wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?

casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2

Sunil K. Agrawal
Department of Mechanical Engineering,
Mechanical Systems Laboratory,
Design and Optimization of a
University of Delaware,
Newark, DE 19716 Cable Driven Upper Arm
Venketesh N. Dubey
School of Design, Engineering and Computing,
Exoskeleton
Bournemouth University,
Fern Barrow, This paper outlines the design of a wearable upper arm exoskeleton that can be poten-
Poole, BH12 5BB, UK tially used to assist and train arm movements of stroke survivors or subjects with weak
musculature. In the last 10 years, a number of upper arm training devices have emerged.
However, due to their size and weight, their use is restricted to clinics and research
John J. Gangloff, Jr. laboratories. Our proposed wearable exoskeleton builds upon our research experience in
wire driven manipulators and design of rehabilitative systems. The exoskeleton consists
Elizabeth Brackbill of three main parts: (i) an inverted U-shaped cuff that rests on the shoulder, (ii) a cuff on
the upper arm, and (iii) a cuff on the forearm. Six motors mounted on the shoulder cuff
Ying Mao drive the cuffs on the upper arm and forearm with the use of cables. In order to assess the
performance of this exoskeleton prior to use on humans, a laboratory test-bed has been
Vivek Sangwan developed where this exoskeleton is mounted on a model skeleton, instrumented with
sensors to measure joint angles. This paper describes the design details of the exoskel-
Department of Mechanical Engineering, eton and addresses the key issue of parameter optimization to achieve a useful workspace
Mechanical Systems Laboratory, based on kinematic and kinetic models. The optimization results have also been motivated
University of Delaware, from activities of daily living. 关DOI: 10.1115/1.3191724兴
Newark, DE 19716

1 Introduction A vast number of people are affected by arm conditions due to

degeneration of muscles that result in profound muscle weakness
With improvements in health care and standard of living, life
or impaired motor control such as, people suffering from muscular
expectancy has dramatically increased, and people are living
diseases like spinal muscular atrophy 共SMA兲, Duchenne muscular
longer. Further, due to busy lifestyle and changing economic situ- dystrophy 共DMD兲, and limb girdle. They have problems in lifting
ations, birth rates have markedly slowed down. This has caused the arms against gravity. Consequently, they are not able to per-
noticeable shifts in aging population in many demographic loca- form various ADL independently 关3兴. External mechanical support
tions. According to current statistics from the Department of can provide assistance to the movement of the arm when adequate
Health and Human Services of the United States, the older popu- motor control is lacking. This would greatly improve the handling
lation, persons 65 years or older, numbered 37.3 million in 2006 capability in people with functional deficits. However, none of the
and will be about 71.5 million by the year 2030 关1兴. In a statistics existing exoskeleton designs have the potential to be lightweight
from the United Kingdom, it has been projected that population of and easy to be worn continuously. Some well-known designs to
older people above 60 years in certain locations, e.g., South West evaluate and treat upper extremity functions for post-stroke reha-
regions, could reach up to one-third in 20 years compared with bilitation of patients employing robotic technology are ARM 关4兴,
just under one-quarter today, and in some regions, the proportion MIME 关5兴, and MIT-MANUS 关6兴. However, none of these were
of older people could even reach 45% or more 关2兴. This has led to meant to be portable support devices. In recent years, there has
a spurt of interest in helping the quality of life of growing aging been a continuous surge of interest in developing upper limb ex-
population. oskeleton for training and assistance. New developments of such
One of the essential requirements of older people is to meet devices are keenly pursued. A 7DOF powered exoskeleton was
their activities of daily living 共ADL兲. However, at this age, they developed as a therapeutic and diagnostics device for human
suffer from a number of bodily weaknesses and disabilities. In the power amplifications 关7兴. However, this device has a rigid me-
U.S. alone, 7.7 million people are living after stroke, and 700,000 chanical structure, which can only lend itself for training pur-
people experience a stroke each year. The functional capability of poses. Wearability and continuous use as a support device is not
this group can be enhanced by augmenting their arms with exter- possible; most existing/recent exoskeletons fall under this cat-
nal mechanical supports in the form of exoskeletons, which can be egory 关8–11兴. To make exoskeletons lighter and wearable, new
used for training, as well as assistance. However, for regular use, designs are reported that are cable based; however, the final sys-
these devices need to be lightweight, wearable, and should pro- tems still appear to be bulky with actuators and controllers
vide anatomically consistent motions. This condition precludes 关12,13兴; other designs have been limited to modeling and simula-
usage of many existing upper limb exoskeletons since they be- tion 关14,15兴.
Our initial design, shown in SOLIDWORKS™ 共Fig. 1兲, is inspired
come bulky in their design. This paper presents a design of wear-
by the natural arm with the goal to make it lightweight and wear-
able, cable driven exoskeleton that can be worn and operated in a
able. A 4DOF arm models the shoulder and elbow motions. The
natural way and considers parameter optimization of the exoskel-
exoskeleton has cuffs on the upper and forearm to which the
eton to maximize the operating workspace based on its kinematic
cables are connected. An inverted U-shaped shoulder cuff is
model. mounted over the scapula that holds the motors. The arm cuffs
have channel sections, which hold circular tubes that can be
Manuscript received January 6, 2009; final manuscript received May 28, 2009; inflated/deflated for sufficient grip on the contact parts. The ex-
published online August 31, 2009. Review conducted by Just Herder. oskeleton uses six small motors, four for the shoulder joint and

Journal of Medical Devices Copyright © 2009 by ASME SEPTEMBER 2009, Vol. 3 / 031004-1
 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 1 Model of the exoskeleton on a skeleton arm

two for the elbow. A minimum of four cables is required to table, and 共iii兲 it can be optimized for global tasks, as well as
achieve three degrees of freedom motion at the shoulder 关16兴. The specific tasks pertaining to ADLs. On the flip side, it is difficult to
skeleton has an encoder at each rotational joint axis to record ensure positive tension in the cables over its workspace. That
motion during experimentation. provides challenging scientific questions, which will be studied in
This paper is organized as follows. Section 2 presents the de- future research.
sign details of the exoskeleton, followed by the kinematic model Due to the nature of cable driven systems, only pulling forces
of the exoskeleton in Sec. 3. Details of optimization technique are on the cables are allowed. Hence, it is necessary to keep the de-
presented in Sec. 4. Laboratory experiments and design evaluation sign modular to make adjustments in motor mounting and cable
are given in Sec. 5, followed by discussion and conclusions in attachment points to optimize the design. In order to achieve this,
Sec. 6. the shoulder cuff has been designed with tracks and groves to
permit motor placement at different radial 共r兲 as well as angular
2 Details of the Design positions 共⌿兲, as shown in Fig. 3. In addition, due to the size of
An experimental prototype was developed, based on the above motors and physical constraints of the design to accommodate six
model, as shown in Fig. 2, which uses a plastic skeleton. As seen, motors on the shoulder cuff, radial fins have been used, which can
this uses a total of three cuffs: The first cuff sits on the shoulder, hold motors on either sides of the plate. The motors can also be
one cuff wraps around the upper arm, and the third one goes onto flipped by 90 deg on the radial fins to further allow adjustments on
the forearm. The arm motion is controlled by cables attached to the cable attachment points 共as shown by the first motor on the left
the cuffs that are driven by motors mounted to the top of the hand side in Fig. 3兲. Up to two motors can be placed on a fin,
shoulder cuff. The prototype has been machined from aluminum. which provides flexibility in the attachment. A design question is
One of the main objectives of the experimental exoskeleton is to how to optimize these parameters for maximizing the workspace.
track the angular position and velocity of the arm while executing
motion sequence. The natural ball-and-socket joint of the shoulder 3 Exoskeleton Kinematic Model
was replaced by a shoulder joint, which has three axes of rotation, The kinematic parameters of the exoskeleton arm are shown in
to which joint encoders were fixed. Fig. 4 and the DH-parameters are listed in Table 1. The exoskel-
The novel features of our exoskeleton design are as follows: 共i兲 eton has four links and it provides 4DOF 共3 at the shoulder and 1
It is modular and can be easily adjusted with optimized param- at the elbow兲. As seen in the figure, Body 1 is the cross bar
eters, 共ii兲 it is lightweight, cable driven, self-contained, and por- holding the arm, Body 2 is the U-link connecting the cross bar to

Fig. 2 Exoskeleton fitted on a plastic arm with cable driven

cuffs Fig. 3 Shoulder cuff design with adjustable motor placement

031004-2 / Vol. 3, SEPTEMBER 2009 Transactions of the ASME

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 4 Parameters for kinematic modeling

the arm, and Body 3 and Body 4 are the upper and lower arms,
respectively. The axes and sense of rotation of the joints in initial
configuration of the DH-model are shown in Fig. 5.

4 Workspace Optimization
The workspace characterization of this cable driven system is
essential to investigate 关17,18兴. Ideally, the exoskeleton should be
able to perform reaching tasks under gravity and inertial loads of
the arm. However, due to unilateral property of cables to only
pull, it can only achieve a subset of all reachable locations. In
other words, the exoskeleton will have a static and dynamic work-
space smaller than the reachable workspace. The dynamic model
of the exoskeleton was developed using the Lagrangian formula-
tion
D共q兲q̈ + C共q,q̇兲q̇ + g共q兲 = J共q兲TT共t兲 共1兲
where q = 共␪1 , ␪2 , ␪3 , ␪4兲T are the generalized coordinates, D共q兲 is
the 共4 ⫻ 4兲 inertia matrix, C共q , q̇兲 is the vector of nonlinear cen-
tripetal terms, g共q兲 is the vector of gravity terms, J共q兲 is the Jaco-
bian relating the cable attachment points, and T共t兲 is a six dimen- Fig. 6 Flow chart of optimization process; the whole arm is
split into shoulder and elbow joint cables. Initially, positive ten-
sional cable tension vector. Due to complexity of the dynamical
sion is ensured in the elbow cables and then later at the
model, we only provide a functional form of the equation here. shoulder.
Detailed models were developed in MAPLE™ and MATLAB®, and
are being used in MATLAB for numerical simulation, and have the
form given in Eq. 共1兲.
This model was used for static workspace evaluation by setting
space. Later, this technique was extended for whole arm param-
q̇ = 0 and q̈ = 0. In the first step, this model was used for shoulder eter optimization. The above equation, in statics, can be written in
workspace evaluation. Parameter optimization was performed to the general form
characterize its useful workspace, i.e., the tension in the cable
remains positive for the range of motion within the useful work- AT = B 共2兲
where
Table 1 DH-parameters of the exoskeleton arm A = 关a1,a2, . . . ,a6兴 苸 R4⫻6
Here, A = J共q兲T and B = g共q兲. B = 关V1 , V2 , . . . , V4兴T 苸 R4⫻1 is the
Link/Parameters ai ␣i di ␪i vector of gravity terms.
1 0 90 0 ␪1
T = 关T1 , T2 , . . . , T6兴T 苸 R6⫻1 is the cable tension vector, and the
2 0 90 0 ␪2 solution for cable tension is given by
3 0 90 d3 ␪3
4 a4 0 0 ␪4 T = T̄B + N共A兲m̄ 共3兲
where T̄ = A 共AA 兲 is the pseudoinverse of matrix A, N共A兲 is the
T T −1

null space of matrix A, and m̄ modulates tension in cables 关19兴.

To simplify the workspace computation of the whole arm, Eq.
共2兲 is split into its component shoulder and elbow matrices, as
shown in Eq. 共4兲. The elbow degree-of-freedom represented by
Eq. 共5兲, uses the last component equation of Eq. 共2兲. Initially,
positive tension is ensured in the elbow cables using Eq. 共5兲; this
gives elbow cable tension 共T5 , T6兲. Then, Eq. 共6兲 is used to ensure
positive tension in the cables for the shoulder joint, as shown in
Eq. 共7兲. If positive tension is not possible for the elbow cables, it
records this as an infeasible arm position, and the program moves
Fig. 5 Zero configuration and axis of rotation „a… as viewed on to a new test point. The whole arm optimization process is
from the back and „b… from the side of a human model shown in the flow chart 共Fig. 6兲. This procedure has been used

Journal of Medical Devices SEPTEMBER 2009, Vol. 3 / 031004-3

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 7 A sketch of a feasible region for m for a 3DOF shoulder
joint driven by four cables

throughout to characterize the workspace and for exoskeleton pa-

rameter optimization
A ST S 艛 A ET E = B 共4兲
where AS 苸 R , 3⫻4
TS共T1 , . . . , T4兲 苸 R 4⫻1
, AE 苸 R 4⫻2
, and
TE共T5 , T6兲 苸 R2⫻1.
A共4,5兲T5 + A共4,6兲T6 = B共4兲 共5兲 Fig. 8 Cable attachment points on shoulder and upper arm
cuffs with optimization parameters as shown with respect to
ASTS = 共B − AETE兲 = Bs 共6兲 the local coordinate frame and the sense of rotation on the
plane of the cuffs
TS = T̄SBS + N共AS兲m 共7兲
where T̄S is the pseudoinverse of AS, and BS = 共B − AETE兲 苸 R3⫻1.
cuff. These are given by ⌿1 = 关30, 60, −60, −30兴. ⌿2 is the corre-
4.1 Shoulder Parameter Optimization. For the shoulder sponding angular orientation of the cable attachment points on the
joint alone, Eq. 共2兲 has a smaller size, i.e., A is a 共3 ⫻ 4兲 matrix, B upper arm cuff given by ⌿2 = 关135, 45, −45, −135兴. The cuff’s ra-
is a 共3 ⫻ 1兲 vector, and T is a 共4 ⫻ 1兲 vector. Since A is nonlinear, dii are r0 = 0.1 m and r1 = 0.05 m. The shoulder cuff is placed at
it is hard to get analytical solution for positive cable tension using the reference plane 共i.e., no shift along the y 0 axis兲, and the upper
the pseudoinverse. Hence, this is done numerically by modulating arm cuff is located at 0.15 m along the arm. With these settings,
the null space parameter m. The feasible solution of m is charac- the model calculates the cable tension and, if negative, attempts to
terized by a convex region bounded by four linear inequalities on make it positive using the modulator m in Eq. 共8兲. Figure 9共a兲
parameter m. With one extra cable for a 3DOF joint, the four shows the qualitative workspace for a unit arm connected to the
linear inequalities in m are given by shoulder joint, higher the color map order, better in the workspace

冤冥冤 冥
feasibility. The workspace is hemispherical about the shoulder
T̄1 n11 joint 共at the origin of the coordinate frame兲, and the corresponding
feasible point count 共out of 8000 possible points兲 is shown in Fig.
T̄2 n21 9共b兲. The gray dots are the feasible solutions, and the black dots
+ mⱖ0 共8兲
T̄3 n31 are infeasible. Clearly the workspace is small and discontinuous at
this setting.
n41
T̄4 In order to maximize the feasible workspace, the model formu-
lation is subjected to the MATLAB fmincon optimization function,
The feasible region FA of m is described by the common interval
typically to solve nonlinear programming problems such as com-
bounded by four linear inequalities, as shown in Fig. 7. Here pi is
putation of continuous trajectories for control 关19兴, with the fol-
the solution point when each component of Eq. 共8兲 is an equality.
lowing lower and upper bounds of shoulder angular positions. The
If FA is empty, the tension constraints cannot be met 关19兴.
lower and upper bounds 共LB and UB兲 of a sample attachment
As discussed earlier, there are several variables in the shoulder
point on the shoulder cuff is shown in Fig. 10. Only angular
cuff design that may be considered for maximizing the workspace
positions of cable attachment points were considered for optimi-
keeping the cable tension positive for the range of arm motion.
zation, this helps in the fast convergence of the optimization pro-
However, some of these parameters cannot be changed due to
cess, as well as facilitates flexibility in the physical assembly of
physical limitations of the design. Most problems of constrained
the components of the exoskeleton. The objective function is to
nonlinear multivariable optimization benefit from good starting
minimize the number of infeasible points in each run
guesses, simpler cost functions, and less stringent termination
criteria—this improves the execution efficiency and can help lo- ⌿1_lower = 关15,45,− 75,− 45兴
cate local minima. With this in mind, only the angular spacing
between the attachment points on the shoulder and upper arm ⌿1_upper = 关45,75,− 45,− 15兴
cuffs were initially considered for optimization. Figure 8 shows
the initial setting of the attachment points with the reference axis ⌿2_lower = 关90,0,− 90,− 180兴
for each cuff. As can be seen, there are a large number of variables
responsible for the performance of the exoskeleton, e.g., angular ⌿2_upper = 关180,90,0,− 90兴
attachment angles on the cuffs 共⌿1 , ⌿2兲, radial positions 共r0 , r1兲, Figure 11 shows the optimized result for the above case, which
and axial positions along 共y 0兲 the arm. The following are the offers the following angles 共rounded to the closest integer兲 for
initial settings for the exoskeleton range of operation 关20兴. cable attachment points for the given joint angle range. The work-
The range of shoulder joint angles 共refer to Fig. 4兲 is as follows: space has significantly improved, and the number of feasible
−80⬍ ␪1 ⬍ 80, 10⬍ ␪2 ⬍ 170, and −60⬍ ␪3 ⬍ 60. points count has increased from 5124 to 7830 共out of 8000兲. Once
⌿1 is the angular orientation 共in degrees兲 of the cable attach- the optimized solution is obtained, the effect of change in the
ment points 共1, 2, 3, 4兲, as shown on the shoulder cuff. This is radial position of the attachment points is evaluated. It has been
with respect to the local coordinate frame and the sense of rotation found that it has a marginal effect on the workspace, slightly
starting at an angle 共⌿1 = 0兲 on the plane containing the shoulder improving with the increasing radius

031004-4 / Vol. 3, SEPTEMBER 2009 Transactions of the ASME

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 9 Feasible workspace at the initial setting: „a… qualitative „darker and better… and „b… quantitative „gray, feasible; black,
infeasible points… obtained by testing the feasibility conditions for a unit arm placed at the coordinate frame passing
through „ArmX, ArmY, and ArmZ… end positions

⌿1_optimized = 关15,75,− 46,− 15兴 ⬍ ␪4 ⬍ 90 共refer to Fig. 4兲; this adequately covers the range of
motions to carry out most ADLs. Angular orientation 共in degrees兲
⌿2_optimized = 关158,0,− 15,− 180兴 of the cable attachment points on the shoulder cuff are ⌿1
= 关30, 60, −60, −30, 75, −75兴. Due to the operational requirement
4.2 Whole Arm Optimization. Based on the formulation in of the elbow joint, the plane containing the elbow cables must be
Fig. 6, optimization for the whole arm was carried out with the
perpendicular to the joint axis at all times. As a result, angular
following initial settings. Range of joint angles on shoulder and
elbow is −70⬍ ␪1 ⬍ 70, 10⬍ ␪2 ⬍ 100, −20⬍ ␪3 ⬍ 20, and −70 attachment of elbow cables on the upper and lower arm are fixed
and not subjected to optimization. This is given 共with reference to
the local frame兲 by ⌿2_elbow = 关90, −90兴 and ⌿3 = 关180, 0兴. There-
fore, angular orientation 共in deg兲 of the cable attachment points on
the upper arm cuff are ⌿2 = 关135, 45, −45, −135兴. The cuff radius
at the shoulder, upper arm, and forearm are r0 = 0.1 m, r1
= 0.05 m, and r2 = 0.05 m, respectively, and the lengths of the
upper and forearms are 0.3 m and 0.25 m, respectively. The shoul-
der cuff is placed at the reference plane, and the upper and lower
arm cuffs are located at 0.15 m along their respective arms from
the preceding joint. Due to the design constraints, the following
are the lower and upper bound, set on the orientation of the cable
Fig. 10 Upper and lower bounds of an attachment point on the
shoulder cuff

Fig. 11 Optimized workspace due to the joint motion at the shoulder: „a… qualitative and „b… quantitative „gray, feasible;
black, infeasible points; 7830 feasible points achieved out of 8000 possible points…

Journal of Medical Devices SEPTEMBER 2009, Vol. 3 / 031004-5

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 12 Exoskeleton workspace: „a… initial setting and „b… optimized result „gray, feasible; black, infeasible points…, 8390
feasible points achieved out of 10,000 possible points

attachment points. Figure 12共a兲 shows the workspace of the ex- exoskeleton design is modular, the optimized parameters can be
oskeleton at initial setting, and the optimized workspace is shown incorporated for improved performance. In order to examine this,
in Fig. 12共b兲 simulated test points were created for picking up food from a table
top, and taking it to the eating position. This requires two end
⌿1_lower = 关15,45,− 75,− 45,60,− 90兴
positions of the arm, which are used to get angle range for shoul-
der and elbow joints. In the current implementation, the exoskel-
⌿1_upper = 关45,75,− 45,− 15,90,− 60兴
eton is a 4DOF manipulator for position control, i.e., a redundant
system with one extra degree-of-actuation. This requires one of
⌿2_lower = 关90,0,− 90,− 180兴
the joint angles to have a specified value in order to get the rest of
⌿2_upper = 关180,90,0,− 90兴 the joint angles from the inverse kinematic formulations. A typical
example considered to test this capability with the specified arm
As seen in Fig. 12, the workspace has improved significantly, and length is given below
the number of feasible point count increased from 3341 to 8390
共out of 10,000兲. The center of the workspace is located at the Pick up point coordinates = 关0.25,0.14,− 0.21兴
origin of the coordinate frame. The infeasible points 共black dots兲
are mostly seen at the top of the shoulder joint, this is intuitively Drop off point coordinates = 关0.036,0.23,− 0.012兴
agreeable since the cable driven arm will have singular configu- For the above points, the angle variation from the inverse kine-
ration at these locations. The optimization gives the following matic formulation has the following joint angle range, based on an
orientation angles 共rounded to the closest integer兲 for cable attach- interpolated straight line motion of the arm between these points:
ment points for the given joint angle range:
⌿1_optimized = 关15,75,− 75,− 45,90,− 60兴

⌿2_optimized = 关180,0,0,− 180兴

The cable attachment points are shown graphically in Fig. 13 from
initial to an optimized condition. The top picture is the initial
setting of the exoskeleton represented by the left hand line
diagram—the central red line is the extended arm, and black lines
represent the cuffs. The optimized configuration is shown in the
right hand line diagram, which can be reflected back into the
original model for the intended application. The effect of change
in radial position of the cable attachment points is further studied
on the optimization result, and contrary to the shoulder optimiza-
tion, it has been found that the workspace improves with decreas-
ing radius.

5 Design Evaluation
Looking at the optimized workspace, we observed that the
workspace is continuous within the useful range of arm operation.
This setting can be used to perform ADL such as eating from a
table top, pushing objects over a horizontal plane, or lifting from
the ground. These activities involve motion of the arm in a limited Fig. 13 Optimization results for cable attachment points from
the initial to the final configuration; the top plate shows the
range, i.e., full three-dimensional arm capability is not required model in the initial configuration represented by the left hand
for such activities. Under these conditions, if the range of the arm line diagram. The optimized arrangement is shown in the right
motion is known a priori, the system can be optimized only for hand line diagram reflected into the new model setting—the
this range. This may allow better parameter optimization of the central line is the arm in the extended position and black lines
exoskeleton for the specified range of the arm motion. Since the represent the cuffs.

031004-6 / Vol. 3, SEPTEMBER 2009 Transactions of the ASME

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 14 Workspace for ADL: „a… initial setting, „b… global optimization, and „c… ADL specific optimization

30 ⬍ ␪1 ⬍ 80, 10 ⬍ ␪2 ⬍ 30, − 10 ⬍ ␪3 ⬍ − 40, − 10 ⬍ ␪4 q̇共t兲. Using the Jacobian relation l̇共t兲 = J共q兲q̇共t兲, the cable speed
⬍ − 40 profiles l共t兲 = 关l1共t兲 , l2共t兲 , . . . , l3共t兲兴 are found, which serve as motor
speed commands after scaling.
Passing these angle range to the optimization algorithm, the fol- The experimentally obtained joint angle profiles are shown in
lowing angular attachment angles are computed, which are given Fig. 16共a兲, in solid line. The dashed lines are generated from the
below. The corresponding workspaces from initial to the global simulation of the same task by integrating Eq. 共1兲. The results
optimization, and to the specific case of ADL, are shown in Fig. show a good match between the experiment and simulation. Simi-
14. The ADL based workspace has improved and was found to be larly, Fig. 16共b兲 shows a comparison between the wrist coordi-
even better than the global optimization for the given range of nates from the experiment and that of the simulation, which also
motion show a good match. Other experiments also show similar results.
⌿1_optimized = 关20,75,− 45,− 15,75,− 60兴 Overall, the upper arm exoskeleton is capable of moving the wrist
from an initial point to a desired final point if the tension profiles
⌿2_optimized = 关180,90,0,− 180兴 are suitably determined. This is currently an open question, and
other experiment aspects are currently being considered.
A photograph of the laboratory experimental setup was shown in
Fig. 2. The control of the upper arm exoskeleton is through a
dSpace DS1103 board, which does the data acquisition and con- 6 Discussion and Conclusions
trol for the system. The motors are Maxon EC45 flat motors, While experiments are being conducted with the exoskeleton,
which are lightweight but permits a large power output. The motor there are several open questions that need to be addressed. These
amplifiers provide closed-loop speed control, and the encoders are are as follows. 共i兲 How to evaluate and control the reaction forces
mounted on the joints of the plastic skeleton. A schematic of the at the joints of the wearer. 共ii兲 Due to the joint limits and positive
implementation architecture is shown in Fig. 15. In the experi- tension requirements of the cables, there exist some singularity
ment, the motors, arm cuffs, and cable reels were mounted accord- voids in the exoskeleton workspace—how to improve this and
ing to the optimization results of Sec. 4.2 for the workspace. control the exoskeleton at these configurations. 共iii兲 Optimization
The following experiment run focuses on achieving point to of the exoskeleton workspace to achieve full range of motion of
point motion of the wrist point to emulate activities of daily liv- the arm. Currently, these questions are outside the scope of the
ing. In the experiment, the arm is assumed initially to be in equi- present paper and we hope to address these in the near future.
librium, with initial and final points chosen as xi = 关 In this paper, design and optimization of a cable driven upper
−0.146, 0.054, −0.612兴 and xd = 关−0.239, −0.148, 0.566兴 in the arm exoskeleton were presented. Our design is motivated by the
base frame. Cable-tension profiles of the six cables are chosen to natural arm, and uses a cable driven parallel mechanism. Cable
move the wrist from xi to xd. The choice of positive tension pro- based parallel designs require additional motors than the number
files still remains an open research question and is beyond the of degrees-of-freedom, since cables can only transmit power in
scope of this paper. Using this tension profile, Eq. 共1兲 is integrated tension. This motivates optimizing the design parameters to maxi-
numerically for 5 s to find joint angle profile q共t兲 and joint rate mize the workspace. It was shown that optimization significantly

Fig. 15 Schematic of the implementation architecture on the exoskeleton in „a… hardware and „b…
software

Journal of Medical Devices SEPTEMBER 2009, Vol. 3 / 031004-7

 wnloaded from http://asmedigitalcollection.asme.org/medicaldevices/article-pdf/3/3/031004/5928351/031004_1.pdf?casa_token=a7gzSUjzqDYAAAAA:ZWRK-q5o_gjgfSZV255fbfmRnupN2299-aK3wkio1ee7m_mwr4yMXYzGDa4mFb2LxYplzOU by Chinese University Of Hong Kong user on 26 June 2
Fig. 16 „a… Joint angle: simulation „dashed… versus experiment „solid…. „b… Wrist point coordinates in base frame: simula-
tion „dashed… versus experiment „solid…

improved the workspace. Initially, the optimization was applied to 54共4兲, pp. 443–446.
关7兴 Perry, J. C., and Rosen, J., 2006, “Design of a 7 Degree-of-Freedom Upper-
the shoulder joints. Subsequently, whole arm optimization was Limb Powered Exoskeleton,” BioRob 2006—The First IEEE/RAS EMBS In-
carried out. This was performed in two steps: first, evaluating the ternational Conference on Biomedical Robotics and Biomechatronics, Pisa,
elbow joint, and then bringing the coupling of the elbow cables to Italy, Feb. 20–22.
the shoulder joints. However, with a limited number of actuators, 关8兴 Sanchez, R. J., Liu, J., Rao, S., Shah, P., Smith, R., Rahman, T., Cramer, S. C.,
the designed exoskeleton cannot be functionally compared with a Bobrow, J. E., and Reinkensmeyer, D. J., 2006, “Automating Arm Movement
Training Following Severe Stroke: Functional Exercises With Quantitative
natural arm, which has a large number of tendons and muscles. Feedback in a Gravity-Reduced Environment,” IEEE Trans. Neural Syst. Re-
The optimization technique is further applied to a specific activity habil. Eng., 14共3兲, pp. 378–389.
of daily living. This fits very well with our modular design, which 关9兴 Rahman, T., Sample, W., Jayakumar, S. M., King, M., Wee, J. Y., Seliktar, R.,
can be tailored to a specific application. Preliminary experiments Alexander, M., Scavina, M., and Clark, A., 2006, “Passive Exoskeletons for
Assisting Limb Movement,” J. Rehabil. Res. Dev., 43共5兲, pp. 583–590.
show a good match between the results obtained by experiments 关10兴 Herder, J. L., Vrijlandt, N., Antonides, T., Cloosterman, M., and Mastenbroek,
and simulations. P. L., 2006, “Principle and Design of a Mobile Arm Support for People With
Muscular Weakness,” J. Rehabil. Res. Dev., 43共5兲, pp. 591–604.
Acknowledgment 关11兴 Carignan, C., Liszka, M., and Roderick, S., 2005, “Design of an Exoskeleton
With Scapula Motion for Shoulder Rehabilitation” Proceedings of the IEEE
The research platform was supported by internal funds from the International Conference on Advanced Robotics 共ICAR兲, Seattle, WA, pp.
University of Delaware. The research was also supported by 524–531.
关12兴 Yang, G., Ho, H. L., Chen, W., Lin, W., Yeo, S. H., and Kurbanhusen, M. S.,
World Class University 共WCU兲 program through the Korea Sci- 2004, “A Haptic Device Wearable on a Human Arm,” IEEE Conference on
ence and Engineering Foundation funded by the Ministry of Edu- Robotics, Automation and Mechatronics, Vol. 1共1兲, pp. 243–247.
cation, Science and Technology 共Grant No. R32-2008-000-10022- 关13兴 Mustafa, S. K., Yang, G., Yeo, S. H., and Lin, W., 2006, “Optimal Design of a
0兲. The support for V.N.D. was provided by the Global Research Bio-Inspired Anthropocentric Shoulder Rehabilitator,” Applied Bionics and
Biomechanics, 3共3兲, pp. 199–208.
Award from the Royal Academy of Engineering of England and 关14兴 Weihai, C., Quanzhu, C., Jianbin, Z., and Shouqian, Y., 2006, “Kinematics
Bournemouth University 共UK兲. Control for a 7-DOF Cable-Driven Anthropomorphic Arm,” Proceedings of the
IEEE/RSJ International Conference on Intelligent Robots and Systems,
Beijing, China.
References 关15兴 Zheng, M., Weihai, C., Shouqian, Y., and Xingming, W., 2006, “Kinematic
关1兴 National Center for Health Statistics 共USA兲, 2006, http://www.cdc.gov/nchs/ Analysis for a 7-DOF Modular Hybrid-Driven Manipulator,” Proceedings of
data/hus/hus06.pdf#001. the IEEE International Conference on Industrial Informatics, pp. 1363–1368.
关2兴 Office for National Statistics Projections 共UK兲, 2006, http:// 关16兴 Ullrich, N. G., Di Lieto, G., Salsedo, F., and Bergamasco, M., 2001, “Design
www.statistics.gov.uk/focuson/olderpeople/. and Optimization of a Purely Rotational 3-DOF Haptic Device,” Proceedings
关3兴 Lucieer, P., and Herder, J. L., 2005, “Design of an Adjustable Compensation of the Tenth IEEE International Workshop on Robot and Human Interactive
Mechanism for Use in a Passive Arm Support,” ASME International Design Communication.
Engineering Technical Conference, Long Beach, CA, Sept. 24–28, Paper No. 关17兴 Oh, S.-R., and Agrawal, S. K., 2006, “Feasible Workspace of a Set Point
DETC2005-85442. Controller for a Cable Suspended Robot With Input Constraints and Distur-
关4兴 Reinkensmeyer, D. J., Hogan, N., Krebs, H. I., Lehman, S. L., and Lum, P. S., bances,” IEEE Trans. Control Syst. Technol., 14共4兲, pp. 735–742.
1999, “Rehabilitators, Robots, and Guides: New Tools for Neurological Reha- 关18兴 Oh, S.-R., and Agrawal, S. K., 2006, “Computationally Efficient Feasible Set
bilitation,” Biomechanics and Neural Control of Movement, J. Winters and P. Points Generation and Control of a Cable Robot,” IEEE Trans. Rob. Autom.,
E. Crago, eds., Springer, Berlin. 22共3兲, pp. 551–558.
关5兴 Lum, P. S., Reikensmeyer, D. J., and Lehman, S. L., 1993, “Robotic Assist 关19兴 Oh, S.-R., and Agrawal, S. K., 2005, “Cable Suspended Planar Robots With
Devices for Bimanual Physical Therapy: Preliminary Experiments,” IEEE Redundant Cables: Controllers With Positive Tensions,” IEEE Trans. Rob.
Trans. Rehabil. Eng., 1, pp. 185–191. Autom., 21共3兲, pp. 457–464.
关6兴 Aisen, M. L., Krebs, H. I., 1997, “The Effect of Robot-Assisted Therapy and 关20兴 Anthropometric Source Book, 1978, Vol. 1: Anthropometry for Designers,
Rehabilitative Training on Motor Recovery Following Stroke,” Arch. Neurol., NASA Reference Publication 1024.

031004-8 / Vol. 3, SEPTEMBER 2009 Transactions of the ASME