IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007 367
Design and Validation of a Rehabilitation Robotic
Exoskeleton for Tremor Assessment and Suppression
E. Rocon, J. M. Belda-Lois, A. F. Ruiz, M. Manto, J. C. Moreno, and J. L. Pons
Abstract—Exoskeletons are mechatronic systems worn by
a person in such a way that the physical interface permits a
direct transfer of mechanical power and exchange of informa-
tion. Upper limb robotic exoskeletons may be helpful for people
with disabilities and/or limb weakness or injury. Tremor is the
most common movement disorder in neurological practice. In
addition to medication, rehabilitation programs, and deep brain
stimulation, biomechanical loading has appeared as a potential
tremor suppression alternative. This paper introduces the robotic
exoskeleton called WOTAS (wearable orthosis for tremor as-
sessment and suppression) that provides a means of testing and
validating nongrounded control strategies for orthotic tremor
suppression. This paper describes in detail the general concept
for WOTAS, outlining the special features of the design and
selection of system components. Two control strategies developed
for tremor suppression with exoskeletons are described. These two
strategies are based on biomechanical loading and notch ﬁltering
the tremor through the application of internal forces. Results
from experiments using these two strategies on patients with
tremor are summarized. Finally, results from clinical trials are
presented, which indicate the feasibility of ambulatory mechanical
suppression of tremor.
Index Terms—Human–robot interface, orthotic tremor suppres-
sion, rehabilitation robotics, tremor.
HE SCIENTIFIC community is becoming more and more
interested in rehabilitation robotics. From a robotic per-
spective, wearable robots are mechatronic systems worn by a
person in such a way that the physical interface permits a direct
transfer of mechanical power and exchange of information. A
wearable robot is designed to match the shape and function of
the human body. Segments and joints correspond to some extent
to those of the human body while the system is externally cou-
pled to the person. Initially, the primary applications of these
robotic mechanisms were teleoperation and power ampliﬁca-
tion. Later, exoskeletons have been considered as rehabilitation
Manuscript received September 27, 2006; revised March 29, 2007; accepted
April 15, 2007. The work presented in this paper has been carried out with
the ﬁnancial support from the Commission of the European Union, within
Framework 5, speciﬁc RTD programme “Quality of Life and Management of
Living Resources”, Key Action 6.4 “Aging and Disabilities”, under Contract
QKL6-CT-2002-00536, “DRIFTS—Dynamically Responsive Intervention for
E. Rocon, A. F. Ruiz, J. C. Moreno, and J. L. Pons, are with the Biomedical
Engineering Group at Consejo Superior de Investigaciones Cientíﬁcas, Madrid
28500, Spain (e-mail: email@example.com).
J. M. Belda-Lois is with the Instituto de Biomecánica de Valencia (IBV),
Valencia 46022, Spain.
M. Manto is with Hôpital Erasme, Brussels 1070, Belgium.
Color versions of one or more of the ﬁgures in this paper are available online
Digital Object Identiﬁer 10.1109/TNSRE.2007.903917
and assistive devices for disabled or elderly people by means
of upper and/or lower limb orthosis. One important and speciﬁc
feature of wearable robotics is the intrinsic interaction between
human and robot. This interaction is twofold: ﬁrst, cognitive, be-
cause the human controls the robot while it provides feedback
to the human; secondly, a biomechanical interaction leading to
the application of controlled forces between both actors.
On the one hand, a typical example of the cognitive inter-
action is the one being developed through the EMG control
of robotic prostheses. Here, the human myoelectrical signals
are used to develop control commands to drive an intelligent
prosthesis. Force feedback can be implemented by a number of
means. On the other hand, a classical example of biomechanical
interaction is the exoskeleton based functional compensation of
human gait. Here, the robotics exoskeleton applies functional
compensation by supporting human gait, i.e., by stabilizing the
stance phase. The fact that a human being is an integral part of
the design is one of the most exciting and challenging aspects in
the design of biomechatronics wearable robots. It imposes sev-
eral restrictions and requirements in the design of this sort of
Tremor is characterized by involuntary oscillations of a part
of the body. The most accepted deﬁnition is as follows: “an
involuntary, approximately rhythmic, and roughly sinusoidal
movement” . Tremor is a disabling consequence of a number
of neurological disorders. Although the most common types
of tremor were subject to numerous studies, their mechanisms
and origins are still unknown. Tremor, the most common of all
involuntary movements, can affect various body parts such as
the hands, head, facial structures, tongue, trunk, and legs. Most
tremors, however, occur in the hands.
Tremor is a disorder that is not life-threatening, but it can be
responsible for functional disability and social inconvenience.
More than 65% of the population with upper limb tremor have
serious difﬁculties performing daily living activities , .
In many cases, tremor intensities are very large, causing total
disability to the affected person. The overall management of
tremor is directed towards keeping the patient functioning in-
dependently as long as possible while minimizing disability.
It has been established in the literature that most of the dif-
ferent types of tremor respond to biomechanical loading –.
In particular, it has been clinically tested that the increase of
damping and/or inertia in the upper limb leads to a reduction
of the tremorous motion , . This phenomenon gives rise
to the possibility of an orthotic management of tremor. An or-
thosis is a wearable device that acts in parallel to the affected
limb. In the case of tremor management, the orthosis must apply
a damping or inertial load to a selected set of limb articulations.
As a wearable device, it must exhibit a number of aesthetic and
1534-4320/$25.00 © 2007 IEEE
Authorized licensed use limited to: CSIC. Downloaded on May 27,2010 at 10:09:17 UTC from IEEE Xplore. Restrictions apply.
368 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
functional characteristics. Aesthetics is more directly related to
size, weight and appearance of the exoskeleton. Functionality is
related to generating required torque and velocity while main-
taining the robustness of operation.
In the framework of the DRIFTS (dynamically responsive
intervention for tremor suppression) project , the WOTAS
exoskeleton was presented with three main objectives: moni-
toring, diagnosis, and validation of nongrounded tremor reduc-
tion strategies . This paper presents the development and val-
idation of such a platform. In the next section, the biomechanics
of the upper limb is studied. In third section, the description of
the WOTAS exoskeleton is given. Next, two novel nongrounded
control strategies for suppression of tremor by means of an or-
thotic (wearable) exoskeleton are presented. Both are based on
biomechanical loading, but one is active and the other is passive.
1) Tremor reduction through impedance control. This strategy
modiﬁes the stiffness, damping and mass properties of the upper
limb in order to suppress tremor. 2) Notch ﬁltering at tremor
frequency. This strategy implements an active noise ﬁlter at the
tremor frequency taking advantage of the repetitive characteris-
tics of tremor. Section IV describes the clinical experiments for
system validation. Finally, the conclusions and future work of
this study are given.
An orthosis is deﬁned as a medical device that acts in parallel
to a segment of the body in order to compensate some dysfunc-
tion. The main function of the arm is to position the hand for
functional activities. The hand must be able to reach any point
in the space, especially any point on the human body, in such
a way, that the person can manipulate, draw on, and move ob-
jects to or from the body. Therefore, the kinematic chain formed
by the shoulder, elbow, forearm, wrist, and hand, has a high de-
gree of mobility. In this way, the upper limb is one of the most
anatomic and physiologically complex parts of the body.
The upper limb is very important because it is able to ex-
ecute cognition-driven, expression-driven, and manipulation
activities. Furthermore, it intervenes in the exploration of the
environment and in all reﬂex motor acts. For this reason, any
alteration or pathology that affects the upper limb motion
range, muscle power, sensibility, or skin integrity will alter
its operation. The concept of WOTAS is to develop an active
upper limb exoskeleton based on robotics technologies capable
of applying forces to cancel tremor and retrieve kinematic
information from the upper limb.
A. Mechanical Design
WOTAS was developed to provide a means of testing
nongrounded tremor reduction strategies. WOTAS follows
the kinematic structure of the human upper limb and spans
the elbow and wrist joints, see Fig. 4. It exhibits three de-
grees-of-freedom corresponding to elbow ﬂexion–extension,
forearm pronation–supination, and wrist ﬂexion–extension.
In the ﬁnal design, WOTAS restricts the movement of wrist
adduction–abduction. This strategy was chosen because this is
the upper limb movement with the least impact in daily living
Fig. 1. Scheme of the pronation–supination control.
The mechanical design of the joints for elbow ﬂexion–exten-
sion, and wrist ﬂexion–extension are similar to other orthotic
solutions and is based upon the behavior of those physiolog-
ical joints as hinges. For orthotic purposes, ﬂexion–extension
movement is considered as a pure rotational movement. There-
fore, this axis of rotation should be used for the rotational ac-
tuator. The axis of rotation for the elbow joint is placed in the
line between the two epicondyles. The axis of rotation for the
wrist joint is located in the line between the capitate and lunate
bones of the carpus. The mechanical design for the control for
the pronation supination movement is more complex and it is
1) Pronation–Supination Control: The pronation–supina-
tion movement of the forearm is a rotational movement of the
forearm on its longitudinal axis which engages two joints that
are mechanically connected: the upper radioulnar joint (which
belongs to the elbow) and the lower radioulnar joint (which
belongs to the wrist) . There are two bones in the forearm
that make this movement possible.
• The ulna is the bone that remains ﬁxed during the prona-
tion–supination movement. It constitutes the main part of
the elbow, in particular the olecranon.
• The radius is the moving bone in the pronation and supina-
tion. It rotates in proximal part (close to the elbow) and
moves distally along the axis formed by the ulna bone (see
Both bones have a shape approximately pyramidal and they
are placed in such a way that the base of the radius is in the tip
of the ulna and vice-versa.
The WOTAS platform controls the pronation–supination
movement with the rotation control of a bar parallel to the
forearm. This bar is ﬁxed very close to the olecranon (Fig. 1,
point B). Thus the bar is ﬁxed to the ulnar position at elbow
level. The distal ﬁxation of the bar is made at the head of
the radius, although the bar is maintained in the ulnar side in
order to minimize the excursion of the system. This ﬁxation is
explained later in the support design section.
2) Design of the Support System: There are no static or-
thoses that achieve tremor suppression due to the intrinsically
Authorized licensed use limited to: CSIC. Downloaded on May 27,2010 at 10:09:17 UTC from IEEE Xplore. Restrictions apply.
ROCON et al.: DESIGN AND VALIDATION OF A REHABILITATION ROBOTIC EXOSKELETON 369
Fig. 2. Structure of WOTAS support.
dynamics characteristics of tremor. In these cases, the tremor
suppression mechanism tends to lose their alignment instead of
To determine the points of the upper limb where the dynamic
forces would be applied, i.e., the points where the arm supports
would be placed for the interface between the actuators and the
arm, a number of biomechanical and physiological considera-
tions of the upper limb have to be observed, such as : 1) the
forces on the arm tissues must stay in acceptable limits; 2) the
application of forces on the arm that have minimal interference
of elbow and wrist movement; 3) the interaction of the robotic
device with the arm, i.e., where the forces will be applied on the
upper limb and how the load will be transmitted to the person for
optimum comfort. To respond to these issues, a biomechanical
study was done of the upper limb . The aim of this study
was to determine the limits of comfort regarding pressure, so
that there is an upper limit to the total force that can be applied
safely to the upper limb. This study analyzed two key aspects:
the person’s perception of the pressure and the maximum pres-
sure tolerance thresholds . The ﬁrst aspect is important to
select the appropriate strategy to apply to the load on the body.
For the development of the mechanical structure, different
types of materials for the securing or support elements between
the orthotic device and the arm were considered. The mechan-
ical conditions of these elements are critical because they must
ergonomically couple the upper limb, and also the rigidity of the
material must be greater than the rigidity of the underlying tis-
sues. To securely ﬁx the structure it was decided to use supports
made from thermoplastic. With this type of material, supports
are obtained that adapt to the morphology of each user’s arm,
Fig. 2. Each support has at least three contact points per seg-
ment and thus misalignments are avoided between the orthosis
and limb . Velcro straps were ﬁxed to the fabric in order to
tighten the support to the arm. The ﬁxation to the wrist is mainly
placed over the radial side for the reason that the wrist follows
the movement of the head of the radius. Both the distal tip of the
pronation–supination control and the proximal tip of the wrist
control are ﬁxed to the ulnar side of the wrist. The ﬁxation to
the hand is very similar to that of the wrist. In addition, a tex-
tile substrate was used to compress the soft tissues and enhance
performance of the ﬁxation supports.
The system aims to allow both monitoring of tremor data and
implementation of tremor suppression strategies. Therefore, it
is equipped with kinematic (angular velocity) and kinetic (in-
teraction force between limb and orthosis) sensors.
Tremor force, position, velocity, and acceleration are the re-
quired information to implement the two control strategies. The
Fig. 3. Full Wheatstone bridge and gauges (R1, R2, R3, R4) mounted on
types of sensors were restricted to the following sensors: go-
niometry, gyroscopes, and accelerometers. MEMS gyroscopes
were selected as a promising technology. The main advantages
of using gyroscopes are : they measure rotational motion
(human motion is rotational about joints), they are not inﬂu-
enced by gravity, both frequency and amplitude information ac-
curate down to direct current (dc) (zero frequency), based on
mathematical operations it is possible to obtain angular dis-
placement and acceleration, high signal-to-noise ratio, high dy-
namic range, and solid state gyros do not inﬂuence motion of
subject being measured.
An analysis of commercial solid-state gyros was performed.
Two alternatives from the Japanese manufacturer Murata were
selected: the GYROSTAR ENC-03J and its surface mounted
device (SMD) version, and the GYROSTAR ENC-03M.
Since gyroscopes provide absolute angular velocity in its
active axis, the combination of two independent gyroscopes,
placed distallly and proximally to the joint of interest, is
required. In order to perform the required treatment of the
sensor output to integrate them into WOTAS architecture
some electronics have been developed: a band pass ﬁlter with
a low cutoff frequency of 0.3 Hz in the sensor output and a
higher cutoff frequency of 25 Hz. This is the frequency band
we considered relevant to the application . The concept of
using gyroscopes as an ungrounded method to assess tremor
variables was evaluated and approved. The system is light,
portable, small, inexpensive, and does not lead any discomfort
to subjects .
Since no backdrivable actuators were chosen for the appli-
cation (see next section), it was also decided to use force sen-
sors as a means of implementing impedance feedback control
strategies. Strain gauges in a full Wheatstone bridge have been
selected as force sensors. The gauges measure the torque ap-
plied by the motors on the WOTAS structure, therefore, they
are mounted on the structure so that they only measure the force
perpendicular to the motor axis
, thus their measurement is
not affected by forces caused in undesired directions, Fig. 3. The
extensiometric gauges are connected to a Wheatstone Bridge
circuit in a combination of four active gauges (full bridge).
The strain gauge system was characterized and the sensitiv-
ities of the system were derived in the three planes. The tests
performed showed that the system has a very low sensitivity to
Authorized licensed use limited to: CSIC. Downloaded on May 27,2010 at 10:09:17 UTC from IEEE Xplore. Restrictions apply.
370 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
orthogonal forces. The system presented a thermal drift, how-
ever, after a few minutes the output voltage becomes stable and
suitable for operation.
Before the selection of speciﬁc actuators to suppress tremor,
an estimation of the required torque and power was performed.
This was achieved through an analysis of kinematic tremor data
obtained from a study performed by the authors . From this
torque estimation of effort that the exoskeleton structure must
support, duralumin was selected as the material to construct the
exoskeleton structure. This material was selected in order to
build a lightweight structure with sufﬁcient rigidity to support
Based on this study, a number of candidate actuators were
evaluated . The analysis was restricted to the following ac-
tuators: electro active polymers (EAPs), electro and magneto
rheological ﬂuids (ERF-MRF), dc motors, shape memory actu-
ators (SMAs), pneumatic muscles, and ultrasonic motors.
The actuator technology should have a high-power density al-
lowing the implementation of a compact and light solution suit-
able for wearable devices. Based on this criteria, both dc motors
and ultrasonic motors can be regarded as best alternatives for
exerting tremor suppression forces. The former are well-known
off the-shelf technologies of easy integration in advanced con-
trol schemes but bulky. The latter are less ﬂexible but offer very
compact solutions due to their dynamic range (low velocity high
torque motors). Both alternatives were evaluated in prototypes
and dc motor was the technology selected for the ﬁnal version
of WOTAS. The main problem with the ultrasonic motors was
their poor response at low speed, therefore, leading to problems
to track user’s slow voluntary motion .
Owing to the problems encountered with ultrasonic motors,
a new WOTAS device was constructed using dc motors as an
actuation element. The dc motor selected to activate WOTAS
articulations was a Maxon dc ﬂat brushless motor EC45. In
order to match the speed and torque of the dc motor to the
application requirements, a harmonic transmission drive was
used. In particular, the drive selected for our application was the
HDF-014-100-2A. This conﬁguration, based on ﬂat dc motors
and pancake transmissions, is able to provide a maximum torque
of 8 N.m, nevertheless the maximum torque was electronically
limited to 3 N.m in order to guarantee the safety of the user.
Fig. 4 illustrates the ﬁnal conﬁguration of WOTAS system acti-
vated by dc motors.
The total weight of the ﬁnal system is roughly 850 g. A pro-
tocol for testing the system was performed to evaluate the us-
ability and the range of workspace allowed to a normal user.
The system was used in the laboratory to perform a wide va-
riety of manoeuvres in free mode. These preliminary tests suc-
cessfully showed the correct operation of the system and the ca-
pability of the system to access the workspace, without affecting
the normal range of motion of the user .
D. Control Architecture
The WOTAS control architecture basically consists of three
components: 1) the exoskeleton, with its structure, sensors and
Fig. 4. Final version of WOTAS for the control of three human upper-limb
movements: ﬂexion–extension of the elbow, ﬂexion–extension of the wrist and
pronation–supination of the forearm.
actuators; 2) a control unit, responsible for executing the algo-
rithms in real time to suppress the tremor and the acquisition
card for the interface between the sensors, actuators and the con-
troller; 3) a remote computer, which in our case executes an ap-
plication developed for the interface between the system and the
doctor who is using it.
This control architecture interfaces the control algorithms
with the orthosis. The WOTAS circuitry and sensors serve
two functions: 1) to obtain the position, angular velocity and
acceleration signals needed for control, data collection, and
evaluation and 2) to generate the power signal to activate the
The control of the entire active orthosis was implemented in
the MatLab RT environment. This environment provides math-
ematically complex control strategies in real time. The inter-
face between the MatLab environment and the active orthosis
is based on a standard data acquisition board.
In order to provide an interface to all the control strategies, a
software application was developed in C language. It commu-
nicates with the low level controller (either by TPC/IP, wired
serial link or BlueTooth) using Dynamic Link Libraries (DLLs)
developed by the authors . This interface monitors signals
and tune controller parameters in real time during the control
strategy execution. It has ﬁve main objectives: 1) monitoring
and validation of algorithms for tremor suppressionl; 2) data
analysis (statistics, algorithms performance, etc.); 3) storage of
user information such as clinical and anthropometrics data; 4)
comparison between different control strategies; 5) extraction
of user parameters: joint position, velocity, and acceleration;
tremor frequency (algorithm implemented in the Target PC);
tremor torque; and power (based on an upper limb biomechan-
ical model) .
It was also possible to save all the information retrieved by
the sensors for future ofﬂine analysis.
ROCON et al.: DESIGN AND VALIDATION OF A REHABILITATION ROBOTIC EXOSKELETON 371
III. CONTROL STRATEGIES FOR
The approach to suppress the pathological tremor within this
study is to assist the limb with compensatory technology in
order to decrease the amplitude of tremor. Brieﬂy, the control
system should work as follows: sensors coupled to the limb
measure its motion, an error cancelling algorithm performs a
real-time discrimination of the undesired component of motion,
tremor information is input to the controller in order to gen-
erate the desired actuator action to suppress the tremor. This
concept can be approached either by ambulatory orthotic de-
vices or nonambulatory solutions among which we can consider
wheelchair mounted devices. The former approach is character-
ized by selective tremor suppression through internal forces at
particular joints, while the latter relies on a global application
of external forces that leads to overall tremor reduction. Both
ambulatory and nonambulatory concepts for tremor suppression
can be found in the literature. Furthermore, both concepts can
be implemented through passive and active systems, , .
In active concepts, the system generates an equal but oppo-
site motion to the tremor, actively compensating and effectively
subtracting the tremor from the overall motion. This force is
generated by the system’s motors as a result of a control algo-
rithm. In passive concepts, a mechanical damper is used, thus
the dissipative force usually results from viscous friction or in-
ertia provided by the damper .
One of the main drawbacks of the systems described in lit-
erature is that the dissipative force is also loading the patient’s
voluntary motion. As a consequence, the user feels a mechan-
ical resistance to the motion. Even though in adaptive systems
this could be avoided, ﬁltering out the voluntary motion elim-
inates the resistance to the voluntary motion. For a successful
adaptive tremor absorption mechanism, a means for intelligent
detection of tremor versus voluntary motion is required. To this
end, a model of the tremor motion is proposed in next section.
Two novel control approaches were developed, one passive and
one active, based on biomechanical loading for tremor suppres-
sion by means of wearable exoskeletons .
• Tremor reduction through impedance control—imple-
ments an impedance control, i.e., the stiffness, damping,
and mass properties of the upper limb can be modiﬁed to
study its effects on tremor.
• Notch ﬁltering at tremor frequency—based on noise reduc-
tion techniques and implements an active noise ﬁlter at the
A. A Model of Tremor
In addition to being important to distinguish between desired
and undesired motion, the analysis of the tremor signal, both
in terms of frequency and amplitude, is relevant to assess the
stationary characteristics of tremor, i.e. frequency drift and am-
plitude variation. This information is important when designing
control strategies to counteract tremor.
A number of estimation algorithms have been developed for
tremor suppression. As a ﬁrst approach, we used robust algo-
rithms based on IEEE-STD-1057, which is a standard for ﬁt-
ting sine waves to noisy discrete-time observations. In partic-
ular, the weighted-frequency Fourier linear combiner (WFLC)
developed by Riviere  in the context of actively counter-
acting physiological tremor in microsurgery was implemented.
The WFLC is an adaptive algorithm that estimates tremor using
a sinusoidal model, estimating its time-varying frequency, am-
plitude, and phase. The WFLC can be described by (1). It as-
sumes that the tremor can be mathematically modelled as a pure
sinusoidal signal of frequency
plus harmonics and com-
putes the error
between the motion and its harmonic model
In its recursive implementation, see (2) and (3), the WFLC
can be used online to obtain estimations of both tremor fre-
quency and amplitude 
The WFLC algorithm was evaluated in signals measured in
patients suffering tremor, . In the completed trials, the algo-
rithm was able to estimate the tremor movement of all the pa-
tients with accuracy always lower than 2
. The main disadvan-
tage of the WFLC is the need for a preliminary ﬁltering stage to
eliminate the voluntary component of the movement . This
ﬁltering stage introduces an undesired time lag for our system
when estimating tremor movement, this time lag introduces a
time delay that could considerably affect the implementation of
the control strategies for tremor suppression.
The solution adopted was the development of an algorithm
capable of estimating voluntary and tremorous motion with a
small phase lag. The tremor literature , , , indicates
that voluntary movements and tremor movements are consid-
erably different. Voluntary movements are slower while tremor
movements are brusquer. This indicates that adaptive algorithms
to estimate and track movement would be useful when sepa-
rating the two movements with an appropriate design. The un-
derlying idea is to design the ﬁlters so that they only estimate
the less dynamic component of the input signal, which in our
case we consider to be voluntary movement, thereby ﬁltering
out the tremor movement. Thus, to estimate voluntary move-
ment and tremor movement, the development of a two-stage al-
gorithm is proposed to estimate voluntary movement and tremor
movement with a minimum time lag, see Fig. 5.
In the ﬁrst stage, a set of algorithms was considered for the
estimation of the voluntary motion: two point-extrapolator, crit-
estimator, Benedict–Bordner esti-
mator, and Kalman ﬁlter. These algorithms implement both esti-
mation and ﬁltering equations. The combination of these actions
allows the algorithm to ﬁlter out the tremorous movement from
the overall motion at the same time it reduces the phase lag intro-
duced, . The equation parameters were adjusted to track the
movements with lower dynamics (voluntary movement) since
372 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
Fig. 5. Two-stage tremor modelling: ﬁrst, the low frequency content voluntary
motion is estimated, secondly, the voluntary motion estimation is subtracted
from the original motion, eventually, tremor frequency and amplitude are
tremors present a behavior characterized by quick movements
The algorithms evaluated were two degrees of freedom esti-
mators i.e., they assume a constant velocity movement model.
This assumption is reasonable since the sample period is very
small compared to the movement velocities , i.e., the sample
period adopted was 1 ms and the voluntary movement estimated
occurs in a bandwidth lower than 2 Hz. The performance of
these algorithms were compared based on their accuracy when
estimating voluntary movements of tremor time series from pa-
tients. The result of such analysis indicated that Benedict–Bor-
dner ﬁlter presents the best results with the lowest computational
cost, . This estimation algorithm is a
ﬁlter with the fol-
lowing tracking update equations:
prediction equations 
The tracking update equations or estimation equations [(4)
and (5)] provide the joint angular velocity and position. The es-
timated position is based on the use of the actual measurement
as well as the past prediction. The estimated state contains all the
information we need from the previous measurements. The pre-
dicted position is an estimation of
based on past states and
prediction, (6) and (7), and take into account the current mea-
surement by means of updated states. The Benedict–Bordner
estimator is designed to minimize the transient error. There-
fore, it responds faster to changes in movement velocity and it is
slightly under-damped . The relation between ﬁlter param-
eters are deﬁned by (8)
In the second stage, the estimated voluntary motion is re-
moved from the overall motion and it is assumed that the re-
maining movement is tremor. After this, we use the WFLC in
order to estimate tremor parameters. In this stage, the algorithm
estimates both the amplitude and the time-varying frequency of
the tremorous movement.
Fig. 6. Modelling of tremor as a sinusoidal non-voluntary motion: velocity
signals (gray) obtained from gyroscopes, estimation of voluntary movement
(black) and estimation of tremorous movement (red).
This algorithm was evaluated with data obtained from 40 pa-
tients suffering from different tremor diseases. The estimation
error of the ﬁrst stage was 1.4
1.3 . The second stage
algorithm has a convergence time always smaller than 2 s for all
signals evaluated and the mean square error (MSE) between the
estimated tremor and the real tremor (obtained ofﬂine by means
of manual decomposition based on classical ﬁlter techniques),
after the convergence, is smaller than 1
. The combination of
both techniques resulted in a very efﬁcient algorithm with small
processing cost for estimating in real time the voluntary and the
tremorous components of the overall motion . Fig. 6 illus-
trates the performance of the algorithm when splitting voluntary
and tremorous movements of a patient of essential tremor.
B. Tremor Reduction Through Impedance Control
The impedance of a system comprises three components, i.e.,
stiffness, damping and inertia . There is evidence  that all
three components modify the biomechanical characteristics of
tremor at the upper limb, which in general, can be described by
a second-order system , .
Our approach consists in changing the biomechanical charac-
teristics of the musculo-skeletal system by means of selecting
the appropriate modiﬁed value of damping and inertia of the
musculo-skeletal system in order to reduce the amplitude of the
tremorous movement, see Fig. 7. Unlike other approaches in the
literature, the control scheme is conceived so that the effect of
the suppression load on voluntary motion is minimized.
The control approach is based on a dual control loop, Fig. 7.
The value of force applied by the exoskeleton over the upper
, is calculated based on the summation of the effects of
This closed-loop control architecture uses the information
from the gyroscopes
on the data treatment block to distin-
guish between voluntary and tremor motion,
, from the overall
ROCON et al.: DESIGN AND VALIDATION OF A REHABILITATION ROBOTIC EXOSKELETON 373
Fig. 7. Control strategy to modify the biomechanical parameters of the upper limb.
Fig. 8. Active tremor suppression control strategy.
movement (as explained in the previous section). The angular
velocity information from the estimated tremorous component,
, is subsequently multiplied by the coefﬁcients and ,
which describe the reference inertia and damping characteris-
tics of the upper limb. This process deﬁnes the actual impedance
, of the system which is set to reduce the tremor. This
impedance force should tend to vanish as tremorous motion is
The lower loop of Fig. 7 serves the goal of minimizing the
effect of the orthosis on the voluntary motion. In this case,
force sensors measure the interaction force between orthosis
and the limb,
. Under ideal circumstances, a user free of
pathologic tremor should feel no force from the orthosis, i.e.,
no loading should be a consequence of voluntary motion. In
order to achieve this, the interaction force
is ﬁltered so that
just the force opposing voluntary motion
is fed back in the
lower branch of the control loop.
The control strategy proposed has an adaptive behavior so
that constantly (in real time) it updates the tremor amplitude
estimate. Thus the system can respond to the changes produced
by the control strategy on the tremor, .
C. Notch Filtering at Tremor Frequency
Tremor frequency varies according to the particular neuro-
logical disorder being considered. In particular, while essential
tremor takes place in the frequency range between 5 and 8 Hz,
rest tremor is usually found at a slightly lower frequency range,
In addition, for a given type of tremor, its main frequency
varies from patient to patient, but tends to be quite stable for a
particular subject. This property should be exploited when de-
signing a control strategy to counteract tremor. In particular,
repetitive control can handle periodic (repetitive) signals and
disturbances. Repetitive control can be regarded as a subset of
learning control since the control action is determined using the
stored error values from preceding periods. Even though repet-
itive approaches can handle periodic signals (tremor), it is not
free from some common problems: tight stability conditions,
poor response to nonperiodic and nonharmonic signals and poor
noise characteristics. As reported by Inoue , a stabilizing
compensation and smoothing of the control signal over periods
can be used to overcome the above mentioned problems.
The control strategy designed and proposed for active control
of the pathological tremor manages the exoskeleton to generate
a motion equal but opposite to the tremor, based on the real
time estimation of the involuntary component of motion. The
control strategy that controls the velocity of each articulation of
WOTAS is illustrated in Fig. 8.
As illustrated in Fig. 8, this control strategy is also composed
by a dual control loop. The upper control loop is responsible for
tremor suppression while the lower control loop should mini-
mize the inﬂuence of the control strategy on the voluntary move-
ment, by means of an admittance control .
374 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
In the upper loop, the algorithm presented in Section III-A
estimates the angular velocity of tremorous movement,
value is delayed by
seconds. is deﬁned by (10), where
is the sample frequency of the control strategy and is the
estimated tremor frequency
The velocity amplitude of the tremorous movement delayed
samples is fed-back to the inner velocity control loop of
WOTAS exoskeleton. This makes the orthosis replicate the ve-
locity amplitude of the tremorous movement but opposite to the
tremor, implementing a notch ﬁlter at tremor frequency and ac-
tively compensating the tremorous movement of the user .
The following equation deﬁne the upper control loop of the ac-
As in the previous control approach, the quality of the tremor
suppression control approach is strongly dependent on accurate
estimation and tracking of tremor frequency. In  it is shown
that the system is stable when the algorithm for tremor estima-
tion is stable.
In summary, the operation of WOTAS system for tremor sup-
pression is based on the identiﬁcation of tremorous motion out
of tremorous motion measured by kinematic and kinetic sensors.
Adaptive algorithms help identify and distinguish tremorous
motion from the voluntary one. This information is then used
to establish a proper physical interaction (modiﬁcation of the
combined human-exoskeleton articular impedance or applying
forces opposite to tremor) that should result in tremor reduction.
Another very important characteristic in the implementation
of the proposed control strategies is that the control strategy pro-
posed is based on an articular control approach because it is sim-
pler and also makes it possible to implement individual control
loops in each joint with a high dynamic range . Furthermore,
the fact that each exoskeleton joint tries to suppress the tremor
generated in its corresponding anatomical joint is interesting be-
cause it guarantees reducing the tremor in each joint. Thus, the
problem of coupling the tremor between the upper-limb joints
is successfully tackled .
In studies done by the authors, the behavior and contribution
of each joint in the upper-limb tremor were evaluated, . This
work has shown that in most patients the tremor movement “dis-
places” along the kinematic chain of the arm when its effects
are reduced (by applying biomechanical loads) on one of the
arm joints. However, the study of tremor behavior when its ef-
fects are cancelled in different joints of the arm has still not been
properly studied . This aspect led to devising active and inde-
pendent control strategies in each joint. Accordingly, if the can-
cellation of the tremor in one of the joints increases the tremor
in the other joint, the algorithm responsible for controlling the
adjacent joint will identify the increased tremor and try to re-
duce the tremor generated by coupling the upper-limb joints.
The aim is thus that the active behavior of tremor reduction in
each joint reaches equilibrium, thereby decreasing the coupling
effects of the upper-limb joints. In addition, the range of move-
ment of each articulation of the exoskeleton is limited by the
control strategies in such a way that they never exceeds the nat-
ural range of motion of the user. This was implemented in order
to guarantee user safety.
In order to evaluate the performance of the device devel-
oped to suppress tremor, we conducted an experimental phase
involving 10 patients suffering from different tremor diseases.
These experiments were conducted in Hôpital Erasme, in Bel-
gium, and Hospital General Universitário de Valencia, in Spain,
and were led by a neurologist. The protocol, the results, and the
analysis of the results are presented in the following sections.
Ten users participated in these experiments (three women,
mean age 52.3 years). Users presented different pathologies but
the majority was affected by essential tremor (ET). ET was mod-
erate in users 1, 3, 4, and 7 and severe in users 2, 5, and 6. User 8
suffer from multiple sclerosis, user 9 from posttraumatic tremor
and user 10 is affected by a mixed tremor. All users provided
their informed consent. The investigation was approved by the
ethical committee of both hospitals. All the experiments were
The users still exhibited tremor despite a regular intake of
the drugs conventionally administered for tremor, even at high
doses for some of the users.
B. Materials and Methods
Three different people were present during the
• A computer operator: In charge of setting the parameters
of the systems and recording the signals.
• A medical doctor: In charge of supervision of the condition
of the trials and the state of the user.
• An experimenter: In charge of ﬁtting and removing the or-
thosis and interacting with the user to perform the trials.
During the experiments neither the user, nor the experimenter
or the medical doctor knew when the systems were applying
a suppressing strategy or when it was operating in monitoring
mode. Just the computer operator knew when the systems were
applying the suppression strategy. For formal purposes, we con-
sider this arrangement equivalent to a double-blind trial in order
to reduce the placebo effects in the experimentation phase .
Three different tasks were selected to be performed by the
users: keep the arm outstretched, point the nose with a ﬁnger,
and keep the arm in a rest position. These tasks have been pre-
viously used to characterize tremor movement .
During the experiments, WOTAS operated basically in its
three different control modes as follows.
1) Monitoring mode. WOTAS operate in free mode (no force
is applied on the upper limb) and monitor tremor parame-
ters of the users.
2) Passive Control mode. WOTAS is able to change biome-
chanical characteristics of upper limb, such as viscosity or
inertia, in order to suppress tremor (Section III-B).
ROCON et al.: DESIGN AND VALIDATION OF A REHABILITATION ROBOTIC EXOSKELETON 375
3) Active Control mode. WOTAS is able to apply oppo-
site forces to the tremorous movement based on a real
time estimation of the involuntary component of motion
The order in which the modes have been applied has been al-
ternated, as well as the order in which the users have executed
the tasks. In each experimental session, three repetitions of each
task were realized. This approach was adopted in order to avoid
interactions in the analysis, as well as learning effects . The
number of repetitions has been chosen in order to have an ex-
perimental session not longer than 1 h.
D. Data Analysis
The data analyzed were the output voltage coming from the
gyroscopes placed on the active orthosis. This output voltage
was sampled at a 2000 Hz rate. The data has been ﬁltered using
a Kernel Smoothing algorithm and a gaussian window 51 points
width. The ﬁgure of merit adopted to quantify the reduction
achieved by the exoskeleton is the ratio between the signal an-
alyzed in monitoring mode
, and the signal analyzed in
, both in passive or active modes, (12).
Therefore, the reduction of tremor was measured with the users
under the same conditions: with the orthosis placed on the upper
limb. As a result, the estimated reduction was the remaining
tremor in suppression mode referred to tremor at monitoring
The parameter selected to compare the tremor level is the
Power contained in the frequency band from 2 to 8 Hz . It
can be deﬁned as follows:
is the signal in the time domain, is the length of the
is the sampling period and, and are the lower
and upper limits of the range of interest.
E. Results and Discussion
The effects of adding effective viscosity were investigated
for the upper limb during the execution of the different tasks.
During the trials, some users were able to identify when the
system was operating in suppression mode, relating to the clin-
ician either “now the system is suppressing my tremor” or “now
it is not.”
Fig. 9 illustrates the performance of WOTAS when operating
in suppression mode for all subjects in this experiment. No-
tice that the efﬁciency of the exoskeleton improves with tremor
power increase. An statistical analysis has been made to char-
acterized the tremor suppression. The statistical analysis has
been made using R . A second-order polynomial ﬁt has been
made with the natural logarithms of power spectra in free and
suppression mode, see red line on Fig. 10. As showed in Table I,
Fig. 9. Tremor reduction ( axis) achieved by WOTAS operating in sup-
pressing both in active (blue markers) and passive (red markers), mode.
represents users’ tremor energy with WOTAS in monitoring mode.
Fig. 10. Polynomial ﬁtting to the data.
the adjusted of the ﬁtting is 0.44 and all the coefﬁcients of
the ﬁtting are statistically signiﬁcant at 0.05.
From this ﬁtting it is possible to estimate where the orthosis
suppress tremor efﬁciently, based on ﬁnding the points where
the ﬁtting crosses the line
(green line on Fig. 10). This
method allows us to identify a lower limit for efﬁcient tremor
suppression, this limit is roughly 0.15
. In Fig. 9, we
can check that the robotic exoskeleton has a minimum tremor
376 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
ESULTS OF THE
Fig. 11. Mode effect on the Ln(PSD).
suppression limit, i.e., if the spectral density of tremor move-
ment is below the lower limit identiﬁed, around 0.15
WOTAS exoskeleton is ineffective in suppressing the tremor.
In order to verify the hypothesis that over the lower limit
) WOTAS is efﬁcient in tremor suppression,
an univariate analysis of variance (ANOVA) analysis was
performed. This analysis was made for all movements which
tremor in free mode was above the threshold obtained by the
regression curve. The hypothesis is that the orthoses is efﬁcient
for severe cases of tremor, when tremor is small (under the
threshold of 0.15
) the system has a negligible contribu-
tion. Five from the six users had movements in which the PSD
of tremor was above the threshold and these are the measure-
ments included in the analysis. The analysis was based on two
factors: system in monitoring mode and system in suppression
mode. Users were included to the model and the dependent
variable was the Neperian Logarithm of the power spectral
density—Ln(PSD). The results of this ANOVA analysis are
shown in Table II. According to the table, the effects due to the
mode (active versus monitoring) as well as due to the user are
statistically signiﬁcant. Moreover, as we can see in Fig. 11, the
active mode reduces the mean value of the Ln(PSD).
The user also have a signiﬁcant effect on the Ln(PSD) but not
on the interaction mode-user as showed in Table II. However,
WOTAS demonstrated its effect in reducing the tremor compo-
nent in all users with tremor superior to the threshold (Fig. 12).
Fig. 12. Effect of the user in the reduction of tremor with WOTAS.
Fig. 13. This ﬁgure illustrates the oscillations of the elbow of tremor with
WOTAS in a monitoring mode and suppressing modes in user 2. Note the strong
reduction in amplitude of tremor when suppressing actions are applied. Oscil-
lations are expressed in rad/sec.
The results also indicated that the range of reduction in tremor
energy for signals above this orthosis operational limit ranges
from 3.4% (percentile 5) to 95.2% (percentile 95) in relation to
energy in monitoring mode.
These reductions can be appreciated in Figs. 13 and 14. These
ﬁgures illustrate the effects of WOTAS on tremorous movement
using both strategies. Fig. 13 illustrates the time series corre-
sponding to the tremorous movement of the elbow joint of user
2 while the arm is outstretched. The top part of the ﬁgure shows
the time signal with WOTAS in the
monitoring mode. Notice
that for both passive and active modes the amplitude of tremor
is clearly smaller than in the monitoring mode.
Fig. 14 illustrates the same reduction in the frequency do-
main. The power spectrum density (PSD) have been obtained
from the part of the signal with tremor. The top part of the ﬁgure
illustrates the PSD of the tremorous movement with WOTAS
ROCON et al.: DESIGN AND VALIDATION OF A REHABILITATION ROBOTIC EXOSKELETON 377
Fig. 14. This ﬁgure illustrates the associated power spectral density (PSD) of
tremor WOTAS in a monitoring mode (upper part) and suppressing modes in
user 2. Note the strong reduction in the PSD of tremor when suppressing actions
are applied. PSD is expressed in
operating in monitoring mode. It is possible to see a clear peak
of tremor activity close to 4 Hz. In the middle, ﬁgure shows the
PSD while WOTAS was operating in active mode. Notice that
the energy associated to tremor activity has been substantially
reduced. In the low part of the ﬁgure the peak of energy corre-
sponding to the tremorous activity when WOTAS is in passive
mode also presents a clear reduction. These results indicate that
WOTAS is able to suppress tremor in addition to validate both
control strategies proposed, active and passive, for control of or-
thotic suppression of tremor. Notice that the frequency of tremor
does not change when the exoskeleton is working in suppression
A detailed analysis of the data showed that the active sup-
pression strategy (81.2% mean power reduction) presents higher
levels of tremor suppression compared to passive suppression
strategy (70% mean power reduction). This suggests a better
performance of the active mode in tremor suppression.
The previous section presents results of the clinical trials re-
lated to the evaluation of the exoskeleton for tremor suppres-
sion presented in the paper. The exoskeleton performance was
assessed in 10 users suffering from different kinds of tremors at
the upper-limb. Two strategies for tremor reduction have been
tested: viscous friction and notch ﬁltering.
The analysis of the videos from the experiments demon-
strated that in the majority of users there was no tremor
displacement to proximal joints of the upper limb. Neverthe-
less, one user in ten presented an increment in tremor activity
at shoulder level. The authors believe that future work should
be performed in order to investigate and deﬁne the proﬁle of
the users affected by this phenomenon.
The results of the experiments indicated that the device could
achieve a consistent 40% of tremor power reduction for all users,
being able to attain a reduction ratio in the order of 80% tremor
power in speciﬁc joints of users with severe tremor. In addition,
the users reported that the exoskeleton did not affect their vol-
untary motion. These results indicate the feasibility of tremor
suppression through biomechanical loading. Nevertheless, the
users reported that the exoskeleton could not be considered as a
solution to their problem since it is bulky and heavy. The users
considered that the use of such device should cause social ex-
clusion. This was expected since the exoskeleton was developed
as a platform to evaluate the concept of mechanical tremor sup-
pression and not as a ﬁnal orthotic solution. The main wish ex-
pressed by the potential users was the possibility of hiding the
exoskeleton under clothing .
The results also indicated a superior performance of active
tremor suppression over passive tremor suppression. However,
the authors believe that this could be due to the fact that the
value of viscosity added to the movement was the same for all
users. The customization of viscosity or inertia added to the
upper limb according to the biomechanical characteristics of
each user should improve the efﬁciency of passive tremor sup-
pression strategy , .
It was noticed that the degree of tremor reduction was depen-
dent upon the power associated with tremor. There are lower
limits for robotic tremor suppression that should be determined
for the mechanical impedance of the contact exoskeleton-skin,
and the particular morphology of tremor.
During the trials two users spontaneously reported that they
felt a decrease in the amplitude of their tremorous movement
and consequently they felt more conﬁdent about the execution
of the task. This indicates that the visual feedback of a smooth
movement has a positive impact in the user. This fact is very im-
portant and future research will be performed in order to eval-
uate this phenomenon with more users of different pathologies.
This paper presented a robotic exoskeleton able to mon-
itor, diagnose and control tremor in subjects. This robotic
exoskeleton is equipped with kinematic and kinetic sensors for
the measurement and calculation of joint angular displacement,
velocity and acceleration, as well as interaction forces between
the limb and orthosis. In addition, it could also apply dynamic
force to the articulations of the upper limb by means of a set
of ﬂat dc motors in combination with pancake gears. The big
innovations of the WOTAS exoskeleton are its portability, its
non invasiveness, and that it provides direct information from
each joint of the upper limb. The exoskeleton was evaluated
with 10 users and validated the concept of tremor suppression
through wearable robotics.
Both proposed strategies for tremor suppression (viscous fric-
tion and notch-ﬁltering) have produced signiﬁcant reduction of
tremor amplitude. The results indicated that notch-ﬁltering of
tremor is more efﬁcient than viscous friction, however further
research should be performed in order to validate this statement.
Some aspects of mechanical tremor suppression were not
evaluated in this study, further trials with a larger number of
users should be held in order to evaluate other different aspects
378 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 15, NO. 3, SEPTEMBER 2007
of WOTAS operation. The exoskeleton will help to analyze
the effect of biomechanical loading on the upper limb motion
and quantify its effects on fatigue. In addition, the carryover
of attenuation effects after the impedance is removed must be
subject to study.
Another very important issue is the interface of the ex-
oskeleton with the upper limb. It was detected that there are
lower limits for robotic tremor suppression determined by the
mechanical impedance of the contact exoskeleton-upper limb.
These limitations make the device useless for users with mod-
erate to low tremor. There are several biomechanical aspects
regarding the transmission of forces from the actuators to the
limb that should be solved in order to improve the exoskeleton
performance. Another aspect that requires further research is
the effect of visual feedback on the capacity of the user to
perform a task.
The capacity of applying internal dynamic forces to the upper
limb opens widely the application ﬁeld of WOTAS exoskeleton.
It could be applied in different areas of the rehabilitation robotic
ﬁeld, for instance, it could provide restoration or maintenance of
motor function to different joints of the upper limb. Most of the
powered orthoses designed to date are nonambulatory devices
. There is a need in the rehabilitation area of ambulatory
devices able to apply dynamic forces to the upper limb.
 A. Anouti and W. Koller, “Tremor disorders: Diagnosis and manage-
ment,” Western J. Med., vol. 162, no. 6, pp. 523–530, 1998.
 R. Elble and W. Koller, Tremor. Baltimore, MD: Johns Hopkins
Univ. Press, 1990.
 E. Rocon, J. Belda-Lois, J. Sanchez-Lacuesta, and J. L. Pons, “Patho-
logical tremor management: Modelling, compensatory technology and
evaluation,” Technol. Disability, vol. 16, pp. 3–18, 2004.
 J. Kotovsky and M. Rosen, “A wearable tremor-suppression orthosis,”
J. Rehabil. Res. Develop., vol. 35, no. 4, 1998.
 M. Rosen, A. Arnold, I. Baiges, M. Aisen, and S. Eglowstein, “Design
of a controlled-energy-dissipation orthosis (cedo) for functional sup-
pression of intention tremors,” J. Rehabil. Res. Develop., vol. 32, no.
1, pp. 1–16, 1995.
 B. Adelstein, “Peripheral mechanical loading and the mechanism of
abnormal intention tremor,” M.S. thesis, Massachusetts Inst. Technol.,
 M. Manto, M. Topping, M. Soede, J. Sanchez-Lacuesta, W. Harwin, J.
Pons, J. Williams, S. Skararup, and L. Normie, “Dynamically respon-
sive intervention for tremor suppression,” IEEE Eng. Med. Biol., vol.
22, no. 3, pp. 120–132, May/Jun. 2003.
 E. Rocon, A. Ruiz, J. Pons, J. Belda-Lois, and J. Sánchez-Lacuesta,
“Rehabilitation robotics: A wearable exo-skeleton for tremor as-
sessment and suppression,” in Int. Conf. Robot. Automat., 2005, pp.
 Guides to the evaluation of permanent impaired Amer. Med. Assoc.,
 I. Kapandji, The Physiology of the Joints: Upper Limb. New York:
Churchill Livingstone, 1983, vol. 1.
 E. Rocon, J. Belda-Lois, J. Sánchez-Lacuesta, A. Ruiz, and J. Pons,
“Estimation of biomechanical characteristics of tremorous movements
based on gyroscopes,” presented at the Conf. Asistive Technol.—From
Virtuality to Reality, Lille, France, 2005.
 E. Rocon, “Reducción activa de temblor patológico de miembro
superior mediante exoesqueletos robóticos,” Ph.D. dissertation, Univ.
Politécnica Madrid, Madrid, Spain, 2006.
 K. Tong and M. H. Granat, “A practical gait analysis system using
gyroscopes,” Med. Eng. Phys., vol. 21, pp. 87–94, 1999.
 E. Rocon, A. Ruiz, and J. Pons, “On the use of rate gyroscopes for
tremor sensing in the human upper limb,” in Proc. Int. Conf. Eurosen-
sors XIX, 2005, p. MP30.
 E. Rocon, A. F. Ruiz, and J. L. Pons, “On the use of ultrasonic motors
in orthotic rehabilitation of pathologic tremor,” in Proc. ACTUATOR
2004 Conf., 2004, pp. 387–390.
 A. F. R. Olaya, “Sistema háptico para monitorización e intervenciónen
pacientes con temblor patológico,” M.S. thesis, Escuela Técnica Supe-
rior de Telecomunicación, Univ. Politécnica Madrid, Madrid, Spain,
 E. Rocon, A. F. Ruiz, and J. L. Pons, “Biomechanical modelling of
the upper limb for robotics-based orthotic tremor suppression,” Appl.
Bionics Biomechan., vol. 2, no. 2, pp. 81–85, 2005.
 C. Riviere, S. Reich, and N. Thakor, “Adaptive fourier modeling for
quantiﬁcation of tremor,” J. Neurosci. Methods, vol. 74, pp. 7–87, 1997.
 C. Riviere, “Adaptive Suppression of Tremor for Improved Human-
Machine Control,” Ph.D. dissertation, Johns Hopkins Univ., Baltimore,
 K. Mann, F. W. Werner, and A. K. Palmer, “Frequency spectrum anal-
ysis of wrist motion for activities of daily living,” J. Orthopedic Res.,
vol. 7, no. 2, pp. 304–306, 1989.
 Y. Bar-Shalom and X. Li, Estimation and Tracking: Principles, Tech-
niques, and Software. Norwood, MA: Artech House, 1998.
 E. Brookner, Tracking and Kalman Filtering Made Easy. New York:
 N. Hogan, “Impedance control: An approach to manipulation: Part
iiiand iii,” J. Dynamics Syst., Measurem. Control, vol. 107, no. 1, pp.
 S. Pledgie, K. E. Barner, S. K. Agrawal, and T. Rahman, “Tremor sup-
pression through impedance control,” IEEE Trans. Rehabil. Eng., vol.
8, no. 1, pp. 53–59, Mar. 2000.
 E. Rocon, A. F. Ruiz, and J. Pons, “Theoretical control discussion on
tremor suppression via biomechanical loading,” in Proc. AAATE2003
Conf.—Assistive Technol.—Shaping Future, 2003, pp. 832–836.
 T. Inoue, “Practical repetitive control system design,” in Proc. 29th
IEEE Conf. Decision Control, 1990, vol. 3, pp. 1673–1678.
 N. Tsagarakis and D. Caldwell, “Development and control of a soft-ac-
tuated exoskeleton for use in physiotherapy and training,” Autonomous
Robots, vol. 15, pp. 21–33, 2003.
 J. Belda-Lois, M. Vivas, A. Castillo, F. Peydro, J. Garrido, J. Sanchez-
Lacuesta, R. B. R, R. Poveda, and J. Prat, “Functional assessment of
tremor in the upper-limb,” in Proc. 8th Congress Eur. Federation Res.
Rehabil., 2004, p. MP30.
 R Development Core Team, R: A language and environment for sta-
tistical computing R Foundation Stat. Comput., Vienna, Austria, 2006
[Online]. Available: http://www.R-project.org
 G. Johnson, D. Carus, G. Parrini, S. S. Marchese, and R. Valeggi, “The
design of a ﬁve degree-of-freedom powered orthosis for the upper
limb,” in Proc. Instrum. Mech. Eng., 2002, vol. 25, p. H.
E. Rocon, photograph and biography not available at the time of publication.
J. M. Belda-Lois, photograph and biography not available at the time of
A. F. Ruiz, photograph and biography not available at the time of publication.
M. Manto, photograph and biography not available at the time of publication.
J. C. Moreno, photograph and biography not available at the time of
J. L. Pons, photograph and biography not available at the time of publication.