-
8/3/2019 01173329
1/6
Power Assist Control for Leg with HAL-3Based on Virtual Torque
and Impedance Adjustment
s.Lee*,Y. a * **Doctoral Program of System and Information
Engineering, University of Tsukuba,lee0golem.kz.tsukuba.ac.jp**
Institute of Systems and EngineeringMechanics,University of
Tsukuba,
sankai0golem.kz.tsukuba.ac.jpTsukuba, Ibaraki, 305-8573,
Japan
Abstmct- This paper describes a method ofpower assist control
for lower body based on neu-romuscular signa), s-EMG (surface
ElectroMyo-gram/Myoelectricity) , nd impedance adjustmentaround
knee joint with the assist system, HAL (Hy-brid Assistive Leg) -3we
have developed. Virtu alTorque calculated by s-EMG enabled the HA
L3 t obe operated as o intention of th e experimental sub-ject put
on HAL, and assist the motion of lowerbody by predicting the moment
around joints. Be-sides, th e operator was able to swing the leg
lighterby reducing the inertia and viscous friction aroundjoint of
the subject and HAL-S. In order to verifythe proposed method,
experiments for simple mo-tion was performed with im pedance values
found byparameter identiflcation with FUS (Recursive LeastSquar e)
method. The evaluation of assisted motionwas done by Assist
EtRciency (AB) alculated froms-EMG in nearly proportion to the
operator's muscleforce. Th e results showed th e response of
operationalsignal into actuato r with impedance adjustment
wasimproved dramatically, and the amplitudes of s-EMGwere reduced
signiflcantly, then we could conflrm theavailabilityof mpedance
adjustment.
Kegwonis- HAL-S, Virtual Torque, impedance ad-justment, Assist
EfRciency(AE).1. INTRODUCTION
Nursing care and rehabilitation are required to be im-proved in
accordance with aging in several country. It isimpo rtan t to
enable physically weak person i.e., the oldand t he disabled to
take care of themselves in th at soci-ety. From he point of view of
locomotion, the spheres ofthe disabled person who have some
disorders like asneu-romuscular diseases, or the aged person who
have mus-cular atrophy are restricted in spite of using
wheelchairdue to stairway or unlevel ground, and it is desirable
forsuch people to walk by themselves with respect to
theirrequirement to move, burden of caregiver, and effective-ness
of rehabilitation . Nevertheless, only few attemptshave been made
at another device to assist the leg'smovement for such people. By
the way, recent progressof robotics technology brings a lot of
benefits in manyother fields like welfare, medicine as well as in
the in-dustry. In particular, integratin g human s and
roboticmachine into one system offer the opportunities for
cre-ating new assistive technologies th at can be used in
suchfields 111, [2]. W e ave developed the s EM G based
ex-oskeleton system for lower body, HAL in recent years
for the context asmentioned above (31, [4], [5], [SI. Thepower
assist control of HAL has been performed wen-tially based on s-EMG
(surface-ElectroMyogam,/ My-oelectricity) of flexor and extensor
muscle. HAL cancognize the operator's condition by th e sensors,and
en-ables the operator (experimental subject put on HAL)do a motion
such as walking to be assisted by trans-ferring the intention of
operator to motivity of actua-tor through s-EMG. Our objective in
this research isto realize the more effective assisted motion based
ons-EMG with the impedance adjustment around kneejoint. This paper
describes the hardware configurationof HAL 3 in Section11,the
control method with s-EMGand impedance adjustment based on
estimated param-eters in Section 111, the evaluation method derived
froms-EMG and experimental results to verify the proposedapproach
in Section IV.
Back Fnck
semora
Fig. 1. HAL Hybrid kssistive Leg) -3
11. HAL (HYBRID SSISTIVE EG) -3Fig. 1. shows the system overview
of the exoskele-
ton t ype power assist system, HAL (Hybrid AssistiveLeg) -3 we
developed, and Fig. 2. indicates the systemconfiguration [7].All
devices for control of HA L3 suchas CPU board (PentiumII 566MHz,
PCI665VRE) ,mc-tor driver (TITECH Driver PC-0121-1, 750W, f 8V,*BA)
,measuring unit, and power source (battery, ac-tuator: 9V x 3,
controller: 1OV x 1) are contained in
0 002IEEE SMC"PlB3
http://lee0golem.kz.tsukuba.ac.jp/http://sankai0golem.kz.tsukuba.ac.jp/http://sankai0golem.kz.tsukuba.ac.jp/http://lee0golem.kz.tsukuba.ac.jp/
-
8/3/2019 01173329
2/6
backpack. A experimenter can adjust the experimentalsettings and
moni tor the operators condition constantlywith palm-top computer
through wireless LAN. Besides,HAL can recognize the operators
condition from sen-sors installed on main body of HAL-3 for
detecting thes-EMG, angle of joint, and foot-grounding and
workautomatically along to operators conditions. So HALcan be
operated alone without external operational en-vironment. Harmonic
drive gear and DC servo motor(Maxon, 150W, 24V) are adopted as the
actuators ofHAL, which have sufficient motivity to support
calcu-lated-on he basis of the maximum torque in standing upmotion
and angular velocity in walking motion (Table
Fig.2. System configuration
TABLE IDATAOF KNEE AND HI P JOINT BASED ON MOTION ANALYSIS
Irod/seclhip joint I 2.2 I 3.8 700kneejoint I 2.5 1 5.0 800
111. CONTROL ETHODA. Virtual Torque
The predictive power assist control with operatorsintention t o
move can be realized by Virtual Torque withadequately processed
s-EMG (surface ElectroMyoGram/ Myoelectricity) signal. The s-EMG is
electric signalthat is generated on he surfaceof muscle through
nervecells just after the controller of human (Central NerveSystem)
convey motor commands to muscle [8]. Andthe torque around the
humansjointcan be estimated bymeasuring the s-EMG because its
amplitude is nearly inproportion t o th e force generated by muscle
contraction.The procedure to detect th e s-EMG and the
relationship
Diffntial Amplifier/Gbws-EMG Signal
SurfaceElectroda
Fig. 3. (a)The procedure to generate the myosignal
(b)therelationship between s-EMG nd muscle force
between myosignal and muscle force measured by forcesensor are
shown in Fig. 3. So the necessary the timingof joints movement and
muscle power can be calculatedindirectly and previously.
As cEMG as it is inappropriate to use itself forcontrol, the raw
signal is filtered through RMS (RootMean Square) filtering process
after amplified and fil-tered through the analog circuit [5].
where r(t) is raw signal of s-EMG, and T s integral p eriod.
Then, th e control of HAL with operators animusto move can be
performed by converting the myosignalbalance of flexor muscle and
extensor muscle to oper-ational signal of actuator . Actually,
estimated torq uearound joint, called Virtual Torque [3], is found
as fol-lOWS
whereI;tirtuot(t) = KfE/ t= ( t )-KeJLt(t) (2)
rwirtuot(t) irtual Torque;K f ,K, Conversion factor from s-EMG
to torque;Eft=(t), .,t(t) Filtered sig nal of s-EMG at flexor an
dextensor.The conversion factors were determined by trial and
er-
ror in former research, and now, hese are also
decidedautomatically through simple motion by Neural Net-work or
Recursive Least Square algorithm [SI, [7].
-
8/3/2019 01173329
3/6
B . Impedance Adjustment Around Knee JointThe harmonic drive
gear and DC servo motor with
relatively high torque-to-weight ratio is adopted as th
eactuator of HAL, but it cannot be operated with flex-ibility in an
ordinary way. Then, we regard that it ispossible to perform the
more effective assisting control ifthe actuator can regulate t he
characteristics around itsjoints according to a motion as human
enables his/herjoints to be flexible or stiff by adjusting the
strain ofmuscle. If muscle force around knee joint is not
gen-erated, and external force works to same direction astorque of
actuator, the lower thigh of operator put onHAL3 can be represented
by llink pendulum model.The motion equation is expressed in this
case as equa-tion(3),
where I and D are inertia and viscousco&uent aroundknee
joint respectively, C ( e ,g) s non-linear erm in-cluding gravity
and coulomb friction, 7 is torque ofactuator, and Fe is external
force. When the targetimpedance, M , B, K found based on operator's
condi-tion in contacting the environment, and target postureis set,
the motion equation can be expressed by equa-tion(4)
whereM , B, K target value of inertia, viscous
coefficient,elastic coefficient;00 angle of joint in target
posture.In order to regulate the impedance, compensationtorque
generated by actuator is determinedas
d2e de~ c o m ( I -M )? + D- B ) x+K(0o - )dt
Hence, Virtual Torque with impedance adjustmentaround knee oint
is decided as follows
raum = v,i+tuai + 7cornd 2 e de= K f E f l z - KeEczr + ( I - M
) - +(D B ); i ldt 2
Fig. 4. shows block diagram of these process. Thepower assist
control according to operator's intentionwith variable impedance,
inertia, viscous friction, stiff-ness,can be performed by feedback
of detected angle ofjoint, and calculated angular velocity and
acceleration.Non-linear term is omitted in this figure.
IV . EXPERIMENTA . Performance Indices
AS amplitude of s-EMG is nearly commensurate withthe torque
around joint, the assisted motion by HAG 3
Fig. 4. Virtual Torque with impedance adjustment aroundknee
joint
can be evaluated with form asEMG, ,H -EMG,"A (7)A E = E M G , ,
H
E M G a W a= i* MGA..i .t(t)dt ( 8 )(9)MG,,H = -lT
MGHurnan(t)dtwhereA E Assist Efficiency[%];EMG, ,A averageof s-EMG
(with assist by HAL) ;EMG, ,H average of s-EMG (without assist by
HAL);EMGa.. i . t ( t ) s-EMG with assist by HAL) ;E M G H , , , ,
( ~ ) s-EMG ignal (without assist by HAL);T measuring time.
Assist Efficiency is consideredas the case of flexor andextensor
separately. that is,AEfl Assist Efficiency of flexor muscle;A E , ,
Assist Efficiency of extensor muscle.
e.g., A E f l = 60% means that HAL generates 60% oftorque
required to perform the same motion human havedone without assist
of HAL by flexor muscle.B. Pammeter Identification around Knee
Joint with Re-
Parameters around knee oint must be acquired in or-der t o apply
the control method mentioned above. RLSmethod is adopted for
parameter estimation in this re-search. So inertia, viscous
coefficient, and mass of jointcan be calculated easily all at once
(In this paper, massof lower thigh doesn't be used to control) .
Discretetransfer function of operator's lower thigh is as
follow-ing form [9].
cursive Least Square (RLS)Method
T 2 / 2 1(z-' + z-')1+ ( D T / I - ) 2-1 + (mglT2/I+ 1- DT / I )
- ~
(10)
-
8/3/2019 01173329
4/6
whereT(Z) torque generated by actuator (input) ;@(z) angle of
knee joint (output) ;T sampling time (lomsec) .Parameters is
represented by coefficient of followingARX model derived from
equation(l0)
Inertia[kg. ma]ViscousCoef.
e ( k )+ a16(k - 1) +a d ( k - 2)= bl.r(k- ) + 62.r(k- ) +e(k)
(11)m2
Lower-Thigh Lower Thigh0.0185 0.0179
0.41 0.48
I = L261
So parameters can be found by coefficient a1 - b. fthese
parameters are unidentified, estimated coefficientvector and
recursive vector are represented by
8 = (a1 a2 bl b2) (15)cp(k) = [-e(k - 1) - ( k - 2) r ( k - 1)
7(k - 11 (16)
(17)Then , estimated coefficient vector is found as followsq k
)= &(]E- 1) +~ ( k )q k )- pT(l c )s (k- 111 (18)
P(k)=
where X is forgetting factor (0 < X < 1) . Fig. 5.shows
experimental data of parameter estimation. Theexperimental subject
(operator) is 25-years-old, phys-ically unimpaired male(height:
176cm, weight: 68kg).The operator is seated putting on HAL-3 with
weak-ness of muscles, and actuator of knee joint generate thetorque
randomly for irregular angle pattern. The lengthbetween joint and
COG is determined as 1=0.13[m][7],initial value of forgetting
factor and recursive vector ar eselected as X=0.998, P(0) = AI, A =
12000 in thisexperiment. Table 2 shows found parameters of bot
hlower thigh. These parameters was confirmed as ade-quate value by
[lo] including exoskeleton.C.Ezperimental Results for Simple Motion
Assisted b y
Fig. 6. shows setu p of experiment for simple mo-tion. The
operator makes his lower thigh move up anddown as o set target
angle and initial angle with assistof H A G 3 in this experiment
(at lsec interval) . Samemotion has done without HAG3 before this
experiment
Virtual Torque with Impedance Adjustment
W Y U Y Y
....................... ...........;.hcnk... I ........... - - -
-: -Pl.cc-. Mu s i........ 4.le
3....................................................................--.....--.----..,
.__ :-.--...- ...-.. _
........... j ..........: .......... .......... j ..........
,-.-----;--_._:_____:- -_ -_------e le m y U Y Y-l=l
Fig. 5. Experimental result of parameter estimation (leftlower
thigh)
TABLE I1ESTIMATEDARAMETER OF OPERATORSLOWERTHIGH(INCLUDING
XOSKELETON)
Parameter 11 Right I Left
to compare the amplitude of s-EMG and calcula te theAssist
Efficiency. Fig. 7.indicate the experimental re-sults. Fig. 7.-(a)
represents the amp litude of filtereds-EMG and the angle of knee
oint in the case of no mo-tion aid by HAL, and 7.-(b) shows he case
of motionassisted by HAL with only Virtual Torque. Though thepower
assist for same motion is performed according toset target angles,
amplitude of s-EMG in Fig. 7.-(c) isremarkably reduced compared to
Fig. 7.-(b), i.e., HAL-3 assists the torque of operator by 86%
concerning theextensor of knee in the case of inertia-viscosity
compen-sation, and operator can move the lower thigh with lesspower
with intention to move. In Fig. 7., e.g., 30%compensation of
inertia means reducing 30% of amountfrom the estimated inertia.
-
8/3/2019 01173329
5/6
Fig. 6. Experimental setup for simple motion
compensationcompensation
TABLE I11ASSISTEFFICIENCYA E ) N THE EXPERIMENT OR
SIMPLEASSISTED MOTION--6% 89%Assisting Method 11 AEfi I AEe,t2% 57
o.-+ inertia (30%)and viscosity 11 83% I 86%
D. Response Analysis of Operational SignalFig. 8. -(a), (b), (c)
show the response properties
of assisting methods. Shaded portions indicate the re-gion where
the angle of knee varies from initial angleto targe t. Operational
signal of actuator rises gently atdotted circle in the case of only
Virtual Torque (Fig. 8.-(a)) ,and the time to peak is delayed for
about 150mseccompared to the raw signal of s-EMG due to
filteringprocess. Th at causes a difference of response
betweenhuman subject's leg and exoskeleton, and can make
theoperator feel discomfort for assisted motion althoughthe
actuator is controlled as o operator's intention. Onthe contrary,
if inertia and viscous friction around kneejoint are compensated,
the change of operational signalbecomes steep almost as soon as
angle initiates falling,asFig. 8.-(b), (c). Notably, the variation
of operationalsignal precede that of the angle by 120 - 130msec
inFig. 8. -(c). Hence, the operator can perform the in-tended
motion of lower thigh including the exoskeletonsmoothly by assist
of HAL with less muscle force. Com-pensation ratio of inertia can
be regulated as to opera-tor's condition or external environment
(e.g., grounding,not grounding) .
V. DISCUSSIONSta tic friction around knee joints are not
compensatedin this research because it is require to restrain the
knee
joint adequately by the exoskeleton when the actuatorof each
joint is in stationary state, e.g., in standing.Assist Efficiency
subject to stay unchanged fairly from70% of compensated inertia,
and the motion of lower
0 1 2 3 4 5 6Time[scc](c)
Fig. 7. Experimental result for simple motion (a)no assistwith
HAL, (b)Virtual Torque, (c)+ inertia (30%) andviscosity
compensation, (d)+ inertia (60%)and viscositycompensation
-
8/3/2019 01173329
6/6
,
iij8E -2d-8!st
0 0.5 I 1.5 2 2 5Timc[ak](a)
2
68I
o $Y2
2
0 0.5 I IS 2 2.5Timc[scJ
Fig. 8. The Response properties of assisting meth-ods(a)Virtual
Torque, (b)+ inertia (30%) and viscositycompensation, (c)+ inertia
(60%) and viscosity compen-sation
thig h become unstab le from approx. 95 % . Percentageof inertia
to be compensated is found based on opera-tors condition, and also
can be adjusted for operator tomove the lower thigh lighter. 30% -
0% is suitable byoperators opinion in this experiment.
VI. CONCLUSIONThe paper has proposed the power assist control
withHAG3 using s E M G and impedance adjustment aroundknee oint. We
have developed the power assisting sys-tem, HAG3 which can generate
enough output to as-
sist the motions at each joint, have a interactive per-formance
by sensor system and wireless LAN. VirtualTorque based on sEMG is
adopted as basic controlmethod of HAL-3, and we confirm th e
assisting motionas t o intention of operator can be realized. For
moreeffective power assist control, we suggest the impedance
adjustment around knee oint. Parameters required forimpedance
adjustment like as inertia, viscous frictionare estimated by
Recursive Least Square method. Theevaluation of assisted motion is
carried out by AssistEfficiency calculated from average of s-EMG as
it s am-plitude is almost in proportion to torque generated
bymuscle force. Th e experimental result for simple motionwith
impedance adjustment shows the operational sig-nal of actuator is
improved, an d remarkable decrease ofrequired torque at knee oint
despite t he same motionas only with Virtu al Torque.We willexpand
this method up to hip joint, and carryout the adaptive power assist
control to the operatorby auto-regulation of each parameter for
whole joint oflower body as uture works. And in near future, we
planto perform the experiment with t he old or the disabledperson
who have functional disorder in the lower body.AcknowledgmentsThis
research was partially supported by the Ministry
of Education, Science, Sports an d Culture of Japan
,Grant-in-Aid for Scientific Research (A) an d (B).
REFERENCESK. Kiguchi, S. Kariya, K. Watanabe, K. bumi, T.Fukuda,
An Exoskeletal Robot for Human Elbow Mo-tion Support-Sensor Fusion,
Adaptation, and Control,IEEE ltnnsactions on Systems, Man, and
Cybernetics-Part B : Cybernetics ,vol. 31, No.3, pp.353-361,
2001.J. Rosen, M. Brand, M.B. Fushs, M. Arcan, AMyosignal-Based
Powered Exoskeleton System, IEEEThnsactions on Systems, Man, and
Cybernetics-PartA:Systems and Humans ,vol. 31, No. 3, pp.210-222,
2001.J. Okamura, H. Tanaka,Y. ankai, EMG-based Proto-type Powered
Assistive System for Walking Aid, AsianSymposium on Industrial
Automation and Robotics(ASIAR99) ,Bankok, Thailand, pp.229-234,
1999.K. Kasaoka, Y. Sankai, Predictive Control Es-timating
Operators Intention for Stepping-up Mo-tion, IEEE/RSJ International
Conference on Intelli-gent Robots and Systems (IROS2001) , Maui,
Hawaii,S.Lee, Y. ankai, Power h i s t Control for Walking Aidby HAL
B a d on Phase Sequence and EMG, Interna-tional Conference on
Control, Automation and Systems(ICCAS2001) ,Cheju, Korea,
pp.353-357, 2001.H.Kawamoto,Y. ankai, EMG-Based Hybrid AasistiveLeg
for Walking Aid Using Feedforward Controller In-ternational
Conference on Control, Automation and Sys-tems (ICCASZOOi) ,Cheju,
Korea, pp.190-193, 2001.T. Nakai, S.Lee, H. Kawamoto, Y. ankai,
Developmentof Power Assistive Leg for Walking Aid using EMG
andLinux, Asian Synpoaium Industrial Automation andRobotics
(ASIAR2000) , Bankok, Thailand, pp.12951310,2000.
u s , pp.i57ai583,2001.
[SI Stephen J. Piazza, Scott L. Delp, Three-DimensionalDynamic
Simulation of Total Knee Replacement MotionDuring a Step-up Task,
Journcl of Biomechanical En-gineering, ASME ,Vo1.123, pp599,
2001.[9] B. Shahian, M. Hassul Control System Design UsingMatlab
,Prentice-Hall, Inc, Simon & Schuster Company,Englewood Cliffs,
New Jersy,US, 1993.[lo] B. I. Kaleps, C.E. Clauser, et,
Investigation into theMassDistribution Properties of the Human Body
and itsSegments, Ergonomica ,vol. 27, No.12, pp.1225-1235,1984.
