A Neurofuzzy Algorithm-Based Advanced Bilateral Controller for Telerobot Systems

Transcription

1 ICASE: he Insttute o Control, Automaton and Systems Engneers, KOREA Vol., No., March, A Neurouzzy Algorthm-Based Advanced Blateral Controller or elerobot Systems Dong-hyu Cha and Hyung Suc Cho Abstract: he advanced blateral control algorthm, whch can enlarge a relected orce by combnng orce relecton and complance control, greatly enhances worablty n teleoperaton. In ths scheme the maxmum boundares o a complance and a orce relecton gan guaranteeng stablty and good tas perormance greatly depend upon characterstcs o a slave arm, a master arm, and an envronment. hese characterstcs, however, are generally unnown n teleoperaton. It s, thereore, very dcult to determne such maxmum boundary o the gan. he paper presented a novel method or desgn o an advanced blateral. he actors aectng tas perormance and stablty n the advanced blateral were analyzed and a desgn gudelne was presented. he neurouzzy complance model (NFCM-based blateral control proposed heren s an algorthm desgned to automa tcally determne the sutable complance or a gven tas or envronment. he NFCM, composed o a uzzy logc (FLC and a rule -learnng mechansm, s used as a complance. he FLC generates complant motons accordng to contact orces. he rule-learnng mechansm, whch s based upon the renorcement learnng algorthm, trans the rule-base o the FLC untl the gven tas s done successully. Snce the scheme allows the use o large orce relecton gan, t can assure good tas perormance. Moreover, the scheme does not requre any pror nowledge on a slave arm dynamcs, a slave arm and an envronment, and thus, t can be easly appled to the control o any telerobot systems. hrough a seres o experments eectveness o the proposed algorthm has been vered. Keywords :eurouzzy, telerobot, blateral, orce relecton, complance I. Introducton A telerobot system s generally composed o a master arm controlled by a human operator and a slave arm whch duplcates the moton o the master arm and perorms actual wors n a remote ste. In a typcal telerobot system wth no orce relecton or complance control, a st slave arm strctly o l- lows the moton o a master arm. o acheve more complex tass, however, such a system may be unsutable because o nsucent normaton on the worng envronment. he orces exerted by a slave arm nteractng wth the envronment contan a lot o normaton on the teleoperaton processes or worng envronments. he use o orce normaton, thereore, can greatly mprove the tas perormance. wo major technques that utlze the contact orces are complance control and blateral control []. In complance control systems []-[], the contact orces are not relected to an operator but used or complance control o a slave arm. An operator assgns reerence trajectory to a slave arm through a master arm and the orces are ed bac to a complance. he complance, then, generates correctve motons and these are supermposed to the reerence trajectory, and as a result, a moded reerence trajectory s generated. In blateral control systems [6]-[7], the contact orces are relected to a human operator va a master arm so that he/she can correct hs/her motons accordng to these orces. Such a system s also called as a orce relecton system and the orce relecton can sgncantly enhance the tas perormance. he larger the orce relecton gan s, the bgger the relected orce becomes, however, the gan should Manuscrpt receved: May,, Accepted: Aug. 7,. Dong-hyu Cha: Dept. o Control and Measurement Engneerng, Korea Polytechnc Unversty (dhcha@pu.ac.r Hyung Suc Cho: Dept. o Mechancal Engneerng, KAIS (hscho@ lca.ast.ac.r hs wor was supported by Korea Research Foundaton Grant (KRF-99--E6. be bounded nto a certan value because o the stablty problem [8]. o enlarge the relected orce, many advanced blateral control algorthms have been p roposed by combnng the b lateral control and the complance control [8]-[]. In these schemes, a slave arm becomes more complant by adoptng complance control technque, and thereore, t s possble to enlarge the relected orce. hus, good tas perormance can be assured. In spte o the prevous studes, there stll remans unsolved problems such as: how to nd out the sutable complance o a slave arm and how to mplement t under the uncertan characterstcs o a slave arm and worng envronments, whch are nevtable n most o teleoperaton tass. In an advanced blateral control system, the complance greatly aects the tas perormance. he larger the complance s the bgger relected orce can assure. But too large complance can cause stablty problem. he maxmum boundary o the complance guaranteeng stablty greatly depends upon the orce relecton gan, characterstcs o the envronment and the slave arm. In normal practce, thereore, t s very dcult to determne such maxmum boundary o the complance. o overcome these dcultes the proposed algorthm automatcally determnes the complance n an advanced blateral. he neurouzzy complance model(nfcm- based blateral proposed heren can nds out a sutable complance or a gven tas or envronment. he NFCM, composed o a uzzy logc (FLC and a rule - learnng mechansm, s used as a complance n ths scheme. he FLC generates complant motons accordng to contact orces whle the rule-learnng mechansm, whch s based upon the renorcement learnng algorthm []-[], trans the rule -base o the FLC untl the gven tas s done successully. Snce the scheme allows the use o large orce relecton gan, t can assure good tas perormance. Moreover, the scheme dose not requre any pror nowledge on a slave

2 ransactons on Control, Automaton, and Systems Engneerng Vol., No., March, arm dynamcs, a slave arm and an envronment, and thus, t can be easly appled to the control o any telerobot systems contactng wth an envronment havng unnown stness. he eectveness o the proposed algorthm has been vered thorough a seres o experments. he paper s organzed as ollows: a telerobot system used n the study s descrbed n secton. In secton, the actors aectng tas perormance and stablty are brely dscussed to nd a desgn gudelne o an advanced blateral. he structure o the NFCM -based blateral s descrbed n secton. he expermental results and dscussons are presented n secton. Fnally, some conclusons are made n secton 6. II. Descrpton o a telerobot system Fg. shows a telerobot system developed at the Laboratory or Control Systems and Automaton n the Korea Advanced Insttute o Scence and echnology. he system conssts o a orce relectve master arm, a slave arm and a system contro l- ler. he master arm has a vertcal artculated structure wth three degrees o reedom (d.o.. hree D.C. servo motors wth harmonc drves are mounted n each axs. o reduce rcton, each motor s drectly connected to the rotatonal shat o each axs. o enhance maneuverablty, balancng weghts are attached to the second and thrd axes. hree axs -control boards are developed as torque s. he slave arm s a sxd.o. ndustral robot (Samsung, FARA A-U wth vertcal artculated structure. A sx-axes orce/torque sensor (AI, F / s attached at the end o the slave arm to measure the orce/moment. o provde vsual normaton o the contact orce, a orce level ndcator composed o 6 lght emttng dodes (LED wth a bar graph arrangement s nstalled at the operator ste. Although the human operator can eel the contact orce through the master arm, addton o vsual eedbac o orce can sgncantly enhance tas perormance durng the experments. o montor the slave envronment, a CCV camera s nstalled at the celng o the worng ste. he telerobot system mplemented by a PC86/DX-66 s composed o a master arm nematcs rou- Force level ndcator orque X m Zm Master nematcs Gravty compensaton Montor Master arm Y m X M Jacoban X Worspace R mappng _ Relected orce (F r X C elerobot system X R Slave nverse nematcs Complance Force relecton gan (K Slave arm Vdeo camera Xs Z s Slave arm Y s Contact orce (Fs Fg.. Structure o the telerobot system. F/ sensor tne, a worspace mappng routne, a slave arm nverse nematcs routne, a complance and a orce rele cton. Snce the master and slave arms have n ematcally derent structures, worspace mappng s requred so that a poston o the master arm, X M, can be approprately transormed to the correspondng space o the slave arm, X R. he detaled algorthm s smlar to that descrbed n []. he detaled structures o the complance and the orce relecton are descrbed n the ollowng secton. he master and slave s are connected wth a man PC through the PC-A bus va two parallel I/O boards. III. Factors aectng tas perormance and stablty Fg. shows a typcal bloc dagram o an advanced blateral. he conssts o a complance contro ller and a orce relecton gan. he complance consttutes an nner eedbac loop, whle the orce relecton loop orms an outer eedbac loop. When contact orce occurs, the complance receves ths orce as nput and generates correspondng correctve moton, X C, as an output. hs correctve moton s supermposed on the master arm trajectory, X R, and as a result, a moded reerence trajectory, X R s generated. At the same tme, the contact orce s also relected bac to the human operator through the master arm ater scalng by a orce relecton gan,. Human operator command ( F h F m Relected orce ( F r Master & human hand Force relecton gan X R K H X R Xc Slave arm Complance Contact orce ( F s Actual slave poston ( X s Z e Envron -ment Fg.. A typcal bloc dagram o the advanced blateral control.. Eects o the complance he eect o the complance can be seen to decrease the eectv e stness o a slave arm. he hgher the gan o the complance s, the lower the eectve stness o the slave arm becomes. However, the gan o the complance should be lmted because too large gan maes the system unstable. hs eect can be shown concsely as n the ollowngs: generally, the slave arm, envronments and the complance can be consdered as nonlnear systems. I we use the small gan theorem [] or stablty o the complance loop, whch orms a nner eedbac loop, the sucent condton can be obtaned by [ S( X R e ] γ [ H ( Z ( X ] S ( X R γ ( X γ S γ,, X X R S L L where S (, H( and Z e ( denote unctons or slave, complance controlller and envronments, respectvely, whle denotes the uncton norm and ( denotes the proje c- ton operator. he above equatons mean that the system s BIBO e e (

3 ICASE: he Insttute o Control, Automaton and Systems Engneers, KOREA Vol., No., March, (Bounded Input Bounded Output stable the loop gan s less than unty [6]. o show the eect o the complance contro l- ler more clearly, let the DC gans o S, H andze be denoted by S, H andze, respectvely, and assume that these gans are remaned unchanged durng the teleoperaton. hen, the DC loop gan, G DC should be less than unty or the system to be stable,.e., Namely, GDC = H Ze S ( e H ( Z S From these equatons, we can see that a complance contro l- ler must be determned consderng the characterstcs o the slave arm and the envronments, and small complance would be better to ensure stablty. Also, note that the equaton ( does not contan any terms about a orce relecton gan or a master arm. It means that we can desgn a complance contro l- ler wthout consderng or constructng a orce relecton loop.. Eects o the orce relecton gan and the complance Fg. shows a smpled bloc dagram o an advanced blateral control system, where Ψ denotes a complance loop,.e., a resultant uncton o the slave, the envronment and the complance. In ths system, the contact orces measured by the orce/torque sensor are relected bac to the human operator ater scalng by a orce relecton gan,. hs orms a closed loop system and causes a trade-o between tas perormance and stablty. o concsely nvestgate the eect o the complance, let us ollowng the smlar procedure o the prevous secton. Let M ( denote a uncton or master ncludng human hand, and the DC gans o M and Ψ be denoted M Ψ by and, respectvely. hen, the DC loop gan, G DC should be less than unty or the system to be stable,.e., Human operator command ( F h F m Master & X R _ human hand Relected orce ( F r Force relecton gan K Contact orce ( F s Fg.. A smpled bloc dagram o the advanced blateral Control. Ψ GDC = Ψ M ( Ψ Consderng denotes the DC gan o a complance loop, t can be represented as ollows: S Z e Ψ = ( S Ze H Substtutng ths nto equaton ( yelds S Z e M GDC = (6 S Z H e he above equaton can be represented as S Z H e (7 S Ze M hese equatons show that to ensure stablty the orce relecton gan must be lmted nto certan boundary and determned under the consderaton o characterstcs o the complance, master, the slave, and the envronment. Large gan,, can realze bg relected orce, but should be lmted to satsy the stable condton o equaton (7. hereore, we can see that there s a trade-o between stablty and tas perormance. From the above results we can get a useul desgn gudelne o an advanced blateral as ollows: Step : desgn a complance whch satses the stable condton o equaton (. Note that the equaton does not contan any terms about a orce relecton gan or a master arm. It s only related wth a slave arm and an envronment. hs means that we can desgn a complance wthout consderng or constructng a orce relecton loop. Step : desgn a orce relecton gan whch satses the stable condton n equaton (7. I we desgn a complance as large as possble n the step, the upper bound o can be maxmzed, whch allows the use o the largest. IV. he Neurouzzy complance (NFCC- based control algorthm. Structure o the NFCM-based control Fg. shows overall structure o the proposed scheme. In ths scheme the NFCM, whch generates correctve moton nstead o a conventonal complance, automatcally determnes the sutable complance or a gven tas or envronment. As shown n the gure, the NFCM conssts o two parts. he rst one s a uzzy logc (FLC whch receves contact orce normaton as nputs and generates correctve motons as outputs. he second one s a rule learnng mechansm whch s composed o two neuron-le elements: an assocatve crtc neuron(acn and an assocatve learnng neuron(aln. he mechansm realzes the renorcement learnng algorthm, and thus, trans rule base o the FLC teratvely untl the gven tas can be successully perormed so that no urther changes n rules are necessary. When the tranng has been completed the NFCM can realze sutable complance or a gven tas. he alure detector receves contact orces and generates an external renorcement sgnal whch s used n the rule learnng mechansm. he contact orce s also relected bac to the human operator through the master arm ater scalng by a orce relecton gan,. he scheme does not requre any pror nowledge on the slave arm dynamcs, slave arm and the envronment. hus, t can be easly used to the complance control o a slave arm.. Constructon o the complance Desgn o the uzzy logc he FLC conssts our modules: a uzzy decoder, a rule base, a uzzy reasonng and a deuzzer. Beore constructng the FLC, let us dene the coordnates or orce and correctve

4 ransactons on Control, Automaton, and Systems Engneerng Vol., No., March, Operator command ( Fh _ F X X m R R Master & human hand _ X c Relected orce ( Fr Force relecton gan K Deuzzer Slave arm Fuzzy Reasonng Rule Base ALN ACN NFCM Fuzzy Decoder Falure detector Actual slave poston ( X S Envron Z e -ment Contact orce ( FS Fg.. he bloc dagram o the proposed algorthm. moton wth respect to the end eector coordnates, { O e }, where ts orgn pont, O e, s n the locaton o tool center pont as shown n Fg.. hs enables us to derve the FLC wthout regardng the geometry o worng envronment. he nput and output varables are contact orces, absolute values o change o contact orces and correctve motons. hey are dened by ( t = [ ( t = x ( t, ( t, ( t, m ( t, m = [ ( t, ( t, ( t, x ( t ( t y y z z x x ( t, y y ( t, m ( t] ( t, c( t = [ c ( t, c ( t, c ( t, θ ( t, θ ( t, θ = [ c ( t, c ( t, c ( t, c ( t, c ( t, c ( t] 6 z 6 z ( t] ( t] where, m, c θ and (=x,y,z denote the orce, moment, translaton and rotaton n -drecton wth respect to the end eector coordnates, { O e }, respectvely, whle t denotes t-th samplng tme. X s Z s Y s X e Z e O e Y e Fg.. Denton o an end eector coordnate { O e }. a Fuzzy decoder : he uzzy decoder nspects the ncomng system state and actvates the correspondng rules n para llel. hese unctons are accomplshed through uzzcaton. From now on, uzzy varables are dscrmnated rom ther crsp varables by addng the tlde sgn( above the varables. hen, the nput uzzy varables are obtaned by scalng rom crsp nputs and they are as ollows = g ( t = g ( t c = g c ( t o ( =,,..,6 where g, g, g o denote scalng actors or,, o, respectvely. b Rule base : In ths study, the output varables o the (8 (9 uzzy rule taes the orm o crsp value, whch decreases the calculaton tme consderably, and thus, protable n real tme applcaton. he rule base taes the orm: RuleR : IF s F, s F, s F, s F, s F, 6 s F6, s F, o so HEN c s C ( =,,.., 6 ( =,,.., N ( where the subscrpt and the superscrpt denote the -th drecton and the -th rule, respectvely. Also, F, F, O and C are uzzy subsets correspondng to the varables,, o and c, respectvely. Note that C s not a uzzy value but a crsp one. In the rst, the value o C s chosen arbtrarly and s learned usng the rule learnng mechansm descrbed later. c Fuzzy reasonng : Fuzzy reasonng s perormed usng the Mn -Max operaton. Let µ ( c be the membershp uncton or a subset o the output whch s the result o the -th C rule. hen t can be obtaned by µ c =[ ( ( ( ( ( o µ µ µ µ µ C ( F F F 6 6 F O µ C ( c ] ( where denotes Mn operaton. From the results o the N rules, the nal membershp uncton µ ( c C can be obtaned by µ c = [ µ ( c µ ( c ( µ c ] ( ( N C C C C where denotes Max operaton. d Deuzzer : In the proposed control scheme, a crsp correctve moton s requred. he deuzzer yelds crsp correctve moton rom the nerred uzzy correctve moton. he crsp correctve moton, c, can be obtaned by c = Deuzy{ µ ( c } ( g c where g c are scalng actors or c and Deuzy{ } denotes the deuzzer operator. Rule learnng mechansm he objectve o rule learnng s to learn the output lngustc values o C n the rule base. Such a learnng capablty s perormed by the rule learnng mechansm, whch conssts o two neuron-le elements: the assocatve crtc neuron and the assocatve learnng neuron. he rule learnng mechansm s based upon the renorcement learnng algorthm []-[], and the detaled algorthm s smlar to that descrbed n []. Falure detector he alure detector receves contact orces as nputs and generates an external renorcement sgnal as an output. When a gven tas s perormed successvely, the external Z s Slave arm Y s C F/ sensor reerence trajectory wall base plate Fg. 6. A wall contactng tas

5 ICASE: he Insttute o Control, Automaton and Systems Engneers, KOREA Vol., No., March, renorcement sgnal has a value more than zero, and n the opposte case, ts value s less then zero. Other sensory no r- maton except contact orce also may be used to generate the renorcement sgnal. he objectve o ths detector s to determne whether the tas s perormed successvely or not. hereore, constructon o a alure detector s rather heurstc and completely depends on a gven tas.. Constructon o the orce relecton he relected orce s evaluated by multplyng the contact orce by the orce relecton gan, as shown n Fg.. he resultant s then converted to the torque appled at each axs o master arm through the transpose o the master Jacoban. he orce relecton gan should satsy the stable condton n equaton (7. V. Experments and Results o nvestgate eectveness o the proposed scheme, a seres o experments has been carred out usng the telerobot system shown n Fg.. A wall contactng tas shown n Fg. 6 was perormed as an example tas. A steel peg was attached at the end o the slave arm, and a plastc wall was located n the lower sde. he reerence trajectory was assgned through the master arm as shown n the gure. Durng the experments the drecton o the end eector s xed to a constant (-Z S drecton wth respect to the global reerence coordnates. Among the contact orces, the three dmensonal moments are small compared to the three dmensonal orce. In ths study, thereore, or smplcty the proposed control scheme s mplemented or the three drectonal orces only.. Constructon o an NFCM Inputs or the NFCM are the three drectonal orces,, the change n the orces,, and one-step-ahead correctve motons, o, whle the outputs are the correctve motons, c, ( = x, y, z. he rule base s as ollows : Rule R : IF s F, s G, s G, s F, o s O HEN c s C ( R : IF s G, s F, s G, s F, o s O HEN c s C R : IF s G, s F, s G, s F, o s O HEN c s C ( =,.., where denotes the -th rule. he nput lngustc values are dened as shown n Fg. 7. able shows the scalng actors able. he scalng actors or the nput and output varables. g g g g g g g o g o g o g c g c g c g, g, go and gc or. Each set o rules, R, R and s composed o ( rules. R, he goal o the experment s to mae the slave arm as complant as possble so as to maxmze the orce relecton gan. hus, a alure detector s desgned to have three rules as ollows Rule : IF Rule : ELSE Rule : ELSE IF (t > (t - (t- > del success max where (t s dened as ollows max HEN HEN al al ( (t = (6 where max = N and del max = N, whle the value or success =. and al = -.. Rule checs whether the learnng process s proceedng to mnmze the contact orce, and Rule checs whether the contact state s stable or not.. Learnng o the NFCM Snce we can desgn a complance wthout consderng or constructng orce relecton, the contact orce was not relected to the human operator durng the learnng pro cess. In other words, only the complance was used. Experments were conducted n the ollowng manner: a human operator perormed the wall contactng tas many tmes. Du r- ng the operaton three trajectores shown n Fg. 8 were stored n the computer, and these were subsequently assgned to the slave arm as reerences n the order o (-(-(-(-(-(. µ G. ( -.. µ F ( NB PB. (b Lngustc values o G (a Lngustc values o F µo (o ( µ F. -.. (d Lngustc values o O SM.. BG (c Lngustc values o F Fg. 7. Denton o the uzzy lngustc values. A learnng operaton conssts o a tranng tas and an evaluaton tas. At the rst operaton, a tas ollowng the trajectory ( s perormed as a tranng tas. Durng ths tas the contact orce occurs as the peg contacts the wall. he uzzy logc, then, generated correctve moton. At the same tme the rule learnng mechansm s actvated, and thus the output lngustc values, o C, are traned. Ater ths, a

6 ransactons on Control, Automaton, and Systems Engneerng Vol., No., March, Y s (cm able. Changes o the rule base or x-axs. (n the case o SM, and = = = X s (cm ( ( ( C NL F PL O C NL F PL O Contact orce (, N Contact orce (, N Contact orce (, N Z s (cm 6 ( ( ( Y s (cm Fg. 8. Varous reerence trajectores me (t, sec (a Intal rule base (th operaton me (t, sec (b raned rule base (rd operaton me (t, sec (c raned rule base (9th operaton Fg. 9. Z-drectonal contact orces and correspondng correctve motons. tas o ollowng trajectory ( s perormed as an evaluaton tas, to evaluate the eect o rule learnng. Durng the evaluaton tas, the rule learnng mechansm s not actvated, and thus the rules are not changed. At the second and thrd operatons, trajectores ( and ( are assgned as reerences durng the tranng tass, respectvely. At every operaton, the evaluaton tas wth trajectory ( s ollowed by the tranng tas. At the orth operaton, trajectory ( s used agan as a reerence and or the next, trajectores (, (,.. are successvely assgned. hese operatons were repeated untl no urther changes occur n the rule base. c z z c z z c z z Correctve moton (c, mm Correctve moton (c, mm Correctve moton (c, mm he z-drectonal results o the evaluatng tass or the thrd and nnth operatons are shown n Fg. 9. In the gure, the (zero th evaluaton tas represents the results wth the ntal rule base. Fg. shows the maxmum orces generated durng each evaluaton tas, and able (a-(c shows change o the rule bases or x-axs n the case o = SM, = and =. From the table we can see that the rule base s changed as the learnng goes on. Changes o the rule bases or the y and z-axes and other cases on the x-axs are smlar, and thus are not presented here. A careul analyss o these results reveals that, n the early stages, the lngustc values ncrease wth the number o operatons. hereo re, the correctve moton also ncreases. hs yelds a decrease n the maxmum orce. However, all o these become statonary as the number o operatons becomes large. hus, t can be seen that the NFCM can generate sutable complance wth the ad o the rule-learnng mechansm. Another set o experment was perormed usng a steel wall whch s much ster than the plastc one, whle ts d mensons and locaton are ept the same as the above experments. he other condtons were also set dentcally to the above experments. he results are shown n Fg. and able (d. From the results we can see that the learnng process shows a smlar tendency to that o the prevous one: as the operatons goes on, Maxmum contact orce (N C (a Intal rule base NL F PL O (c 9th operaton: plastc wall (b rd operaton: plastc wall C NL F PL : steel : plastc Number o operaton O (d 9th operaton: steel wall Fg.. Change o maxmum contact orces n z-drecton.

7 6 ICASE: he Insttute o Control, Automaton and Systems Engneers, KOREA Vol., No., March, Contact orce (, N Contact orce (, N me (t, sec (a NFCM-based blateral control me (t, sec (b Km s advanced blateral control Fg.. Results o the wall contactng experments. the lngustc values and the correctve moton ncrease, and contact orce decreases. All o these also become statonary as the number o operaton becomes large. In case o steel, however, the contact orce gets larger durng all operatons and the nal lngustc values are smaller than those o the plastc one. hs means that the NFCM can realze smaller complance than that n case o plastc. Snce all the expermental condtons except the materal o the wall are the same as those o the prevous experment, t can be concluded that such de r- ences arse rom the derence n the envronment. he act, whch the NFCM can realze larger complance n plastc wall than that n steel wall, comples wth Wabel's result [7], where t s reported that n order to guarantee stablty or a gven robot poston, the gans o a complance must be decreased as the stness o the envronment ncreases. In other words, or a more complant envronment, a complance can produce more complance, and thus, can acheve a smaller contact orce. hus t can be seen that the NFCM s capable o provdng sutable complance, correspondng to the varous types o envronment.. NFCM-based advanced blateral teleoperaton An NFCM -based blateral was constructed usng the learned NFCM and the orce relecton descrbed n secton.. he orce relecton gan was expermentally set to., wh ch was the maxmum allowable gan n the stable regon. he tas was to mantan a constant contact orce o N n z-drecton to the plastc wall shown n Fg. 8, and the results are shown n Fg. (a. From the gure t can be seen that the contact occurred at about.7 sec. Note that ater the contact, the complance generated correctve moton accordng to the contact orce, and at the same tme the orce was also ed bac to the human operator. Ater. sec contact orces were very close to the desred value, N. Snce t was controlled by a human operator, the results dd not show exact N. o compare the results wth those o another algorthm, the same tas was conducted usng Km's algorthm [8]. he gans o the complance was tuned so as to mae the slave arm as complant as possble. he orce rele cton gan was expermentally set to., whch was also the maxmum allowable gan n the stable regon. From the results shown n Fg. (b t can be seen that the contact orce s a lttle oscllatory, and becomes about N ater sec. From the above results, t can be seen that the proposed algorthm shows better tas perormance. VI. Conclusons he paper presented a novel method or desgn o an advanced blateral. he actors aectng tas perormance and stablty n the advanced blateral were analyzed and a desgn gudelne was presented. he neurouzzy complance model (NFCM-based blateral control proposed heren s an algorthm desgned to automatcally determne the sutable complance or a gven tas or envronment. he NFCM, composed o a uzzy logc (FLC and a rule-learnng mechansm, s used as a complance. he FLC generates complant motons accordng to contact orces. he rule-learnng mechansm, whch s based upon the renorcement learnng algorthm, trans the rule-base o the FLC untl the gven tas s done successully. Snce the scheme allows the use o large orce relecton gan, t can ensure good tas perormance. Moreover, the scheme does not requre any pror nowledge on a slave arm dynamcs, a slave arm and an envronment, and thus, t can be easly appled to the control o any telerobot systems. he eectveness o the proposed algorthm was vered through a seres o experment. In ths study automatc selecton o orce relecton gan s not dealt wth. Development o an algorthm that can automatcally determne both a complance and a orce relecton gan would assure better tas perormance and s now under nvestgatng. Reerences [] D. H. Cha and H. S. Cho, Neural networ and uzzy logc systems: echnques and applcatons n telerobot systems, n Fuzzy heory Systems echnques and Applcatons, Ed. C.. Leondes, Academc Press, pp. 9-79, 999 [] A. K. Bejczy and W. S. Km, Predctve dsplays and shared complance control or tme delayed telemanpulaton, Proc. IEEE/RSJ Int. Worshop on Intellgent Robots and Systems, pp. 7-, 99. [] W. S. Km, Shared complant control: A stablty analyss and experments, Proc. IEEE Int. Con. Robotcs and Automaton, pp. 6-6, 99. [] K. Kosuge, A. Sato and K. Furuta, as-orented control o master-slave manpulators, Proc. Japan-U.S.A. Sym.

8 ransactons on Control, Automaton, and Systems Engneerng Vol., No., March, 7 on Flexble Automaton, pp. 87-9, 99. [] D. H. Cha and H. S. Cho, A neurouzzy model-based complance wth applcaton to a telerobot system, Control Engneerng Practce, vol., no., pp. 9-, 996. [6] B. Hannaord and R. Anderson, Expermental and smulaton studes o hard contact n orce relectng teleoperaton, IEEE Int. Con. Robotcs and Automaton, pp. 8-89, 988. [7] G. J. Raju, G. C. Verghese and. B. Sherdan, Desgn ssues n -port networ models o blateral remote manpulaton, Proc. IEEE Int. Con. Robotcs and Automaton, pp. 6-, 989. [8] W. S. Km, Developments o new orce relectng control schemes and an applcaton to a teleoperator tranng smulator, Proc. IEEE Int. Con. Robotcs and Automaton, pp. -9, 99. [9] B. Hannaord, Stablty and perormance tradeos n blateral telemanpulaton, IEEE Int. Con. Robotcs and Automaton, pp , 989. [] A. A. Goldenberg, and D. Bastas, On the blateral control o orce relectng teleoperaton, Proc. th IFAC World Congress, pp. -, 99. [] D. H. Cha and H. S. Cho, Desgn o a neurouzzy algorthm-based shared or telerobot systems, Robotca, Vol., No., pp. -, 997. [] A. G. Barto, R. S. Sutton and C. W. Anderson, Neuronle adaptve elements that can solve dcult learnng control problems, IEEE rans. Systems, Man and Cybernetcs, vol., no., pp. 8-86, 98. [] C. C. Lee, A sel-learnng rule -based wth approxmate re asonng and neural nets, Proc. IFAC '9 World Congress, vol. 7, pp. 9-6, 99. [] D. H. Cha, Shared control o telerobot systems usng artcal ntellgence algorthms, Ph. D. Dssertaton, Korea Advanced Insttute o Scence & echnology, 99. [] C. A. Desore and M. Vdyasagar, Feed-bac systems: Input-output propertes, New Yor: Academc Press, 97. [6] F. L. Lews, C.. Abdallah and D. M. Dawson, Control o robot manpulators, Macmllan Publshng, 99. [7] B. J. Wabel and H. Kazeroon, heory and experments on the stablty o robot complance control, IEEE rans. Robtotcs and Automaton, vol. 7, no., pp. 9-, 99. Dong-hyu Cha He receved the B.S. degree n Mechancal desgn rom Seoul Natonal Unversty n 98, and the M. S. and Ph. D. degrees n Mechancal Engneerng rom KAIS n 986 and 99, respectvely. From to he was a vstng proessor at Unversty o Pennsylvana. Snce 998 he has been an Assstant Proessor o Control and Measurement Engneerng n Korea Polytechnc Un versty. Hs elds o research are ntellgent robotcs, control and automaton o manuacturng processes. Hyung Suc Cho He receved the B.S. degree n ndustral educaton (Mechancal Engneerng rom Seoul Natonal Unversty, Seoul, Korea, the M.S. degree n mechancal engneerng rom Northwestern Unversty, Evanston, IL, and the Ph.D. degree n mechancal engneerng rom the Un - versty o Calo rna at Bereley, CA, n 97, 97, and 977, respectvely. Hs elds o research are ntellgent control and robotcs, machne vson and ts applcaton, and manuacturng process montorng and control. Dr. Cho s a Fellow o Korea Academy o Scence and echnology, Korea and also a member o ASME, IEEE, SPIE, SME, KAS, KSPE, and KSME.