MEDARM: A rehabilitation robot with 5DOF at the shoulder complex

Abstract

A key approach for reducing motor impairment and regaining independence after stroke is frequent and repetitive functional training. A number of robotic devices have been developed to assist therapists with the labourious task of providing treatment. Although robotic technology is showing significant potential, its effectiveness for upper limb rehabilitation is limited in part by the inability to make functional reaching movements. A major contributor to this problem is that current robots do not replicate motion of the shoulder girdle despite the fact that the shoulder girdle plays a critical role in stabilizing and orienting the upper limb during activities of daily living. To address this issue, a new adjustable robotic exoskeleton called MEDARM is proposed for motor rehabilitation of the shoulder complex. MEDARM provides independent control of six degrees of freedom (DOF) of the upper limb: two at the sternoclavicular joint, three at the glenohumeral joint and one at the elbow. Its joint axes are optimally arranged to mimic the natural upper- limb workspace while avoiding singular configurations and while maximizing manipulability. This mechanism also permits reduction to planar shoulder/elbow motion in any plane by locking all but the last two joints. Electric motors actuate the joint using a combination of cable and belt transmissions designed to maximize the power-to-weight ratio of the robot while maintaining backdriveability and minimizing inertia. Thus, the robot can provide any level of movement assistance and gravity compensation. This paper describes the proposed technical design for MEDARM.

Full text

MEDARM: a rehabilitation robot with 5DOF at the shoulder complex
Stephen J. Ball, Ian E. Brown and Stephen H. Scott
Abstract— A key approach for reducing motor impairment
and regaining independence after stroke is frequent and
repetitive functional training. A number of robotic devices
have been developed to assist therapists with the labourious
task of providing treatment. Although robotic technology is
showing significant potential, its effectiveness for upper limb
rehabilitation is limited in part by the inability to make
functional reaching movements. A major contributor to this
problem is that current robots do not replicate motion of
the shoulder girdle despite the fact that the shoulder girdle
plays a critical role in stabilizing and orienting the upper limb
during activities of daily living. To address this issue, a new
adjustable robotic exoskeleton called MEDARM is proposed
for motor rehabilitation of the shoulder complex. MEDARM
provides independent control of six degrees of freedom (DOF)
of the upper limb: two at the sternoclavicular joint, three
at the glenohumeral joint and one at the elbow. Its joint
axes are optimally arranged to mimic the natural upper-
limb workspace while avoiding singular configurations and
while maximizing manipulability. This mechanism also permits
reduction to planar shoulder/elbow motion in any plane by
locking all but the last two joints. Electric motors actuate
the joint using a combination of cable and belt transmissions
designed to maximize the power-to-weight ratio of the robot
while maintaining backdriveability and minimizing inertia.
Thus, the robot can provide any level of movement assistance
and gravity compensation. This paper describes the proposed
technical design for MEDARM.
Index Terms— rehabilitation, robot, shoulder girdle, stroke
I. INTR ODUCTION
S
TROKE is a leading cause of disability in Canada,
leaving people with motor deficits that limit their ability
to perform activities of daily living. Impairments may involve
loss of a combination of motor, sensory and/or cognitive
functions including weakness, reduced coordination, dimin-
ished ability to think, communicate, and make decisions, as
well as to feel, see and hear. Currently, more than 300,000
Canadians live with the effects of stroke, and there are
50,000 new occurrences every year. 16,000 Canadians die
from stroke every year. Overall, stroke health care costs the
Canadian health care system $2.7 billion every year [6]. The
costs will only increase as the population continues to age.
It is possible to regain partial or complete motor function
through individualized rehabilitation programs. Traditionally,
Manuscript received January 12, 2007. This work was supported by the
Canadian Institutes of Health Research (CIHR), and the Natural Sciences
and Engineering Research Council of Canada (NSERC).
Stephen J. Ball is with Department of Electrical and Computer Engineer-
ing, Queen’s University, Kingston, ON, Canada.
Ian E. Brown is with the Department of Anatomy and Cell Biology, Cen-
tre for Neuroscience Studies, Queen’s University, Kingston, ON, Canada.
Stephen H. Scott is with the Department of Anatomy and Cell Biol-
ogy, Centre for Neuroscience Studies, Queen’s University, Kingston, ON,
Canada.
therapists taught compensatory movements such as using
redundant joints of the body (i.e. using trunk motion to
assist with reaching) or fractionation of movements into
several simpler movements [5]. While compensation allows
patients to rapidly regain some degree of independence, a
strong reliance on compensation promotes learned non-use
of the impaired limb which slows or inhibits functional
recovery [12]. Current rehabilitation programs tend to focus
instead on reducing the degree of permanent disability.
Recovery of motor function has recently been linked to motor
learning that occurs during repetitive, frequent and inten-
sive movements [16]. This increased sensorimotor activity
takes advantage of neural plasticity, which is the ability of
adjacent areas of the brain to reorganize and compensate
for lost function in other brain regions. It is agreed that
exercising and practicing a variety of functional multi-joint
movements with the impaired limb is an important part of
therapy for stroke patients because it increases their ability
to perform activities of daily living [5]. Therefore, typical
therapy programs include the use of a variety of techniques
such as restraint, gravity compensation, manual guidance,
and progressive-resistive exercise.
Although these conventional techniques are effective, a
significant drawback is that they require strenuous manual
labour and extensive one-on-one attention from therapists.
As such, recovery is severely limited by staffing, time, and
budget constraints, and it is becoming more difficult to give
patients the time and attention that they require for maximum
recovery. Even without these limits, therapists can tire or
injure easily when manually moving heavy limbs, and with
no means to quantitatively record progress, it is a challenge to
properly monitor functional ability. Unfortunately, therapists
are often forced to resort to shorter, less intense therapy
programs that focus on teaching compensatory techniques
rather than on recovering motor function [3].
The ever increasing need for motor rehabilitation is strain-
ing the capabilities of the health care system. There is no
doubt that a more efficient system is required to provide the
high quality care that patients need. Robotic systems provide
a unique solution, and they offer a number of significant
advantages over conventional techniques. Robots can make
many precise movements with any level of assistance without
getting tired or making mistakes, allowing therapists to focus
on treatment planning and progress monitoring. Furthermore,
robots can provide a wealth of quantitative measurements
for every movement, which could lead to more sensitive
functional assessment scores. Other opportunities such as
home or group therapy (one therapist with multiple patients)
also become feasible using robotic technology.
1-4244-1264-1/07/$25.00 ©2007 IEEE

There is no question that robots may be ideally suited for
rehabilitation and there are a number of projects currently
underway that are trying to realize their potential for rehabili-
tation of the upper limb after stroke [7], [14]. Some existing
robots for upper limb rehabilitation and assistance include
MIT-MANUS [9], MIME [2], GENTLE/s [11], MULOS [8],
T-WREX [15] and ARMin [13] among others.
A significant drawback of current robotic devices is that
they cannot match the mobility of the human upper limb.
This is particularly true for the shoulder complex because it
has a compact arrangement of five major degrees of freedom
(DOF): two at the sternoclavicular joint and three at the
glenohumeral joint. The glenohumeral joint can be approxi-
mated as a ball-and-socket joint and has been replicated in
some current devices. However, the shoulder girdle has been
neglected, despite its importance in stabilizing and orienting
the upper limb. Without direct control at the sternoclavicular
joint, it is not possible to prevent compensatory movements,
nor is there a means to properly regain strength and coor-
dination of the shoulder girdle. It seems that a next logical
step in assessing the usefulness of robotic technology in a
rehabilitation setting is to develop a robot that can more
closely match the mobility of the human shoulder.
II. DESIGN OBJECTIVES
There is no current robotic device that can independently
control all five DOF at the shoulder complex. As such,
MEDARM (Motorized Exoskelton Device for Advanced
Rehabilitation of Motor function) has been designed with
shoulder complex control as its fundamental goal. It is
intended to be used for assessment in addition to therapy.
Figure 1 shows MEDARM for the right upper limb. The
design objectives have been described previously [1], but the
key goals can be summarized as follows.
• Independent control of five DOF of the shoulder com-
plex (2DOF at the sternoclavicular joint, 3DOF at the
glenohumeral joint), and 1DOF at the elbow, with a
workspace similar to a typical upper limb workspace.
• Actuators and exoskeleton sized for users from 1.4 m
to 2.0 m in height, and weighing up to 115 kg.
• High backdriveability, low mass and inertia to minimize
influence on natural motion.
• Allow simplification of the mechanism to 2DOF shoul-
der/elbow motion in any plane.
• Avoid singular configurations across the workspace.
• Maximize manipulability across the workspace.
• Safe and comfortable for users with motor impairments.
• Quick and easy to set up a patient.
III. MEDARM SPECIFICATIONS
MEDARM is an exoskeleton. An exoskeleton design was
chosen because it is the only way to independently control
all DOF in the shoulder complex, otherwise the redundancy
of the joints would make it impossible to isolate motion of
the glenohumeral joint from motion of the shoulder girdle.
Another advantage of the exoskeleton design is that grasping
a handle is not necessary, leaving the hand available for
Fig. 1. MEDARM system consists of a 6DOF robotic exoskeleton mounted
onto a support structure. The motors and electronics are mounted underneath
the robot, and external gravity compensation is provided by a motorized
vertical cabling system. A movable chair is used to bring the user into
alignment with the system.
functional training. The design phase of the MEDARM
project is now complete, as shown in Figure 1. The robot
is mounted to a support structure, and the user is wheeled
into position using a movable chair. There is space for the
operator to get beside the exoskeleton during the set-up
procedure. MEDARM consists of two main subsystems: the
shoulder/elbow mechanism (4DOF to move the upper arm
and forearm), and the shoulder girdle mechanism (2DOF
to move the glenohumeral joint relative to the torso). The
following describes the technical details these subsystems.
A. Shoulder/Elbow Mechanism
The shoulder/elbow mechanism (Figure 2a) is a 4DOF
mechanism consisting of a 3DOF spherical joint centred at
the user’s glenohumeral joint and a single rotary joint at the
elbow. It is actuated entirely by a cable-drive transmission
which is powered by five electric motors located on the
base of the system. The overall gear ratio for each of the
four joints is 8 so that the robot maintains backdriveability
without the need for force/torque sensors. The lightweight
mechanism is attached to the lateral side of the user’s arm
using two adjustable inflatable arm cuffs, which are the only
points of physical attachment to the user.
1) Joint Axis Orientation: Glenohumeral motion is
achieved using a spherical joint made from three intersecting
revolute joint axes. A problem with spherical joints is that it
is always possible to reach a singular configuration (where
one DOF is lost) by rotating the second joint so that the three
axes become coplanar. The order and relative orientation of
these three axes was optimized to ensure that the system
does not reach singularity within the user’s workspace (as
specified in [1]), that manipulability is maximized, and that
collision with the user or itself does not occur over the entire
workspace. The optimization process is described below.
1-4244-1264-1/07/$25.00 ©2007 IEEE

b
a
α
β
Glenohumeral
Joint #1 (
θ
1
)
Glenohumeral
Joint #3 (
θ
3
)
Elbow
Joint (
θ
4
)
Glenohumeral
Joint Centre
Glenohumeral
Joint #2 (
θ
2
)
Link 1
Link 2
Link 3 Link 4
s
3
s
5
s
2
s
4
s
1
r
1
r
2
r
3
r
4
ξ
1
θ
1
τ
1
θ
2
τ
2
θ
3
τ
3
θ
4
τ
4
ξ
4
ξ
2
ξ
5
ξ
3
Glenohumeral
Joint #1
Glenohumeral
Joint #2
Glenohumeral
Joint #3
Elbow
Joint
Fig. 2. The shoulder/elbow mechanism is cable-driven. (a) A CAD drawing
of the mechanism showing the final joint orientations and cable system. (b)
A simplified planar schematic representation of the optimal cable routing
structure. Each of the five cables is denoted by a different colour. Each joint
has a separate pulley for each cable that passes by the joint. Symbols s,
ξ, r, τ and θ represent cable displacement, cable force, pulley radius, joint
torque and joint angle respectively.
The first step in choosing the orientations of the axes
was to reduce the number of possible configurations by
considering the design objectives. In order for the last two
joints of the exoskeleton to operate in planar mode, it was
necessary to make the last (third) joint axis in the spherical
joint parallel to the elbow joint axis (see Figure 2a). With the
third joint axis orientation chosen, it was straightforward to
determine that the second joint axis should be perpendicular
to the third axis (and in the horizontal plane) in order to
avoid singularities in the workspace. This configuration also
has the added benefit of allowing basic flexion/extension or
adduction/abduction motions to be controlled using a single
joint axis.
To determine the optimal first joint axis orientation, it was
necessary to develop a simple iterative procedure to calculate
the box product, M, at each configuration in the workspace.
The box product is defined as:
M = z
1
× (z
2
× z
3
) (1)
where z
i
are the unit vectors corresponding to the joint
axes. When M = 1, the joint axes are orthogonal and
manipulability is maximized. When M = 0, the joint axis
are coplanar and a degree of freedom is lost (i.e. singular
y
z
x
a
c
b
1.0
0.75
0.5
y-direction
0.0
-1.0
0.2
-0.8
0.25
0.4
x-direction
-0.6
z-direction
0.6
-0.4
0.8
-0.2
1.0
0.0
0.0
right glenohumeral
joi
nt centre
d
α=45
o
β=40
o
00.5
1
x-direction
–1
–0.5
0
0.5
1
z-direction
Manipulability
Singular
Abduction
Angle
M
max
θ
2(M=0)
90
180
0
1
Abduction Angle (deg)
Box Product Maximum
Singular Abduction
Angle (
θθ
2
(M=0)
)
200 40 60 80
αα (deg) αα (deg)
0
20
40
60
80
ββ (deg)
ββ (deg)
140
o
120
o
Maximum
Manipulability (M
max
)
200 40 60 80
0
20
40
60
80
0.8
0.95
Fig. 3. (a) The coordinate frame used for joint orientation optimization cal-
culations. The octant shaded in green approximates the humeral workspace.
(b) A plot of M for a given combination of α and β as θ
2
is varied to obtain
the singular abduction angle (θ
2(M =0)
) and the maximum manipulability
(M
max
). (c) Plots of θ
2(M =0)
and M
max
for all combinations of α and
β. The range of α and β combinations that provides a suitable compromise
between θ
2(M =0)
and M
max
is shown by contour lines, and the overlap
is highlighted. (d) M
max
plotted radially over the workspace. Points closer
to the origin are configurations that are closer to singularity.
configuration). The procedure is summarized as follows, and
step-by-step results are shown in Figure 3:
1) The orientation of the first joint axis was defined rela-
tive to the second joint axis in terms of two variables
(α and β, as shown in Figure 2a).
2) With θ
1
and θ
3
fixed, manipulability (M) was calcu-
lated for a combination of α and β as θ
2
was varied
(corresponds to abduction, as shown in Figure 3b).
3) The singular abduction angle (θ
2(M =0)
) and the maxi-
mum manipulability (M
max
) were calculated and plot-
ted for all combinations of α and β (Figure 3c).
4) A range of α and β combinations that reached a
compromise between high M
max
and large θ
2(M =0)
was revealed (i.e. M
max
> 0.8 and θ
2(M =0)
> 120
◦
).
The following iterative procedure was then used to
select a combination of α and β within this range.
5) M was calculated for the spherical joint workspace
for a given α and β (all three joints varied across their
ranges of motion, as shown in Figure 3d).
6) If any points were within 15
◦
of singularity (M <
0.3) or if the exoskeleton could collide with the user,
1-4244-1264-1/07/$25.00 ©2007 IEEE

the process was repeated from step 5 using a different
combination of α and β.
7) The process was repeated until there were no singular-
ities in the workspace, and the manipulability was as
high as possible over the workspace.
The angles that provided the optimal spherical joint axes
arrangement are α = 45
◦
and β = 40
◦
. The maximum ma-
nipulability is 0.85 and averages 0.75 across the workspace.
2) Cable-Drive System: The joints of the shoulder/elbow
mechanism are actuated by electric motors with an open-
ended cable drive transmission. Cable-driven mechanisms
have a high power-to-weight ratio because all the motors can
be placed on the fixed base of the system. This significantly
reduces the size, mass and inertial properties of the robot, and
helps to reduce the motor torque output requirements. Open-
ended cable systems distribute loads across several cables,
which also reduces the the actuator requirements. Overall
the mechanism becomes more transparent to the user.
Additional transformations are required to control open-
ended cable-drive systems [17]. This is because the number
of joints needing control (n) is less than the number of
actuators (m). Cable systems can apply force through tension
only, so it is necessary to have an antagonistic cable routing
scheme for motion capability in both rotational directions
at each joint. As such, a minimum of n + 1 cables are
necessary for complete control of n joints. It is necessary
to have a positive tension in all cables at all times to prevent
the cables from becoming slack. Furthermore, since the
cables are routed along the entire length of the mechanism
through a series of pulleys, their motion affects multiple
joints, allowing loads at the joints to be distributed among
the actuators. Ultimately, joint angles and torques are related
to the length displacement and the forces in the cables.
The choice of cable routing scheme has a significant effect
on the performance of the device. In fact, for this 4DOF
system (n = 4) actuated by five motors (m = 5), there are
11 possible unique cable routing structures, all of which can
be described in matrix form. This structure matrix can be
analyzed to obtain quantitative measures related to efficiency
and actuator torque requirements [10]. Figure 2b illustrates
the optimal routing scheme which has minimized antagonism
between cables and hence the most even distribution of forces
across the cables, and also has the lowest peak forces.
3) Joint Design: Each joint of the exoskeleton requires a
low friction bearing system that provides rigidity against all
forces and non-axial moments. In addition to withstanding
non-axial gravitational and inertial moments during motion,
the joints must withstand substantial non-axial moments
resulting from forces applied by the cables and pulleys. Four-
point contact bearings are highly resistant to these moments
and therefore need not be used in pairs. Use of four-point
contact bearings in MEDARM has resulted in a thin and
lightweight exoskeleton.
B. Shoulder Girdle Mechanism
The shoulder girdle mechanism (Figure 4) provides 2DOF
about the sternoclavicular joint centre: elevation/depression
Sternoclavicular
Joint #2
Sternoclavicular
Joint #1
Sternoclavicular Joint Centre
Glenohumeral Joint Centre
Linear
Adjustment
Fig. 4. A CAD drawing of the shoulder girdle mechanism. The two joint
axes intersect at the user’s sternoclavicular joint. The second joint is a
translation along a curved track, producing a rotation about the vertical axis.
The joint is driven by a hinged linkage system. A single linear adjustment
shifts the entire shoulder/elbow mechanism to align the spherical joint with
the user’s glenohumeral joint centre.
and protraction/retraction. The entire mechanism is located
behind the user, and there is an adjustment to account for
users of different size. The mechanism supports the complete
shoulder/elbow system including the user’s arm, and as a
result must be structurally strong.
The first joint axis is fixed to the base structure behind
the user, with its axis pointing forward in the horizontal
plane. It is a conventional rotary joint that provides eleva-
tion/depression motion. The second joint axis is vertically
aligned, and intersects the first joint axis through the user’s
sternoclavicular joint centre, allowing protraction/retraction
motion. It is not a typical rotary joint; it is a curved track
system on which a carriage travels. The low-friction carriage
supports the entire cable drive system and is driven by a
hinged linkage system. The resulting mechanism operates
like a 4-bar linkage without requiring any structural elements
near the user’s sternoclavicular joint (see Figure 4). Both
joints are driven by electric motors with timing belts, and
operate with gear ratios of 5 and 6.25 respectively.
The benefits of this track system are significant. First,
it facilitates placing equipment behind the user rather than
above their head, which is safer and more comfortable for
the user, and also easier for the operator to set up. Second,
the hinged driving linkage doubles as a routing system for
the cables from the shoulder/elbow mechanism by guiding
them through to the base of the robot without any non-linear
changes in cable length. Any change in cable length as a
result of shoulder girdle motion is easily accounted for in
the cable length transformations.
The weight of this mechanism is substantial, and puts
high static torque requirements on the first shoulder girdle
joint. To assist the motor at this joint, an external gravity
compensation system is employed. A vertical motorized
cable is mounted directly above the end of the curved track
1-4244-1264-1/07/$25.00 ©2007 IEEE

(see Figure 1), and applies a vertical force on the track to
offset the gravitational forces on the MEDARM.
C. User Attachment and Alignment
To function correctly, the exoskeleton must be aligned with
the sternoclavicular and glenohumeral joints of the user, and
adjusted to fit arms of different lengths. A harness attaches
the user’s torso to a moveable chair which provides the
three translational adjustments necessary to align the user’s
sternoclavicular joint with the mechanism’s fixed shoulder
girdle joint centre. Once aligned, the chair is locked to the
main structure.
The mechanism must next be aligned with the user’s
glenohumeral joint centre. As before, three spatial adjust-
ments are required. However, this can be achieved by a
single manual linear adjustment because the redundancy of
the exoskeleton structure can be used to make the remaining
alignments. This linear adjustment shifts the cable-drive
system relative to the carriage in the direction approximately
aligned with the horizontal projection of the clavicle (see Fig-
ure 4) and is then clamped to the carriage. Thus modifying
the position of the mechanism’s glenohumeral joint centre
is achieved through the linear adjustment and the 2DOF
provided by the shoulder girdle. The 3DOF spherical joint
of the shoulder/elbow mechanism automatically compensates
by rotating until the mechanism is properly aligned with the
user’s limb.
This adjustment scheme has the benefit of simplifying
the structure of the shoulder girdle joint, and also the
set-up procedure. Otherwise, three consecutive translational
adjustments would be required, making the system sig-
nificantly larger, heavier and more complicated. Relying
on the shoulder/elbow mechanism to compensate for the
adjustment tends to push the shoulder joint away from its
optimal configuration, decreasing the range of motion of
the mechanism in some directions. However, the adjustment
range is typically small (2
◦
or 3
◦
at most), so the singularities
and manipulability of the mechanism will not be significantly
altered. Another issue that arises when adjusting a cable-
drive system is that it is necessary to maintain tension in the
cables at all times. Adjusting the link length must not change
the cable length, otherwise tension would be lost. Routing
the cables along the hinged driving linkage ensures that the
cable length does not change and that tension is maintained.
The exoskeleton system attaches to the user in two places:
the upper arm and the forearm (Figure 5a). These attach-
ments are to keep the exoskeleton aligned with the limb at
all times. The proposed design is to strap the limb into a rigid
half-cylindrical trough using an inflatable Velcro strap similar
to a blood pressure cuff. Once strapped in, the cuff is inflated
to provide a secure fit that is customized to the user. On the
lateral side of the cuffs is a single rigid connection to the
exoskeleton structure. The cuff will be attached to the subject
before connecting to the exoskeleton, which is easier for the
operator, and more comfortable for the user. An important
difference from many previous arm cuff designs is that the
arm cuff does not have a fixed size cuff through which the
a
b
Slider
Quick
Release
Clamp
Upper
Arm
Cuff
Forearm
Cuff
Fig. 5. (a) A CAD drawing illustrating the arm cuff attachments and
adjustments. Each cuff has two translational adjustments to correctly align
the limb segments relative to the mechanism structure: perpendicular to
the link (small arrows) and parallel to the link (hollow arrows). A fifth
adjustment (large arrow) moves the location of the elbow joint to change
the length of the upper limb link. (b) A close-up of the cuff attachment,
showing the quick-release clamp.
user must put their arm. This allows simpler set up, and also
is compatible with a larger variety of arm sizes.
A total of five adjustments are required to ensure that
the user’s arm is properly aligned with the exoskeleton (see
Figure 5a). Each cuff is adjustable along the length of the
exoskeleton (for limbs of different length) and perpendicular
to the exoskeleton (for limbs of different width). The cuff is
attached by inserting it into a slider which can move freely
along the exoskeleton. A single quick-release clamp (similar
to those used to clamp bicycle components) simultaneously
clamps the cuff to the slider and the slider to the exoskeleton
(see Figure 5b). To accommodate users with different arm
lengths, a similar slider and clamp is used to locate the elbow
joint along the upper arm link. A passive hinged guide was
added to the upper arm link of the robot to ensure that tension
is not lost when adjusting the arm length.
Exoskeleton type devices always require more set up time
than their end-effector type counterparts. However, given its
mobility and adjustability, MEDARM has a relatively simple
set up procedure. In fact, once the chair is locked in place,
only four clamps are required to secure all eight adjustments.
This will keep set up time to a minimum, allowing the user
to receive a longer therapy session.
IV. DYNAMIC MODEL AND SIMULATION
To make appropriate choices for the eight electric motors
required to actuate MEDARM, a dynamic model of the ex-
oskeleton and the human limb has been created in MATLAB
based on the robot toolbox [4]. The model was also used
to specify a number of other design parameters including
bearing strength, joint gear ratios, and cable load capacity.
The model takes the form of a standard rigid-body ma-
nipulator, and assumes that the cable dynamics are not
1-4244-1264-1/07/$25.00 ©2007 IEEE

TABLE I
MAXIMUM TORQUE OUTPUT FOR EACH JOINT.
Motor Static Torque (Nm) Peak Torque (Nm)
Shoulder Girdle #1 ±24 ±73
Shoulder Girdle #2 ±30 ±91
Glenohumeral #1 +39, −26 ±60
Glenohumeral #2 +39, −26 ±60
Glenohumeral #3 +39, −26 ±60
Elbow ±13 +40, −30
significant. Dynamic parameters of the exoskeleton including
lengths, masses and inertial properties are estimates from
CAD drawings. The same properties of the human upper
limb were calculated from anthropometric data tables based
on user height and weight [18] and are fully integrated
into the model. The model was adapted to account for the
external gravity compensation system, and includes estimates
of viscous and static friction. Given a trajectory for each
joint, the model calculates the joint torques required to
achieve that motion. The cable forces required to generate
these joint torques are then calculated using the torque
resolver technique [10]. The final output is the torque outputs
for all eight motors, the force in each cable, and all forces
and non-axial moments at each joint.
To get an estimate of the peak dynamic motor torques
for non-contact applications, the model was used to simulate
various reaching movements with a peak end-point velocity
of 1.0 m/s. Anthropometric limb measurements were chosen
to meet the maximum design requirements. Movements
included single joint movements through the full range of
motion of each joint, and a range of typical multi-joint reach-
ing movements such as reaching towards the face or chest
from a relaxed position. The most demanding positions for
the exoskeleton system in terms of static torque requirements
are those in which the arm is raised to the horizontal plane
with the elbow fully extended. The gravitational component
of the joint torques is the most significant contribution,
and produces the largest stresses on the motors in static
situations, so each position was held for one second to
facilitate measurements of peak static torque.
Motors and gear ratios were selected based on the results
of these simulations. The motors have built-in high resolution
encoders capable of measuring joint angle in increments of
0.003
◦
. Each motor also incorporates an electric brake to
guarantee that the mechanism will not collapse during a
power failure. The brakes also ensure that the cables remain
in tension when the power is turned off. The simulations
also enabled selection of a braided stainless steel cable of
appropriate size, and also joint bearings with sufficient load
capabilities. The overall torque capabilities of each joint of
the exoskeleton are shown in Table I, and are a result of the
limits of both the motors and the cable strength.
V. CONCLUSIONS AND FUTURE WORK
This paper describes the design of a new robot for assess-
ment and rehabilitation of upper-limb motor function after
stroke. MEDARM is designed to be a versatile machine with
the potential to improve motor function of the proximal upper
limb. It is hoped that MEDARM will assist in prevention of
compensatory movements, while encouraging more natural
coordination for stroke patients of any degree of motor
impairment. The robot provides a large range of adjustments,
so it can easily accommodate users of varying shape and
size. MEDARM’s main advantage, however, is that it can
independently monitor and control all five of the main DOF
of the shoulder complex.
Efforts are currently focused on construction and testing
of a simpler 3DOF planar version of MEDARM to test out
the main principles of operation of the MEDARM design
including the curved track and cable-drive transmission.
REFERENCES
[1] S. J. Ball, I. E. Brown, and S. H. Scott, “Designing a robotic exoskele-
ton for shoulder complex rehabilitation,” in 30th Canadian Medical
and Biological Engineering Conference (CMBEC30), Submitted for
Publication, 2007.
[2] C. G. Burgar, P. S. Lum, P. C. Shor, and H. F. Machiel Van der
Loos, “Development of robots for rehabilitation therapy: the Palo Alto
VA/Stanford experience,” J. Rehabil. Res. Dev., vol. 37, no. 6, pp. 663–
673, 2000.
[3] M. C. Cirstea and M. F. Levin, “Compensatory strategies for reaching
in stroke,” Brain, vol. 123, no. 5, pp. 940–953, 2000.
[4] P. I. Corke, “A robotics toolbox for MATLAB,” IEEE Robot. Automat.
Mag., vol. 3, no. 1, pp. 24–32, 1996.
[5] B. H. Dobkin, “Strategies for stroke rehabilitation,” Lancet Neurol.,
vol. 3, no. 9, pp. 528–536, 2004.
[6] Heart and Stroke Foundation of Canada. (2007) Stroke statistics.
[Online]. Available: http://www.heartandstroke.ca
[7] S. Hesse, H. Schmidt, C. Werner, and A. Bardeleben, “Upper and
lower extremity robotic devices for rehabilitation and for studying
motor control,” Curr. Opin. Neurol., vol. 16, no. 6, pp. 705–710, 2003.
[8] G. R. Johnson, D. A. Carus, G. Parrini, M. S. Scattareggia, and
R. Valeggi, “The design of a five-degree-of-freedom powered orthosis
for the upper limb,” Proc. Inst. Mech. Eng [H.], vol. 215, no. 3, pp.
275–284, 2001.
[9] H. I. Krebs, N. Hogan, M. L. Aisen, and B. T. Volpe, “Robot-aided
neurorehabilitation,” IEEE Trans. Rehabil. Eng, vol. 6, no. 1, pp. 75–
87, 1998.
[10] J. J. Lee, “Tendon-driven manipulators: Analysis, synthesis, and con-
trol,” Ph.D. Thesis, University of Maryland, 1991.
[11] R. Loureiro, F. Amirabdollahian, M. Topping, B. Driessen, and
W. Harwin, “Upper limb robot mediated stroke therapy - GENTLE/s
approach,” Autonomous Robots, vol. 15, no. 1, pp. 35–51, 2003.
[12] S. M. Michaelsen, R. Dannenbaum, and M. F. Levin, “Task-specific
training with trunk restraint on arm recovery in stroke: randomized
control trial,” Stroke, vol. 37, no. 1, pp. 186–192, 2006.
[13] M. Mihelj, T. Nef, and R. Riener, “ARMin - toward a six DoF upper
limb rehabilitation robot,” in Proc. IEEE/RAS-EMBS International
Conference on Biomedical Robots and Biomechatronics (BIOROB’06),
2006, pp. 1154–1159.
[14] D. J. Reinkensmeyer, J. L. Emken, and S. C. Cramer, “Robotics, motor
learning, and neurologic recovery,” Annu. Rev. Biomed. Eng, vol. 6,
pp. 497–525, 2004.
[15] R. Sanchez, D. Reinkensmeyer, P. Shah, J. Liu, S. Rao, R. Smith,
S. Cramer, T. Rahman, and J. Bobrow, “Monitoring functional arm
movement for home-based therapy after stroke,” in Proc. IEEE An-
nual Conference of Engineering in Medicine and Biology Society
(EMBC’04), vol. 2, 2004, pp. 4787–4790.
[16] R. Teasell, N. Foley, J. Jutai, T. Doherty, J. Bitensky, S. Bhogal,
and M. Speechley, “Stroke rehabiliation evidence-based review: Upper
extremity interventions,” Monograph on the Internet, 2003.
[17] L.-W. Tsai, Robot Analysis: The Mechanics of Serial and Parallel
Manipulators. New York, U.S.A.: John Wiley & Sons, Inc., 1999.
[18] D. A. Winter, Biomechanics and Motor Control of Human Movement,
2nd ed. New York, U.S.A: John Wiley & Sons, Inc., 1990.
1-4244-1264-1/07/$25.00 ©2007 IEEE