A soft wearable robot for tremor assessment and suppression

Abstract

Tremor constitutes the most common motor disorder, and poses a functional problem to a large number of patients. Despite of the considerable experience in tremor management, current treatment based on drugs or surgery does not attain an effective attenuation in 25 % of patients, motivating the need for research in new therapeutic alternatives. In this context, this paper presents the concept design, development, and preliminary validation of a soft wearable robot for tremor assessment and suppression. The TREMOR neurorobot comprises a Brain Neural Computer Interface that monitors the whole neuromusculoskeletal system, aiming at characterizing both voluntary movement and tremor, and a Functional Electrical Stimulation system that compensates for tremulous movements without impeding the user perform functional tasks. First results demonstrate the performance of the TREMOR neurorobot as a novel means of assessing and attenuating pathological tremors.

Full text

Page 1

A Soft Wearable Robot for Tremor Assessment and Suppression
J.A. Gallego, E. Rocon, J. Ib´ a˜ nez, J.L. Dideriksen, A.D. Koutsou, R. Paradiso, M.B. Popovic,
J.M. Belda–Lois, F. Gianfelici, D. Farina, D.B. Popovic, M. Manto, T. D’Alessio, and J.L. Pons
Abstract—Tremor constitutes the most common motor dis-
order, and poses a functional problem to a large number
of patients. Despite of the considerable experience in tremor
management, current treatment based on drugs or surgery
does not attain an effective attenuation in 25 % of patients,
motivating the need for research in new therapeutic alterna-
tives. In this context, this paper presents the concept design,
development, and preliminary validation of a soft wearable
robot for tremor assessment and suppression. The TREMOR
neurorobot comprises a Brain Neural Computer Interface
that monitors the whole neuromusculoskeletal system, aiming
at characterizing both voluntary movement and tremor, and
a Functional Electrical Stimulation system that compensates
for tremulous movements without impeding the user perform
functional tasks. First results demonstrate the performance of
the TREMOR neurorobot as a novel means of assessing and
attenuating pathological tremors.
I. INTRODUCTION
Tremor is defined as a rhythmic activity of a body part.
Although we all exhibit a certain degree of tremor -the so
called physiological tremor-, there are several pathologies
that have associated very disabling tremors. This patholog-
ical tremors constitute the most common motor disorder,
affecting 15 % of people with ages ranging between 50
and 69 years [1]. Moreover, among people suffering from
upper limb tremors, 65 % of them report severe disability to
perform their activities of daily living [2]. In addition to that,
the mechanisms underlying the different forms of tremor are
yet not understood, which often leads to misdiagnosis and
subsequent treatment problems [3].
Tremor is typically managed with drugs, and with surgery
(gamma knife thalamotomy) or implantation of a Deep Brain
The work presented in this paper has been carried out with the financial
support from the Commission of the European Union, within Framework
7, specific IST programme “Accessible and Inclusive ICT”, Target outcome
7.2 “Advanced self-adaptive ICT-enabled assistive systems based on non-
invasive Brain to Computer Interaction (BCI)”, under Grant Agreement
number ICT-2007-224051, “TREMOR: An ambulatory BCI-driven tremor
suppression system based on functional electrical stimulation.”
J.A. Gallego, E. Rocon, J. Ib´ a˜ nez, A.D. Koutsou, and J.L. Pons are with
the Bioengineering Group, Consejo Superior de Investigaciones Cient´ ıficas,
CSIC, Spain, e-mail: gallego@iai.csic.es.
J.L. Dideriksen, F. Gianfelici, and D.B. Popovic are with the Center for
Sensory-Motor Interaction, Department of Health Science and Technology,
Aalborg University, Denmark.
R. Paradiso is with SMARTEX s.r.l, Italy.
M.B. Popovic is with UNA Systems, Serbia.
J.M. Belda–Lois is with Instituto de Biomec´ anica de Valencia, Spain.
D. Farina is with Dept. Neurorehab Eng, Bernstein Center for Computa-
tional Neuroscience, Georg–August University, G¨ ottingen, Germany.
D.B. Popovic also is with Faculty of Electrical Engineering, University
of Belgrade, Serbia.
M. Manto is with Service de Neurologie, Hˆ opital Erasme-ULB, Bruxelles,
Belgium.
T. D’Alessio is with Biolab, University of Rome TRE, Rome, Italy.
Stimulator (DBS) in those patients refractory to medication.
However, the drugs used often induce side effects, or may
be contraindicated, and surgery is associated with a risk of
hemorrhage and psychiatric manifestations, [4]. This causes
that around 25 % of patients suffering from pathological
tremor do not benefit from an effective reduction of their
symptoms, and motivates research in alternative means of
compensating for tremors.
In this context, the authors validated, both functionally
and clinically, tremor suppression based on the application of
biomechanical loads with an upper limb robotic exoskeleton,
[5]. This work relied on the known effects of impedance
modulation in tremor, [6], and provided with up to 80 %
tremor attenuation in severe cases, without impeding the
user to perform functional tasks. Nevertheless, patients were
reluctant to use such an anesthetic and bulky solution as a
robotic exoskeleton.
Here we present the design, development, and first re-
sults of a novel soft wearable robot for tremor assess-
ment and suppression based on biomechanical loading.
The TREMOR neurorobot monitors the neuromusculoskele-
tal system to characterize both concomitant voluntary and
tremulous movements based on a Brain Neural Computer
Interface (BNCI), and then stimulates upper limb muscles
to compensate functionally for the tremor. The TREMOR
neurorobot is implemented as an active garment, fulfilling
users’ expectances in terms of comfort and cosmetics, and
incorporates arrays of sewn electrodes to apply selective
biomechanical loads through Functional Electrical Stimula-
tion (FES). Notice that attenuation of writs tremors based
on muscle stimulation has been previously evaluated in [7],
demonstrating the feasibility of the approach.
This paper is organized as follows. First, we review the
design of the neurorobot and its development, together with
a brief overview of the different experimental protocols
employed in tests with users. Afterwards, we describe both
the cognitive and physical human–robot interaction, summa-
rizing the BNCI algorithms for characterization of voluntary
and tremulous movements, and the control strategies to com-
pensate for the tremor. Next, we provide first experimental
results showing the performance of the different components.
The paper ends with a discussion of the current results
obtained, and an outline of current research.
II. CONCEPT DESIGN FOR THE SOFT WEARABLE ROBOT
The TREMOR neurorobot aims at functionally compensat-
ing for upper limb tremors, hence it incorporates sensors and
actuators to control the next degrees of freedom: 1) elbow
2011 IEEE International Conference on Robotics and Automation
Shanghai International Conference Center
May 9-13, 2011, Shanghai, China
978-1-61284-385-8/11/$26.00 ©2011 IEEE2249

Page 2

Multimodal BCI
EEG
EMG
IMU
Multi-channel
FES array
Inverse dynamics
muscle model
pHRI
cHRI
Fig. 1. Concept design of the TREMOR neurorobot, showing both physical
and cognitive interfaces (pHRI and cHRI respectively).
flexion/extension, 2) forearm pronation/supination, 3) wrist
flexion/extension, and 4) wrist abduction/adduction. Previous
studies, [8], indicate that the former three have the largest
impact on disability, thus they constitute our main focus.
As all wearable robots, the major characteristic of the
TREMOR neurorobot is its strong interaction with the user,
[9]. This interaction is both physical and cognitive, and
happens in a bidirectional manner. In our case, the cognitive
Human–Robot Interface (cHRI) is built upon a BNCI that
assesses the generation, transmission, and execution of both
voluntary and tremorous movements. On the other hand,
the physical Human–Robot Interface (pHRI) comprises a
multichannel FES system that selectively drives the muscles
based on the output of the control algorithm, Fig. 1. The
TREMOR neurorobot takes the shape of an active garment
(a sleeve) that incorporates arrays of sewn electrodes for
both electrical stimulation and recording, [10], satisfying
users aesthetical’ preferences and usability requirements.
Moreover, stimulation through electrode matrices allows us
to implement techniques to minimize fatigue, discomfort and
painful sensations, [11], at the same time that we enhance
the selectivity of the simulation.
The BNCI comprises recording of electroencephalo-
graphic (EEG) and electromyographic (EMG) activity, to-
gether with motion capture with inertial measurement units
(IMUs). Each sensor modality aims at extracting certain in-
formation, following a hierarchical integration scheme. First,
a real–time EEG classifier is in charge of detecting user’s
intention to perform a voluntary movement, triggering the
system. Next, processing of sEMG information yields tremor
onset and an estimation of its frequency and phase. Finally,
IMUs track instantaneous tremor amplitude and frequency
at each joint. The use of multiple sensor modalities also
permits us implementing fusion and redundancy techniques
to enhance the dependability of the system. One example
of redundancy is the use of EMG activity to detect the
occurence of a motor command, compensating for eventual
EEG classification errors. An example of sensor fusion is the
use of machine learning techniques to adjust the parameters
of the EEG clasisifier after the execution of a movement, as
detected with IMUs. Another interesting example of integra-
tion of different modalities is employing sEMG detection of
tremor to feed the controller in real–time with which muscles
are responsible for the tremorous movement, while IMUs
are used to precisely characterize tremor features at a joint
level. Identification of the muscles that are contributing to the
tremulous movement with sEMG is crucial for subsequent
compensation with FES; therefore adequate modulation of
the stimulation cannot be attained with IMUs alone.
The physical interface of the TREMOR neurorobot aims
at taking advantage of biomechanical loading as a selective
means for compensating for tremor. Hence, the system will
filter out the tremor, leaving concomitant voluntary motion
unaltered, based on the application of FES through the
array of sewn electrodes. Command signals to drive the
stimulators are provided by a control strategy that relies on
the information of the BNCI described above, and on an in-
verse dynamic model of the musculoskeletal system. We are
currently evaluating two tremor suppression strategies: one
based on impedance control, the other on learning control
techniques, [5]. Notice that these strategies are implemented
independently in every joint because tremors are known to
be local phenomena, and prone to migrate to proximal joints
when being mechanically atenuated, [12].
III. METHODS
The soft wearable robot described above has been imple-
mented as an experimental platform that comprises both the
BNCI and the multichannel FES system, interfacing with
the user through the active garment. A master computer run-
ning Neutrino Real–Time Operating System (QNX Software
Systems, Ontario, Canada) is in charge of synchronizing
and integrating the information from the different sensor
modalities, and driving the FES units. This platform has been
built and validated with patients in an iterative fashion, and
in its current form comprises the following systems: 1) a
16 channel EEG amplifier (gUSBamp, Guger Technologies
OG, Graz, Austria), 2) a 128 channel surface EMG amplifier
(EMG-USB, OT Bioelettronica, Torino, Italy), 3) a motion
capture system based on IMUs (TechMCS, Technaid SL,
Madrid, Spain), 4) multichannel, individually controllable,
electrical stimulators (Tremuna, UNA Systems, Belgrade,
Serbia), and 5) an e-textile (SMARTEX, Prato, Italy). Fig. 2
shows a control subject wearing the TREMOR neurorobot.
The first results presented in this paper correspond to
different experimental sessions carried out in the Erasmus
Hospital (Brussels, Belgium) and in the General Hospital
of Valencia (Valencia, Spain), and were collected following
three protocols. The first protocol aimed at identifying and
characterizing the different neurophysiological phenomena
associated to movement preparation in the case of tremor
patients. The second protocol was designed to activate the
different types of tremor based on well described clinical
tasks, so that the BNCI–based algorithms parameterize its
features. The last protocol aimed at evaluating tremor sup-
pression strategies based on FES. Between both locations,
17 patients suffering from the most common types of tremor
have participated in this study, after giving written informed
consent. The group of patients comprises the three major
types of pathological tremor, namely Parkinsonian patients,
2250

Page 3

(a) (b)
Fig. 2.
part of the active garment showing the stimulation and recording sewn electrodes.
(a) A control subject wearing the TREMOR neurorobot, including the EEG system and the active garment with IMUs. (b) Detail of the inner
essential tremor patients, and patients suffering from cere-
bellar disorders. Experimental protocols were approved by
the Ethical Commitees of both hospitals.
IV. COGNITIVE HUMAN–ROBOT INTERACTION
This section describes briefly the different algorithms
implemented in the BNCI to characterize both voluntary and
tremorous movements.
A. Detection of Movement Intention
Detection of movement intention is carried out based on a
BCI–switch that anticipates the execution of self initiated vo-
litional movements in the special case of tremor patients. The
proposed algorithm evaluates desynchronization in the alpha
and beta bands at the motor cortex during movement plani-
fication and execution. This neurophysiologic phenomenom
is known as Event Related Desynchronization (ERD), [13],
and is vastly described on the literature, although, to the
authors’ knowledge, has never been systematically analyzed
for tremor patients. Fig 3 shows an example of ERD in an
essential tremor patient. Movement starts at time 0, and ERD
is evident in the 12 and 22 Hz frequency bands already 2 s
before the motion is executed.
The single trial BCI–switch predicts movement intention
by combining two Bayesian Classifiers, based on the next
ERD features: 1) power in some frequency bands decreases
some seconds before the movement is executed, and 2) this
decay tends to be stable during a certain time interval, [13].
Both classifiers take as input the spatially filtered signal -by
means of a Laplacian filter- at the three preferred positions,
C3, Cz and C4.
B. Detection of Tremor Onset
Analysis of sEMG information serves to detect the onset of
tremor, and gives a first estimate of its frequency. Application
of the Iterated Hilbert Transform (IHT), a multicomponent
Fig. 3.
map has been obtained after spatial filtering of the EEG signal around the
contralateral motor cortex. Cold colours correspond to low power.
Desynchronization map for an essential tremor patient. The ERD
AM–FM decomposition method [14] on the raw sEMG sig-
nal serves to extract the tremorours component from the vol-
untary activity component, without any a priori information
about tremor frequency, its spectrum, and its nonstationary
dynamics. Moreover, as the correlation between the tremor
component and the oscillatory movement is significantly
high, this algorithm lets us obtain an estimate of tremor
frequency.
C. Estimation of Tremor Features
The algorithm to estimate instantaneous tremor features
at every joint first obtains joint rotation by differential
measurement with two IMUs placed distally and proximally
with respect to it. Then it estimates, from the raw motion,
the volitional and tremulous components of the rotation.
Voluntary movement and tremor are separated based on the
fact that they occur in different frequency bands, and can be
considered to be additive, [15].
Volitional movement is then modelled as a first order
2251

Page 4

0 2040 6080100
time(s)
120140160 180200
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
arbitrary units (A.U.)


Trigger
Classifier output
Fig. 4.
combined output of both classifiers, red lines to the execution of movements.
Example of detection of movement intention with the proposed algorithm in an essential tremor patient. The blue surface corresponds to the raw
process that is estimated by a Critically Dampened Filter,
a special type of g–h filter. By removing the estimated
voluntary component from the raw motion, we obtain an
estimation of the tremor. This estimation of tremor is then
fed into two adaptive algorithms. The first algorithm is
the Weighted Frequency Fourier Linear Combiner (WFLC),
which is employed to track tremor frequency based on a LMS
approach, [16]. Afterwards, a Kalman Filter that incorporates
a harmonic model of the process, estimates tremor amplitude
taking WFLC frequency estimation into account.
V. PHYSICAL HUMAN–ROBOT INTERACTION
Due to the differences among symptoms in the different
types of tremor, we are evaluating two control strategies for
the selective application of biomechanical loading through
FES. The first of them is based on modification of the
apparent impedance of the limb, whereas the second one
relates to noise canceling techniques. The reason for this
is two fold. On the one hand, a previous study suggests
that the latter strategy is more efficient, although it was
only compared with addition of constant -nonmodulated-
viscosity, [5]. On the other hand, even with modulated
impedance, as Parkinson’s disease carries increased limb
rigidity, we believe that a noise cancelling approach will
still yield more tremor attenuation for these patients. Both
strategies are described next.
The impedance modulation strategy relies on modifying
the apparent joint stiffness and viscosity so that tremor is nat-
urally filtered out, leaving concomitant voluntary movement
unaffected. Considering that a human joint may be simply
modelled as a second order system, [17], increased rigidity or
damping decreases the cut off frequency of the low pass filter
response of the joint, hence allowing for tremor attenuation.
This is similar to the cocontraction strategy employed by
healthy subjects to stabilize the upper limb in a variety of
situations, [18].
The noise canceling like approach generates an oscillation
with the same amplitude but opposite phase to the tremor,
compensating for it. This strategy is implemented as a repet-
itive controller which memory block is adapted following
the instantaneous estimation of tremor frequency. According
to learning control theory, if the prediction of perturbation
frequency is accurate enough, the system will reach to zero
steady state error, [19]. Nevertheless, as tremor is a dynamic
phenomenon, we correct the prediction of tremor frequency
based on the instantaneous amplitude, as it indicates which
muscle to stimulate.
?2
?1
0
1
raw sEMG (a.u.)
?0.2
0
0.2
0.4
0.6
tremor (a.u.)
500100015002000
time (ms)
2500 3000 3500
0.05
0.1
0.15
0.2
0.25
voluntary activity (a.u.)
Fig. 5.
Top panel shows the raw EMG signal. Middle and bottom panels show the
estimation of tremor and voluntary activity obtained with the IHT algorithm.
Decomposition of tremor recorded with surface EMG electrodes.
VI. FIRST RESULTS
This section presents results obtained with the BNCI
algorithms and the tremor suppression strategies presented
above.
A. Detection of Movement Intention
The performance of the EEG classifier is evaluated follow-
ing an event like approach, thus the output of the classifier
2252

Page 5

is compared with a certain threshold to translate the proba-
bilistic output of the Bayessian classifiers to a binary signal.
The figures of merit employed are the precision and recall.
Fig. 4 shows an example of detection of movement
intention in a tremor patient. The figure shows the raw
combined output of both classifiers, and the execution of
movements. Results with 4 tremor patients suffering from
different pathologies yield a precision of 46.5 ± 10.66 %,
and a recall of 69.5 ± 14.76 %. Average anticipation of the
Bayesian classifier is 320±141 ms, proving the ability of the
proposed algorithm to predict the execution of a voluntary
movement, [20].
B. Detection of Tremor Onset
Extraction of tremor information from sEMG has been
validated with a novel sEMG tremor model [21] and with
data from tremor patients. The first analysis serves to infer
a series of relationships that cannot be obtained from real
data, such as the correlation of the tremor estimation with the
central tremor oscillator, and the amount of voluntary muscle
contraction. Performance of the proposed algorithm with
patients data is evaluated by computing the crosscorrelation
with the information of tremor provided by the IMUs, and the
phase advance of the sEMG tremor estimation with respect
to movement.
Model results show significant correlation between the
imposed voluntary activation and that estimated with the IHT,
and also between imposed tremor intensity and its estimation.
Results with signals collected from 4 tremor patients yield
a crosscorrelation of 0.62 ± 0.15, and a phase advance of
13.0 ± 9.9 ms. Fig. 5 shows an example of detection of
tremor from sEMG. The proposed algorithm provides not
only tremor onset, but also an estimation of its shape, and
the voluntary activity exerted by the muscle.
C. Estimation of Tremor Features
Both stages of the algorithm to estimate tremor parameters
have been evaluated stepwise. First, voluntary movement
tracking is assessed using the Kinematic Tracking Error
(KTE), a figure of merit that accounts for the estimation error
and the variance of the estimator, [22]. Next, the performance
of the tremor estimation algorithm is evaluated by comput-
ing the Filtered Mean Square Error with Delay Correction
(FMSEd), a metric specifically proposed for tremor tracking
filters, [23]. Estimation of frequency estimation is validated
by comparing it with the spectrogram.
Analysis of data from five patients suffering from tremor
with different aetiologies, provides with a KTE of 0.223 ±
0.082 rad/s, and a FMSEdof 0.001±0.002 rad/s, [22]. Fig.
6 shows an example of estimation of voluntary movement,
instantaneous tremor amplitude and frequency in a parkin-
sonian patient performing a postural test. Accurate tracking
is obtained.
D. Attenuation of Tremor
Validation of tremor suppression strategies is being carried
out in an iterative fashion. First, we have evaluated tolerance
Fig. 6.
task performed by a parkinsonian patient. Top panel shows total movement
(black) and the estimation of voluntary movement (red). Middle and bottom
panels show estimation of tremor amplitude and frequency (overimposed to
the movement spectrogram) respectively.
Estimation of instantaneous tremor parameters during a postural
of tremor patients to FES in terms of discomfort and pain.
Afterwards, we have studied the role that muscle FES
induced fatigue plays in tremorogenic muscles. Next we
have tested the impedance control strategy in open loop, i.e.
with constant stiffness and damping increase. Current work
focuses on evaluation of both strategies in closed loop.
Preliminary results indicate that, as expected, results vary
considerably among patient groups. First trials suggest that
modification of joint impedance is effective in the suppres-
sion of tremor. Fig. 7 shows attenuation of wrist tremor in
an essential tremor patient while performing a postural task.
In this trial, tremor attenuation reaches approximately 60 %
of the peak to peak value. Experimental results show migra-
tion of tremor to proximal joints when tremor is cancelled
out through FES, supporting the need of independent joint
controllers.
VII. DISCUSSION
Previous sections presented the design, and development
of the TREMOR neurorobot, a novel soft wearable robot
that aims at characterizing and suppressing for upper limb
tremors with different aetiology.
First results demonstrate the ability of the cHRI to predict
user’s intention to perform a volitional movement, to detect
the presence of tremor from sEMG, and to estimate its
instantaneous amplitude and frequency out of kinematic in-
formation. Moreover, we are integrating a number of features
that will serve to enhance the reliability of the neurorobot,
for example: 1) taking advantage of the sEMG algorithm to
estimate voluntary movement activity to compensate for BCI
based classification errors, 2) using the frequency estimation
2253

Page 6

02468 1012 14 161820
?1.5
?1
?0.5
0
0.5
1
time (s)
wrist angular velocity (rad/s)
FES on
Fig. 7.
keeping his arms outstrechted. During FES a considerable attenuation is
observed, although tremor reappears when stimulation ceases.
Compensation of tremor in the case of an essential tremor patient
obtained by the IHT algorithm as an initial guess for the IMU
based algorithm to track tremor features, 3) implementing
machine learning techniques to adjust the parameters of
the Bayesian classifier online, based on the execution of a
voluntary movement.
Regarding FES based tremor suppression, though the
results are yet preliminary, we can draw a number of conclu-
sions from them. First, impedance modulation is mandatory
to attain a significant degree of tremor attenuation. In this
regard, we are working on a model based adaptive controller
to estimate the apparent impedance of the joint. Other
approaches based in impedance adaptation of the muscular
system will also be explored, [24]. Second, determining
which muscles contribute to joint tremor (from sEMG) is
mandatory for efficient cancellation with FES.
VIII. CONCLUSIONS
This paper presents the concept design, development, and
preliminary validation of a soft wearable robot for tremor
assessment and suppression. The TREMOR neurorobot com-
prises a cHRI based on a BNCI that monitors the whole
neuromusculoskeletal system, aiming at characterizing both
voluntary movement and tremor, and a pHRI that relies
on FES to compensate for tremulous movement without
impeding the user’s functional movements. The soft robot
interfaces with the user through sewn electrodes incorporated
in an active garment, which takes the shape of a sleeve, thus
aimed at satisfying users’ aesthetic and cosmetic preferences.
First results with patients suffering from different types of
tremor, show the ability of the TREMOR neurorobot to
model the generation, transmission, and execution of both
volitional and concomitant tremulous movements. Moreover,
these tests demonstrate the ability of the system to attain
considerable attenuation of tremors in some patients.
REFERENCES
[1] G.K. Wenning, S. Kiechl, K. Seppi, J. M¨ uller, J. H¨ ogl, M. Saletu,
G. Rungger, A. Gasperi, J. Willeit, and W. Poewe, “Prevalence of
movement disorders in men and women aged 50-89 years (bruneck
study cohort): A population-based study,” Lancet Neurol, vol. 4,
no. 12, pp. 815–820, 2005.
[2] E. Rocon, J. M. Belda-Lois, J. M. Sanchez-Lacuesta, and J. L.
Pons, “Pathological tremor management: Modelling, compensatory
technology and evaluation,” Tech Disab, vol. 16, pp. 3–18, 2004.
[3] E. D. Louis, “Essential tremors: a family of neurodegenerative
disorders?” Arch Neurol, vol. 66, no. 10, pp. 1202–1208, 2009.
[4] D. K. Binder, G. Rau, and P. A. Starr, “Hemorrhagic complications
of microelectrode–guided deep brain stimulation,” Stereotact Funct
Neurosurg, vol. 30, pp. 28–31, 2003.
[5] E. Rocon, J. M. Belda-Lois, A. F. Ruiz, M. Manto, J. C. Moreno, and
J. L. Pons, “Design and validation of a rehabilitation robotic exoskele-
ton for tremor assessment and suppression,” IEEE Tans Neural Syst
Rehab Eng, vol. 15, no. 3, pp. 367–378, 2007.
[6] G. C. Joyce and P. M. H. Rack, “The effects of load and force on
tremor at the normal human elbow joint,” J Physiol, vol. 240, pp.
375–396, 1974.
[7] A. Prochazka, J. Elek, and M. Javidan, “Attenuation of pathological
tremors by functional electrical stimulation. i: Method.” Ann Biomed
Eng, vol. 20, no. 2, pp. 205–224, 1992.
[8] J. M. Belda-Lois, E. Rocon, J. J. S´ anchez-Lacuesta, A. F. Ruiz, and
J. L. Pons, “Functional assessment of tremor in the upper limb,” in
Euro Conf Advancement Assistive Techn in Europe, 2005.
[9] J. L. Pons, Ed., Wearable Robots: Biomechatronic Exoskeletons. John
Wiley & Sons, Ltd, 2008.
[10] R. Paradiso, L. Caldani, M. Pacelli, F. Negro, and D. Farina, “E-textile
platforms for rehabilitation,” in Conf Proc IEEE Eng Med Biol Soc,
2010.
[11] A. Popovic-Bijelic, G. Bijelic, N. Jorgovanovic, D. Bojanic, M. B.
Popovic, and D. B. Popovic, “Multi-field surface electrode for
selective electrical stimulation.” Artif Organs, vol. 29, no. 6, pp.
448–452, 2005. [Online]. Available: http://dx.doi.org/10.1111/j.1525-
1594.2005.29075.x
[12] M. Manto, E. Rocon, J. L. Pons, J. M. Belda-Lois, and S. Camut,
“Evaluation of a wearable orthosis and an associated algorithm for
tremor suppression.” Physiol Meas, vol. 28, no. 4, pp. 415–425, 2007.
[Online]. Available: http://dx.doi.org/10.1088/0967-3334/28/4/007
[13] G. Pfurtscheller and F. H. Lopes da Silva, “Event-related eeg/meg
synchronizationanddesynchronization:
Neurophysiol, vol. 110, no. 11, pp. 1842 – 1857, 1999. [On-
line]. Available: http://www.sciencedirect.com/science/article/B6VNP-
3XJSSG7-1/2/919fb705601eee998c08fc2160dc2f99
[14] F. Gianfelici, G. Biagetti, P. Crippa, and C. Turchetti, “Multicompo-
nent am–fm representations: An asymptotically exact approach,” IEEE
Trans Audio Speech Lang Process, vol. 15, no. 3, pp. 823–837, 2007.
[15] P. O. Riley and M. J. Rosen, “Evaluating manual control devices for
those with tremor disability,” J Rehabil Res Dev, vol. 24, no. 2, pp.
99–110, 1987.
[16] C. N. Riviere, R. S. Rader, and N. V. Thakor, “Adaptive canceling of
physiological tremor for improved precision in microsurgery,” IEEE
Trans Biomed Eng, vol. 45, no. 7, pp. 839–846, 1998.
[17] E. Rocon, R. A. F., and P. J. L., “Biomechanical modelling of the upper
limb for robotics-based orthotic tremor suppression,” Appl Bionics
Biomech, vol. 2, no. 2, pp. 81–85, 2005.
[18] N. Hogan, “Adaptive control of mechanical impedance by coactivation
of antagonist muscles,” IEEE Trans Autom Control, vol. 29, no. 8, pp.
681 – 690, aug 1984.
[19] S. Hara, Y. Yamamoto, T. Omata, and M. Nakano, “Repetitive control
system: a new type servo system for periodic exogenous signals,” IEEE
Trans Autom Control, vol. 33, no. 7, pp. 659 –668, jul 1988.
[20] J. Ib´ aez, J. I. Serrano, M. del Castillo, and L. Barrios, “An asyn-
chronous bmi system for online single-trial detection of movement
intention,” in Conf Proc IEEE Eng Med Biol Soc, 2010.
[21] J. L. Dideriksen and D. Farina, “An integrative model of the surface
emg in pathological tremor,” in Conf Proc IEEE Eng Med Biol Soc,
2010.
[22] J. A. Gallego, E. Rocon, J. O. Roa, J. C. Moreno, and J. L. Pons, “Real-
time estimation of pathological tremor parameters from gyroscope
data,” Sensors, vol. 10, no. 3, pp. 2129–2149, 2010.
[23] J. G. Gonzalez, E. A. Heredia, T. Rahman, K. E. Barner, and G. R.
Arce, “Optimal digital filtering for tremor suppression,” IEEE Trans
Biomed Eng, vol. 47, no. 5, pp. 664–673, 2000.
[24] K. P. Tee, D. W. Franklin, T. Milner, M. Kawato, and E. Burdet,
“Concurrent adaptation of force and impedance in the redundant
muscle system,” Biol Cybern, vol. 102, pp. 31–44, 2010.
basicprinciples,”Clin
2254