative control strategies based on impedance and admittance architectures. The device can work in three therapy modes: passive mobilization, game therapy, and task-oriented training. This paper presents the technical components of the new version ARMin II, the therapy modes, the control strategy for a new example of a game therapy, and clinical results of a pilot study with 11 chronic stroke patients and of single case studies conducted with three chronic stroke patients.

I. INTRODUCTION

R

ECENT studies estimate that stroke affects more than 700’000 people in the U.S. each year [1]. Of these individuals, 60-70% will be alive one year after the insult, resulting in a prevalence of 3 million stroke patients [2]. The major symptom of stroke is acute sensory hemiparesis [3]. With the available rehabilitation techniques, motor recovery of the lower extremities is usually faster and more pronounced than motor recovery of the upper extremities [4]. Regarding stroke patients with severe upper extremity paresis, only 18% regain full arm function [5]. However, arm function is essential to cope with tasks of daily living. Several studies prove that sensorimotor arm therapy has positive effects on the rehabilitation progress of stroke patients (see [6] for review). The goal of this therapy is to induce long-term brain plasticity and to improve functional outcomes. Critical factors are that the therapy is intensive [7,8], of long duration [9,10,11], highly repetitive [12], and task-oriented [13]. Regarding these criteria, one-to-one manually-assisted training has several limitations. The training is laborintensive and therefore expensive. The disadvantageous Manuscript received February 4, 2007. This work was supported in part by the Swiss Research Foundation NCCR on “Neural Plasticity and Repair” and by the Gottfried and Julia Bangerter-Rhyner Foundation. Tobias Nef, Matjaz Mihelj, Gabriela Kiefer, Christina Perndl and Robert Riener (nef/[email protected]) are with the Sensory-Motor Systems Laboratory of the ETH and University Zurich, Switzerland and with Balgrist University Hospital, Zurich, Switzerland. Roland Müller is with the Spinal Cord Injury Center of the Balgrist University Hospital, Zurich, Switzerland.

1-4244-1320-6/07/$25.00 (c)2007 IEEE

68

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

Axis 2 and axis 4 are driven by a dc motor with a Harmonic Drive gear unit (Harmonic Drive Inc., Japan) with a reduction of 1:100. This motor/gear combination is characterized by low backlash (transmission accuracy a<0.04°) and good back-drivability (no-load back driving torque T= 2.8 Nm). Axis 3 is also driven by a dc motor with a Harmonic Drive gear unit with a reduction of 1:30. The output shaft of the gear box is connected to a tooth belt that is fixed to the two ends of a curved rail (Fig. 2).

Sanches et al. [30] performed a controlled clinical trial with a passive exoskeleton device that allowed to measure arm position and to compensate for gravity forces. Another passive exoskeleton is under development by Stienen et al [31]. II. METHODS A. Specifications of ARMin To perform task-oriented [13] training, the robot must be able to move the shoulder, the elbow, as well as some distal joints. The range of motion (RoM) of the arm must correspond to the RoM of the human arm. For good performance of model-based patient-cooperative control strategies [32], the robot must have low inertia, low friction, and negligible backlash. Furthermore, the motor/gear combination needs to be back-drivable. The required velocities and accelerations have been determined by measuring the movements of a healthy subject during two tasks (eating soup and manipulation of a coffee cup) and are summarized in Table 1. Safety of the patient and the therapist must be guaranteed at any time.

Fig. 2. Axis 3: Internal/external shoulder rotation. The dc motor (M2) drives the Harmonic Drive gear unit with the output shaft connected to a toothed wheel driving the tooth belt.

Axis 5 for pronation and supination of the lower arm is realized by a cable-driven half cylinder rotary module. Wrist flexion/extension (Axis 6) is implemented with a dc motor connected to a ball spindle. Axis 1 is actuated by a linear module with a ball spindle incorporated. This ball spindle is driven by a dc motor and moves the point A (Fig. 1) in vertical direction. The linear displacement results in a rotation of the upper arm around the y-axis and in a vertical rotation of the upper arm around the point S. Considering S as immobile in space would simplify the kinematic design, because this assumption would reduce the shoulder complex to a spherical joint composed of 3 individual axes intersecting at the center of the glenohumeral joint [28]. The human shoulder complex has at least 5 DoF, and the vertical position of the glenohumeral joint depends on the elevation angle of the upper arm [33]. Therefore, the height of S should also change with the elevation angle in order to get an ergonomic shoulder movement and to avoid misalignment between the robotic rotation axis and the human rotation axis. A linkage system that connects the slide of the linear drive (axis 1) with a linear bearing allows changing the height of point S depending on the arm elevation angle. More details about this mechanism is described in [34]. For a rotation of 80°, the vertical movement of S is around 80 mm. The actual position of each axis is measured with a digital

Fig. 1. Mechanical structure of ARMin. Axis 1: Vertical shoulder rotation, Axis 2: Horizontal shoulder rotation, Axis 3: Internal/external shoulder rotation, Axis 4: Elbow flexion/extension, Axis 5: Pro/supination of the lower arm, Axis 6: Hand flexion/extension.

B. Kinematics, Actuation and Sensors An exoskeleton structure was selected for the mechanical structure of the robot (Fig. 1). The patient is sitting in a wheelchair, while his arm is placed inside an orthotic shell.

TABLE I REQUIREMENTS FOR ARMIN Axis Axis 1: Vertical shoulder rotation Axis 2: Horizontal shoulder rotation Axis 3: Internal/external shoulder rotation Axis 4: Elbow flexion/extension Axis 5: Pro/supination of the lower arm Axis 6: Wrist flexion/extension

1-4244-1320-6/07/$25.00 (c)2007 IEEE

ROM -60°…60° -45°…135° -50°…95° 0°…135° -90°…90° -50°…79°

69

Velocity 71°/s 60°/s 150°/s 91°/s 62°/s 55°/s

Acceleration 2

103°/s 129°/s2 245°/s2 116°/s2 58°/s2 43°/s2

Torque 20Nm 20Nm 10Nm 20Nm 8Nm 5Nm

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

actively weight and friction- compensated, so that the therapist just feels the resistance of the human arm and not the one of the robot. By feeling the resistance of the human arm the therapist can select an appropriate RoM. Once recorded, the trajectories are smoothed and then repeated. During the repeat sequence, the robot is position-controlled. In order to help the patient to know the position and orientation of his arm, an avatar doing the same movement is presented on the graphical display (Fig. 4).

encoder as well as a potentiometer-based sensor. The encoder allows high resolution (0.0009°) position sensing and the additional potentiometer provides initial absolute position information and serves as redundant sensor to detect malfunction of a position sensor. A six degree of freedom force/torque sensor (JR3 inc., USA) measures the interaction force between the human and the robot (Fig. 1). In addition, motor current measurement is used for torque estimation. The overall setup is shown in Figure 3. A computer monitor is used for visual feedback and two loudspeakers are used for acoustic feedback.

Fig. 4: Graphical scenario for three different therapy modes: left: passive movement therapy, middle: ball game therapy and right: training of task-oriented movements.

The purpose of the game therapy is to motivate the patient with simple games presented on the audiovisual display. One example of a game is the ball game presented in Figure 4. This game scenario includes a ball rolling down an inclined table and a hand connected to a handle. The ball is reflected by the walls and the patient’s task is to catch the ball with the handle. Whenever the patient is able to catch the ball, he doesn’t get any support by the robot. Only in case when he cannot catch the ball he gets support from the robot [35]. The color of the handle changes depending on the patient’s performance. Another example for a game therapy is the labyrinth game, with the control strategy described in the next section. In the task-oriented training mode, the patient is asked to grasp an object by approaching it with the virtual hand and to put it onto a table (fig. 4). The control strategy for these kind of tasks is based on the minimal intervention principle, which allows an efficient exploitation of task space redundancies and results in user driven movement trajectories and is described in [36,37]

Fig. 3: ARMin overall setup. A healthy person is sitting in the wheelchair, having the right arm connected to ARMin and looking onto a monitor showing therapy scenarios.

A counterweight of 14.5 kg, positioned behind the robot, is connected via a nylon cord with the linear slider of axis 1 (point A in fig. 1) and the friction of axis 3 (static part: 2.8 Nm) prevents the robot from collapsing in case of power interruption and the therapist or the patient can still move the robot with some additional force as the motor/gear units are back-drivable.

D. Control Strategy for the Labyrinth Game A labyrinth with the dimension n × m is displayed to the patient (Fig. 5). The actual position of the robot in joint angle coordinate system is given by:

G (1) q = ( q1 ,q2 ,q3 ,q4 ,q5 ,q6 )T with qk being the actual joint angle of axis k . A red sphere with the radius r on the display indicates the actual hand G position h derived from the actual position of axis i and axis j :

C. Three Therapy Modes ARMin allows three different therapy modes. The goal of the passive movement therapy is to prevent joint degeneration and to preserve joint mobility. In this mode, the robot moves the patient’s arm while the patient remains passive. As the RoM of the patient’s arm depends on the pathology, and differs from patient to patient, a teach-andrepeat protocol has been implemented. Thus, the therapist can first move the human arm and the robot while the position data is recorded. During the teaching, the robot is

1-4244-1320-6/07/$25.00 (c)2007 IEEE

q −q ⎛ ⎞ d x ⋅ i i min ⋅ n ⎟ qi max − qi min G ⎛ x⎞ ⎜ ⎟ ( 1 ≤ i ≤ 6 , 1 ≤ j ≤ 6 ,i ≠ j ) h =⎜ ⎟=⎜ ⎝ y ⎠ ⎜ d ⋅ q j − q j min ⋅ m ⎟ ⎜⎜ y ⎟⎟ q j max − q j min ⎝ ⎠

70

(2)

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

All axis combinations are possible, but i = 2 or i = 4 and j = 1 are the most intuitive ones. The range of motion can be adjusted to the patient’s needs by selecting appropriate maximum and minimum values. The factors d x and d y are

control (Fig. 6). The reference force has been selected as a linear function depending on the overlap of the sphere with the wall and for non-boundary positions ( 1 < x < n,1 < y < m ) the force is given by:

either equal to 1 or -1 and define the direction of the movement on the screen. The four axes that are not involved are position controlled and kept in a static position (Fig. 6).

G Fref

⎛ ⎧ a( x0 ,y0 +1) ⋅ ( r − x + x0 ) ⋅ k x / r ⎜ ⎪⎪ ⎜ Fx = ⎨ 0 ⎜ ⎪ a ⎜ ⎪⎩ ( x0 + 2 ,y0 +1) ⋅ ( − x − r + x0 + 1) ⋅ k x / r =⎜ ⎧ a( x0 +1,y0 ) ⋅ ( r − y + y0 ) ⋅ k y / r ⎜ ⎪⎪ ⎜ 0 ⎜ Fy = ⎨ ⎪ a ⎜ ⋅ − y − r + y0 + 1) ⋅ k y / r x0 +1,y0+ 2 ) ( ( ⎪ ⎩ ⎝

⎞ if x < r + x0 ⎟ if ( x ≥ r + x0 ) and ( x ≤ x0 + 1 − r ) ⎟ ⎟ if x > x0 + 1 − r ⎟ ⎟ ⎟ if y < r + y0 ⎟ if ( y ≥ r + y0 ) and ( y ≤ y0 + 1 − r ) ⎟ ⎟ if y > y0 + 1 − r ⎠

(4) with x 0 = floor(x) and

and there selection depends on the patient’s performance. A patient with poor ability to control the position of his hand will need a lot of guidance and, therefore, high factors k x and k y . A healthy person would be able to pass the

The labyrinth topology is randomly calculated in such way that there is always a path (path width = 1) between the start

G

labyrinth without touching the walls and k x and

point hstart and the endpoint hend . The matrix M , given by ⎛ a1,1 a1,2 " a1,n ⎞ ⎧1 if field(i, j) occupied ⎜ ⎟ M=⎜ # # ⎟ with a i, j = ⎨ ⎩ 0 if field(i, j) free ⎜a ⎟ ⎝ m,1 a m,2 " a m,n ⎠

(5)

Boundary positions need to be treated separately. The factors k x and k y are positive numbers that scale the force

Fig. 5: Labyrinth game scenario. The red circle indicates the actual position of the human arm. The walls are indicated with white cubes and the yellow path is the trajectory where the patient can pass without feeling resistance. The patient can collect points by catching the yellow/black spheres.

G

y0 = floor(y)

k y could

therefore be set to zero. The game is over when the patient either reaches the end position or when he enters into an occupied field. Therefore, if (6) a x ,y = 1

(3)

0

0

a nasty sound is produced to indicate that the game is over and the game starts again. If the patient reaches the end point, then he will be awarded by a nice sound, a new labyrinth is generated, and the game starts again.

reflects the topology of the labyrinth. Axes i and j are first position controlled and guide the patient’s arm to the start

G

point hstart . Once there, the robot control is changed to force

Fig. 6: Labyrinth control schema with friction and gravity compensation, position control for the fixed axes, position or force control (depending on the state machine) for the active axes 1 and 2.

1-4244-1320-6/07/$25.00 (c)2007 IEEE

71

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

E. Clinical Evaluation Most of the clinical evaluation has been performed with the earlier version of ARMin [16]. This simpler version has only axis 1-4 implemented and the shoulder complex is simplified to a three DoF ball and socket joint, what means that there was no link between axis 1 and point S (Fig. 1). Actuators, sensors and software are similar. A pilot study was carried out to validate the patient comfort and handling, i.e. whether the robot is adjustable to different patient sizes, whether the cuffs are comfortable for the patients, and whether the patients are able to perform the movement tasks. Furthermore, the patient acceptance was interrogated. This pilot study included 11 patients with a cumulative training duration of more then 76 hours. A possible benefit on the motor performance was not assessed in the pilot study, but in the subsequent single case studies with three chronic stroke patients (14-40 months post-stroke). Thereby, after a baseline phase, the three patients trained their impaired upper extremity with ARMin three or five times a week during eight weeks. A single training session consisted of 20 minutes of mobilization and 40 minutes of ball game training. The cumulative training duration was 88 hours. Active range of motion, isometric muscle strength and the score in the ball game were measured with ARMin and the Fugl-Meyer score [38] was assessed by an experienced therapist.

0.9 0.8

Normalized Interacting Force

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

1

2

3

4

5 Time [s]

6

7

8

9

10

Fig. 8: Interacting force between the robot and the patient. Red: a “bad” patient with high interacting force (performance measure = 74.92). Blue: a “good” patient with low interaction forces (performance measure = 13.74).

B. Clinical Evaluation The pilot study showed that the teach-and-repeat method enables the therapist to select an appropriated movement for the mobilization of the patient. Figure 9 shows typical trajectories for passive mobilization. It was possible to adjust the robot to patients with body sizes varying from 155cm to 192cm. But taller subjects are possible (maximum estimated: 210cm). It took 5 minutes to position a patient in the robot and additional 10 minutes to change the robot’s kinematics from the left arm to the right arm and vice-versa.

III. RESULTS A. Control Strategies The force field of the labyrinth with the topology presented in Figure 5 is shown in Figure 7. By exerting enough force, the user can pass trough a wall and can enter into an occupied field. This case is detected and treated as a “game over” with immediate restart. A good patient would pass the labyrinth without touching the walls (Fig. 8) while a patient with poor control touches the walls several times. The work that the force field exercises onto the patient is calculated by integrating the interaction force over time and serves as performance measure to evaluate the patient. In addition to that, whenever the patient touches the wall, a nasty sound is produced.

The patient acceptance was assessed on a questionnaire and the patient rated the robot with 5.5 point out of a scale from 1 to 6. All patients wanted to continue the robot-supported therapy. The three patients participating in the single case study, where highly motivated for the ARMin therapy and not a single therapy session was canceled.

100 80 60

position [°]

40 20 0 -20 -40 -60 -80

4

6

8

10 time [s]

12

14

16

18

Fig. 9: Teach- and repeat trajectories for passive mobilization with a chronic stroke patient. The thin lines show the recorded trajectories and the thick line represents the smoothed trajectories that are repeated by the robot. Axis 1: red, Axis 2: green, Axis 3: blue, Axis 4: black

Fig. 7: Force field for the labyrinth training scenario. The Z coordinate shows the absolute value of the interacting Force that would act if the patient would pass the x,y position. The radius of the ball r has been selected to be equal to 0.3.

1-4244-1320-6/07/$25.00 (c)2007 IEEE

2

The investigation of the effects showed that intensive training with this device can induce improvements in

72

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands

coordination of arm movements, isometric muscle strength, active range of motion, and functional tasks. However, the single patients did not improve in all parameters, and the ones in which they improved differed. Thus, the effects of ARMin training seem to be quite individual. The improvements in the Fugl-Meyer Assessment (17 (+3), 33 (+3), 21 (+4.2)) are in a similar range as in other studies with robotic devices [20,22]. These improvements in FuglMeyer score are remarkable in chronic stroke subjects. Nevertheless, we observed that the carry-over into activities of daily living (Barthel index) was rather small. This means that the subjects regained movement abilities but did not change their way how to cope with activities of daily living. Therefore, future work will focus on the training of activities of daily living.

REFERENCES [1]

[2]

[3] [4] [5]

IV. CONCLUSION AND OUTLOOK

[6]

The ARMin robot was built with four active DoF in the first prototype and has then been extended with two additional DoF for the forearm in order to allow training of taskoriented movements and an additional coupled-DoF to accommodate the vertical movement of the center of rotation of the shoulder joint. The modular design of the ARMin robot that allows various combinations of proximal and distal arm training modes will also serve as platform for the search of the best arm rehabilitation practice. The functionality has been proven during the pilot study, and the results of the three single case studies suggest that this sort of therapy has positive effects on the movement performance of chronic stroke patients. For the future, a controlled clinical study with a larger amount of patients is in preparation.

[7] [8] [9]

[10] [11] [12]

[13]

ACKNOWLEDGMENT We thank Dr. Gery Colombo from Hocoma AG, Volketswil, Switzerland for his contribution to this work. We also thank the occupational therapist and Prof. Dr. med. Volker Dietz of the Balgrist University Hospital, Zürich, as well as Stéphane Kühne for his contribution to this work.

[14] [15] [16] [17]

[18] [19]

[20] [21]

1-4244-1320-6/07/$25.00 (c)2007 IEEE

73

R. Wayne et al. “Heart Disease and Stroke Statistics – 2007 Update. A Report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee.” Circulation published by the American Hearth Association 2006. 7272 Greenville Avenue, Dallas, Texas. M. Kelly-Hayes, J.T. Robertson, J.P. Broderick, P.W. Duncan, L.A. Hershey, E.J. Roth, W.H. Thies, C.A. Trombly, “The American Heart Association Stroke Outcome Classification: executive summary. Circulation 97(24), pp. 2474-2478, 1998. H.C. Maeurer, H.C. Diener, “Der Schlaganfall”, Georg 1996. H.S. Jørgensen, H. Nakayama, H.O. Raaschou, T.S. Olsen, „Recovery of walking function in stroke patients: the Copenhagen Stroke Study.” Arch Phys Med Rehabil 1995, 76(1):27-32. H. Nakayama, H.S. Jørgensen, H.O. Raaschou, T.S. Olsen, ”Recovery of upper extremity function in stroke patients: the Copenhagen Stroke Study.” Arch Phys Med Rehabil 1994, 75(4):394-8. T. Platz, “Evidenzbasierte Armrehabilitation: Eine systematische Literaturübersicht.“ Nervenarzt 2003, 74:841-849. G. Nelles “Cortical reorganization- effects of intensive therapy.“ Archives of Physical Medicine and Rehabiliation 2004, 22:239-244. G. Kwakkel, R.C. Wagenaar, T.W. Koelman, G.J. Lankhorst, J.C. Koetsier, “Effects of intensity of rehabilitation after stroke. A research synthesis.“ Stroke 1997:28, 1550-1556. A. Sunderland, D.J. Tinson, E.L. Bradley, D. Fletcher, H.R. Langton, and D.T. Wade, ”Enhanced physical therapy improves recovery of arm function after stroke. A randomised clinical trial.“ J Neurol Neursurg Psychiatry 1992, 55:530-535. G. Kwakkel, R.C. Wagenaar, J.W.E. Twisk, G.J. Langkhorst, and J.C. Koetsier. „Intersity of leg and arm training after primary middlecelebral artery stroke: a randomised trial.” Lancet, 1999, 35:191-196. G. Kwakkel, B.J. Kollen, and R.C. Wagenaar, „Long therm effects of upper and lower limb training after stroke: a randomised trial.” J. Neurol. Neurosurg. Psychiat., 2002, 72:473-479. C. Butefisch, H. Hummelsheim, P. Denzler, and K.H. Mauritz, „Repetitive training of isolated movements improves the outcome of motor rehabilitation of the centrally paretic hand.” J Neurol Sci 1995, 130:59-68. N.A. Bayona, J. Bitensky, K. Salter, and R. Teasell, “The role of taskspecific training in rehabilitation therapies.” Top Stroke Rehabil 2005, 12:58-65. T. Nef, M. Mihelj, G. Colombo, R. Riener, “ARMin – Robot for Rehabilitation of the Upper Extremities.” IEEE ICRA International Conference on Robotics and Automation 2006, 3152-3157. T. Nef, R. Riener, “ARMin- design of a novel arm rehabilitation robot.” IEEE 9th international conference on rehabilitation robotics, ICORR 2005, 57-60. T. Nef, G. Colombo, R. Riener, “ARMin - Robot for Movement Therapy of the Upper Extremities.” Automatisierungstechnik at 2005, 53(12):597-606. R. Riener, L. Lünenburger, S. Jezernik, M. Anderschitz, G. Colombo, V. Dietz “Cooperative subject-centered strategies for robot-aided treadmill training: first experimental results.” IEEE Transactions on Neural Systems and Rehabilitation Engineering 13:380-393. R. Riener, T. Nef, G. Colombo,”Robot-aided Neurorehabilitation for the Upper Extremities.” Medical & Biological Engineering & Computing 2005b, 43:2-10. G.B. Prange, M.J. Jannink, C.G. Groothuis-Oudshoorn, H.J. Hermens, M.J. IJzerman, “Systematic review of the effect of robot-aided therapy on recovery of the hemiparetic arm after stroke.”, J Rehabil Res Dev, 2006 Mar- Apr;43(2):171-84. S.E. Fasoli, H.I. Krebs, M. Ferraro, N. Hogan, B.T. Volpe, “Does shorter rehabilitation limit potential recovery poststroke?” Neurorehabil Neural Repair 2004;18(2):88-94. D.J. Reinkensmeyer, L.E. Kahn, M. Averbuch, A. McKenna-Cole, B.D. Schmit, W.Z. Rymer, „Understanding and treating arm movement impairment after chronic brain injury: progress with the ARM guide.” J Rehabil Res Dev. 2000; 37(6):653–62.

Proceedings of the 2007 IEEE 10th International Conference on Rehabilitation Robotics, June 12-15, Noordwijk, The Netherlands [22] P.S Lum, C.G. Burgar, P.C. Shor, ”Evidence for improved muscle activation patterns after retraining of reaching movements with the MIME robotic system in subjects with poststroke hemiparesis.” IEEE Trans Neural Syst Rehabil Eng. 2004;12(2):186–94. [23] S. Hesse, C. Werner, M. Pohl, S. Rueckriem, J. Mehrholz, M.L. Lingnau. “Computerized arm training improves the motor control of the severely affected arm after stroke: a singe-blinded randomized trial in two centers.” Stroke. 2005;36(9):1960–66. [24] S. Coote, E. Stokes, B. Murphy, W. Harwin, ”The effect of GENTLE/s robot-mediated therapy on upper extremity dysfunction post stroke.” In: Proceedings of the 8th International Conference on Rehabilitation Robotics; 2003 Apr 23–25; Daejeon, Korea. Daejeon (Korea): Human-Friendly Welfare Robot System Engineering Research Center; 2003. p. 59–61. [25] C. Fanin, P. Gallina, A. Rossi, U. Zanatta, S. Masiero, “NeRe- Bot: a wire-based robot for neurorehabilitation.” In: Proceedings of the 8th International Conference on Rehabilitation Robotics; 2003 Apr 23–25; Daejeon, Korea. Daejeon (Korea): Human-Friendly Welfare Robot System Engineering Research Center; 2003. p. 23–26. [26] G. Arz, A. Toth, G. Fazekas, D. Bratanov, N. Zlatov, “Threedimensional anti-spastic physiotherapy with the industrial robots of “REHAROB.””, In: Proceedings of the 8th International Conference on Rehabilitation Robotics; 2003 Apr 23–25; Daejeon, Korea. Daejeon (Korea): Human-Friendly Welfare Robot System Engineering Research Center; 2003. p. 215–18. [27] J. Dewald, M.D. Ellis, B.G. Holubar, T. Sukal, A.M. Acosta.”The robot application in the rehabilitation of stroke patients” [Abstract]. Neurol Rehabil. 2004;(4):S7 [28] J. C. Perry, J. Rosen, “Design of a 7 Degree-of-Freedom Upper-Limb Powered Exoskeleton”, BioRob 2006, The first IEEE/RAS-EMBS International conference on Biomedical Robotics and Biomechatronics, Pisa, Tuscany, Italy, Feb 29-22, 2006. p. 805-810. [29] C. Carignan, M. Liszka, "Design of an arm exoskeleton with scapula motion for shoulder rehabilitation," Advanced Robotics, 2005. ICAR '05. Proceedings., 12th International Conference on , vol., no.pp. 524531, July 18-20, 2005 [30] R. Sanchez, J. Liu, S. Rao, P. Shah, R. Smith, T. Rahman, S.C. Cramer, J.E. Bobrow, D.J. Reinkensmeyer, “Automating Arm Movement Training Following Severe Stroke: Functional Exercises With Quantitative Feedback in a Gravity-Reduced Environment.” IEEE Transactions on neural systems and rehabilitation engineering, Vol.14, No.3, Sept 2006. [31] A.H.A. Stienen, E.E.G. Hekman, F.C.T. Van der Helm , G.B. Prange, M.J.A. Jannink, A.M.M. Aalsma, H. Van der Kooij, “Dampace: dynamic force-coordination trainer for the upper extremities” in Proc ICORR’07, Noordwijk, the Netherlands, Jun 13-15 2007 [32] R. Riener, M. Frey, M. Bernhardt, T. Nef, G. Colombo, "Humancentered rehabilitation robotics," Rehabilitation Robotics, 2005. ICORR 2005. 9th International Conference on , vol., no.pp. 319- 322, 28 June-1 July 2005. [33] J. Lenarcic, M. Stanisic, "A humanoid shoulder complex and the humeral pointing kinematics," Robotics and Automation, IEEE Transactions on , vol.19, no.3pp. 499- 506, June 2003 [34] M. Mihelj, T. Nef, R. Riener, „ARMin II – 7 DoF rehabilitation robot: mechanics and kinematics“, IEEE International Conference on Robotics and Automation (ICRA 2007), to be published [35] T. Nef, M. Mihelj, R. Riener, “ARMin – A Robot for PatientCooperative Arm Therapy“, submitted to Medical & Biological Engineering & Computing. [36] M. Mihelj, T. Nef, R. Riener, “A novel paradigm for patientcooperative control of upper limb rehabilitation robots”, Advanced Robotics 2007:21(8), to be published [37] F. Oldewurtel, M. Mihelj, T. Nef, R. Riener, „Patient-Cooperative Control Strategies for Coordinated Functional Arm Movements”, Proceedings of the European Control Conference 2007 to be held in Kos, Greece on July 2-5, 2007 [38] A.R. Fugl-Meyer, “The effect of rehabilitation in hemiplegia as reflected in the relation between motor recovery and ADL function.” Proceeedings of the International Rehabilitation Medicine Association II, Mexico City. Mexico City, IRMA, 1976, pp 683-699

1-4244-1320-6/07/$25.00 (c)2007 IEEE

74