ArticlePDF AvailableLiterature Review

Robot-assisted movement training for the stroke-impaired arm: Does it matter what the robot does?


Abstract and Figures

Robot-assisted movement training improves arm movement ability following acute and chronic stroke. Such training involves two interacting processes: the patient trying to move and the robot applying forces to the patient's arm. A fundamental principle of motor learning is that movement practice improves motor function; the role of applied robotic forces in improving motor function is still unclear. This article reviews our work addressing this question. Our pilot study using the Assisted Rehabilitation and Measurement (ARM) Guide, a linear robotic trainer, found that mechanically assisted reaching improved motor recovery similar to unassisted reaching practice. This finding is inconclusive because of the small sample size (n = 19), but suggest that future studies should carefully control the amount of voluntary movement practice delivered to justify the use of robotic forces. We are optimistic that robotic forces will ultimately show additional therapeutic benefits when coupled with movement practice. We justify this optimism here by comparing results from the ARM Guide and the Mirror Image Movement Enabler robotic trainer. This comparison suggests that requiring a patient to generate specific patterns of force before allowing movement is more effective than mechanically completing movements for the patient. We describe the engineering implementation of this "guided-force training" algorithm.
Content may be subject to copyright.
Volume 43, Number 5, Pages 619–630
August/September 2006
Journal of Rehabilitation Research & Development
Robot-assisted movement training for the stroke-impaired arm: Does it
matter what the robot does?
Leonard E. Kahn, PhD;
Peter S. Lum, PhD;
W. Zev Rymer, MD, PhD;
David J. Reinkensmeyer, PhD
Sensory Motor Performance Program, Rehabilitation Institute of Chicago, Chicago, IL;
Departments of Biomedical
Engineering and Physical Medicine and Rehabilitation, Northwestern University, Evanston, IL;
Hunter Holmes
McGuire Department of Veterans Affairs Medical Center, Richmond, VA;
Biomedical Engineering, The Catholic
University of America, Washington, DC;
Departments of Mechanical and Aerospace Engineering and Biomedical
Engineering, University of California, Irvine, CA
Abstract—Robot-assisted movement training improves arm
movement ability following acute and chronic stroke. Such
training involves two interacting processes: the patient trying to
move and the robot applying forces to the patient’s arm. A fun
damental principle of motor learning is that movement practice
improves motor function; the role of applied robotic forces in
improving motor function is still unclear. This article reviews
our work addressing this question. Our pilot study using the
Assisted Rehabilitation and Measurement (ARM) Guide, a lin
ear robotic trainer, found that mechanically assisted reaching
improved motor recovery similar to unassisted reaching prac
tice. This finding is inconclusive because of the small sample
size (n = 19), but suggest that future studies should carefully
control the amount of voluntary movement practice delivered to
justify the use of robotic forces. We are optimistic that robotic
forces will ultimately show additional therapeutic benefits
when coupled with movement practice. We justify this opti
mism here by comparing results from the ARM Guide and the
Mirror Image Movement Enabler robotic trainer. This compari
son suggests that requiring a patient to generate specific pat-
terns of force before allowing movement is more effective than
mechanically completing movements for the patient. We
describe the engineering implementation of this “guided-force
training” algorithm.
Key words: arm movement, control strategies, motor control,
motor learning, movement training, reaching, rehabilitation,
rehabilitation therapy, robotics, stroke.
Robotic technology could partially automate move-
ment training following injury to the central nervous sys-
tem (CNS). Rehabilitation therapists spend significant
time using hands-on therapy during stroke rehabilitation.
Hands-on techniques, such as active-assist exercise, are
advocated in practice guidelines and standard texts [1–3].
Robotic devices, because of their programmable force-
producing ability, can replicate some features of a thera
pist’s manual assistance, allowing patients to semiautono-
mously practice their movement training. However,
robotic devices can also implement novel forms of
mechanical manipulation impossible for therapists to
emulate because of limited speed, sensing, strength, and
repeatability of the therapist’s neuromuscular system.
Abbreviations: ARM = Assisted Rehabilitation and Measure-
ment (Guide), CNS = central nervous system, MIME = Mirror
Image Movement Enabler, MIT = Massachusetts Institute of
Address all correspondence to David J. Reinkensmeyer,
PhD, Associate Professor; Department of Mechanical and
Aerospace Engineering, 4200 Engineering Gateway
(EG3225), University of California, Irvine, CA 92697-3975;
949-824-5218; fax: 949-824-8585. Email:
DOI: 10.1682/JRRD.2005.03.0056
JRRD, Volume 43, Number 5, 2006
Novel forms of manipulation may ultimately enhance
movement recovery beyond current possibilities.
Since the 1997 pioneering study of Massachusetts
Institute of Technology (MIT)-Manus [4], the number of
research groups developing robotic therapy devices has
rapidly increased. As reviewed in this issue and elsewhere
[5–9], devices have been developed for automating train
ing for arm movement following stroke, gait and posture
following stroke and spinal cord injury, and wrist and fin
ger movement following stroke. Initial results are promis-
ing: patients who receive more therapy with a robotic
device recover more movement ability [9–10]. The bene
fits of robot-assisted therapy are comparable with or bet-
ter than that of conventional therapy [11–12].
This article, however, offers an interim, yet critical,
analysis of our early experiences with robot-assisted ther
apy. We argue that a substantial gap exists in the rationale
for widespread implementation of robot-assisted therapy
in rehabilitation clinics because a key question remains
unanswered: “Is the expense of an actuated device
needed to achieve therapeutic benefit?” Put another way,
“Could similar benefits be achieved with simpler, less
expensive, nonrobotic technology that facilitates move
ment practice?” Nonrobotic technology includes exercise
machines such as hand cycles, low-cost movement moni
toring, and virtual reality systems, and passive antigrav-
ity devices such as mobile arm supports and overhead
slings. Clearly, a mechanical device that measures move
ment for directing rehabilitation is only made more
expensive and less safe by adding robotic actuators.
Thus, this question of the benefits of robotic actuators is
practically and economically important for rehabilitation
technologists and the clinicians and patients they serve.
This question is also scientifically interesting, because
answering it requires understanding how sensory motor
activity influences CNS recovery. The answer will refine
rehabilitation therapists’ actions during conventional,
one-on-one therapy, as well as help determine the fate of
robot-assisted therapy.
We first explain why we think that this question
remains unanswered, then we review two studies from
our laboratories that provide clues to its answer. We focus
our discussion on movement training of the arm follow
ing stroke, although similar issues are likely relevant for
gait and hand training and for other CNS disorders. More
comprehensive reviews of robotic therapy have been
published elsewhere [5–9].
Background: Initial Results with Robot-Assisted
Therapy for Arm after Stroke
The first robotic system to receive extensive clinical
testing was MIT-Manus, a 2 degree-of-freedom robot
manipulator that assisted patients in tabletop arm move
ments. In a first report, 20 stroke patients received four to
five 1-hour sessions a week for up to 9 weeks with MIT-
Manus beginning on average 3 weeks after a single stroke
[13]. The device assisted planar pointing and drawing
movements with an impedance controller. A control
group received 1-hour a week of “sham” therapy in which
they used the less-impaired limb in the robot, or the robot
interacted passively with the more-affected limb.
Compared with the control group, the patients who
received robotic therapy had reduced shoulder and elbow
motor impairment according to the Motor Status Score.
The groups still statistically differed in motor impairment
at a 3-year follow-up [14]. These findings were con
firmed with larger samples of 56 and 96 patients [15].
Chronic stroke subjects who performed assistive and
resistive exercises with MIT-Manus also improved arm
movement ability [16]. This pioneering work indicates
that supplemental robotic therapy can improve recovery
in acute and chronic stroke patients.
The second key study of robot-assisted therapy for
the arm after stroke used the Mirror Image Movement
Enabler (MIME) device [6]. The MIME is a 6-degree-of-
freedom, industrial robot manipulator (PUMA 560 [Uni
mation, Inc, Connecticut, no longer in existence]) that
applies forces to the paretic limb through a customized
forearm splint. The robot moves the forearm through a
large range of positions and orientations in three-
dimensional space. A six-axis sensor measures the forces
and torques between the robot and the paretic limb. Sev
eral modes of robot-assisted movement have been imple-
mented with MIME, including passive, active-assisted,
and active-constrained, as well as a bimanual mode in
which MIME continuously moves the impaired limb to
the mirror image position of the unimpaired limb as mea
sured with a digitizing linkage.
The initial clinical testing of MIME compared the
effectiveness of robot-assisted therapy with equally
intensive conventional therapy [11]. In this study,
chronic stroke subjects received 24 1-hour sessions
over 2 months. The robot group practiced shoulder and
elbow movements assisted by MIME in all four of its
modes. The control group received conventional treatment
and 5 minutes of robot exposure each session. At the
KAHN et al. Robot-assisted arm movement training
conclusion of training, the robot group had statistically
larger improvements in the Fugl-Meyer score, a common
clinical motor impairment scale. The robot group also
had larger gains in strength and reach extent. At the 6-
month follow-up, the groups no longer differed in Fugl-
Meyer score; however, the robot group improved more in
the self-care and transfers sections of the Functional
Independence Measure. These results suggest that robot-
assisted therapy can be comparable with, or perhaps more
effective than, conventional rehabilitation therapy.
Critique of Robot-Assisted Therapy Study Designs
While the initial MIT-Manus and MIME studies are
important because they demonstrated the feasibility and
potential of robot-assisted therapy, the studies did not
address which components of the robot-assisted therapy
produced the observed therapeutic benefits. For example,
which was most beneficial—the process of the patient
trying to move or of the robot applying forces to the
patient’s arm—remains unclear.
In the initial MIT-Manus study, the patients who
received robot-assisted therapy also received more move
ment practice time compared with the control group.
Whether these patients would have improved as much if
the robot motors were turned off during their extended
practice time is uncertain.
The MIME study also did not completely control the
amount of movement practice in the robot and control
groups. The groups received a matched therapy duration,
but how much of that time the patient actually practiced
movement is unclear. The finding that robot-assisted
therapy was comparable with or better than conventional
rehabilitation therapy is encouraging. However, this find
ing could also be an indictment of conventional therapy
rather than proof of benefit of robotics; conventional
therapy may not have maximized movement practice.
Two recent studies using MIT-Manus with chronic
stroke subjects have begun to address the relative roles of
robotic forces and movement practice. One study of 46
subjects found no significant differences in movement
improvements for groups that received robotic assistive
or resistive forces [17]. A possible explanation is that the
form of robotic forces (assistive or resistive) did not mat
ter as much as the extended movement practice, i.e., as
much as the subjects trying to move. Another study with
30 subjects examined performance-based progressive
therapy, an adaptive robot control strategy in which the
robot intervened less if the patient was more capable
[9,18]. Impairment scores improved more with this strat
egy than in the previous MIT-Manus studies, especially
for moderately impaired subjects for whom the robot pre
sumably intervened less. Again, one interpretation is that
the key element for stimulating recovery relates to subject
effort; i.e., the progressive therapy mode may have opti
mized the subjects’ abilities to accomplish the task while
maximizing their efforts to activate damaged descending
pathways and thereby reorganize those pathways.
Unassisted goal-directed practice is the predominant
way for children, Olympic athletes, and stroke patients to
learn a motor skill. Hands-on assistance may be used in
limited circumstances for a limited time, for example, for
safety during dangerous tasks or to demonstrate a desired
movement. A wide body of literature has promoted the
idea that “repetitively trying to achieve a goal” is impor
tant for motor learning (for a comprehensive review, see
Schmidt and Lee [19]). In fact, repetitive goal-directed
effort is so useful that even mental rehearsal of move
ment can improve arm motor impairment following
stroke [20–22]. How robotic forces and quantitative feed
back to the user might improve motion for persons with
neurological injury beyond the benefits of repetitive
practice remains unclear.
We developed a device called the Assisted Rehabili-
tation and Measurement (ARM) Guide that measures and
applies assistive or resistive forces to linear reaching
movements across a wide workspace [23]. The ARM
Guide consists of a hand piece that is attached to a linear
track and actuated by a DC servomotor. The track can be
oriented at different yaw and pitch angles to allow reach
ing to different workspace regions. The device is stati-
cally counterbalanced so that it does not gravitationally
load the arm. We used the ARM Guide to directly test the
role of movement practice versus robotic forces in
retraining reaching following chronic stroke.
Study Design
We used the ARM Guide to help train reaching in
chronic stroke subjects (n = 10, mean time poststroke =
6.3 years) [24–25]. Subjects in the robot-trained group
performed reaching movements under their own power
with active assistance from the device. The active assis
tance algorithm required subjects to initiate movement.
JRRD, Volume 43, Number 5, 2006
The targeted normative movements were along a straight-
line path (linear track of the ARM Guide) and followed
the bell-shaped velocity profile typical of unimpaired
reaching movements. To emphasize the importance of
subjects moving under their own efforts, we incorporated
a 1 cm dead-band in the position trajectory that allowed a
subject a small margin of error along the planned path
before the motor provided assistance. Outside this dead-
band, the motor assisted the subject in maintaining the
correct trajectory with a force proportional to a weighted
sum of the position and velocity errors. To further moti
vate subjects to exert effort through the entire movement,
we encouraged them to minimize the magnitude of robot
assistance, which was provided graphically after every
five movements. Control subjects (n = 9, mean time post
stroke = 8.5 years) practiced reaching without ARM
Guide assistance. The two groups were matched by
impairment level according to the arm section of the
Chedoke-McMaster Stroke Assessment [26]. We included
subjects who scored between 2 and 5, inclusive, on this
7-point scale.
Both training groups performed equal numbers of
reaching movements (50 reaches) to identical targets, par
ticipated in sessions of equal duration (45 minutes a ses-
sion, three times a week for 8 weeks), and received
graphical performance feedback throughout each session.
However, only the robot-trained group received mechani
cal assistance to complete the desired movement.
We evaluated the subjects’ movement ability using
various outcome measures, including measures of reach
ing and functional task performance [24–25]. The most
sensitive measures were derived from a test of the sub
jects’ supported reaching ability. At the beginning of each
of the 24 training sessions, subjects in both groups
reached as far and fast as possible along the horizontal
ARM Guide track and we measured their range and max
imum speed of movement.
Summary of ARM Guide Study Results
Both groups significantly improved their maximum
speed during supported reaching gradually across train
ing sessions, but improvements did not significantly dif-
fer between the groups (Figure 1). Significant
improvements with training, but not between training
methods, were also observed for the maximum range of
supported movement. Of particular interest to both clini
cians and patients was that these gains translated into
improved performance of functional tasks (decreased
time to complete each task) on the Rancho Los Amigos
Functional Test of the Hemiparetic Upper Extremity [27].
Tasks included practical activities of daily living like zip
ping a jacket, putting a pillow in a pillowcase, and wring-
ing a rag. Limb stiffness and movement range of the hand
as it reached in free space to a target did not change after
training. The discrepant results between the measures
might be explained by the fact that reaching out from the
body in free space is a relatively demanding task that
requires substantial shoulder strength. Many functional
tasks are achievable with partial support of the arm rest
ing on a table or with the arm closer to the body.
Discussion of ARM Guide Study
In this study, subjects who received robotic assistance
exhibited improved arm movement ability similar to that
of subjects who received no assistance. This result sur
prised us: the active-assist exercise received by the robot-
Figure 1.
Comparison of supported fraction of speed (FS) for robot-trained
group and control group, which trained without robot assistance. FS is
speed during fast-as-possible reaching along Assisted Rehabilitation
and Measurement Guide by hemiparetic arm, divided by speed o
contralateral arm, with device oriented horizontally. Ensemble
averages across subjects in each group and regression lines (robot-
trained: R
= 0.79, p < 0.001; control: R
= 0.86, p < 0.001) shown for
24 training sessions over 8-week period. Adapted from Kahn LE,
Zygman ML, Rymer WZ, Reinkensmeyer DJ. Robot-assisted reaching
exercise promotes arm movement recovery in chronic hemiparetic
stroke: A randomized controlled pilot study. J Neuroengineering
Rehabil. 2006;3:12.
KAHN et al. Robot-assisted arm movement training
trained group forcefully extended the arm as each subject
exerted effort to reach forward. We thought such stretch
ing would at least improve movement ability by reducing
passive tone and spasticity, and perhaps, by providing
novel somatosensory input that stimulated neural reorga
nization. The observed gains, possibly indicative of such
plasticity, were anticipated to be greater than those
attained from repetitive reaching without assistance.
One possible explanation for the lack of a differential
benefit for active-assist exercise is that the sample size
was small and a difference went undetected. Longitudinal
power analysis is a powerful technique for calculating the
probability of such an error, given data from many mea
surements over time [28]. We applied longitudinal power
analysis techniques to the maximum supported velocity
data from each training session (Figure 1) and found that
the study had an 80 percent chance to detect a 30 percent
difference in the maximum supported velocity. Thus, any
difference between training techniques for this measure
was likely incremental rather than dramatic. Incidentally,
we believe that maximum supported velocity of ballistic
reaching is a promising measure of recovery because it is
simple, inexpensive, quantitative, functionally relevant,
and most importantly, sensitive to movement ability
across the wide range of impairment levels used in this
study. This measure does not appear to have either floor
effects, because even very impaired patients can slide
their arms along a support, or ceiling effects, because
less-impaired patients can learn to move faster.
Another possible conclusion from these results is that
the “robot” in our “robot-assisted therapy” program was
superfluous. However, this study tested only one possible
form of robotic therapy using a specific device with a
small sample of chronic stroke subjects over a limited
number of repetitions. Thus, while these results confirm
that movement practice is a primary stimulant for move
ment recovery (both groups did make movement gains)
and lead us to postulate that robotically assisting comple
tion of a movement for a chronic stroke subject does not
have a strong therapeutic benefit within the number of
repetitions tested, another pattern of robotic forces or a
greater level of exposure to robotic forces could possibly
act synergistically with the process of movement practice.
In the first therapeutic study of the MIME, robot-
assisted therapy resulted in larger gains in reach extent
than conventional therapy. Reach extent was defined as
the distance that a patient could reach unassisted toward a
target (i.e., unsupported active range of motion of reach
ing). Reach extent was also an outcome measure in the
ARM Guide study, allowing comparison [29]. The key
finding was that subjects who participated in nonrobotic
therapy (unassisted reaching in the ARM Guide study or
conventional rehabilitation therapy in the MIME study)
and subjects who received active assistance from the
ARM Guide did not improve their reach extent. Only
subjects who received movement training with the
MIME improved their reach extent (Figure 2).
Three possible causes for this difference in reach
extent are therapy intensity, kinematics of practiced
movements, and therapy modes. The MIME subjects
experienced 24, 50-minute sessions over 8 weeks, com
pared with 24, 45-minute sessions over 8 weeks for ARM
Guide subjects. Subjects performed movements at
Figure 2.
Comparison of improvement in reaching extent for treatment groups
from Mirror Image Movement Enabler (MIME) and Assisted Reha-
ilitation and Measurement (ARM) Guide studies. Only MIME group
improved reach extent. Error bars indicate standard deviation.
icant difference between groups, p < 0.05. Adapted from Kahn LE,
Reinkensmeyer DJ. Selection of robotic therapy algorithms for the
pper extremity in chronic stroke: Insights from MIME and ARM
Guide results. Proceedings of the International Conference on Reha-
ilitation Robotics; 2003 Apr 23–25; Daejeon, Korea. Daejeon
(Korea): Human-Friendly Welfare Robot System Engineering
Research Center; 2003; p. 208–10.
JRRD, Volume 43, Number 5, 2006
approximately the same frequency during these sessions.
Consequently, we believe that this dissimilarity in ther
apy intensity was too small to cause the highly significant
improvements with the MIME but not the ARM Guide.
Another possible explanation is that the movements
practiced with the MIME differed from other protocols.
The MIME accommodates fully naturalistic arm postures
during reaching because of its 6 degrees of freedom,
while the ARM Guide constrains arm movement to a lin
ear path. However, we assessed reaching movements to
obtain the reach extent outcome measure of interest here.
The devices showed a small difference in kinematics for
reaching movements, since reaching movements follow
approximately straight-line paths. Thus, different arm
postures during training seem unlikely to account for the
different reach extent outcomes.
Lastly, two modes of active therapy were included in
the MIME study but not the ARM Guide study. First,
the training of bimanual mirror movements may have
provided a unique stimulus for recovery of bilateral or
ipsilesional neuromotor pathways. Recent bimanual ther
apy studies have shown some transfer to unimanual tasks
[30–32]. However, preliminary data from a controlled
study with the MIME comparing bimanual to unimanual
therapy suggest that the end benefits achieved by each
mode are similar [33]. Second, in active-constrained
mode, a force sensor measured the direction of force gen
erated by the subject’s hand at the interface between the
hand and the robot. If the force vector had a component in
the desired direction (i.e., toward the target), then the robot
moved in that direction with a velocity proportional to
force. If the force was misdirected, however, the robot
stopped moving toward the target and a programmed
impedance allowed the robot to deflect slightly in the
direction of the force providing visual feedback of the mis
direction. This training mode forced subjects to not only
activate muscles to move the limb but also to try to acti
vate muscle groups in appropriate combinations, depend-
ing on the desired target and the limb configuration.
The active-constrained mode used by MIME is one
example of “guided-force training” that requires a patient
to generate specific patterns of force to move. We
hypothesize that guided-force training accounts for the
difference between the two studies because it helped sub
jects efficiently relearn the sensory motor transforma-
tions required for reaching. Essentially, the robot halted
the subjects’ movements when it sensed grossly incorrect
muscle activation patterns, forcing subjects to attempt to
generate a more normative pattern at each troublesome
workspace position.
Generating the correct pattern required lifting the arm
against gravity, which likely strengthened the arm. More
over, the nature of the task required significant attention
and effort, both of which predict the amount of motor
learning in a task [19,34–35]. If the subjects correctly
sequenced the learned patterns, they were rewarded with a
smooth, uninterrupted movement toward the target. The
active-assist mode used exclusively with the ARM Guide
helped the subjects move through their full passive range
of motion but did not allow the subjects to systematically
decompose and revise incorrect muscle activations. This
mode also did not penalize subjects for allowing the
device to support the arm as they moved.
Individuals with a stroke often display anisotropic or
nonuniform patterns of weakness: when asked to gener
ate isometric force in a specific direction, they are sur-
prisingly strong in some directions but dramatically weak
in others [36]. Anisotropic weakness may arise from con
strained muscle-activation pathways [36–37] or muscle
strength imbalances [38]. Either way, requiring subjects
to exercise in their weakest directions makes sense for
enhancing the strength required for function [36].
Guided-force training provides directionally targeted
strength training across the arm’s workspace. This tech
nique would be extremely challenging for a therapist
implement. Implementing guided-force training non-
robotically may be possible with mechanical guides and
ratcheting mechanisms, but robotic devices provide a
convenient and flexible way to initially assess its value.
Given the inconclusive result of our first ARM Guide
study regarding therapeutic benefits of robotic forces, we
have begun a second ARM Guide study to test whether
guided-force training differentially benefits chronic
stroke subjects, compared with unassisted reaching exer
cise [39]. This section describes the engineering design
and initial testing of the guided-force training algorithm.
Therapeutic testing of the algorithm is still under way
and will not be reported.
KAHN et al. Robot-assisted arm movement training
Guided-Force Training Algorithm
Guided-force training promotes conscious shaping of
endpoint forces in the hemiparetic arm. We make the sub
ject aware of the forces generated against the ARM Guide
by measuring those forces with a six-axis force transducer
attached to the hand piece. During a single trial, the user
attempts to reach forward at a comfortable speed to the
end of his or her range of motion. A graphical interface on
a computer monitor represents the current position of the
ARM Guide and a target position at the end of the passive
range of motion. Throughout the movement, the force
transducer is monitored for off-axis forces (i.e., forces
perpendicular to the desired movement direction) above a
10 N threshold, a value selected based on data from unim
paired subjects. Once such a force is detected, the motor
locks the position of the hand piece and the user receives
real-time graphical feedback of the error. The user is
instructed to relax, then begin a new reaching movement
from the current position taking into account the force-
error feedback. When the force transducer detects appro
priate forces toward the target, the graphical cue is
removed and the motor unlocks the hand piece, which
allows the user to progress toward the target. Essentially,
the control algorithm breaks the reach into a series of dis
crete movements bounded by error events. Our goal is to
force users to learn how to generate the correct muscle
activations at troublesome workspace positions.
One issue we encountered in designing the guided-
force training program was that the target population of
chronic stroke patients exhibits a wide range of arm
impairment levels. Some subjects can move through a
large range of motion at a high velocity, while others
have severe range and velocity limitations with stronger
coupling of motions between the elbow and shoulder
(e.g., elbow flexion is uncontrollably elicited with shoul
der abduction). To account for varying movement abili-
ties, we designed the training such that the threshold at
which off-axis forces lock the hand piece can be raised so
that the reaching task is achievable but still provides
challenge and feedback to the subject. Furthermore,
adapting the level of assistance provided during guided-
force training on a patient-specific basis is desirable
because each patient could practice movements with
maximal achievable reaching range and speed and mini
mal robotic assistance. In our experience, making the
task “difficult but doable” is important for maximizing
subject motivation.
We adapted the training to each subject’s ability by
adding a velocity-dependent assistive force from the
motor. The difference between the maximum velocity of
the previous reach and a desired velocity (quantified as
the maximum speed of a self-paced “comfortable” move
ment with the ipsilesional arm unattached to the device)
determines a coefficient b that specifies the assistive (or
resistive) force for the next movement:
where b
is the velocity assistance coefficient for trial i,
is an adaptation constant determining increment size,
(t) is the velocity profile of the previous trial, and
(t) is the velocity profile of the ipsilesional arm. Dur-
ing trial i, the coefficient is multiplied by the instanta-
neous velocity to determine motor output:
Thus, the motor does not assist movement unless the
subject initiates movement. Even in subjects who have
very little ability to move their paretic arm, the assistance
provided destabilizes the subjects’ movement, increasing
the range and speed. As the motor assistance increases
the subject’s movement velocity, the algorithm lowers the
velocity assistance coefficient b
from trial to trial until
the actual and desired peak movement speeds are equal.
If the subject is able to move faster than the desired
velocity, the algorithm decreases the velocity assistance
and eventually resists movement if b
becomes negative.
The algorithm is compatible with the guided-force train
ing because the assistance is velocity-dependent. Thus, if
the subject’s movement is stopped because of an errant
off-axis force, the assistance is terminated. When the
subject generates forces in the correct direction, the assis
tance turns on smoothly.
Example Use of Guided-Force Training Algorithm
To illustrate the feasibility of the guided-force train-
ing and adaptive assistance, we had two chronic stroke
subjects and an unimpaired subject practice 15 reaching
movements to a single target in a single session. To
examine repeatability of the algorithm, we asked one of
the chronic stroke subjects to practice 15 movements to
the same
target for 4 days. As shown in Figure 3, the
coefficient of assistance converged to a consistent value
for all three subjects. The coefficient for severely
i 1
c max v
t()[]{}max v
i 1
t()[], 1()+=
t() b
t().= 2()
JRRD, Volume 43, Number 5, 2006
impaired stroke subject CL01 repeatedly approached
similar values on multiple visits to the laboratory when
this subject was reaching to the same target. The coeffi
cient converged on a lower value for subject AL01 who
displayed greater functional ability on a clinical scale (3
out of 7 on the Chedoke-McMaster Stroke Assessment)
than subject CL01 (2 out of 7). The coefficient became
negative for an uninjured subject who was able to move
at a velocity greater than the desired velocity. Figure 3
also shows the desired effect of allowing a severely
impaired subject to gradually reach faster and farther by
automatically increasing assistance based on the sensed
error and the algorithm given by Equations (1) and (2).
Discussion of Guided-Force Training Algorithm
These results demonstrate the feasibility of a guided-
force training paradigm with adaptive assistance. This
type of training requires the subject to generate norma
tive force patterns to move, but it also adapts to the sub-
ject’s ability based on velocity measurements. Thus, even
severely impaired subjects achieve a sense of satisfaction
in completing the desired reaching task with short-term
practice (tens of reaches), although they still must work
very hard at trying to move. The algorithm is different
from the original MIME active-constrained mode
because it requires a more focused force generation
(within a forward-facing cylinder in force space rather
than just some force component toward the target), which
adds the potential advantage of requiring greater atten
tion from the subject. Guided-force training also departs
from the MIME because it adapts the assistance (or resis
tance) based on the subjects performance. We are cur-
rently testing this training algorithm in a study that
compares its benefits to a repetition-matched unassisted
reaching exercise and a time-matched conventional ther
apy. The current guided-force training paradigm with the
ARM Guide combines visual, haptic, and proprioceptive
cues. Future studies will examine the roles of these sen
sory cues as well as possible synergistic effects. Recent
studies suggest that manipulating tactile/haptic [40–44]
and visual cues [45–46] can enhance learning.
Distinguishing the benefits of the two interacting pro-
cesses highlighted here—movement practice and applica-
tion of robotic forces—is critical for determining the
future of robot-assisted therapy. If movement practice is
the dominant stimulus for movement recovery, then
robotic actuators may turn out to be technological orna
mentation. The question remains whether complex, and
potentially expensive, devices are essential for maximiz
ing the learning and recovery capabilities of the injured
CNS, or if less complex—and likely less expensive—
machines without actuation [47–48] that facilitate optimal
Figure 3.
Adaptive guided-force training algorithm. (a) Convergence of assis-
tance over multiple reaching trials along Assisted Rehabilitation and
Measurement Guide. Values for stroke subject CL01 consistently
converged around single value (~1.1) over multiple sessions on dif-
ferent days. Values for single session also shown for stroke subject
AL01 and unimpaired subject. (b) Increasing assistance from device
allowed subject to reach farther. Trajectories shown for stroke subject
CL01’s reaching trials 1, 7, and 15 on Day 4. Reprinted with partial
alteration by permission from Kahn LE, Rymer WZ, Reinkensmeyer
DJ. Adaptive assistance for guided force training in chronic stroke.
Proceedings of the 26th Annual IEEE Engineering in Medicine and
Biology Society Meeting; 2004 Sep 1–5; San Francisco, CA. Piscat-
away, NJ: IEEE press; 2004. p. 2722–25. (© [2004] IEEE.)
KAHN et al. Robot-assisted arm movement training
forms and amounts of practice will be the most viable
solution. In the interim, robotic devices are useful tools
for rigorously addressing the role of external forces on
neurorecovery. These devices also allow strictly con
trolled comparison of training effects across intervention
methods, severity levels, lesion locations, time postinjury,
sex, and virtually any other patient parameter. They will
provide a useful tool for understanding the specific mech
anisms of neuroplasticity that support motor learning
after stroke, although this area has been largely unad
dressed in the context of robotic therapy.
We can clarify the role of robotic forces by including
control groups that do not use robotic forces. For exam
ple, the control group that received a matched amount of
reaching practice but no robotic forces in the ARM Guide
study provided insight into whether the specific form of
robotic forces was necessary for motor gains. Our con
clusion is that robotically finishing a movement for a
chronic stroke subject did not add value beyond the con
current movement practice. Thus, our advice for clini-
cians considering bearing the additional costs of robotic
technology is to wait for further controlled testing, partic
ularly if they will use that technology only to finish
patient-initiated movements.
Nonetheless, based on our comparison of the MIME
and ARM Guide results, we are optimistic that robotic
forces will be found useful. Guided-force training tech
niques hold promise for improving arm movement ability
after stroke because they directly address anisotropic
weakness, a key impairment after stroke. Guided-force
training uses robotic forces to focus and intensify patient
effort and attention. We expect the principle of using
robotic forces to optimize subject effort to be important
in the future development of robot-assisted therapy.
This material was based on work supported in part
by the U.S. Department of Education National Institute
on Disability and Rehabilitation Research (NIDRR)
Field-Initiated grant H133G80052; NIDRR Rehabilita
tion Engineering Research Center on Rehabilitation
Robotics and Telemanipulation grant H133E020724;
National Institutes of Health, National Institute of Child
Health and Human Development, Institutional National
Research Service Award (Rymer); and Department of
Veterans Affairs merit review grant B2056RA.
The authors have declared that no competing inter-
ests exist.
1. American Physical Therapy Association. Guide to physical
therapist practice. 2nd ed. Alexandria (VA): American
Physical Therapy Association; 2001.
2. Bobath B. Adult hemiplegia: Evaluation and treatment. 2nd
ed. London (England): William Heinemann Medical
Books, Ltd; 1978.
3. Ryerson SD. Hemiplegia. In: Umphred DA, editor. Neuro-
logical rehabilitation. 3rd ed. St. Louis (MO): The C. V.
Mosby Company; 1995. p. 681–721.
4. Aisen ML, Krebs HI, Hogan N, McDowell F, Volpe BT.
The effect of robot-assisted therapy and rehabilitative train-
ing on motor recovery following stroke. Arch Neurol.
1997;54(4):443–46. [PMID: 9109746]
5. Hesse S, Schmidt H, Werner C, Bardeleben A. Upper and
lower extremity robotic devices for rehabilitation and for
studying motor control. Curr Opin Neurol. 2003;16(6):
705–10. [PMID: 14624080]
6. Riener R, Nef T, Colombo G . Robot-aided neurorehabilita-
tion of the upper extremities. Med Biol Eng Comput. 2005;
43(1):2–10. [PMID: 15742713]
7. Volpe BT, Krebs HI, Hogan N. Robot-aided sensorimotor
training in stroke rehabilitation. Adv Neurol. 2003;92:429–33.
[PMID: 12760210]
8. Reinkensmeyer DJ, Emken JL, Cramer SC. Robotics,
motor learning, and neurological recovery. Ann Rev
Biomed Eng. 2004;6:497–525. [PMID: 15255778]
9. Hogan N, Krebs HI. Interactive robots for neuro-rehabilita-
tion. Restor Neurol Neurosci. 2004;22(3–5):349–58.
[PMID: 15502275]
10. Hesse S, Schulte-Tigges G , Konrad M, Bardeleben A,
Werner C. Robot-assisted arm trainer for the passive and
active practice of bilateral forearm and wrist movements in
hemiparetic subjects. Arch Phys Med Rehabil. 2003;
84(6):915–20. [PMID: 12808550]
11. Lum PS, Burgar CG , Shor PC, Majmundar M, Van der
Loos M. Robot-assisted movement training compared with
conventional therapy techniques for the rehabilitation of
upper-limb motor function after stroke. Arch Phys Med
Rehabil. 2002;83(7):952–59. [PMID: 12098155]
12. Werner C, Von Frankenberg S, Treig T, Konrad M, Hesse
S. Treadmill training with partial body weight support and
an electromechanical gait trainer for restoration of gait in
subacute stroke patients: A randomized crossover study.
Stroke. 2002;33(12):2895–2901. [PMID: 12468788]
JRRD, Volume 43, Number 5, 2006
13. Krebs HI, Hogan N, Aisen ML, Volpe BT. Robot-aided
neurorehabilitation. IEEE Trans Rehabil Eng. 1998;6(1):
75–87. [PMID: 9535526]
14. Volpe BT, Krebs HI, Hogan N, Edelstein L, Diels CM,
Aisen ML. Robot training enhanced motor outcome in
patients with stroke maintained over 3 years. Neurology.
1999;53(8):1874–76. [PMID: 10563646]
15. Volpe BT, Krebs HI, Hogan N, Edelstein L, Diels C, Aisen
ML. A novel approach to stroke rehabilitation: Robot-aided
sensorimotor stimulation. Neurology. 2000;54(10):1938–44.
[PMID: 10822433]
16. Fasoli SE, Krebs HI, Stein J, Frontera WR, Hogan N.
Effects of robotic therapy on motor impairment and recov-
ery in chronic stroke. Arch Phys Med Rehab. 2003;
84(4):477–82. [PMID: 12690583]
17. Stein J, Krebs HI, Frontera WR, Fasoli SE, Hughes R,
Hogan N. Comparison of two techniques of robot-aided
upper limb exercise training after stroke. Am J Phys Med
Rehabil. 2004;83(9):720–28. [PMID: 15314537]
18. Ferraro M, Palazzolo JJ, Krol J, Krebs HI, Hogan N, Volpe
BT. Robot-aided sensorimotor arm training improves out-
come in patients with chronic stroke. Neurology. 2003;
61(11):1604–7. [PMID: 14663051]
19. Schmidt RA, Lee TD. Motor control and learning: A
behavioral emphasis. 4th ed. Champaign (IL): Human
Kinetics; 2005.
20. Page SJ, Levine P, Sisto S, Johnston MV. A randomized
efficacy and feasibility study of imagery in acute stroke.
Clin Rehabil. 2001;15(3):233–40. [PMID: 11386392]
21. Stevens JA, Stoykov ME. Using motor imagery in the reha-
bilitation of hemiparesis. Arch Phys Med Rehabil. 2003;
84(7):1090–92. [PMID: 12881842]
22. Dijkerman HC, Ietswaart M, Johnston M, MacWalter RS.
Does motor imagery training improve hand function in
chronic stroke patients? A pilot study. Clin Rehabil.
2004;18(5):538–49. [PMID: 15293488]
23. Reinkensmeyer DJ, Kahn LE, Averbuch M, McKenna-
Cole A, Schmit BD, Rymer WZ. Understanding and treat-
ing arm movement impairment after chronic brain injury:
Progress with the ARM Guide. J Rehabil Res Dev.
2000;37(6):653–62. [PMID: 11321001]
24. Kahn LE, Averbuch M, Rymer WZ, Reinkensmeyer DJ.
Comparison of robot-assisted reaching to free reaching in
promoting recovery from chronic stroke. In: Mokhtari M, edi-
tor. Integration of assistive technology in the information age.
Amsterdam (the Netherlands): IOS Press; 2001. p. 39–44.
25. Kahn LE, Zygman ML, Rymer WZ, Reinkensmeyer DJ.
Robot-assisted reaching exercise promotes arm recovery in
chronic hemiparetic stroke: A randomized controlled pilot
study. J Neuroengineering Rehabil. 2006;3:12.
26. Gowland C, Stratford P, Ward M, Moreland J, Torresin W,
Van Hullenaar S, Sanford J, Barreca S, Vanspall B, Plews
N. Measuring physical impairment and disability with the
Chedoke-McMaster Stroke Assessment. Stroke. 1993;
24(1):58–63. [PMID: 8418551]
27. Wilson DJ, Baker LL, Craddock JA. Functional test for the
hemiparetic upper extremity. Am J Occup Ther. 1984;
38(3):159–64. [PMID: 6711667]
28. Diggle P, Heagerty P, Liang KY, Zeger SL. Analysis of lon-
gitudinal data. 2nd ed. Oxford (England): Oxford Univer-
sity Press; 2002.
29. Kahn LE, Lum PS, Reinkensmeyer DJ. Selection of robotic
therapy algorithms for the upper extremity in chronic
stroke: Insights from MIME and ARM Guide results. Pro-
ceedings of the 8th International Conference on Rehabilita-
tion Robotics; 2003 Apr 23–25; Daejeon Korea. Daejeon
(Korea): Human-Friendly Welfare Robot System Engineer-
ing Research Center; 2003. p. 208–10.
30. Luft AR, McCombe-Waller S, Whitall J, Forrester LW,
Macko R, Sorkin JD, Schulz JB, Goldberg AP, Hanley DF.
Repetitive bilateral arm training and motor cortex activa-
tion in chronic stroke: A randomized controlled trial.
JAMA. 2004;292(15):1853–61. [PMID: 15492483]
tum in: JAMA. 2004;292(20):2470.
31. Cauraugh JH, Kim S. Two coupled motor recovery proto-
cols are better than one: Electromyogram-triggered neuro-
muscular stimulation and bilateral movements. Stroke.
2002;33(6):1589–94. [PMID: 12052996]
32. Whitall J, McCombe Waller S, Silver KH, Macko RF.
Repetitive bilateral arm training with rhythmic auditory
cueing improves motor function in chronic hemiparetic
stroke. Stroke. 2000;31(10):2390–95. [PMID: 11022069]
33. Lum PS, Burgar CG , Van der Loos M, Shor PC,
Majmundar M, Yap R. The MIME robotic system for
upper-limb neuro-rehabilitation: Results from a clinical
trial in subacute stroke. Proceedings of the IEEE 9th Inter-
national Conference on Rehabilitation Robotics; 2005 Jun
28–Jul 1; Chicago, IL. Piscataway (NJ): IEEE Press; 2005.
p. 511–14.
34. Robertson IH, Ridgeway V, Greenfield E, Parr A. Motor
recovery after stroke depends on intact sustained attention:
A 2-year follow-up study. Neuropsychology. 1997;11(2):
290–95. [PMID: 9110335]
35. Pohl PS, Filion DL, Kim SH. Effects of practice and unpre-
dictable distractors on planning and executing aiming after
stroke. Neurorehabil Neural Repair. 2003;17(2):93–100.
[PMID: 12814054]
36. Dewald JP, Beer RF. Abnormal joint torque patterns in the
paretic upper limb of subjects with hemiparesis. Muscle
Nerve. 2001;24(2):273–83. [PMID: 11180211]
37. Dewald JP, Pope PS, Given JD, Buchanan TS, Rymer WZ.
Abnormal muscle coactivation patterns during isometric
torque generation at the elbow and shoulder in hemiparetic
subjects. Brain. 1995;118(Pt 2):495–510. [PMID: 7735890]
KAHN et al. Robot-assisted arm movement training
38. Lum PS, Burgar CG , Shor PC. Evidence for strength
imbalances as a significant contributor to abnormal syner-
gies in hemiparetic subjects. Muscle Nerve. 2003;27(2):
211–21. [PMID: 12548529]
39. Kahn LE, Rymer WZ, Reinkensmeyer DJ. Adaptive assis-
tance for guided force training in chronic stroke. Proceed-
ings of the 26th Annual IEEE Engineering in Medicine and
Biology Society Meeting, Vol 4; 2004 Sep 1–5; San Fran-
cisco, CA. Piscataway (NJ): IEEE Press; 2004. p. 2722–25.
40. Liu J, Cramer SC, Reinkensmeyer DJ. Learning to perform
a new movement with robotic assistance: Comparison of
haptic guidance and visual demonstration. J Neuroengi-
neering Rehabil. 2006;3(1):20.
41. Patton JL, Mussa-Ivaldi FA, Rymer WZ. Altering move-
ment patterns in healthy and brain-injured subjects via cus-
tom designed robotic forces. Proceedings of the 23rd
Annual IEEE Engineering in Medicine and Biology Soci-
ety Meeting. Vol. 2; 2001 Oct 25–29; Istanbul, Turkey. Pis-
cataway (NJ): IEEE Press; 2001. p. 1356–59.
42. Feygin D, Keehner M, Tendick F. Haptic guidance: Experi-
mental evaluation of a haptic training method for a percep-
tual motor skill. Proceedings of the 10th International
Symposium on Haptic Interfaces for Virtual Environment
and Teleoperator Systems; 2002 Mar 24–25; Orlando, FL.
Piscataway (NJ): IEEE Press; 2002. p. 40–47.
43. Henriques DY, Flanders M, Soechting JF. Haptic synthesis
of shapes and sequences. J Neurophysiol. 2004;91(4):
1808–21. [PMID: 14645382]
44. Emken JL, Reinkensmeyer DJ. Robot-enhanced motor
learning: Accelerating internal model formation during
locomotion by transient dynamic amplification. IEEE
Trans Neural Syst Rehabil Eng. 2005;13(1):33–39.
[PMID: 15813404]
45. Wei Y, Bajaj P, Scheidt R, Patton JL. Visual error augmen-
tation for enhancing motor learning and rehabilitative
relearning. Proceedings of the IEEE 9th International Con-
ference on Rehabilitation Robotics; 2005 Jun 28–Jul 1; Chi-
cago, IL. Piscataway (NJ): IEEE Press; 2005. p. 505–10.
46. Brewer BR, Fagan M, Klatzky RL, Matsuoka Y. Perceptual
limits for a robotic rehabilitation environment using visual
feedback distortion. IEEE Trans Neural Syst Rehabil Eng.
2005;13(1):1–11. [PMID: 15813400]
47. Sanchez R, Reinkensmeyer DJ, Shah P, Liu J, Rao S, Smith
R, Cramer S, Rahman T, Bobrow J. Monitoring functional
arm movement for home-based therapy after stroke. Pro-
ceedings of the 26th Annual IEEE Engineering in Medicine
and Biology Society Meeting, Vol. 7; 2004 Sep 1–5; San
Francisco, CA. Piscataway (NJ): IEEE Press; 2004.
p. 4787–90.
48. Lum PS, Taub E, Schwandt D, Postman M, Hardin P,
Uswatte G. Automated Constraint-Induced Therapy Exten-
sion (AutoCITE) for movement deficits after stroke.
J Rehabil Res Dev. 2004;41(3A):249–58. [PMID: 15543442]
Submitted for publication March 2, 2005. Accepted in
revised form August 16, 2005.
... Patient-passive control robots mainly deliver automated practical movements to patients, and patient-active control robots can monitor and evaluate the physical parameters and performance of voluntary motion of patients [24] and then provide assistance as needed to complete the movement initiated by patients [25]. In the latter strategy, patients pay more attention to and put more effort into the training and more actively participate in the practice [26], which is essential for improving cortical activity, excitability and motor performance [[ [ 27 ]]] . Active participation is influenced by the level of impairment, the mechanical properties of the robot, the control strategies, the training mode of the robot, the instructions of the therapist and various other factors, therefore, we conducted a subgroup analysis to investigate the effect of training mode and impairment level on the superiority of RAT. ...
... The movement practice and application of robotic force are two interacting processes of RAT, and which process is more beneficial is controversial. A previous study [26] found that robotically finishing a movement for a patient with stroke did not show better improvement of function than usual movement practice, and using robotic forces to assist patients to complete correct movements could focus and intensify patients' effort and attention to the treatment, achieving better outcomes [50]. Active participation of the patients is critical for neuroplasticity, motor learning and rehabilitation [50,51], and studies have found that rehabilitation treatment integrated with patients' voluntary movement could facilitate the recovery of lost motor ability [16,52]. ...
Full-text available
Objective: To investigate the effect of robot-assisted therapy (RAT) on upper limb motor control and activity function in poststroke patients compared with that of non-robotic therapy. Methods: We searched PubMed, EMBASE, Cochrane Library, Google Scholar and Scopus. Randomized controlled trials published from 2010 to nowadays comparing the effect of RAT and control treatment on upper limb function of poststroke patients aged 18 or older were included. Researchers extracted all relevant data from the included studies, assessed the heterogeneity with inconsistency statistics (I2 statistics), evaluated the risk of bias of individual studies and performed data analysis. Result: Forty-six studies were included. Meta-analysis showed that the outcome of the Fugl-Meyer Upper Extremity assessment (FM-UE) (SMD = 0.20, P = 0.001) and activity function post intervention was significantly higher (SMD = 0.32, P < 0.001) in the RAT group than in the control group. Differences in outcomes of the FM-UE and activity function between the RAT group and control group were observed at the end of treatment and were not found at the follow-up. Additionally, the outcomes of the FM-UE (SMD = 0.15, P = 0.005) and activity function (SMD = 0.32, P = 0.002) were significantly different between the RAT and control groups only with a total training time of more than 15 h. Moreover, the differences in outcomes of FM-UE and activity post intervention were not significant when the arm robots were applied to patients with severe impairments (FM-UE: SMD = 0.14, P = 0.08; activity: SMD = 0.21, P = 0.06) or when patients were provided with patient-passive training (FM-UE: SMD = - 0.09, P = 0.85; activity: SMD = 0.70, P = 0.16). Conclusion: RAT has the significant immediate benefits for motor control and activity function of hemiparetic upper limb in patients after stroke compared with controls, but there is no evidence to support its long-term additional benefits. The superiority of RAT in improving motor control and activity function is limited by the amount of training time and the patients' active participation.
... 21 Studies have shown that robotic rehabilitation helps in improving the deterioration of the hemiparetic upper extremity after chronic stroke. 8,17,[22][23][24][25][26][27][28][29][30][31][32] Although the number of papers published on robotic therapy has increased rapidly, with existing studies in the literature, it is not possible to make an exact decision in the acute term. 14,[33][34][35][36] Moreover, robotic rehabilitation studies performed on acute stroke patients have some limitations. ...
Objective The aim of this study was to investigate the effects of EMG-driven robotic rehabilitation on hand motor functions and daily living activities of patients with acute ischemic stroke. Materials & Method A preliminary randomized-controlled, single-blind trial rectuited twenty-four patients with acute ischemic stroke (<1 month after cerebrovascular accident) and randomly allocated to experimental group (EG) and control group (CG). Neurophysiological rehabilitation program was performed to both EG and CG for 5 days a week and totally 15 sessions. The EG also received robotic rehabilitation with the EMG-driven exoskeleton hand robot (Hand of Hope®, Rehab-Robotics Company) 15 sessions over 3 weeks. Hand motor functions (Fugl-Meyer Assessment-Upper Extremity (FMA-UE) and Action Research Arm Test (ARAT)), activities of daily living (Motor Activity Log (MAL)), force and EMG activities of extensor and flexor muscles for the cup test were evaluated before treatment (pretreatment) and after the 15th session (posttreatment). Results Eleven patients (59.91 ± 14.20 yr) in the EG and 9 patients (70 ± 14.06 yr) in the CG completed the study. EG did not provide a significant advantage compared with the CG in FMA-UE, ARAT and MAL scores and cup-force and EMG activities (p > .05 for all). Conclusion In this preliminary study, improvement in motor functions, daily living activities and force were found in both groups. However, addition of the EMG-driven robotic treatment to the neurophysiological rehabilitation program did not provide an additional benefit to the clinical outcomes in 3 weeks in acute stroke patients.
... Robot systems designed for collaborations with human workers, often termed collaborative robots, can carry out tasks that are difficult to perform ergonomically when unassisted or otherwise uneconomical to automate [5,6], or that require close proximity to human operators [7]. Robotic systems are also developed to facilitate physical rehabilitation [8][9][10][11][12] and sport training. ...
Interaction control can take opportunities offered by contact robots physically interacting with their human user, such as assistance targeted to each human user, communication of goals to enable effective teamwork, and task-directed motion resistance in physical training and rehabilitation contexts. Here we review the burgeoning field of interaction control in the control theory and machine learning communities, by analysing the exchange of haptic information between the robot and its human user, and how they share the task effort. We first review the estimation and learning methods to predict the human user intent with the large uncertainty, variability and noise and limited observation of human motion. Based on this motion intent core, typical interaction control strategies are described using a homotopy of shared control parameters. Recent methods of haptic communication and game theory are then presented to consider the co-adaptation of human and robot control and yield versatile interactive control as observed between humans. Finally, the limitations of the presented state of the art are discussed and directions for future research are outlined.
... Various causes such as sedentary lifestyles, aging populations and low birth rates predict a troubling future regarding the capacity of public healthcare systems [10,11]. The clinical use of these devices can reduce the waiting lists in rehabilitation services by allowing to follow more patients at the same time and lower the cost of each therapy [12][13][14]. ...
Introduction: Stroke is the leading cause of long-term disability in developed countries. Due to population aging, the number of people requiring rehabilitation after stroke is going to rise in the coming decades. Robot-mediated neurorehabilitation has the potential to improve clinical outcomes of rehabilitation treatments. A statistical analysis of the literature aims to focus on the main trend of this topic. Areas covered: A bibliometric survey on post-stroke robotic rehabilitation was performed through a database collection of scientific publications in the field of rehabilitation robotics. By covering the last 20 years, 17429 sources were collected. Relevant patterns and statistics concerning the main research areas were analyzed. Leading journals and conferences which publish and disseminate knowledge in the field were identified. A detailed nomenclature study was carried out. The time trends of the research field were captured. Opinions and predictions of future trends that are expected to shape the near future of the field were discussed. Expert opinion: Data analysis reveals the continuous expansion of the research field over the last two decades, which is expected to rise considerably in near future. More attention will be paid to the lower limbs rehabilitation and disease/design specific applications in early-stage patients.
... At initial exposure to the reversal task, the EA group stopped less frequently, and reached their targets more quickly. Other researchers have observed that subject will stop-and-think in the event of large movement errors, perhaps to evaluate recent movements and sub-movements and then re-plan movement strategies [37] [36]. For our results, we speculate that because the EA group perceives their mistakes more clearly, they require less resetting and iterative attempts at performing straight line reaching movements (Fig. 3). ...
Motor control, error-augmentation and limit-push
... Traditional controllers [28][29][30] assisted patients along pre-determined trajectories with very small allowable deviations. The brittleness of these systems in terms of the allowable motions and deviations severely limits their efficacy as rehabilitative tools [31] and patients feel like they are fighting the robotic device for control. ...
Full-text available
This dissertation presents the design and evaluation of iART : an Intelligent Assistant for Robotic Therapy. iART is a robot-assisted therapy system designed for home-based upper limb stroke rehabilitation. Stroke is the leading cause of motor impairments and serious long term disability in the United States. These impairments severely limit a patient's ability to lead a normal independent life, and requires them to participate in hospital-based stroke rehabilitation. Recent years have seen the advent of robotic rehabilitation systems as a home-based alternative to hospital-centric stroke therapy. These systems comprise of a robotic device that assists/resists a patient's movements as they perform virtual therapy exercises. In this dissertation, we describe a novel intelligent robotic therapy system that can provide adaptive assistance to patients as they perform virtual therapy tasks. iART comprises of five robot-assisted therapy games/tasks along with an artificial intelligence (AI) agent that adapts the degree of robotic assistance based on a patient's performance. As with any traditional robotic rehabilitation system, iART enables a therapist to remotely monitor a therapy session and suggest changes. Additionally, iART employs an AI that uses surface electromyography (sEMG) and data from the robotic device to monitor a patient's performance/engagement levels in realtime and adapt the system accordingly. The realization of an AI agent to monitor a patient is the key contribution of this work. The dissertation also proposes the use of LSTM-based imitation learning and reinforcement learning towards the realization of an adaptive robotic therapy assistant. This dissertation is divided into three parts. The first part includes a description of existing robot/haptics-based stroke rehabilitation systems. It also introduces the key components of iART and provides a preliminary evaluation of the system. The concept of mental engagement in therapy is introduced as well. Part two delves deeper into the study of mental engagement and its importance towards the success of robotic rehabilitation. It describes a robot-based and an sEMG-based methodology adopted in iART towards monitoring and ensuring patient engagement. Part three explores two novel mechanisms for adaptive assistance viz. learning from demonstration and reinforcement learning. The applications of these paradigms in robotic rehabilitation are fairly nascent and this dissertation serves as one of the initial forays into these domains.
Arm and hand motor impairments are frequent after a neurological injury. Motor rehabilitation can improve hand and arm function in many cases, but in the current healthcare climate, the time and resources devoted to physical and occupational therapy after injury are inadequate. This represents an opportunity for technology to be introduced that can complement rehabilitation practices, provide motivating task training and allow remote supervision of exercise training performed in the home. Over the last decades, many research groups have been developing robotic devices for exercise therapy, as well as other methods such as electrical stimulation of muscles or vagus nerve stimulation. Robotic devices tend to be expensive and recent studies have raised some doubt as to whether assistance to movements is always preferable as it can reduce salience and engagement. This chapter reviews the evidence for spontaneous recovery, the means and mechanisms of conventional rehabilitation interventions, the advent of affordable passive devices and other treatment modalities that can be used in combination with passive devices. It is argued that task practice on passive devices, in some cases remotely supervised over the internet or augmented with functional electrical stimulation (FES), is now an affordable and important modality of occupational and physical therapy. Passive devices offer numerous opportunities in the field of neurological rehabilitation to support arm and hand motor recovery.
Brain injury often results in a partial loss of the neural resources communicating to the periphery that controls movements. Consequently, the signals that were employed prior to injury may no longer be appropriate for controlling the muscles for the intended movement. Hence, a new pattern of signals may need to be learned that appropriately uses the residual resources. The learning required in these circumstances might in fact share features with sports, music performance, surgery, teleoperation, piloting, and child development. Our lab has leveraged key findings in neural adaptation as well as established principles in engineering control theory to develop and test new interactive environments that enhance learning (or re-learning). Successful application comes from the use of robotics and video feedback technology to augment error signals. These applications test standing hypotheses about error-mediated neuroplasticity and illustrate an exciting prospect for rehabilitation environments of tomorrow. This chapter highlights our works, identifies our acquired knowledge, and outlines some of the successful pathways for restoring function to brain-injured individuals.
A rehabilitation robot is a device that has been proving its positive effectiveness in the process of helping patients recover quickly after a stroke. Researching, designing, and manufacturing robot models in general and upper limb rehabilitation robots in particular are very practical. In this study, we proposed to combine the use of an algorithm and a physical modeling method to shorten the calculation process and design an upper limb rehabilitation robot. First, an exoskeleton upper limb rehabilitation robot model (UExosVN) was briefly described. Next, in turn, all the important problems including inverse kinematics, inverse dynamics for this robot model were proposed and solved by using optimization algorithms and physical modeling methods. The model was evaluated in the critical movement of daily operations. The results after the testing process have proven the accuracy and effectiveness of the proposed methods.
Dynamic control of an intrinsically compliant robot is paramount to ensuring safe and synergistic assistance to the patient. This paper presents an impedance controller for the rehabilitation of stroke patients with compromised wrist motor functions. The control design employs a Koopman operator-based autodidactic system identification model to predict the anatomical stiffness of the wrist joint during its various degrees of rotational motion. The proposed impedance controller, perceiving the level of the subjects’ participation from their joint stiffness, can modify the applied force. The end-effector robot has a parallel structure that uses four biomimetic muscle actuators as parallel links between the end-effector and the base platform. The controller performance is corroborated by testing the end-effector robot with three healthy subjects.
Full-text available
Recent studies have suggested that physical rehabilitation performed with robotic devices can enhance arm movement recovery following stroke. In these studies, the robotic devices mechanically assisted arm movement as the patient attempted to move. Although this "robotic active assist" technique has shown promise, a key question remains unanswered: is the mechanical assistance provided by the robots necessary, or is it primarily the repetitive movement attempts by the patients that stimulate recovery? We are using a novel robotic device called the Assisted Rehabilitation and Measurement (ARM) Guide to investigate this question. To date, ten subjects have completed twenty-four therapy sessions over an eight-week period, randomized to either a robot exercise group (N = 6) or non-robotic exercise group (N = 4). For the robot exercise group, the ARM Guide mechanically assisted in reaching to a series of targets. For the non-robotic group, the subjects performed unassisted, unrestrained reaching exercises to the same targets for the same number of repetitions as the ARM Guide group. All subjects have been evaluated using a set of clinical and biomechanical measures of arm movement. The ten subjects tested so far have shown improvement in the measures after completion of both exercise programs. However, the amount of improvement has been comparable for the robot and free reaching groups. Although the subject numbers are currently insufficient to draw a definitive conclusion, these results are suggestive that the repetitive movement attempts by the patient, rather than the active assistance from the ARM Guide, are the primary stimuli to recovery.
Conference Paper
Full-text available
In this paper we investigate a use of haptics for skills training which we call haptic guidance. In the haptic guidance paradigm, the subject is physically guided through the ideal motion by the haptic interface, thus giving the subject a kinesthetic understanding of what is required. Subjects learned a complex 3D motion under three training conditions (haptic, visual, haptic and visual) and were required to manually reproduce the movement under two recall conditions (with vision, without vision). Performance was measured in terms of position, shape, timing, and drift. Findings from this study indicate that haptic guidance is effective in training. While visual training was better for teaching the trajectory shape, temporal aspects of the task were more effectively learned from haptic guidance. This supports a possible role for haptics in the training of perceptual motor skills in virtual environments
Full-text available
Our goal is to apply robotics and automation technology to assist, enhance, quantify, and document neurorehabilitation. This paper reviews a clinical trial involving 20 stroke patients with a prototype robot-aided rehabilitation facility developed at the Massachusetts Institute of Technology, Cambridge, (MIT) and tested at Burke Rehabilitation Hospital, White Plains, NY. It also presents our approach to analyze kinematic data collected in the robot-aided assessment procedure. In particular, we present evidence 1) that robot-aided therapy does not have adverse effects, 2) that patients tolerate the procedure, and 3) that peripheral manipulation of the impaired limb may influence brain recovery. These results are based on standard clinical assessment procedures. We also present one approach using kinematic data in a robot-aided assessment procedure.
The primary purpose of this study was to examine practice effects on the planning and execution of an aiming movement after tight versus left stroke. A secondary purpose was to investigate the effects of a distractor that appeared randomly on motor performance after stroke. Right-hand dominant individuals, 15 with right stroke (right-sided brain damage), 16 with left stroke, and 30 without stroke, performed aiming movements to targets. Those with stroke used the ipsilesional upper extremity (UE). Right and left comparison groups used the light and left UE, respectively, Reaction time (RT) and movement time (MT) were collected to represent movement planning and execution, respectively. Individuals with right stroke improved RT with practice. Individuals with left stroke did not improve RT with practice and made more errors than their comparison group. Those with left stroke achieved faster MT with practice, but MT remained slower than their comparison group. There were no effects of the distractor on RT or MT Adults with left stroke have persistent deficits in movement planning and execution. Further studies are needed to determine bow the performance of older adults, with or without stroke, is affected by an unpredictable visual distractor.
This study presents the results of a novel paradigm for characterizing abnormal coordination in subjects with hemiparesis. Subjects generated maximum voluntary torques (MVTs) isometrically in four randomly ordered blocks consisting of elbow flexion/extension, shoulder flexion/extension, shoulder abduction/adduction, and shoulder external/internal rotation. A 6–degree-of-freedom (DOF) load cell was used to measure torques in secondary DOFs at the elbow and shoulder, as well as in the torque direction the subject was attempting to maximize. This allowed characterization of the multijoint torque patterns associated with the generation of MVTs in the eight directions examined. Significant differences were found between the torque patterns exhibited by the paretic limb of the hemiparetic group (n = 8) and those observed for the nonparetic limb and control group (n = 4). Potential neural and biomechanical mechanisms underlying these abnormal torque patterns are discussed along with implications for the functional use of the paretic limb. © 2001 John Wiley & Sons, Inc. Muscle Nerve 24: 273–283, 2001