IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO.

3, JUNE 2010 293

Universal Haptic Drive: A Robot for

Arm and Wrist Rehabilitation
Jakob Oblak, Imre Cikajlo, and Zlatko Matjac̆ić

Abstract—In this paper we present a universal haptic drive (NDT) [7], isolated selective movement of reaching, wrist ori-
(UHD), a device that enables rehabilitation of either arm enting and grasping are repetitively practiced. The emphasis is
(“ARM” mode) or wrist (“WRIST” mode) movement in two on reestablishment of range-of-motion (ROM), strength, spas-
degrees-of-freedom. The mode of training depends on the selected
mechanical configuration, which depends on locking/unlocking of ticity reduction, and selectivity of motion. Particularly, in most
a passive universal joint. Actuation of the device is accomplished of the stroke population arm reaching activity confined to a
by utilizing a series elastic actuation principle, which enables horizontal plane and wrist supination and extension movement
use of off-the-shelf mechanical and actuation components. A practice are the core building blocks of neurological rehabilita-
proportional force control scheme, needed for implementation of tion, which are needed to allow progress toward more functional
impedance control based movement training, was implemented.
The device performance in terms of achievable lower and upper whole upper extremity movement. It has been shown that inten-
bound of viable impedance range was evaluated through ade- sity of the upper extremity therapy positively correlates with the
quately chosen sinusoidal movement in eight directions of a planar motor outcome after stroke [6]. Unfortunately, the repetitive na-
movement for the “ARM” mode and in eight directions of a com- ture of therapy, which demands one-on-one manual interaction
bined wrist flexion/extension and forearm pronation/supination with a therapist, is time consuming as well as expensive as the
movement for the “WRIST” mode. Additionally, suitability of the
universal haptic drive for movement training was tested in a series majority of the healthcare budgets assigned to movement reha-
of training sessions conducted with a chronic stroke subject. The bilitation is related to labor costs.
results have shown that reliable and repeatable performance can There is a pressing need for a new, more cost-effective reha-
be achieved in both modes of operation for all tested directions. bilitation approaches. In order to increase the ratio between the
Index Terms—Arm, haptics, neurological rehabilitation, series outcome and cost of rehabilitation, robotic devices are being in-
elastic actuation, wrist. troduced into clinical rehabilitation [8]–[10]. Robot mediated
rehabilitation is a rapidly advancing field. Besides the cost-ef-
fective aspect, robotic devices introduce higher accuracy and re-
I. INTRODUCTION
peatability in performing rehabilitation exercises. Precise mea-
surements of quantitative parameters by means of robotic in-

O VER the past decade, the number of people that suf-

fered from cerebro-vascular accidents or traumatic brain
injuries has substantially increased [1]. The damage to the
strumentation may improve objective monitoring of patient’s
recovery. Furthermore, rehabilitation robots can be combined
with virtual reality environment [11], [12]. This possibility is
central nervous system can lead to impaired motor control of very important, because the patient’s motivation and cognitive
upper and lower limbs, which represent significant difficulties involvement has a great impact on the outcome of rehabilitation.
in performing activities of daily living (ADL). Several studies The current state of the art in the field of upper extremity
indicated that rehabilitation based on task-oriented repetitive rehabilitation robotics includes haptic robotic solutions that
movements is beneficial for improvement of movement abilities have from one to three endpoint degrees-of-freedom (DOF)
[2]–[5]. and were developed for training of the shoulder and elbow:
Upper extremity function is of paramount importance to carry MIT-MANUS [13], GENTLE/S [14], Pneu-WREX [15],
out various ADL. Reaching or approaching toward an object NEREBOT [16], ReoGo [17] and ARM Guide [18] as well
is done by shoulder and elbow, orienting of an object is ac- as for wrist: CRAMER [19], RICE WRIST [20], HAND
complished by wrist, while grasping and releasing of an ob- MENTOR [21], HWARD [22], and BI-MANU-TRACK [6].
ject is carried out by opening and closing a hand. To improve The common denominator of the referenced devices is that they
upper extremity ability after a brain lesion, an early and inten- provide training environment for only one component/activity
sive therapy approach is needed [6]. Within the framework of of upper extremity movement; either arm reaching movement
motor relearning approach and neurodevelopmental technique or wrist movement. Clinical studies have shown that training
results are task specific; motor improvement can be expected
only in the practiced aspects of movement [23]. The upgraded
Manuscript received June 10, 2009; revised August 26, 2009; accepted version of MIT-MANUS with Wrist-Robot [24], ARMIN II
September 22, 2009. First published October 20, 2009; current version pub-
lished June 09, 2010. Development of the universal haptic drive was supported
[25], and REHAROB [26] are rehabilitation robots that enable
by the Slovenian Research Agency and FUNDACION FATRONIK. training of arm and wrist movement, however, these devices
The authors are with the Institute for Rehabilitation, Republic of Slovenia, have up to seven active DOFs, which necessitates complex,
1000 Ljubljana, Slovenia (e-mail: [email protected]). precise, and expensive mechanisms and drives and thus may be
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. affordable only for major rehabilitation centers. Furthermore,
Digital Object Identifier 10.1109/TNSRE.2009.2034162 their use in neurological rehabilitation may take place only
1534-4320/$26.00 © 2010 IEEE
294 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 3, JUNE 2010

Fig. 1. (a) UHD is composed of 1: handle bar, 2: universal joint, 3: force sensor, 4: sliding mechanism, 5: spherical joint, and 6: actuated bar. Switching between
“ARM” and “WRIST” mode can be easily achieved by locking or unlocking universal joint on the actuated bar. In “WRIST” mode, the length difference D of 1
the bar under the universal joint is compensated with sliding mechanism. (b) In “ARM” mode universal joint is locked. Movement in R: right, L: left, B: backward,
and F: forward directions is possible. (c) In “WRIST” mode universal joint is unlocked and forearm is supported to fix the centre of wrist joint. When handle bar
is set in HORIZONTAL position, S: supination/P: pronation, and F: flexion/E: extension can be exercised. Setting the handle bar in VERTICAL position enables
to exercise S: supination/P: pronation and U: ulnar/R: radial deviation.

after the isolated components of selective movement have been In this paper we present the proposed concept, development
adequately reestablished. The penetration of the rehabilitation of a prototype and the results of system performance evalua-
robotics into widespread clinical use may be significantly tion of a 2 DOFs haptic rehabilitation robot that enables planar
facilitated by lower cost hardware. Therefore, there is a need reaching movement and also wrist movement as well as results
for haptic rehabilitation robots that have only few DOFs, thus of limited clinical training tests.
reducing the costs of hardware, while at the same time offering
possibility of training reaching as well as wrist movement, II. DEVICE DESCRIPTION
which are essential prerequisites for rehabilitation of a more
complex functional movement of upper extremity. One ap- A. Mechanism
proach would be to conceive an innovative mechanical design The concept of the universal haptic drive (UHD) is presented
of a rehabilitation robot, which would be reconfigurable in such in Fig. 1. The UHD mechanism is composed of an actuated bar
a way that depending on the mechanical mode of operation inserted into a spherical joint, which is fixed to a base plate.
either reaching or wrist movement would be possible. Upon the actuated bar a series of elements are serially linked: a
OBLAK et al.: UNIVERSAL HAPTIC DRIVE: A ROBOT FOR ARM AND WRIST REHABILITATION 295

TABLE I
ROM IN HUMAN WRIST AND UHD

sliding mechanism that enables linear movement, force sensor, a

passive 2 DOFs universal joint that can be mechanically locked
and a handle bar. If the universal joint is locked the handle bar
moves according to the movement of the actuated bar, which
means that inclination of 15 results in horizontal change in po-
sition of 20 cm, given the selected lengths of mechanical ele-
ments [Fig. 1(a)]. In this mechanical configuration, termed as
“ARM” mode, a subject holding on the handle bar can per-
form quasi-planar reaching movement, as shown in Fig. 1(b).
Expression quasi-planar means that patient’s hand is limited to
move in the workspace with the shape of a spherical surface. The
maximal vertical displacement is at the limits of the achievable
workspace and is not bigger than 3 cm. If the universal joint is
unlocked then the handle bar can rotate in relation to the ac-
tuated bar in 2 DOFs, which means that rotation of 15 of the
actuated bar results in rotation of 45 of the handle bar, given Fig. 2. (a) Actuation of UHD consists of 7: DC motors, 8: elastic springs,
9: linear potentiometers, 10: pulleys, and 6: actuated bar. Motors, springs, and
the selected lengths of mechanical elements [Fig. 1(a)]. In this actuated bar are connected with string wires. (b) By moving actuated bar in
mechanical configuration, termed as “WRIST” mode, a subject horizontal direction A (*), the geometry of the actuation in vertical direction B
holding on the handle bar and having forearm strapped to the changes from R + r to D + d. (c) Resistive force, which appears due to ge-
ometrical design of the UHD actuation, is calculated for three different spring
support can perform wrist movement in 2 DOFs. The first DOF stiffness values over the whole ROM of the UHD.
is pronation/supination of the forearm, while the second DOF
depends on the orientation of the handle bar. When the handle
bar is oriented horizontally, wrist flexion/extension movement maximal continuous force that can be applied on the handle bar
can be performed, while in the vertical orientation ulnar/radial in “ARM” mode is 75 N (300 N * 1/4) and the maximal contin-
deviation movement can be performed, as shown in Fig. 1(c). uous torque provided in “WRIST” mode is 20 Nm (300 N * 1/3
Achievable ROM of UHD in “WRIST” mode and human wrist * 0.2 m).
ROM [24] are presented in Table I. It is important to point out There are two reasons for utilization of series elastic elements
that in “WRIST” mode a length difference D [Fig. 1(a)] of in the actuation. The first reason is related to the geometrical
the mechanism below the universal joint occurs, which is due design of the UHD actuation. The distance as measured from
to fixed forearm support and is compensated with movement of the shaft of the motor to the pulley located on the opposite side
the sliding mechanism. of the actuated bar is affected by the movement of the actu-
ated bar in the perpendicular direction as depicted in Fig. 2(b)
B. Actuation . The change in the lengths of both springs of the
The actuation of the UHD is presented in Fig. 2. It consists actuation system in B direction generates a force that acts on the
of two sets of DC motors with gears and encoders, which are actuated bar in A direction and tends to move the actuated bar
connected in series with elastic springs to the actuated bar by back in “zero position.” The magnitude of this coupling force
means of string wires and pulleys, as shown in Fig. 2(a) and (b). depends on the position of the actuated bar as well as the stiff-
Both motors actuate movement of the actuated bar in the per- ness of used springs. This dependence is illustrated in Fig. 2(c)
pendicular directions A and B. The maximal continuous torque for three different spring stiffness values. One can observe that
of the selected DC motors (Maxon, RE 40, 150 W) with Maxon the greatest coupling between the A and B directions of move-
Planetary Gearhead (GP 52 C, 81:1) is approximately 15 Nm. ment takes place at the limits of the achievable workspace of
The radius of the pulley attached to the gearhead is 5 cm, which the actuated bar. However, this coupling is rather weak and does
means that the maximal force that can be exerted on the actuated not represent an important factor in the design of the actuation
bar in linear direction is 300 N (15 Nm/0.05 m). Therefore, the system.
296 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 3, JUNE 2010

Fig. 3. (a) Model of the actuator, where all parameters and variables are converted from rotational to translational motion. (b) Bode plots of open-loop dynamics
of a series elastic actuator with fixed load. Shown are experimental results as well as simulation results for three different values of spring stiffness.

The second reason to use springs relates to the effect that, load force for frequencies below the natural frequency ,
by introducing an elastic element in series with the motor we which depends on the spring stiffness:
have actually designed a series elastic actuator (SEA), which
provides many benefits in force control. These benefits include
greater shock tolerance, lower reflected inertia, more accurate (4)
and stable force control and the capacity for energy storage
[27]–[30]. However, these benefits come with one shortcoming, In general, the stiffer the spring the larger bandwidth of the
which is related to reduction of the achievable bandwidth. This actuation system can be achieved.
is illustrated by a simple linear model for an actuator with a Since the UHD device is predominantly intended for reha-
series elastic element, where all parameters and variables are bilitation purposes where relatively slow movement can be ex-
converted from rotational to translational motion as shown in pected during training we have experimentally determined that
Fig. 3(a). the spring stiffness value of N/m provides sufficient
Dynamics of the system, where the load is assumed to be actuator bandwidth on one hand while on the other hand pro-
clamped, is determined by the following two equations: vides also sufficient attenuation of backlash and friction non-
linear effects as well as actuator’s own impedance.
(1) Open-loop dynamics of SEA with fixed load used in the UHD
was assessed in experimental measurements. The experiment
(2)
was done by fixing the load of the actuator and sweeping the
frequency spectrum of the actuator input. The damping term
where is the motor’s force converted from the motor torque; ( Ns/m) was determined experimentally by fitting
is the double stiffness of the spring, [see Fig. 3(a)] while model damping term to match simulation results with experi-
and are the positions of the load and motor. Values mental measurements, see Fig. 3(b).
(36,7 kg) and stand for reflected motor mass and reflected
viscous friction in motor and planetary gearhead. By taking C. Control
Laplace transforms and assuming that the output of the actuator Haptic devices used in rehabilitation are typically controlled
is clamped , the transfer function between the output in such a way to 1) provide unhindered movement, i.e., the inter-
force and the motor force is given by action force between the haptic device and the hand of the user
should be as small as possible to give a feeling of “an empty
(3) space;” 2) provide a force “tunnel” along the desired move-
ment trajectory where radial force typically proportionally in-
creases with the radial deviations from the desired trajectory,
The (3) presents open-loop dynamics of a series elastic actu- thus forcing the hand of the user back to the desired trajectory
ator with fixed load. Fig. 3(b) shows the Bode plots of (3), for while force tangential to the desired trajectory also proportion-
three different values of stiffness of the springs. One can observe ally increases if the user is either lagging or leading the desired
good agreement between the input motor force and output trajectory in the tangential direction. Haptic devices can be also
OBLAK et al.: UNIVERSAL HAPTIC DRIVE: A ROBOT FOR ARM AND WRIST REHABILITATION 297

force feedback signal was measured by means of JR3 (JR3,

Woodland, USA) force sensor, which was mounted on the
actuated bar [Fig. 1(a)]. The desired force was calculated from
selected impedance parameters and kinematics of the actuated
bar. The position of the actuated bar was determined by the
positions of the encoders , mounted onto DC motors,
and the lengths of the springs , which were measured
with linear potentiometers [Fig. 2(a) and (b)]. The control
scheme was implemented in Simulink (MATLAB) and run in
real time on xPC target platform with the sampling frequency
of 1 KHz. All the signals from the sensors were assessed with
National Instruments 6025 Data Acquisition Card (National
Instruments, Austin, TX).

III. EXPERIMENTAL EVALUATION—SYSTEM PERFORMANCE

The principle function of the UHD is to ensure a haptic
interface between a user and a computer simulation based
training task. Therefore, the device performance can be es-
timated by measuring how precisely the UHD can exert
required force on a user. Because impedance control strategy
was used, the required force we want to exert on the user’s
hand was set by a selection of the virtual impedance of the
UHD. Virtual impedance is typically bounded with the lowest
achievable impedance, which would ideally be zero, and the
highest achievable impedance, which would ideally be infinite.
Therefore, the quality of a haptic interface can be determined
by examining the performance in “LOW IMPEDANCE” and
“HIGH IMPEDANCE” environment. In “LOW IMPEDANCE”
environment the objective is “zero” force control, meaning that
the interaction between a user and the UHD should be minimal
while moving in the UHD workspace. That situation is typical
for user-in-charge oriented exercises. Opposite situation oc-
curs, when we want the UHD to generate the biggest possible
resistive force. That situation is common for robot-in-charge
mode or when we want to simulate “HIGH IMPEDANCE”
environment. However, conservative requirement for stability
of SEA actuators (in terms of passivity of a human-machine
interface) is that maximal virtual stiffness is limited by the
intrinsic mechanical spring stiffness [30], which is in our
Fig. 4. (a) Photograph of the UHD prototype 1: handle bar, 2: universal joint,
3: force sensor, 4: sliding mechanism, 5: spherical joint, 7: DC motors with
case 4000 N/m. Since, the actuated bar is fixed between two
gears and encoders, 8: elastic springs, 11: forearm support (“WRIST” mode), pretensioned springs, Fig. 3(a), the overall stiffness equals
and 12: user. (b) Schematic diagram of impedance based force control of UHD. 8000 N/m. Ratio of force that is transmitted from the bottom of
the actuated bar to the handle bar is 1:4; see Fig. 1(a). Similar
ratio for movements is 4:1. For this reason, the maximal stably
used for offering continuous partial assistance or partial resis- achievable stiffness at the handle bar is 16 times smaller than
tance to the patient along the desired trajectory. on the bottom of the actuated bar. In the “ARM” mode this
Fig. 4(a) shows the photograph of the actual prototype of value equals approximately to 500 N/m. Similar consideration
the UHD device configured to enable movement training in for “WRIST” mode yields stiffness of 20 Nm/rad.
“WRIST” mode, while Fig. 4(b) shows schematic diagram The UHD performance was verified for both “LOW
of the implemented impedance based force control of UHD. IMPEDANCE” and “HIGH IMPEDANCE” environment.
A conventional proportional force control scheme, which Because the UHD allows performing exercises in “ARM” and
is needed for implementation of impedance control based “WRIST” modes, we measured forces/torques for both modes.
movement training, was implemented independently in each In the conducted experiment we also measured the UHD’s
actuated DOF. Because the coupling between both actuated own impedance, by switching the motors off and monitoring
DOFs that appears due to actuation design is relatively weak, the forces/torques. The UHD performance was examined in
no compensation of this coupling was implemented in the experimental trials involving one healthy subject. Fig. 5(a)
control scheme. The values of the controller gain were the shows single repetitions denoting the full ROM and directions
same for both DOFs and were determined experimentally. The of movement for both modes. The experimental evaluation
298 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 3, JUNE 2010

Fig. 5. (a) Ranges and directions of motion for “ARM” and “WRIST” modes. (b) Device performance in “ARM” mode. Plot shows the force (solid line) that user
can feel on the handle bar, while making similar movements in forward/backward direction (dotted line). In the first 10 s UHD performed “LOW IMPEDANCE”
environment. In time interval from 10 to 20 s “HIGH IMPEDANCE” environment was performed and in the last 10 s, the motors were switched off. (c) Device
performance in “WRIST” mode. Plot shows the torque (solid line) that a user can feel on the handle bar, while making similar supination/pronation movements
(dotted line). In first 10 s UHD performed “LOW IMPEDANCE” environment. In the time interval from 10 to 20 s “HIGH IMPEDANCE” environment was
performed and in the last 10 s, the actuation was switched off.

was carried out by moving (“ARM” mode)/rotating (“WRIST” maximal force was approximately 10 N. On the other hand,
mode) the handlebar in a given direction for approximately forces that appear in “HIGH IMPEDANCE” environment
cm/ with a frequency of approximately 1 Hz while simul- were approximately 40 N, which corresponds well to a desired
taneously measuring the force interaction between the user and virtual stiffness of 500 N/m. Fig. 5(c) shows performance of
the UHD device. This frequency was selected as being close the UHD in “WRIST” mode where the maximal torque that the
to the upper bound of expected frequency range in upper limb user felt in “LOW IMPEDANCE” environment was approxi-
movement rehabilitation. As can be seen from the Fig. 5(b), mately 0.7 Nm, which is much smaller than in the case when
showing the performance of the UHD in “ARM” mode, the the motors were switched off, where the maximal torque was
maximal force that the user felt in “LOW IMPEDANCE” approximately 1.8 Nm. On the other hand, torques that appear
environment was approximately 2.5 N, which is much smaller in “HIGH IMPEDANCE” environment were approximately
than in the case when the motors were switched off, where the 7 Nm, which corresponds to a virtual stiffness of 20 Nm/rad.
OBLAK et al.: UNIVERSAL HAPTIC DRIVE: A ROBOT FOR ARM AND WRIST REHABILITATION 299

TABLE II
DEVICE PERFORMANCE TESTED IN EIGHT DIFFERENT DIRECTIONS FOR BOTH MODES OF OPERATION

Fig. 6. (a) Subject’s tracking performance in “WRIST” mode before and after the training. (b) ROMs before and after the training.

Table II shows the results (mean values and standard devi- since the movement, performed by the tested subject, was rather
ations) for different impedance modes for five repetitions of variable.
movement in each of the eight tested directions for “ARM”
and “WRIST” mode. The qualitative inspection of the results IV. EXPERIMENTAL EVALUATION—CLINICAL TESTS
in various directions shows that the performance of the UHD is The developed UHD device was tested also in a series of
similar in all tested directions for different modes of operation. training sessions in which movement of arm and wrist was prac-
Quantitative analysis of these results would not be appropriate, ticed by a volunteer subject (10 years poststroke), with chronic,
300 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 3, JUNE 2010

Fig. 7. (a) Subject’s tracking performance in “ARM” mode before and after the training. (b) ROMs in each of the tracking directions before and after the training.

right-sided hemiparesis resulting in limited movement abilities was assisted if she lagged behind the required reference trajec-
in shoulder, elbow and wrist as well as grasp (score 5 out of 7 on tory with the force/torque proportional to the selected parame-
Utrecht Arm/Hand Test). The subject gave informed consent for ters of impedance controller. The desired position and actual po-
the experimental tests, which were approved by the Slovenian sition were displayed on computer screen in a form of easily vis-
national ethics committee. The primary objective of these clin- ible circular objects, Fig. 4(b); therefore the training tasks were
ical tests was to determine whether the achievable impedance visually cued. If the subject’s movement was faster than the
range of the UHD device, presented in the previous section, is movement of reference trajectory the UHD provided no assis-
suitable for adequate assistance in movement tracking tasks. tance. Throughout the training sessions the impedance parame-
Overall six training sessions were performed on six consecu- ters were gradually decreased, according to the gradual progress
tive days in both “ARM” and “WRIST” modes of operation. The in patient’s progress in the following ranges: virtual stiffness
training protocol for each session consisted from performing N/m (“ARM” mode); 20–8 Nm/rad (“WRIST”
tracking movement, first of the wrist and followed by the arm. mode) and virtual damping Ns/m (“ARM” mode);
The duration of each training session was approximately 45 min. 0.5–0.25 Nms/rad (“WRIST” mode). Training of each move-
Training of the movement was in form of tracking tasks, which ment modality (three wrist movements and planar arm move-
were for the wrist confined consecutively, only to each single ment in eight directions) consisted of 7 min activity. In the first
DOF: ulnar/radial deviation, pronation/supination and flexion/ 2 min of tracking the subject was instructed to be passive while
extension while training of the arm movement included tracking the impedance controlled UHD device performed tracking of
tasks in eight directions within the plane of movement. ROM in the reference trajectories. In the following 5 min the subject was
tracking tasks was determined such that the subject was com- instructed to follow or if possible to be ahead of the reference
fortable with the extent of movement. Tracking trajectories were trajectory while the UHD provided assistance according to the
composed by using polynomial functions with time as indepen- characteristics of the impedance controller as described above.
dent variable similarly as described in [10]. Impedance con- On the first day and the last day of the training period the
troller assistance implemented during the movement tracking tracking performance of the subject was evaluated in conditions
tasks was also similar to the one described in [10] where the where the impedance controller provided no assistance but was
UHD device opposed to deviations in radial directions in accor- setup to maintain “LOW IMPEDANCE” environment in radial
dance to selected impedance parameters [virtual stiffness and as well as tangential directions relative to reference tracking tra-
damping, Fig. 4(b)], while in the tangential direction the subject jectory directions. The ROM for each trained DOF (for wrist
OBLAK et al.: UNIVERSAL HAPTIC DRIVE: A ROBOT FOR ARM AND WRIST REHABILITATION 301

radial/ulnar deviation, pronation/supination and flexion/exten- impediment for their widespread use. The cost of robotic de-
sion; for arm eight directions in the plane of movement) was vices for rehabilitation could be reduced by 1) using standard
determined as the mean peak value of five repetitions. Range mechanical components that are produced in large quantities
of motions before and after the six training sessions was com- and 2) by designing mechanisms to enable training of various
pared in paired t-test. Level of statistical significance was set to aspects of upper extremity movement. The proposed haptic de-
. vice presented in this paper is made from off-the-shelf mechan-
Fig. 6 shows the tracking performance for each DOF of the ical components, motors and motor drives. Also low cost single
wrist before and after the training. One can observe that in nei- DOF force load cells or even measurements of springs elonga-
ther direction the subject was able to reach the required peak tions could be used in the future instead of multiaxis robotic
excursions (set to ). The most difficulties in movement the wrist force sensor.
subject experienced in the forearm supination and wrist exten- The developed universal haptic drive offers similar quality of
sion, which was markedly improved after the training. a haptic training environment for practicing movement of the
Fig. 7 shows the tracking performance for planar arm move- arm and the wrist, where the mode of training depends only on
ment before and after the training. The results show that the sub- the configuration of mechanical linkage.
ject could perform the required task satisfactorily already be- The UHD device presented in this paper has potentially two
fore the training. The performance after the training was similar major advantages over single-purpose (either reaching move-
where modest improvement can be observed in two of the eight ment or wrist movement) rehabilitation robots [6], [13]–[22]:
tracking directions. Also, after the training one can observe im- i) it enables isolated training of specific movement of arm
proved coordination of movement in the directions 3 and 7. and wrist and ii) by combining mechanical structure with
SEA-based actuators the cost of hardware may be signifi-
cantly reduced thus possibly enabling future development of
V. DISCUSSION rehabilitation robots for home use. Through presented innova-
This paper presented the universal haptic drive, a device that tive approach haptic training environment may become more
can be used for arm as well as wrist rehabilitation. The UHD is readily available to a major portion of stroke population that is
designed in a way that allows performing exercises in two dif- in need of a targeted, specific movement training of both arm
ferent modes, “ARM” mode and “WRIST” mode, which enable and wrist movement. Such targeted, specific movement training
movement in two DOFs, depending only on the mechanical con- must precede attempts of whole upper extremity functional
figuration of the device, while exactly the same controller and movement training that can be facilitated either by a human
the same controller’s gains are used for both modes of operation. therapist or an exoskeleton type of rehabilitation robots with
Results show that entirely acceptable performance can be ex- many DOFs [25]. In the current design of the UHD only
pected in both modes of operation. The frequency and ampli- reaching movement in the horizontal plane is possible. How-
tude of movement selected for experimental evaluation of UHD ever, one can easily imagine that if the handle was displaced
performance (Fig. 5) was rather high and can be regarded to rep- by a firm rod for example laterally from the force sensor, in the
resent the upper limit of practiced movement in rehabilitation. direction parallel to the users’ arm, then the reaching movement
The maximal force that the user feels in low impedance environ- in the forward direction would incorporate also a component of
ment termed as “an empty space” was similar to performance of vertical movement. In further development of UHD device we
other “single-task” haptic devices, for example MIT wrist robot will focus on improvements of a mechanism that will enable
[24]. practicing reaching in needed directions for each individual
SEA-based actuation system utilized in the UHD device user.
successfully enabled generation of LOW IMPEDANCE en- In this paper we have demonstrated that rather simple mech-
vironment in both modes of operation. Generation of HIGH anisms, which enable training of various aspects of upper ex-
IMPEDANCE environment is limited to a stiffness of 500 N/m tremity movement by employing limited number of actuated
for the “ARM” mode and 20 Nm/rad for the “WRIST” mode, DOFs, can be successfully designed and implemented for pur-
which is due to a conservative requirement for stability of SEA poses of neurological rehabilitation.
actuators [30]. However, this should not represent a notable
limitation in rehabilitation where more compliant and thus
gentle guiding in performing a task under training is necessary REFERENCES
as recent clinical results [9] suggest that such an environment [1] T. Truelsen, B. Piechowski-Jóźwiak, R. Bonita, C. Mathers, J. Bogous-
is much more stimulating for relearning of movement as com- slavsky, and G. Boysen, “Stroke incidence and prevalence in europe: A
pared to a rather rigid implementation of a stiff virtual wall. review of available data,” Eur. J. Neurol., vol. 13, no. 6, pp. 581–598,
June 2006.
Also the clinical training tests that were performed with the [2] R. Colombo, F. Pisano, S. Micera, A. Mazzone, C. Delconte, M. C.
poststroke subject have shown that the achievable impedance Carrozza, P. Dario, and G. Minuco, “Robotic techniques for upper limb
range of UHD is sufficient to: i) guide the movement of the sub- evaluation and rehabilitation of stroke patients,” IEEE Trans. Neural
Syst. Rehabil. Eng., vol. 13, no. 3, pp. 313–324, Sep. 2005.
ject’s arm/wrist also in conditions where no effort is provided [3] D. J. Reinkensmeyer, J. L. Emken, and S. C. Cramer, “Robotics, motor
on the part of the user and ii) bring about the training results learning, and neurologic recovery,” Annu. Rev. Biomed. Eng., vol. 6,
similar to other “single-task” devices [2], [5]. pp. 497–525, Aug. 2004.
[4] R. Riener, T. Nef, and G. Colombo, “Robot-aided neurorehabilitation
It is common understanding that today’s haptic devices used for the upper extremities,” Med. Biol. Eng. Comput., vol. 43, pp. 2–10,
in rehabilitation are too expensive, which represents the major 2005.
302 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 18, NO. 3, JUNE 2010

[5] P. S. Lum, C. G. Burger, P. C. Shor, M. Majmundar, and M. van der [25] T. Nef, M. Mihelj, G. Kiefer, C. Perndl, R. Müller, and R. Riener,
Loos, “Robot-assisted movement training compared with conventional “ARMin—Exoskeleton for arm therapy in stroke patients,” in Proc.
therapy techniques for the rehabilitation of upper limb motor function 2007 IEEE 10th Int. Conf. Rehabil. Robotics, Jun. 2007, pp. 68–74.
after stroke,” Arch. Phys. Med. Rehabil., vol. 83, pp. 952–959, 2002. [26] T. G. Fazekas, G. Arz, M. Jurak, and M. Horvath, “Passive robotic
[6] S. Hesse, G. Schulte-Tigges, M. Konrad, A. Bardeleben, and C. movement therapy of the spastic hemiparetic arm with REHAROB:
Werner, “Robot-assisted arm trainer for the passive and active practice Report of the first clinical test and the follow-up system improvement,”
of bilateral forearm and wrist movements in hemiparetic subjects,” in Proc. 2005 IEEE 9th Int. Conf. Rehabil. Robotics, Jun. 2005, pp.
Arch. Phys. Med. Rehabil., vol. 84, no. 6, pp. 915–920, Jun. 2003. 127–130.
[7] L. Richards, C. Hanson, M. Wellborn, and A. Sethi, “Driving motor re- [27] G. Pratt and M. M. Williamson, “Series elastic actuators,” in Proc.
covery after stroke,” Topics Stroke Rehabil., vol. 15, no. 5, pp. 397–411, IEEE Int. Conf. Intell. Robots Syst. 1, 1995, pp. 399–406.
Nov. 2008. [28] W. Robinson, “Design and analysis of series elasticity in closed-loop
[8] W. S. Harwin, J. Patton, and V. R. Edgerton, “Challenges and oppor- actuator force control,” Ph.D. dissertation, Dept. Mech. Eng., Massa-
tunities for robot mediated neurorehabilitation,” in Proc. IEEE, Sep. chusetts Inst. Technol., Cambridge, 2000.
2006, vol. 94, no. 9, pp. 1717–1726. [29] G. Wyeth, “Control issues for velocity sourced series elastic actuators,”
[9] J. M. Hidler, D. Nichols, M. Pelliccio, and K. Brady, “Advances in in Proc. 2006 Australian Conf. Robotics Automat., 2006, (ACRA).
the understanding and treatment of stroke impairment using robotic [30] H. Vallery, J. Veneman, E. van Asseldonk, R. Ekkelenkamp, M.
devices,” Top Stroke Rehabil., vol. 12, no. 2, pp. 22–35, 2005. Buss, and H. van der Kooij, “Compliant actuation of rehabilitation
[10] H. I. Krebs, J. J. Palazzolo, L. Dipietro, M. Ferraro, J. Krol, K. Ran- robots—Benefits and limitations of series elastic actuators,” IEEE
nekleiv, B. T. Volpe, and N. Hogan, “Rehabilitation robotics: Perfor- Robotics Automat. Mag., vol. 15, no. 3, pp. 60–69, Sep. 2008.
mance-based progressive robot-assisted therapy,” Autonomous Robots,
vol. 15, no. 1, pp. 7–20, 2003.
[11] M. K. Holden, T. A. Dyar, and L. Dayan-Cimadoro, “Telerehabilita-
tion using a virtual environment improves upper extremity function in
patients with stroke,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 15,
no. 1, pp. 36–42, Mar. 2007.
[12] A. Heuser, H. Kourtev, S. Winter, D. Fensterheim, G. Burdea, V.
Hentz, and P. Fourducey, “Telerehabilitation using the Rutgers Master
II glove following carpal tunnel release surgery: Proof-of-concept,” Jakob Oblak was born in Novo mesto, Slovenia, in
IEEE Trans. Neural Syst. Rehabil. Eng., vol. 15, no. 1, pp. 43–49, 1982. He received the B.S. degree in electrical engi-
Mar. 2007. neering from the University of Ljubljana, Ljubljana,
[13] H. I. Krebs, N. Hogan, B. T. Volpe, M. L. Aisen, L. Edelstein, and C. Slovenia, in 2006. He is currently pursuing a Ph.D.
Diels, “Overview of clinical trials with MIT-MANUS: a robot-aided in rehabilitation robotics at the Institute for rehabili-
neuro-rehabilitation facility,” Technol. Health Care: Official J. Eur. tation, Ljubljana, Slovenia.
Soc. Eng. Med., vol. 7, no. 6, pp. 419–423, 1999.
[14] S. Coote, E. Stokes, B. Murphy, and W. Harwin, “The effect of
GENTLE/s robot-mediated therapy on upper extremity dysfunction
post stroke,” in Proc. 8th Int. Conf. Rehabil. Robot., Daejon, Korea,
Apr. 23–25, 2003, pp. 59–61.
[15] R. J. Sanchez Jr., E. Wolbrecht, R. Smith, J. Liu, S. Cramer, T.
Rahman, J. E. Bobrow, and D. J. Reinkensmeyer, “A pneumatic robot
for re-training arm movement after stroke: Rationale and mechanical
design,” in Proc. 2005 IEEE 9th Int. Conf. Rehabili. Robot., Jun. 2005, Imre Cikajlo received the Ph.D. degree in robotics
pp. 500–504. and electrical engineering from the University of
[16] G. Rosati, P. Gallina, and S. Masiero, “Design, implementation and Ljubljana, Ljubljana, Slovenia, in 2003.
clinical tests of a wire-based robot for neurorehabilitation,” IEEE From 2003 until 2004 he was a post-doc fellow
Trans. Neural Syst. Rehabil. Eng., vol. 15, no. 4, pp. 560–569, Dec. at the Tohoku University, Japan. Currently he is a
2007. Research Associate at the Institute for Rehabilita-
[17] ReoGo [Online]. Available: http://www.motorika.com tion, Ljubljana, Slovenia and an Associate Professor
[18] L. E. Kahn, M. L. Zygman, W. Z. Rymer, and D. J. Reinkensmeyer, at the University of Nova Gorica, Slovenia. His
“Robot-assisted reaching exercise promotes arm movement recovery research interests include human motion analysis
in chronic hempiaretic stroke: A randomized controlled pilot study,” J. and synthesis, biomechanics, integration of sensory
Neuroeng. Rehabil., Jun. 3–12, 2006. information, control of machine and human move-
[19] S. J. Spencer, J. Klein, K. Minakata, V. Le, J. E. Bobrow, and D. J. ments and rehabilitation robotics. Currently, he is involved in the development
Reinkensmeyer, “A low cost parallel robot and trajectory optimization of rehabilitation devices and procedures that may be used in clinical and/or
method for wrist and forearm rehabilitation using the Wii,” in Proce. home environment through application of telerehabilitation techniques and
2008 IEEE Conf. Biorobot., 2008, pp. 869–874. introduction of virtual reality in rehabilitation.
[20] A. Gupta and M. K. O’Malley, “Design of a haptic arm exoskeleton for
training and rehabilitation,” IEEE/ASME Trans. Mechatronics, vol. 11,
pp. 280–289, 2006.
[21] E. J. Koeneman, R. S. Schultz, S. L. Wolf, D. E. Herring, and J. B. Zlatko Matjac̆ić is currently a Head of Research
Koeneman, “A pneumatic muscle hand therapy device,” in Conf. Proc. Unit at the Institute for rehabilitation, Ljubljana,
IEEE Eng. Med. Biol. Soc., 2004, vol. 4, pp. 2711–2713. Slovenia and an Associate Professor of Biome-
[22] D. Takahashi, L. Der-Yeghiaian, V. H. Le, and S. C. Cramer, “A robotic chanics at University of Ljubljana, Ljubljana,
device for hand motor therapy after stroke,” in Proc. 2005 IEEE 9th Int. Slovenia.
Conf. Rehabil. Robot., Jun. 2005, pp. 17–20. His research interests include human motion
[23] N. Hogan, H. I. Krebs, B. Rohrer, J. J. Palazzolo, L. Dipietro, S. E. Fa- analysis and synthesis, biomechanics and control
soli, J. Stein, R. Hughes, W. R. Frontera, D. Lynch, and B. T. Volpe, of human movement and rehabilitation robotics for
“Motions or muscles? Some behavioral factors underlying robotics as- neurorehabilitation. Currently, his work is focused
sistance of motor recovery.,” J. Rehabil. Res. Development, vol. 43, pp. on development and clinical validation of simple
605–618, 2006. rehabilitation robotic devices that may be used at
[24] H. I. Krebs, B. T. Volpe, D. Williams, J. Celestino, S. K. Charles, D. homes of individual users through application of telerehabilitation techniques.
Lynch, and N. Hogan, “Robot-aided neurorehabilitation: A robot for Dr. Matjac̆ić received prestigious Slovenian national award “The Joz̆ef Stefan
wrist rehabilitation,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 15, Golden Emblem Award” for outstanding contributions made to science in his
pp. 327–335, Sep. 2007. Doctoral thesis.