Procings of th 2001 IEEE Intrnational Confrnc on Robotics & Automation Soul, Kora May 21-26, 2001 Fuzzy-Nuro Control of an Exoskltal Robot for Human Elbow Motion Support Kazuo Kiguchi *1, Shingo Kariya *1, Kigo Watanab *1, Kiyotaka Izumi *1, an Toshio Fukua *2 *1 Dpt. of Avanc Systms Control Enginring Grauat School of Scinc an Enginring Saga Univrsity Saga, 840-8502, Japan E-mail: kiguchi@i.org Abstract W ar vloping xoskltal robots for human (spcially for physically wak popl) motion support. In this papr, w propos a fuzzy-nuro control mtho for a 1DOF xoskltal robot to support th human lbow motion. Th propos controllr controls th angular position an impanc of th xoskltal robot systm bas on vagu biological signals that rflct th human subjct s intntion. Skin surfac lctromyogram (EMG) signals an th gnrat wrist forc by th human subjct uring th lbow motion hav bn us as input information of th controllr. Bcaus of th aaptation ability of th fuzzy-nuro control, th robot is flxibl nough to al with vagu biological signal such as EMG. Th xprimntal rsults show th ffctivnss of th propos controllr. *2 Cntr for Cooprativ Rsarch in Avanc Sci. an Tch. Nagoya Univrsity Nagoya, 464-8603, Japan 1 Introuction A cras in th birthrat an aging ar progrssing in Japan an svral countris. In that socity, it is important that physically wak popl ar abl to tak car of thmslvs. W hav bn vloping xoskltal robots to support motion of th physically wak popl such as lrly prsons [1]. It is important for l prsons to tak car of thmslvs in vryay lif. Rcnt progrss of robotics tchnology brings a lot of bnfits not only in th inustris, but also in many othr fils such as wlfar, micin, or amusmnt. In th fil of wlfar, for xampl, much rsarch has bn carri out for th isabl popl who lost thir original function in orr to support thir motion [2][3], or to mak up thir lost function [4]-[11]. Th lctromyogram (EMG), which contains biological information to unrstan th patint s muscl activitis, can b us as input information for th robotic prosthtic vics. This signal is important to unrstan how th human subjct intns to mov. It is ifficult, howvr, to obtain th sam EMG signal for th sam motion vn from th sam patint sinc th EMG signal is a biologically gnrat signal. Many factors such as fatigu of patints affct th biological signal [12]. Furthrmor, anothr patint gnrats iffrnt lvl of th EMG signal. Consquntly, it is important that th systm has ability to aapt itslf to th physiological conition of ach human subjct on-lin [13][14]. W hav propos a 1DOF xoskltal robot systm for human lbow motion support as th first stp towar an xoskltal robot for th whol boy motion support [1]. This robot systm is mainly suppos to hlp th motion of physically wak popl. Th skin surfac EMG signals an th gnrat forc by th human subjct s wrist uring th human lbow motion hav bn us for input information of th control systm in orr to control th robot systm bas on human subjct s intntion. In this papr, w propos an ffctiv controllr in which both th lbow angl an impanc of th xoskltal robot ar controll in accoranc with th vagu EMG signals an th gnrat wrist forc signal. In orr to mak th controllr al with an aapt to ths vagu signals, fuzzy-nuro control has bn appli an taching systm is introuc for this robot systm. Th ffctivnss of th propos control mtho has bn valuat by xprimnt. 2 Exoskltal Robot Systm W hav sign th 1DOF xoskltal robot for human lbow motion support [1]. This xoskltal robot is suppos to b attach irctly to th latral si of human arm as shown in Fig. 1. This robot consists of two links, a ballscrw riv shaft, a ballscrw support fram, a DC motor, an forc snsors (strain gaugs). Th DC motor rivs th ballscrw riv shaft to mak th link-2 flx or xtn. Th link-2 is flx (or xtn) by contracting (or xpaning) th prismatic joint along th 0-7803-6475-9/01/$10.00 2001 IEEE 3668
DC motor Ballscrw fram Link 1 Link 2 Strain gaugs Link 2 Wrist holr Wrist holr outr covr Wrist holr innr covr Fig. 3 Th strain gaug bas forc snsor Fig. 1 Th attach xoskltal robot Ballscrw fram Link 1 Elctro - Elctro + Elctro groun Link 2 (a) Bicps Fig. 2 Gnration of flxion motion of th robot ballscrw riv shaft in th ballscrw support fram, which is attach to th link-1, as shown in Fig. 2. Th forc gnrat by th human subjct s wrist uring th human lbow motion is masur by th strain gaug bas forc snsor. In this forc snsor, strain gaugs ar attach on th bams btwn th wrist holr outr covr, which is connct to th xoskltal robot, an th wrist holr innr covr, which is connct to th human subjct (Fig. 3). Th signal from th forc snsor is sampl at a rat of 2kHz an low-pass filtr at 4Hz. Th masur forc by ths forc snsors ar us to unrstan th forc xtrnally acting on th human subjct s forarm. Th skin surfac EMG signals of bicps an tricps, which imply th human subjct s intntion, ar anothr Elctro - Elctro Elctro + groun (b) Tricps Fig. 4 Location of lctros input information to control th robot. Th location of lctros on bicps an tricps (2ch for bicps an an- 3669
Tabl 1 Elbow motion pattrns in th pr-xprimnt 1. Flxion/xtnsion motion without a wight in a han 2. Flxion/xtnsion motion with a 7kg wight in a han 3. Hol motion with a 7kg wight in a han 4. Flxion motion with a 18kg wight in a han (ovrwight) 5. Flxion/xtnsion against th fix constraint (a) EMG (b) Wavform lngth of EMG Fig. 5 Sampl of WL an EMG uring lbow motion othr 2ch for tricps) is pict in Fig. 4. Th tails of control mtho ar xplain in Sction 4. Usually, movabl rang of human lbow is btwn 5 an 145 grs. Consiring th safty of th human subjct, th lbow motion of th propos robot is limit btwn 0 an 120 grs in this systm. Th maximum torqu of th robot is also limit for safty. 3 Human Elbow Motion Human lbow is mainly actuat by two antagonist muscls, bicps an tricps. By ajusting th amount of forc gnrat by ths muscls, th lbow angl an impanc can b arbitrary controll [4]. Th muscl activity lvl can b scrib by th EMG signal. In orr to sign th control systm of th xoskltal robot, th skin surfac EMG signals an th gnrat forc in th human subjcts wrist uring th human lbow motion hav bn analyz by th pr-xprimnt. Tabl 1 shows th analyz human lbow motion pattrns in th pr-xprimnt. Th amplifi EMG signals ar sampl at a rat of 2kHz. Wavform Lngth (WL) [5] that quation is shown blow is us to xtract th fatur of EMG signals. WL = N k = 1 x k x k 1 (1) whr x k is th kth sampl voltag valu an N is th numbr of sampls in sgmnt. Th numbr of sampls is st to b 100 in this stuy. An xampl of th WL of bicps obtain in th pr-xprimnt (task 2 in Tabl 1) is shown in Fig. 5. On can s that th bicps ar activat uring th lbow flxion motion from th magnitu of th WL. 4 Control of th Exoskltal Robot Fuzzy-nuro control mtho is propos to control both th angl an impanc of th xoskltal robot bas on both th skin surfac EMG signals of bicps an tricps an th gnrat forc in th human subjcts wrist. So that th robot can b controll in accoranc with th human subjct s intntion. Th fuzzy IF- THEN control ruls of th fuzzy-nuro control ar sign bas on th analyz human subjct s lbow motion pattrns in th pr-xprimnt. Th proprtis of human lbow impanc stui in anothr rsarch [15][16] ar also takn into account. By applying snsor fusion with th skin surfac EMG signals an th gnrat wrist forc, rror motion caus by littl EMG lvls an th xtrnal forc affcting to human arm can b avoi. Th input variabls of th fuzzy-nuro control ar th WL of bicps (2 channls) an tricps (2 channls) an th forc masur by th wrist forc snsor. Thr kins of fuzzy linguistic variabls (ZO: zro, PS: positiv small, an PB: positiv big) ar prpar for th WL of EMG an fiv kins of fuzzy linguistic variabls (NB: ngativ big, NS: ngativ small, ZO: zro, PS: positiv small, an PB: positiv big) ar prpar for th wrist forc ata. Th outputs of th fuzzy-nuro control ar th sir joint angl an impanc of th xoskltal robot. In 3670
Taching Equipmnt Taching Equipmnt Encor Drivr RIF-01 UPP A/D D/A PC FNN Controllr Fig. 6 Th taching quipmnt EMG AMP STRAIN AMP this mtho, impanc control is prform to follow th gnrat sir joint angl using th gnrat sir impanc cofficints. Consquntly, both th angl an impanc of th xoskltal robot ar controll lik human bings o. Th quation of impanc control is writtn as: τ = M ( q&& q&& ) + B ( q& q& ) + K ( q q) (2) whr τ nots torqu comman for th xoskltal robot joint, M is th momnt of inrtia of link 2 an human subjct s forarm, B is th viscous cofficint gnrat by th fuzzy-nuro controllr, K is th spring cofficint gnrat by th fuzzy-nuro controllr, q is th sir joint angl gnrat by th fuzzy-nuro controllr, an q is th masur joint angl of th xoskltal robot. Th torqu comman for th xoskltal robot joint is thn transfrr to th torqu comman for th riving motor. In th fuzzy-nuro controllr, 20 kins of fuzzy IF-THEN ruls ar prpar to gnrat th sir joint angl an impanc of th xoskltal robot. It is important that th controllr aapts itslf to physiological conition of ach human subjct on-lin, sinc th EMG signal is a biologically gnrat signal. In this stuy, th antcnt part an som of th consqunc part (i.. fuzzy ruls for th sir joint angl gnration) of th fuzzy IF-THEN control ruls ar suppos to b ajust. Th back-propagation larning algorithm has bn appli to minimiz th squar rror function writtn blow. 1 2 E = ( q q) 2 E = 0 IF ( q q) is not ZERO IF ( q q) is ZERO (3) Fig. 7 Exprimntal stup whr q is th angl of th sir motion an q is th masur joint angl of th xoskltal robot. Th sir motion of th xoskltal robot, which is rquir for th valuation function of th back-propagation larning, is monstrat by th taching quipmnt (s Fig. 6) attach on th othr arm. In th taching tim, th subjcts ar suppos to gnrat th sam lbow motion in both arms. Consquntly, th iffrnc btwn th right lbow angl an th lft lbow angl is th control rror of th xoskltal robot systm. In orr to avoi th uslss aaptation caus from littl rror, th fuzzy variabl: ZERO is appli to xprss th rror of th xoskltal robot is about zro. 5 Exprimnt In orr to valuat th ffctivnss of th propos control mtho for th xoskltal robot systm, xprimnts hav bn prform with thr halthy human mal subjcts (subjct A an B ar 23 yars ol an subjct C is 24 yars ol). Figur 7 shows th xprimntal stup. Th EMG signals ar sampl at a rat of 2kHz an thn us for calculation of th WL. Th wrist forc is sampl at a rat of 2kHz an low-pass filtr at 4Hz. In this xprimnt, th subjcts ar suppos to carry out flxion/xtnsion lbow motion without any wight at first, grab a havy wight (7kg) whn thy xtn thir lbow at th scon tim, an thn manipulat th wight by han in orr to valuat th ffct of th xoskltal robot systm. Figur 8 an 9 show th xprimntal rsult of EMG signals (ch.1: mial si of bicps) uring lbow motion with an without support of th propos xoskltal robot systm, rspctivly. Th EMG lvls of bicps 3671
(a) Subjct A (a) Subjct A (b) Subjct B (b) Subjct B (c) Subjct C Fig. 8 Exprimntal rsult with support of th xoskltal robot (ch.1: mial si of bicps) (c) Subjct C Fig. 9 Exprimntal rsult without support of th xoskltal robot (ch.1: mial si of bicps) 3672
uring th havy wight manipulation ar suppos to b lowr if th propos xoskltal robot systm supports th human motion proprly. Comparing th rsults in Fig. 9 an 10, on can s that th EMG lvls of bicps uring th havy wight manipulation ar much lowr whn th human motion is support by th xoskltal robot systm. On can also s that th EMG lvls uring flxion/ xtnsion lbow motion without th wight ar almost th sam in Fig. 8 an 9. This mans that th xoskltal robot systm os not constraint th lbow motion of th subjcts. Ths rsults show that th propos xoskltal robot systm can ffctivly support th lbow motion of any subjcts using its aaptation ability. 6 Conclusions Th ffctiv aaptiv control mtho has bn propos for th xoskltal robot for human lbow motion support. Th skin surfac EMG signals of bicps an tricps an th gnrat forc by th human subjcts wrist uring th human lbow motion hav bn us for input information of th controllr of th robot as th signals implying th human subjct s intntion. In orr to control both th angl an impanc of th xoskltal robot accoring to th vagu biological signals, a fuzzynuro control mtho has bn propos. Th xprimntal rsults show th ffctivnss of th propos control mtho. Th propos xoskltal robot is also applicabl to rhabilitation of lbow motion. W woul lik to continu this stuy towar th ralization of a full motion support xoskltal robot. 7 Rfrncs [1] K.Kiguchi, S.Kariya, K.Watanab, K.Izumi an T.Fukua, Dsign of an Exoskltal Robot for Human Elbow Motion Support, Proc. of IEEE/RSJ Intrnational Confrnc on Intllignt Robots an Systms, pp.383-388, 2000. [2] K.Nagai, I.Nakanishi, H.Hanafusa, S.Kawamura, M.Makikawa an N.Tjima, Dvlopmnt of an 8 DOF Robotic Orthosis for Assisting Human Uppr Limb Motion, Proc. of IEEE Intrnational Confrnc on Robotics an Automation, pp.4386-4391, 1998. [3] W.H.Fingr an H.H.Asaa, Dsign an Control of an Activ Mattrss for Moving Brin Patints, Proc. of IEEE Intrnational Confrnc on Robotics an Automation, pp.2044-2050, 1999. [4] C.J.Abul-Haj an N.Hogan, Functional Assssmnt of Control Systms for Cybrntic Elbow Prosthss Part I an II, IEEE Trans. on Biomical Enginring, vol.37, no.11, pp.1025-1047, 1990. [5] B.Hugins, P.Parkr, an R.N.Scott, A Nw Stratgy for Multifunction Myolctric Control, IEEE Trans. on Biomical Enginring, vol.40, no.1, pp.82-94, 1993. [6] K.A.Farry, I.D.Walkr, an R.G.Baraniuk, Myolctric Tlopration of a Complx Robotic Han, IEEE Trans. on Robotics an Automation, vol.12, no.5, 1996. [7] O.Fukua, T.Tsuji, A.Ohtsuka, an M.Kanko, EMG-bas Human-Robot Intrfac for Rhabilitation Ai, Proc. of IEEE Intrnational Confrnc on Robotics an Automation, pp.3942-3947, 1998. [8] D.Nishikawa, W.Yu, H.Yokoi, an Y.Kakazu, EMG Prosthtic Han Controllr using Ral-tim Larning Mtho, Proc. of IEEE Intrnational Confrnc on Systms, Man, an Cybrntics, pp.i-153-i- 158, 1999. [9] J.A.Cozns, Robotic Assistanc of an Activ Uppr Limb Exrcis in Nuro logically Impair Patints, IEEE Trans. on Rhabilitation Enginring, vol.7, no.2, pp.254-256, 1999. [10]C.Pfiffr, K.DLaurntis, an C.Mavroiis, Shap Mmory Alloy Actuat Robot Prosthss: Initial Exprimnts, Proc. of IEEE Intrnational Confrnc on Robotics an Automation, pp.2385-2391, 1999. [11]H.Huang an C.Chn, Dvlopmnt of a Myolctric Discrimination Systm for a Multi-Dgr Prosthtic Han, Proc. of IEEE Intrnational Confrnc on Robotics an Automation, pp.2392-2397, 1999. [12]E.Park an S.G.Mk, Fatigu Compnsation of th Elctromyographic Signal for Prosthtic Control an Forc Estimation, IEEE Trans. on Biomical Enginring, vol.40, no.10, 1993. [13]T.Tsuji, Y.Kawaguchi, an M.Kanko, An Aaptiv Training Mtho for Human-Robot Systms Using Nural Ntworks, Journal of th Robotics Socity of Japan, vol.18, no.5, pp.683-389, 2000. (in Japans) [14]S.Fujii, D.Nishikawa, an H.Yokoi, Dvlopmnt of Prosthtic Han Using Aaptiv Control Mtho for Human Charactristics, Intllignt Autonomous Systms, IOS Prss, pp.360-367, 1998. [15]C.J.Abul-Haj an N.Hogan, An Emulation Tchniqu for Dvloping Improv Elbow-Prosthsis Dsigns, IEEE Trans. on Biomical Enginring, vol.bme-34, pp.724-737, 1987. [16]D.J.Bnntt, J.M.Hollrbach, Y.Xu, an I.W.Huntr, Tim-Varying Stiffnss of Human Elbow Joint During Cyclic Voluntary Movmnt, Exprimntal Brain Rsarch, vol. 88, pp.433-442, 1992. 3673