Active Control of a Robotic Arm with Pneumatic ArtificialMuscle Actuator

Hossein Jahanabadi, Musa Mailah, Mohd Zarhamdy Md Zain

Department of Applied Mechanics, Faculty of Mechanical Engineering,

Universiti Teknologi Malaysia,81310 Skudai, Johor Bahru, Johor Darul Takzim, MALAYSIA

h.jahanabadi@ieee.org, musa@fkm.utm.my, zarhamdy@fkm.utm.my

ABSTRACTThe paper is on the control of a robotic arm actuated by apneumatic artificial muscle (PAM) system. Aconventional proportional-integral-derivative (PID)controller with active force control (AFC) is employed tocontrol the robot arm assigned to operate a prescribedtask. Simulation results show the effectiveness of theproposed control strategy compared to the PID controlonly scheme. The results of the simulation studydemonstrate the intense robustness and effectiveness ofthe proposed control scheme in countering the loadingand operating conditions compared to the PID controlscheme.

INTRODUCTIONPneumatic muscle system was first developed by

McKibben in the 1950's as soft actuators in artificiallimbs and became commercially available in the 1980's(Caldwell et al., 1993). They have since been used asactuators in high-tech robotic applications, in physicaltherapy for functional healing and for strength escalationdevices involving humans, where the devices have to beself-contained and carried for long distances. McKibbenpneumatic artificial muscles (PAM) possess all theadvantages of traditional pneumatic actuators (i.e,cheapness, quickness of response, high power/weight andpower/volume ratios) without the main drawback (i.e.compliance or sponginess). For this reason, they arefinding increased use in robotic systems. PAMtechnology is currently under study for use in exoskeletonsuits to be WOl11 by humans for force and/or mobilityassistance. A difficulty inherent in PAM technology foruse in precision and/or force applications is the difficultyin controlling them precisely. This is because they arenonlinear and time varying (i.e., since they are made offlexible rubber or plastic, their characteristics vary withtemperature and PAM temperature varies with use) (Lilly,2003). Due to their high nonlinearities, advanced controlstrategies like adaptive control (Tanaka et al., 1999),adaptive control based on neural networks (Tao andKokotovic, 1996), nonlinear pID control (Thanh and Ahn,2006) have been investigated and explored.

In this paper, AFC with PID scheme is proposed in orderto improve the control performance of the PAMmanipulator. The pID element provides the basic stableand reliable performance while the AFC strategy is usedto suppress the disturbance so as to ensure a robust andeffectiveness scheme is achieved to control themanipulator actuated by highly nonlinear actuators.The paper is structured as follows: First, the McKibbenpneumatic artificial muscle (PAM) is briefly described interms of its construction and operation. The next sectionis on the application of a feedback control method usingAFC with PID to a single link mechanical arm. It isfollowed by modelling and simulation study. The resultsobtained from the study are discussed in the next sectionand finally a conclusion is derived with indication forfuture work and direction of the continuing researchstudy.

PNEUMATIC ARTIFICIAL MUSCLEThe McKibben muscle was invented in the 1950s by

physician Joseph L. McKibben with the initial aim ofdriving an arm orthosis for his poliomyelitic daughter.Even though, at this time, it has been marketed as anorthopedic device (Schulte, 1961). in the 1960s theinvention was abandoned in favour of step motors.However, in the 1980s engineers of the Japanese tyremanufacturer Bridgestone proposed a redesigned andmore powerful version of the McKibben muscle calledthe rubbertuator (Inoue, 1988), that was proposed tomotorize both soft and powerful robot arms(Eskiizmirliler et al., 2002). Such an artificial muscle ­consisting of a braided sheath, according to adouble-helix weaving, surrounding a rubber inner tube ­is mainly characterized by its initial length Ln, initialradius that is depicted in Fig. 1.

••:. '18

Fig.!. McKibben pneumatic actuators relaxed (top) andinflated (bottom)-Shadow Company (Aschenbeck et al.,2006)

- 417 -

(4)

ACTIVE FORCE CONTROLAct ive force control (A FC l strategy was first proposed

by Hewit and co-workers to control a dynam ic system inorder to ensure the sys tem remain stable and robust 111 theexistence of kno wn and unkno wn d isturbances (Husse in e tal., 2000 ). AFC has been demonstrated to be super iorcompared to conventional meth ods in controlling a robotarm {M ailah and Rah im, 2000). Fig, 2 illustrates theprincipl e of AFC concept app lied to a translational system.AFC can be shown to compl ement the basic Newton 'sseco nd law of motion, r.c. for a rotary system:

Where Kp, K1• and Ko are the propo rtiona l, integra l an dderivative gains, respectively. In this study, theZieg ler- Nichols method is employed to tune the PIDparameters, Thu s the gaind ac hieved by thi s method areKp=1, KI=8, Ko"-0.5,

PIO CONTROL LOOPIt is represented by the outermo st loop of the proposed

scheme spec ially designed to provide for pos itionaltrajec tory cont ro l. The tro lley mass d isplacement and theeff ect o f disturbances, friction can be clearly observed atthe controlled variable response (output) of the system. Itis used in co mbination with the AFC scheme since it hasbeen shown that it provides co nsistently stable thou gh notso robu st performance. Also, it is a well-known fact thatPID contro l generally performs e xce llently for a systemwith no o r lillie disturbances and ope rating at a low speed.However, at the co mmencement of adverse cond itio ns,the pe rformance degrad es co nsiderably. The AFC partprovides the extra robustn ess feature through itsdisturban ce rejection capability (Priyand oko et at.. 2008).Usuall y, the p ID system is de signed prior to theimplementati on of AFC, The transfer funct ion of a PIDcontro ller is given as follow s :

(3I

(2)e = (I-OJ - L)llfJ

P = Po ± t. P

Where

The general working principle o f the PAM is similar tothe McK ibben actuat or, thus the same relationshipbetween pressure, force , and co ntractio n can be used . Thefollowing expression of Fas a function of the pressure Pand the cont raction rat io s is as follows (To ndu et al.,2000):

An anta go nist ic pa ir Mc Kibben muscles is illustrated inFig. 2. Th e muscles ca n dr ive a single joint robot link in asimilar way to how the biceps and triceps rotate theforearm arou nd the elbow. However, unlike naturalskeletal muscle s, McK ibben muscle s can not applypassive te nsions s ince the ir braided she lls are notextensible. Both muscle s arc inflated at the same pressurePo and have the same cont racti on ratio q) in the initialstate in o rder 10 keep the robot link in good working order.When the agon ist is inflated at pressure PI, different fromthe anta gon ist pressure P2, an ann rotation of ang le isprodu ced as it is depi cted in Fig .1. A true energet ictran sfo rmation is performed because of the basicpantograph netw ork sha ped by its hel ical weavmg, whichconve rts inner tube c ircumfe rential pressure forces intoan axia l contraction force F

The McKibben muscles used in this study have thefollow ing parameters: Lo= 30 em. rf} = 0.7 em. ~ , = 20and k = 1.30 (Tondu et al., 2000 ), The initia l bra id angle~ is defined as the angle between the muscle axis andeach thread of the braided sheath. before expan sion.

(51

(61

Whe re rd is the es timated d isturbance, IN theestimated inert ia. and Tq is the actuated torque. In thestudy. the actuated torque and acceleration were assumed

Where I Tq is the sum of actuated torque s, I is the tota lmass moment of inertia of the body and (j is theangular acceleration of the rotating mass. In Ar C, we caneffe ctivel y accom modate the d isturbances by obta iningthe measurements of the acceleration and the torque usin gphysical acceleromet er and torque sensor respecti vely.Referring 10 Fig , 3, IN is the es timated inertia of the li nkand f) is the angular acceleration of the arm. Theco ncept of Arc is to use some measured and estimatedvalues of the identified sys tem parame ters namely theactuated torque, acceleration of the body and theestimated inert ia of the bod y.

actuator

(;=0

./ ~ /'

- I , P 7lC!D

- I

Lc,I

Fig. ") Antagonistic Mck ibben muscle(Eskiizmirlile r et at. 200 2)

- 418 -

to perfect ly mode lled while the estimated inerti a IS

obta ined through a crude approximat ion tec hnique .

Fig. 3. AFC scheme app lied to a single link robot armactuated by McKibben pneumatic artific ial muscle

a

FIg. 5 . Respo nse of system controlled by AFC

are suppress ed by the AFC loop depicted in Fig. 8. Thetrend in Fig. 8 produced by the AFC-based scheme isstead ier than the PID contro ller. Therefore, the resultsaffirm that the sys tem is more robu st and effect ive . Figs.9 and 10 exhibit the response of the manipul ator unde rsinuso idal input. In th is case, the jo int position followsthe reference va lue nearly without lag. To eva luate theinfluence of the input, the performance using thesinuso idal input shows marginally better resu lt. Thu s. it isclear that the AFC based schem e manages to suppress thedisturbances effective ly dur ing the arm 's opera tion. Theresults prese nt an important findin g that may assistresea rcher to design and develop a robust tool for a robotarm (or even actual huma n a rm) part icula rly in the even tthe end-e ffec tor or too ling device attached to the wrist issubjected t? various forms of disturbances that includevibration,

"LI ' . ]

.

I~ O••". d Thela. .. . . A~lua ll1'l~lal

~ '1os

o j .' .0 2.5 s

Tim" Is)

0.."" ..... .

•.,-8

SIMULATIONThe simulation was performed using MATLAB and

Simulink software packa ges. The McKibben model,single link robotic manipulator, PID control ler and a lsoAFC are modelled in the Sirnulink envi ronment. Thi s isdepicted in Fig. 4. In addition. the step and sinusoida linputs were applied 10 the system to test the systemresponse.

-

Fig. 4. Simu link diagram of the proposed sys tem

s

(

~ 15 II, I

0.'

The following parameters were used in the simulationstudy:Length of the arm (L): 4\ 0 (mm)Initial pressure (Po): 3 (ba r)Radius of pulley (R): 15 (mm)Pretension of PAM: 10 %Disturbance (step input): 5 N

Addit ionall y, the step input is co nsidered as adisturb ance in the simulation study wh ich can be replacedas a friction in the bea ring and PAM.

RESULT AND DISCUS SIONAr C was implemented on the sing le link actuated by

PAM. Figs. 7 throu gh 10 shows the results obtained fromthe simulation study with two different inputs which arestep input and s inus respectivel y.As can be seen in Figs. 7 to 8 with reference to thedesired angle and actual angle, it is very obvious that theArC method is much more robu st and preci se than thePID con trol. The response curve generated by theAFC-ba sed scheme is far less than the PID control. Inaddition, in Fig. 7 clearly shows the fluctuations which

FIg. 6 . Respon se of sys tem contro lled by PID

..,r /1\r s,• ";

e [ .t,...."'M T~~T 'l

• AI:lU/II T~'l

~ ·1 _.;,s , , , ,.,.".,· 2 ~

T,,,,,, Isl

Fig. 7. Response of sys tem co ntrolled by AFC

- 419 -

Time ts,

Fig . H. Response of sys tem con tro lled by PID

the IEEE !tA,S· EMOS Inft:ffl,lf.....lJIf C"" '''""fI« "" B 'QnIl!dl(il l ROOolICSlind BIOf1l<!>."ootrt>m,·s, 2006

Tondu B, Lorc~ P. - Modchng and ccmrol of McKlbhcn artificialmu'il:1c robo t aCI UJlo~·· . IEFE COllfrol SY"f "'as , Vol . 20. no I, PI'15 -1 11.2000,

G Pnyandcko. M /;I'li la h, H Janlollud,j, n, " Vehicle aCllVC,u,pen" on system usml; ,l,yhood ada ptive neuro acnv c force cunlrol" ,M « mln i ,'al Sys.(emJ lind Sl[:n;l1 Prt>ccssm~ (20011 1.do, :10 IOI61j ymssp 200ll07.01-l

S B Hussein. H hmaluddUl , M Mailah and A Zalzala. MA Il ybor idInlclhgcn. ACI"e Force Ccmrcner For Robol Arm USing f'·oluhon.ll·)rNcur.l1 Ner...ori,$" , Proc . Inr7 C""/~"!IIC-C (J(/ t ·'"(IIUf"""'f) CO/llf'U1llflOllIC [e 2000), San Drcgc. USA, July 2000.

\-I . Marla h. N IA Rahim, "" Intelligent Acuve Force Conlro l of aRollO! Arm USing f Ully Logic" , !ETE Im..m.1ft' lf/RI lIml" J<"Il" c onIII(dhg<'/l1 5p,<'m.1 IIml fcr:lmvfvgl<!,. -r I NCON 20ll0. Kuab Lumpur.September 2000

r - Onof'e'd ThetaI

a 5 , _. ~...... rl'Wlla

as,i e,

I 05• 0 l---\-+~\-"";-~~~

1~ -I) 5.,.".,·2.5

CONC LUSIONThe efficiency of the proposed AFC-ba~d control

method has been demonstrated o n a single linkmanipulator actuated by PAM Th e robustness andacc uracy of the proposed system was particular lyhighlighted in the Simulat ion study. Th e res ults may serveas a potent ia l too l in aiding the design and de velop mentof a device for usc in a Mecharronics robot arm or eve nhuma n ann . Further researc h co uld be ca rried Qui tocomplemen t the results obtained in the "rudy. This mayincl ude inve stigation of the sys tem subject 10 the usc ofartific ial in telligence (A I) methods. two links act uated byPAM. and varying loading and operating conditio ns suchas di fferent type of dist urbances and speed of o peration.Research on the pract ical impleme ntatio n of the pro posedsys tem is actively on-going and in thc fina l stage ofco mpletion.

ACKNOWLEDG EM ENTSTh e authors wou ld like to acknowledge the Ma laysianMi nistry of Sc ience, Technology and Inno vation(MOSTI) and Unive rsiti 'Ick nolo g i Malaysia (UTM ) forthe ir support in the research work. The resea rch issuppo rted through an c-Scic nccfund gran t ( Project No. :03-0 1-Oh--SFOOO8, Vole No. 79 112)

REFE RENCES

nI

Hoss ein J ah anabadi receivedhis fiSc (20{).J) degree inMecha nical Engi neering fromU nivers ity of Mazandaran. Iran .lie IS cur re ntly a Mastersstudent at the Facult y ofMechanica l Engineering, (F:.vt E),Universi ri Te knologi Mal aysia(UTM). His current interestsinclude mechatronics sys temdesign, robotics , humanoids .

Musa Mailah rece ived his BEn gdegree In MechanicalEngineering from U'I'M in 1988and lat er his MSc and P hD inRobotic Control andMecharronics from theUniversity of Dundee, UK in1992 and 1998 respectively,Currently, he is an Associa teProfessor in the Depar t ment ofApplied Mechan ics, FME, UTM.His cur re nt research interestsare robotic s, mechetron ics andactive force control ofdynamica l systems.

Caldwell D G. Mcd rano-Cerdl . GA .. & Good ", in~ \1 J "BraidedI'nC\lmalie AeIUJI~r Control of a MlIlIl-Jo lnled Malllpub lo r " It EES/I-lCColllcrt:ncc. VOl I. France, I'N )

H I. Lilly, "Adapuve TradIng for Pncumalie \ Iuselc Acuu torsin Bicep and Tncep Cun lil;urallons", IFEE T'RIl'/. J A"crIf,lI S.I .'t~·m.•ilnd R('Iublkw /{)fI Eng ,n,: .:nng , Vol II , No. J. 2(0)

G Tanaka. Y Yamada. T Salab, A Uch,oori. S UChlludo. "" ModelRe ference Adapt ive Control .. "h Mulu -Rale Type: Neural f\ etwor~ forEjec tro-Preumauc 5.:0'0 Sy~em", In Proc IEEE InrclTUfKll'lll1O>nrCf'fflCCQfI COIIrro! Appl,,"f'''''~, Ha.. a ll, pp. 1716 1172 1, 1q<N

G. Tao. P.V. K"l, uluvic. - A ,I;>pl'''';o C<>lIfrnl ot,.Ufcms •..,l h .1,·lu.1formHi .','11.'<" n..nlm<·/In(I~·" ', John WI I ~ y & Sons. Inc , "I e..' Yorl" I 'N~.

T D C. Thanh. K K,Ahn, " N<>nlinear I'ID con trol 10 Imp,,"e Ihecon lml performa nce of:): a\c~ pn(ulnal ,.; ar nflcral muscle rnantpularo ruSing neural n"h. ork " , in Juumil l Qrl< k d 3(romcs. PI'. ~7"_ ~1I7, 201)(,

S ES~ I1~m"I , lcr, N. Fo teSl1cr. B Tond u, C. [);arlO! . ~A model of theccrr~lbr path",'ays appllCd l<> .he cceuot of a sIngle-join '. l'IlobvI a.rmactuated by ~kK ibbcn al1lficial muscles:' In .h>umillol BJOH't! ,c/l1CyM1N!I1C5. Sprmgcr . Vol. ~6, pp 379_39-$.2002

KlIl1 S AsdleT1 bccl" N,col e l. Ker n, Richard J. Bachnunn, Roger DQ umn, "Oc . jgn of a quadruped mbol driven by all musc les". in I'n,>(' fit

- I• •

Mohd Zarhamdy M Zainreceived his n Eng degree inMechanical Engi neering fromUTl\1 in 1999 and later his M&and PhD In ControlEngineering from theUniversity of Sheffield, UK in2002 a nd 2006 respect ively,Curren tly, he is a SeniorLecturer in t he Department ofApplied Mechanics, FM E, UTM.His current research interestsare control. auto mation andmechatronics .

- 420 -