6th International Special Topic Conference on ITAB, 2007, Tokyo

A Study on EMG-Based Control of Exoskeleton Robots for Human Lower-limb

Motion Assist
H.Hel, K.Kiguchil,
'Department of Advanced Systems Control Engineering, Saga University, Japan
Abstract An exoskeleton robot is an external structural adaptation ability for the user's physical and physiological
mechanism with joints and links corresponding to those of the condition and the user's precise motion intension were not
human body. When it is worn, it transmits torques from taken account. In the proposed method, the EMG signal is
actuators through rigid exoskeletal links to the human joints. used as the input information for the controller, in which
We have been developing exoskeleton robots for assisting the the user's motion intention is directly reflected. In order to
motions of physically weak persons such as elderly or disabled
in daily life. In this paper, we propose an electromyogram make the controller adapt to the EMG signals according to
(EMG) based control (i.e., control based on the skin surface the user's physical and physiological conditions, fuzzy-
EMG signals of the user) for the exoskeleton robot to assist neuro control method has been applied. The effectiveness
physically weak person's lower-limb motions. The skin surface of the newly designed control system for the power-assist
EMG signals are mainly used as the input information for the has been evaluated by experiment.
controller. In order to generate flexible and smooth motions
and take into account the changing EMG signal levels II. THE EXOSKELETON ROBOT
according to the physical and psychological conditions of the
user, fuzzy-neuro control method has been applied for the The exoskeleton robot should be adaptable to the
controller. The experimental results show the effectiveness of human lower-limb in terms of segmental lengths and
the designed EMG-based controller for the power-assist. location of center of rotation. Since the exoskeleton robot is
Keywords- Exoskeleton Robot, Lower-Limb Motion,
supposed to be used for daily activities of the user, it should
Power-Assist, Electromyogram Signals be generated flexible and smooth motions.
The architecture of the exoskeleton robot is shown in
I. INTRODUCTION Fig. 1. The exoskeleton robot consists of a waist holder, a
thigh holder, a lower leg holder, two DC motors (Maxon
In the recent society in which the birthrate is DC Motors), two links, a footrest and two force sensors
decreasing and the aging are progressing, it is important (hip and knee force sensor) in one leg. Locations of force
that physically weak persons are able to take care of sensors are shown in Fig. 2. The exoskeleton robot worn by
themselves. We have been developing exoskeleton robots a user is supposed to help the user's lower-limb daily
[1]-[3] for motion assist of physically weak persons such as motions (sitting down, standing up, squatting, ascending
elderly, injured and disabled. The physically weak persons stairs, descending stairs and walking). There is a passive
are able to take care of themselves with the help of the DOF (dorsiflexion/ planterflexion) in ankle joint and two
exoskeleton robot in daily life. The lower-limb motions (hip, active DOF (flexion/extension) in hip and knee joints. Each
knee and ankle motion) are especially important for people DC motor generates the assist torque at each joint.
to perform daily activities. The exoskeletons for ankle
motion assist [4], knee motion assist [5], hip motion, knee
motion and ankle motion assist [6]-[12] have been proposed
for daily use or rehabilitation up to the present. Since
around the 1950s, several exoskeleton leg systems have
been studied and developed [10], and could mainly be used
for two conceptually different applications [8]. 1) Walking
aid for gait disorder persons or aged persons; 2) walking
power augmentation to travel long distances on foot with
heavy loads. Some important daily activities which involve
hip and knee motions are sitting down, standing up,
squatting, walking, ascending stairs and descending stairs.
The electromyogram (EMG) signals are important Front view Side view
information for robotic systems [1]-[3] to understand how Fig. 1 Architecture of exoskeleton robot
the user intends to move. Therefore, the user's EMG signals
are used as main input information for the controller in this Usually, the limitation of human knee movable range
study. The hip and knee forces (i.e., the generated force is 150 degrees in flexion and 0 degree in extension, and the
between the robot and the thigh of the robot user and the limitation of human hip movable range is 110 degrees in
generated force between the robot and the lower leg of the flexion and 30 degrees in extension. Considering the
robot user) are also used as subordinate input information practical application to everyday life and safety, the hip
for the controller. motion limitation of the exoskeleton robot is decided to be
In this paper, we design an EMG-based controller of 110 degrees in flexion and 20 degrees in extension, and the
the exoskeleton robot system for the purpose of human hip knee motion limitation of the proposed robot is 150 degrees
and knee motion support. Although many researches have in flexion and 0 degree in extension. The stoppers are
been using EMG signals to control robots [6] [7] [12], the attached for each motion to prevent the exceeding of the

978-1-4244-1868-8/08/$25.00 (C2008 IEEE. 292

 6th International Special Topic Conference on ITAB, 2007, Tokyo

movable range for safety. In addition, maximum torque of CH1:Tensor fasiae Itae Ch5:Gracilis
Ch2 Rectus fernoris Ch6Vastus medialis
the robot is limited by the hardware and the software for Ch3 Vastus lateralls Ch7 Biceps femoris
safety. Ch4:Adductor longus CHRSotnitpnendinosus

Fig. 2 Location of force sensor Front view Back view

Fig. 3 Location of electrodes
III. HUMAN LOWER-LIMB MOTION
Agonist-antagonist muscles exist in many human joint IV. EMG-BAsED CONTROLLER
such as elbow, wrist, hip, knee, ankle, etc. Such human
joint is usually activated by several muscles. Some of The exoskeleton robot is controlled based on the skin
muscles are bi-articular muscles and the others are uni- surface EMG signals and the force sensor signals with the
articular muscles. Human hip joint generates flexion/ fuzzy-neuro control system [1]-[3]. In this section, a
extension motions which are mainly actuated by the controller for the sitting down, squatting, and standing up
muscles of rectus femoris, tensor fasciae latae, biceps motions of one leg is explained as an example. The same
femoris, and semitendinossus [13]. Many of these muscles controller is used to control the other leg.
are bi-articular muscles that work on both of hip joint and
knee joint. Human knee joint generates flexion/extension PS PB
motions which are mainly actuated by the muscles of biceps
femoris, semitendinossus, gastrocnemius, rectus femoris,
vastus lateralis and vastus medialis. Most of these muscles T)
e
0kA La 3b OA 2.7 5
are also bi-articular muscles. Human ankle joint generates RMS Rms
dorsiflexion/plantarflexion motions which are mainly Tensor fasciae latae Rectus femoris
actuated by the muscles of gastrocnemius, soleus and
tibialis anterior. Here, only gastrocnemius is the bi-articular zo PS. Pf.!
muscle. The muscles activity level can be described by the
EMG signal. In order to design the control system of the &S. 3A PE
;lit
exoskeleton robot, the skin surface EMG signals have been
analyzed in the preliminary experiments [14]. In this study, RM S RM¶S
root mean square (RMS) has been applied as the feature Vastus lateralis Vastus medialis
extraction method of the EMG levels for the fuzzy-neuro
controller. The equation of RMS is written as: Zr n PB

RAIf _[
N ,=0
V2 (1) ,it ,s

urMS
where vi is the voltage value at ith sampling and N is the Adductor longus fi, PS P.X,
.,
Gracilis

number of the samples in a segment. The number of the

samples is set to be 100 and the sampling time is set to be
0.5ms in this study.
Eight kinds of EMG signals (ch. 1: tensor fasciae latae,
ch.2: rectus femoris, ch.3: vastus lateralis, ch.4: adductor
longus, ch.5: gracilis, ch.6: vastus medialis, ch.7: biceps RMS RM$

Biceps femoris Semitendinosus

femoris - short head and ch.8: semitendinosus) are
measured to control the sitting down, squatting, and
standing up motions. The locations of electrodes on lower- NS NB 1IZO PS PB
limb muscles are shown in Fig. 3. Although bi-articular
muscles such as rectus femoris and semitendinosus are
activated for standing up, sitting down, squatting, ascending
stairs and descending stairs motions, the motion can be Force sensor
classified by monitoring the activity level of the uni-
articular muscles. Details can be referred in [ 14]. Fig. 4 Membership function for RMS of each EMG and force sensor

978-1-4244-1868-8/08/$25.00 2008 IEEE. 293

 6th International Special Topic Conference on ITAB, 2007, Tokyo

The initial fuzzy IF-THEN control rules of the fuzzy-

neuro controller are designed based on the analyzed user's
lower-limb motions in the preliminary experiment [14]. In
the controller, the skin surface EMG signals and the force
sensor signals are used, since the generated hip force and
knee force are more reliable when the user activates the
muscles little (when the EMG levels of the user are low),
and the EMG signals are more reliable when the user /
activates the muscles actively (when the EMG levels of the
user are not low). In other words, the exoskeleton robot is
controlled based on the generated hip force and knee force
when the EMG levels of the subject are low, and the
exoskeleton robot is controlled based on the EMG signals
when the EMG levels of the user are not low. Thus, the Fig. 5 Architecture of the fuzzy-neuron controller
exoskeleton robot can be controlled in accordance with the
Table 1 IF-THEN control rules
user's motion intention [1]-[3]. Three kinds of fuzzy IF THEN
linguistic variables (ZO: zero, PS: positive small, and PB: Hip joint:
positive big) are prepared for RMS of the each EMG and Rule 1: ch. 1 is PB and ch.2 is PB and ch.5 is PB
five kinds of fuzzy linguistic variables (NS: negative small, and hipforce sensor is PB rl is 2.5
NB: negative big, ZO, PS, PB) are prepared for force Rule 2: ch.l is PB and ch.2 is PB and ch.5 is PS rl is 2.0
Rule 3: ch.l is PB and ch.2 is PS and ch.5 is PS rl is 2.0
sensors as shown in Fig. 4. The output of the fuzzy-neuro Rule 4: ch. is PS and ch. 2 is PS and ch. 5 is PS rl is 2. 0
control system is the torque command for each motor. Rule 5: ch.l is PS and ch.2 is PS and ch.5 is PB al is 2.0
The architecture of the fuzzy-neuro controller of hip Rule 6: ch. is PS and ch. 2 is PB and ch. 5 is PB
and knee joint motions of one leg is depicted in Fig. 5. It and hipforce sensor is PS rl is 2.0
consists of five layers (input layer, fuzzifier layer, rule layer, Rule 7: ch. 3 is PB and ch. l0 is PB and ch. 13 is PB rl is -2.5
defuzzifier layer, and output layer). In order to realize the Rule 8: ch. 3 is PB and ch. l0 is PB and ch. 13 is PS
and hipforce is NS rl is -2. 0
sitting down, squatting, and standing up motions, 20 fuzzy Rule 9: ch. 3 is PB and ch. l0 is PS and ch. 13 is PS rl is -2. 0
IF-THEN control rules are defined as shown in Table 1. Rule 10: ch.3 is PB and ch. l0 is PS and ch. 13 is PB
It is important that the controller adapts online itself to and hipforce is NB rl is -2. 0
physical and physiological condition of each user. The Rule 11: ch.l is ZO and ch.2 is ZO and ch.3 is ZO
and ch.l0 is ZO and ch.13 is ZO and hipforce is ZO ra is 0.0
adaptation of fuzzy-neuro controller is carried out by Knee joint:
adjusting each weight value of the fuzzy-neuro controller to Rule 12: ch. 6 is PB and ch. l0 is PB and ch. 13 is PB
minimize the evaluation function using the back- and knee force is PB r2 is 1.8

propagation learning algorithm [ 1 ] - [3 ]. Rule 13: ch.6 is PB and ch.10 is PB and ch.13 is PS r2 is 1.5

Rule 14: ch.6 is PB and ch.10 is PS and ch.13 is PS r2 is 1.5

Rule 15: ch.6 is PS and ch.10 is PS and ch.13 is PS r2 is 1.5

Rule 16: ch.6 is PS and ch.10 is PS and ch.13 is PB r2 is 1.5

V. EXPERIMENT Rule 17: ch.6 is PS and ch.10 is PB and ch.13 is PB
and knee force is PS 2 is 1. 5
An experiment was performed with a healthy male Rule 18: ch. 2 is PB and ch. 3 is PB and ch. 4 is PB
subject (27 years old) to show the effectiveness of the and knee force is NS r2 is -1.8

Rule 19: ch.2 is PS and ch. 3 is PB and ch. 4 is PB

assistance of the proposed EMG-based controller. The and knee force is NB r2 is -1.5
experimental setup is shown in Fig. 6. The experimental Rule 20: ch.2 is ZO and ch.3 is ZO and ch.6 is ZO
set-up consists of two motors with encoders, a personal and ch.4 is ZO and ch.10 is ZO and ch.13 is ZO
computer with an interface card (JIF-171-1), a motor and knee force is ZO r2 is 0. 0

drivers (two channels), two force sensors and power

suppliers. Torque commands are sent to the motor drivers
according to the output of the EMG-based controller. The
rotations of the motors are measured by the encoders and
fed back to the personal computer via the interface card.
In the first experiment, sitting down and standing up
motions were performed as examples of daily activities
under different physical conditions (normal and tired
conditions) of the human subject with and without assist of Fig. 6 Experimental setup
the exoskeleton robot. For the experiment without assist of
the robot, the force control is applied to cancel out the force The experimental results of sitting down and standing
between the robot and the user (not to disturb the user's up motion under tired physical condition without and with
motion). assist of the exoskeleton robot are shown in Fig. 8 (a) and
The experimental results of sitting down and standing (b), respectively. Only the results of EMG signals of ch.2
up motion under normal physical condition without and (rectus femoris) and ch.6 (vastus medialis), which represent
with assist of the exoskeleton robot are shown in Fig. 7 (a) the hip and knee muscles are shown. These results show
and (b), respectively. that the activity levels of the muscles were reduced when

978-1-4244-1868-8/08/$25.00 (C2008 IEEE. 294

 6th International Special Topic Conference on ITAB, 2007, Tokyo

the motions were assisted by the exoskeleton robot. to show that the robot is controlled under the motion
Furthermore, one can see that the robots hip and knee intention of the user. Fig. 9 shows the results of the second
angles were smoothly controlled. experiment with and without assist of the robot. Here, only
the result of EMG signal of ch.2 is presented as an example.
Region A and region B show the stopped positions. These
results show that the activity levels of the muscles were
reduced when the motions were assisted by the exoskeleton
robot. Furthermore, one can see that the robots hip and knee
angles were also smoothly controlled.
VI. CONCLUSION
An EMG-based controller is designed to control an
Ch.2 Ch.6
exoskeleton robot for assisting physically weak person's
(a) Without assist
lower-limb motion. The effectiveness of the proposed
EMG-based controller for the power-assist of the lower-
limb motion was verified by experiment with a healthy
human subject.
ACKNOWLEDGMENT

This work was supported by Japan Society of

Promotion of Science (JSPS) Grant-in-Aid for Scientific
Ch.2 Ch.6 Research (C) 19560258.
(b) With assist
Fig. 7 Results of the motions of normal physical condition REFERENCES

[1] K. Kiguchi, T. Tanakda, T. Fukuda, "Neuro-fuzzy control of a robotic

exoskeleton with EMG signals," IEEE Trans. on Fuzzy Systems, vol.12,
no.4, pp.481-490, 2004.
[2] K. Kiguchi, R. Esaki, T. Fukuda, "Development of a wearable
exoskeleton for daily forearm motion assist", Advanced Robotics, vol.19,
no.7, pp.751-771, 2005.
[3] K. Kiguchi, "Active Exoskeletons for Upper-Limb Motion Assist",
Journal of Humanoid Robotics, 2007. (to appear)
[4] S.K.Au, P. Dilworth, H. Herr, "An Ankle-Foot Emulation System for
the Study of Human Walking Biomechanics", Proc. of IEEE International
Ch.2 Ch.6 Con. on Robotics and Automation, pp.2939-2945, 2006
(a) Without assist [5] J.E.Pratt, B.T.krupp, C.J.Morse, "The RoboKnee: An Exoskeleton for
Enhancing Strength and Endurance During Walking", Proc. of IEEE
International Con. on Robotics and Automation, pp.2430-2435, 2004
[6] S.Lee, Y. Sankai, "Virtual impedance adjustment in unconstrained
motion for an exoskeleton robot assisting the lower limb", Advanced
Robotics, Vol.19, No.7, pp773-795, 2005.
[7] H. Kawamoto, Y. Sankai, "Power assist method based on Phase
Sequence and muscle force condition for HAL", Advanced Robotics,
Vol. 19, No.7, pp773-795 (2005) VSP and Robotics Society of Japan 2005.
[8] Y.CanJun, N.Bin, C.Ying, "Adaptive Neuro-Fuzzy Control Based
Development of a Wearable Exoskeleton Leg for Human Walking Power
Augmentation", Proc. of the 2005 IEEE/ASME Intl. Conference on
Ch.2 Ch.6 Advanced Intelligent Mechatronics Monterey, California, USA, 24-28
(b) With assist July,pp467-472, 2005.
Fig. 8 Results of the motions of tired physical condition [9] N. Hata, Y. Hori, "Realization of Robotic-Suit for Walking Assistance",
SICE Annual Conference in Sapporo, Hokkaido, Japan, August 4-6,
abig p elPWte R it iond46
to

pp2266-2271, 2004.
E2MdurG
smo
[10] http:Hgatel.fzi.de/ids/public_html/historical.php
[11] G.Cheron, F. Leurs, A.Bengoetxea, J.P.Draye, M.Destree, B.Dan, "A
~~~~~~~~~~~~~~~~~~~~~~~~~~~~dynamic recurrent neural network for multiple muscles electromyographic
mapping to elevation angles of the lower limb in human locomotion",
Journal of Neuroscience Methods, vol.129, pp.95-104, 2003.
[12] G.S.Sawicki, K.E.Gordon, D.P.Ferris, "Powered Lower Limb
A
the9
A0
lm ''t' E
Tin* jst
2 D _ 20
Orthoses: Application in Motor Adaptation and Rehabilitation", Proc. of
Ch2. Wihu asis Ch2 With assist the 2005 IEEE gth International Conference on Rehabilitation Robotics,
Chicago, IL, USA, June 28-July 1, pp2O6-211 2005.
Fg 9 eut.ftesopn nabtaypsto [13] F.H.Martini, M.J.Timmons, and R.B.Tallitsch, Human Anatomy,
In th seon exeiet th sqatn moinanh Pearson Education, Inc., New Jersey, pp.305-321, 1997.
[14] H.He, K.Kiguchi "A Study on Lower-Limb Muscle Activities during
Daily Lower-Limb Motions" Proc. of The International Special Topic
stadin up motionts soppted storpn in arbilerarinpsthemono Conference on Information Technology Applications in Biomedicine
2007.(to appear)

978-1-4244-1868-8/08/$25.00 (C2008 IEEE. 295