An EMG-based Assistive Orthosis for Upper Limb Rehabilitation

Luis M. Vaca Benitez

, Niels Will

, Marc Tabie

, Steffen Schmidt

, Elsa Kirchner

1,2

and Jan Albiez

Robotics Innovation Center, German Research Center for Artiﬁcial Intelligence (DFKI),

Robert-Hooke-Straße 5, Bremen, Germany

Robotics Lab, University of Bremen, Robert-Hooke-Straße 5, Bremen, Germany

Keywords:

Active Orthosis, Robotic Assistive Rehabilitation, Upper Limb Exoskeleton, EMG Detection, Torque Control.

Abstract:

In this work an upper limb active orthosis for assistive rehabilitation is presented. The design and torque

control scheme of the orthosis that take into account important aspects of human rehabilitation, are described.

Furthermore, ﬁrst results of successful muscle activity detection and processing for the operation of the or-

thosis in two movement directions are presented. The proposed system is the ﬁrst step towards an adaptive

support of patients with respect to the strength of their muscle activity. To allow an adaptive support, differ-

ent methods for EMG analysis have to be applied which allow to correlate muscle activity strength with the

recorded signal and thus enable to adapt the support of the orthosis to the needs of the patient and state of

therapy.

1 INTRODUCTION

The ability to move is one of the most important char-

acteristic of life and is determined by the functions of

our muscles, which are required in tasks like breath-

ing, blood circulation, eating and locomotion. In

other words, in any kind of activity. For human loco-

motion and activities a complex and ﬁne-tuned mus-

culoskeletal system has evolved. However, if func-

tions of the muscular system are impaired or disabled

it can not only have far-reaching consequences on the

personal professional and social life, but also on psy-

che of the affected person and their families.

In Henze (Henze, 2007) it is stated that loosing

the ability to move (even of one single extremity) is

in many cases associated with a drop in independence

and therefore reduces the quality of life of the person.

Generally, motor restrictions are often the result of

neurological disorders.

These can be caused by illness, accidents, or birth

defects. In this context stroke plays a major role, since

it is one of the most common causes of neuromotor

disorders and permanent disabilities in western civi-

lization (Deaton et al., 2011).

The inability to move the affected arm or even to

use it in a coordinated way is a very common and se-

rious consequence of a cerebral stroke. About 40 %

of the affected people suffer from a non-functioning

upper extremity. Therefore, it is not surprising that in

the recent years, several studies and ﬁndings on stroke

rehabilitation were published (Albert and Kesselring,

2012; Platz and Roschka, 2009).

In this ﬁeld, rehabilitation robotics has also made

great progress and is currently subject of many re-

search projects (Loureiro et al., 2011). The aim of

robotic systems in the context of neuromotor rehabil-

itation is to optimize the rehabilitation process and to

support the therapist in labor-intensive therapies.

Further, there is the ambition to re-enable pa-

tients to execute self paced movements of the paretic

extremity using their movement intention. This is

thought to increase the patients motivation and to sup-

port processes that are important for neuronal plastic-

ity

(Brewer et al., 2007).

A central, but time consuming part of stroke reha-

bilitation is the process of re-learning directed hand

and arm movements according to the patients needs.

Hence, we want to introduce a rehabilitation device

for the upper extremity, which supports stroke pa-

tients and therapists in their daily rehabilitation rou-

tine. This device is an active elbow orthosis (see Fig-

ure 1) with one degree of freedom (see Section 3). A

patient-speciﬁc control can be realized by processing

Neuronal plasticity is the ability of brain to reorganize

itself by forming new neural connections. This form of ad-

justment allows the brain to compensate injury and disease

and to adjust activities in response to new situations or to

changes in the environment (Johnston, 2009).

323

M. Vaca Benitez L., Will N., Tabie M., Schmidt S., Kirchner E. and Albiez J..

An EMG-based Assistive Orthosis for Upper Limb Rehabilitation.

DOI: 10.5220/0004371803230328

In Proceedings of the International Conference on Biomedical Electronics and Devices (MHGInterf-2013), pages 323-328

ISBN: 978-989-8565-34-1

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

the patients muscle activity measured with the elec-

tromyogram (EMG) (see Section 5). In the long term,

this device could be used for the entire rehabilitation

process. For this, three different functionalities can

be provided by the active orthosis, namely passive,

active-assisted, and active-resisted modes of opera-

tion (Gomez-Rodriguez et al., 2011). In addition, the

progress of therapy can be evaluated by monitoring

and analyzing the muscle activity via EMG.

Figure 1: The active orthosis is designed to be easily worn

by the user. The carrying system distributes the weight, in-

creasing the comfort.

2 APPLICATION ENVIRONMENT

There are several approaches for EMG-controlled ac-

tive elbow orthotics. For example the mPower 1000

system (Myomo



Inc, Cambridge, USA) is a com-

mercially available EMG-controlled orthosis for the

upper limb. A german research group is develop-

ing an active orthosis for paraplegic persons (Schmitz

et al., 2011). In contrast to these approaches, the pro-

posed device offers a unique combination of sensors

and planned functions.

The therapeutic (long-term) goal of the orthosis is

recovery of lost motor functions of the upper extrem-

ity after neurological diseases. As mentioned before,

the human brain is able to compensate functional im-

pairment. This requires intensive and early training

after, e.g., stroke. Therefore, it is important to design

a system, which motivates the patient for a constant

training.

The device can enable patients to perform the fol-

lowing exercise modalities:

• Early and Intensive Practice. Start of the arm

rehabilitation, e.g, few days after acute stroke

with a high intensity, when indicated.

• Repetitive Practice. Repetitive target movements

across various sequences.

• Task-oriented Training. Exercise oriented on

everyday life situations, e.g., in an exercise-

kitchen.

• Independent Training. Therapeutic treatments

with intermittent supervision by the therapist.

These therapy modalities are based on established

and evidence-based rehabilitation methods (Platz and

Roschka, 2009).

The goal is to achieve a therapy session compara-

ble to a guided session by a therapist, without having

him at site.

In the early stage of treatment the device can be

used to passively move the patients arm. With ther-

apy in advanced stages the residual muscle activity

will be measurable again. This low residual activity

may not be sufﬁcient for moving the arm, but result in

myoelectric signals. By measuring these signals with

EMG, they can be used to detect the patients move-

ment intent.

Further these signals can be used to move the pa-

tients arm in a self motivated way. This kind of

treatment can support processes that underlay neu-

ronal plasticity. In later stages of treatment the patient

should regain more and more muscle strength. There-

fore, it is planned to adjust the assistance level of the

device via the measured muscle activity, in a way that

higher muscle activity leads to a lower level of assis-

tance.

3 DESIGN AND MECHANICAL

STRUCTURE OF ORTHOSIS

In this section the mechanical design of the active or-

thosis is presented (see Figure 2).

The orthosis is designed with ﬁve degrees of free-

dom, four passive joints are required to compensate

misalignments and one actuated joint to support the

ﬂexion/extension movement of the elbow joint. The

active joint is driven by a 24 V Maxon A-max 22

DC-Motor with a 333:1 Maxon planetary gear and

a 4:1 worm wheel gear. For a natural force interac-

tion, safety reasons and to measure the applied force

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

324

Figure 2: Mechanical design of the active Orthosis. The

red dots represent the positions of the passive degrees of

freedom.

interaction, the actuated joint is compliant. This com-

pliance is generated via serial elasticity in the worm

wheel gear set-up. The worm is axial moveable and

centred in the gear via disc springs. In case a load

is applied, the worm is pushed to one side and thus,

the spring is compressed on this side. The position of

the worm wheel is measured with a Bahluff inductive

sensor. In this way the applied load can be calculated

(see Section 4). The position of the joint is measured

with an IC-Haus-MH position encoder.

Furthermore, the used electronics consist of a

STM32F103VE microcontroller, offering several data

acquisition (GPIO) and communication (USART,

CAN-bus) ports, and a BD6232 PWM H-Bridge

driver. The current used DC-Drive can generate a

torque of about 16 Nm.

To avoid any danger for the user, various safety as-

pects are considered. Therefore, the natural working

range of the human elbow is limited by mechanical

stops. Furthermore, at too high forces the forearm in-

terface will release from the orthosis (similar to the

principle of ski binding).

The active range of motion of the elbow orthosis

corresponds to the anatomic workspace of the human

joint and is individually adjustable to each subject.

Since an additional and unilateral load can repre-

sent a major inﬂuence on, e.g., neurological patients,

the orthosis’ weight with respect to the user must be

kept as low as possible. Therefore, the orthosis’ ma-

terials is a combination of carbon reinforced plastics

and polyamid PA6, for a lightweight, robust and stiff

design. Additionally, a carrying system was devel-

oped, which distributes the weight of the device on

both shoulders.

4 CONTROL SCHEME OF THE

ORTHOSIS

Several research groups have described robotic de-

vices for upper limb rehabilitation and their strategies

to control them in a user-oriented way. In (Rosen

et al., 2001) the torque applied to the elbow joint

of an upper extremity exoskeleton is measured via a

load cell, while the set torque is calculated via mus-

cle models. In a second step, the authors derive four

performance indices, in order to calculate the magni-

tude of support by the exoskeleton from EMG data.

In (Andreasen et al., 2005) an impedance control

scheme is implemented. Two load cells in series es-

timate the joint torque which is fed into a dynamic

impedance function.

In the following the torque control system of the

proposed active orthosis will be presented. This can

be visualized in the simpliﬁed block diagram in Fig-

ure 3.

Figure 3: Block diagram of the torque control loop.

The general control structure is designed to be cas-

caded, while the main- and inner loop of the control

architecture is a torque control loop. As shown in Sec-

tion 3, the DC-drive of the device is provided with

two disc springs performing the serial elasticity of the

drive. These springs deﬂect when load is applied to

the joint. One is used for movements that are upwards

directed and one for movements that are downwards

directed. The inductive sensor detects this deﬂection

d. With these measurements it is possible to obtain a

nearly linear function between spring deﬂection and

the actual torque applied to the joint, τ

. The curve,

shown in Figure 4, was empirically determined ap-

plying several load (torque) values to the joint, and

matching this load values with the resulting spring de-

ﬂection.

The set (desired) joint torque τ

at this point is fed

externally, from a computer via USART port. The dif-

ference between these two torques is the control error

e, which is propagated into an anti-windup PID con-

troller. For more information about anti-windup con-

trollers and methods, please refer to, e.g., (Bohn and

Atherton, 1995). The controller computes the voltage

for the motor needed to reach the desired torque. The

performance of the control system was veriﬁed with

weight discs in order to simulate values for τ

, and

giving the corresponding τ

to the system, resulting

in an acurate balancing of the weights. Furthermore,

the resulting measured torque was compared with the

deﬂection-torque curve depicted above. Figure 5 il-

AnEMG-basedAssistiveOrthosisforUpperLimbRehabilitation

325

Figure 4: The calculated joint torque - deﬂection curve.

Figure 5: Setup for control experiments.

lustrates the experimental setup.

The torque control loop already allows to use the

upper limb orthosis in a free-running mode (with

= 0 Nm), or in other words, to use the device as

a completely passive one.

Alternatively, the orthosis can be manually oper-

ated via two buttons at any time, supplying a con-

stant voltage of ±15 V. This allows corrections and

re-positioning of the joint if needed.

5 DETECTION OF MOVEMENT

INTENTION BY EMG

ANALYSIS

This section describes how muscle activity measured

by the EMG is used as a control signal for the active

orthosis. EMG signals are measured at two muscles

on the upper right limb (same arm at which the or-

thosis is used) named M. biceps brachii and M. tri-

ceps brachii. Ag/AgCl electrodes are used in a bipolar

Figure 6: Resulting signals after preprocessing and thresh-

olding: (top) Raw EMG signals from biceps (blue) and tri-

ceps (green), (middle) ﬁltered and normalized signals, and

(bottom) results of the thresholding algorithm.

setup. The signals are ampliﬁed and digitalized by

a BrainExG MR (Brain Products GmbH, Germany)

ampliﬁer and transfered to a computer with a sam-

pling frequency of 5000 Hz where they are saved or

directly processed.

The acquired signals are ﬁltered. For this, a ﬁlter

that is based on the standard deviation (blind review

Ownpaper, 2013) is used. For this ﬁlter a sliding win-

dow of 100 ms is passed along the signal. The stan-

dard deviation of this window is assigned to the last

sample. This is done consecutively for all samples of

the streamed EMG-signals.

Afterwards the signals are normalized in a range

from 0 to 1. Often the normalization is done using the

isometric or isokinetic maximum voluntary contrac-

tions (Burden and Bartlett, 1999). Here a different

approach is used. Basically the signal is normalized

with the maximum value obtained after ﬁltering so

far. The maximum is initialized with 1.0. If a ﬁltered

sample has a higher value than the maximum, this

value is assigned as the new maximum. The drawback

of this method is, that big artifacts, e.g., produced

due to resistance changes at the electrode site, artiﬁ-

cially reduce the amplitude of the normalized signals.

Therefore, a forgetting factor ε

with ε

= 0.9999 is

used to reduce the maximal value in each time step.

Further, a minimum is deﬁned as one fourth of the

maximal value. This minimum is again degraded with

a forgetting factor ε

with ε

= 0.99999. The mini-

mum is needed to keep the normalization factor in a

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

326

value range, where a clear distinction between base-

line noise and signal during muscle contraction can

be achieved. Without the minimum the normalization

factor could degrade to values lower than the baseline

noise values. This would lead to a value of close to 1.0

after the normalization in ranges where no movement

is performed.

The normalization needs to be calibrated, i.e.,

there has to be a contraction of each muscle to ob-

tain meaningful values from the function. The control

signals for the orthosis are obtained using a threshold

algorithm similar to the on used in (DiCicco et al.,

2004). Two thresholds are used for determining the on

and off phases of each muscle. The on threshold was

set to 0.4 and the off threshold to 0.3. This hystere-

sis is used to prevent continuous on and off switch-

ing if the normalized EMG signal has a value close

to the threshold. For the off threshold an additional

time constrain is used. The mode is only switch from

on to off, if the signal is continuously below the off

threshold for 20 ms. Further a resting time of 500 ms

in between a direction switch has to be maintained.

As a last step we had to decide which of the two mea-

sured muscles should be preferred, in case that both

were active, we decide to set both signals to 0 and

therefore to decide for none of the muscle or move-

ment directions.

With this preprocessing in combination with the

threshold algorithm, we are able to create three dif-

ferent control commands for the orthosis: (1) to ﬂex

the arm (M. biceps brachii is active), (2) to stretch

the arm (M triceps brachii is active), and (3) to relax

(none of both muscles is active). The control signals

are derived on the computer acquiring the EMG data

and send to the orthosis via USART. All described pa-

rameters were chosen by empirical testing on EMG

data recorded with the orthosis attached to the sub-

jects arm running in free-runing-mode see Section 4.

In Figure 6 the processing of EMG data, and the

thresholding result for an arm ﬂexion and extension

are shown. In the top, the raw EMG signals from the

biceps (blue) and triceps (green) are shown, the mid-

dle illustrates the ﬁltered and normalized EMG sig-

nals, and ﬁnally in the bottom, the outcome from the

thresholding is shown. The three vertical lines, dot-

ted, dashed, and solid denote time points, where the

subject was asked to stretch, ﬂex, and relax the arm re-

spectively. The stimuli for those three actions where

presented on a monitor and marked in the EMG data.

The obtained control signals can be used directly to

operate the orthosis, e.g., support the users arm move-

ment.

6 CONCLUSIONS AND

OUTLOOK

In this work we presented an active orthosis, its pos-

sible application, design and mechanics, and control.

To summarize, the system allows to support self ini-

tiated movement that are normally executed by both

upper arm muscles M. biceps brachii or M. triceps

brachii. Patients that are not able to effectively control

both muscles can be supported for individual move-

ments. This can be achieved by detecting EMG onset

activity. If EMG activity onset is detected, the or-

thosis actively executes the directed movement corre-

sponding to the active muscle.

The next developmental step is to adapt the

strength of support with respect to the strength of

muscle activity. To allow this, different methods

for EMG analysis have to be applied. These meth-

ods (Ajiboye and Weir, 2005) allow to correlate mus-

cle activity strength with the recorded signal and thus,

enable to adapt the support of the orthosis to the needs

of the patient and state of therapy.

Further, we established a collaboration with a clin-

ical partner. Supported by the medical specialists, the

design of the orthosis will be improved in order to

follow further sanitary and medical guidelines and to

deﬁne proper application scenarios. From the con-

trol side, it is important to ﬁnd a way to measure the

resistance of the human-robot interaction in order to

recognize unvoluntary muscle activity due to spastic-

ity. Further, the possibility of deﬁning torque curves

will be analysed. In speciﬁc, the development of the

applied torque by the user depending on the angular

position of the joint can be observed and deﬁned as a

mathematical function. Using the inverse of this curve

may lead to a more natural movement.

Finally, user-friendliness elements have to be con-

sidered in the future. The possibility for therapist (and

patients) to adapt important parameters online in an

easy way, by means of some sort of intuitive user in-

terface is the goal.

ACKNOWLEDGEMENTS

This work was supported by the German Bundesmin-

isterium fur Bildung und Forschung (BMBF,grant

FKZ 01IW10001).

REFERENCES

Ajiboye, A. and Weir, R. (2005). A heuristic fuzzy logic ap-

proach to emg pattern recognition for multifunctional

AnEMG-basedAssistiveOrthosisforUpperLimbRehabilitation

327

prosthesis control. IEEE Transactions on Neural Sys-

tems and Rehabilitation Engineering, 13(3):280 –291.

Albert, S. and Kesselring, J. (2012). Neurorehabilitation of

stroke. Journal of Neurology, 259:817 – 832.

Andreasen, D. S., Allen, S., and Backus, D. (2005). Ex-

oskeleton with EMG Based Active Assistance for Re-

habilitation. In IEEE 9th International Conference on

Rehabilitation Robotics, pages 333–336.

blind review Ownpaper (2013). b.

Bohn, C. and Atherton, D. (1995). An Analysis Package

Comparing PID Anti-Windup Strategies. IEEE Con-

trol Systems, 15:34–40.

Brewer, B. R., McDowell, S. K., and Worthen-Chaudhari,

L. C. (2007). Poststroke Upper Extremity Rehabil-

itation: A Review of Robotic Systems and Clinical

Results. Topics in Stroke Rehabilitation, 14:22 – 44.

Burden, A. and Bartlett, R. (1999). Normalisation of EMG

amplitude: an evaluation and comparison of old and

new methods. Medical Engineering and Physics,

21(4):247 – 257.

Deaton, C., Froelicher, E. S., Wu, L. H., Ho, C., Shishani,

K., and Jaarsma, T. (2011). The Global Burden of

Cardiovascular Disease. Journal of Cardiovascular

Nursing, 26:5 – 14.

DiCicco, M., Lucas, L., and Matsuoka, Y. (2004). Com-

parison of Control Strategies for an EMG Controlled

Orthotic Exoskeleton for the Hand. In Proceedings of

the 2004 IEEE International Conference on Robotics

and Automation (ICRA ’04).

Gomez-Rodriguez, M., Grosse-Wentrup, M., Hill, J.,

Sch

olkopf, B., Gharabaghi, A., and Peters, J. (2011).

Towards Brain-Robot Interfaces for Stroke Rehabilita-

tion. In Proceedings of the International Conference

on Rehabilitation Robotics (ICORR). Intelligent Au-

tonomous Systems.

Henze, T. (2007). Moderne Rehabilitation nach Schlagan-

fall. NeuroTransmitter, pages 59–67.

Johnston, M. (2009). Plasticity in the Developing Brain:

Implications For Rehabilitation. Developmental Dis-

abilities Research Reviews, 15:94 – 101.

Loureiro, R. C. V., Harwin, W. S., Nagai, K., and Johnson,

M. (2011). Advances in upper limb stroke rehabilita-

tion: a technology push. Medical & Biological Engi-

neering & Computing, 49:1103 – 1118.

Platz, T. and Roschka, S. (2009). Rehabilitative Therapie

bei Armparese nach Schlaganfall. Neurol Rehabil,

pages 81–106.

Rosen, J., Brand, M., Fuchs, M., and Arcan, M. (2001).

A Myosignal-Based Powered Exoskeleton System. In

IEEE Transactions on Sytems, Man, and Cybernetics -

Part A: Systems and Humans, volume 31, pages 210–

222.

Schmitz, B., Wiegand, R., Pylatiuk, C., Rupp, R., and

Schulz, S. (2011). Erste Erfahrungen mit dem Ortho-

Jacket. Orthop

adie-Technik, 4:256 – 261.

BIODEVICES2013-InternationalConferenceonBiomedicalElectronicsandDevices

328