PATH PLANNER FOR OBJECTS, ROBOTS AND MANNEQUINS BY MULTI-AGENTS SYSTEMS OR MOTION CAPTURES

Transcription

1 Author mnusript, publishe in "Interntionl Conferene on Digitl Enterprise ehnology, Proeeings Nntes : Frne of DE2008 (2008)" 5th Interntionl Conferene on Digitl Enterprise ehnology Nntes, Frne Otober 2008 PAH PLANNER FOR OBJECS, ROBOS AND MANNEQUINS BY MULI-AGENS SYSEMS OR MOION CAPURES D. Chblt Institut e Reherhe en Communitions et Cybernétique e Nntes (UMR CNRS 6597) Éole Centrle e Nntes,, rue e l Noë, 4432 Nntes Dmien.Chblt@iryn.e-nntes.fr hl , version 2-22 Ot 2008 ABSRAC In orer to optimise the osts n time of esign of the new prouts while improving their qulity, onurrent engineering is bse on the igitl moel of these prouts. However, in orer to be ble to voi efinitively physil moel without loss of informtion, new tools must be vilble. Espeilly, tool mking it possible to hek simply n quikly the mintinbility of omplex mehnil sets using the numeril moel is neessry. Sine one ee, the MCM tem of IRCCyN works on the retion of tools for the genertion n the nlysis of trjetories of virtul mnnequins. he simultion of humn tsks n be rrie out either by robot-like simultion or by simultion by motion pture. his pper presents some results on the both two methos. he first metho is bse on multi-gent system n on igitl mok-up tehnology, to ssess n effiient pth plnner for mnikin or robot for ess n visibility tsk tking into ount ergonomi onstrints or joint limits. he humn opertor is integrte in the proess optimistion to ontribute to globl pereption of the environment. his opertor oopertes, in rel-time, with severl utomti lol elementry gents. In the seon metho, we worke with the CEA n EADS/CCR to solve the onstrints relte to the evolution of humn virtul in its environment on the bsis of t resulting from motion pture system. An pproh using of the virtul guies ws evelope to llow to the user the reliztion of preise trjetory in bsene of fore feebk. KEYWORDS Multi-gent system, Ergonomis, Motion plnner, Motion pture, Virtul guie.. INRODUCION In n inustril environment, the ess to shrble n globl view of the enterprise projet, prout, n/or servie ppers to be key ftor of suess. It improves the triptyh ely-qulity-ost but lso the ommunition between the ifferent prtners n their implition in the projet. For these resons, the igitl mok-up (DMU) n its funtions re eeply investigte by inustrils. Bse on omputer tehnology n virtul relity, the DMU onsists in pltform of visuliztion n simultion tht n over ifferent proesses n res uring the prout lifeyle suh s prout esign, inustriliztion, proution, mintenne, reyling n/or ustomer support. he igitl moel enbles the erlier ientifition of possible issues n better unerstning of the proesses even, n mybe bove ll, for tors who re not speilists. hus, igitl moel llows eiing before expensive physil prototypes hve been built. Even if evient progresses were notie n pplie in the omin of DMUs, signifint progresses re still wite for plement in n inustril ontext. As mtter of ft, the igitl moel offers wy to explore res suh s mintenne or ergonomis of the prout tht were tritionlly ignore t the beginning phses of projet; new proesses must onsequently be

2 hl , version 2-22 Ot 2008 evelope. hrough the integrtion of mnikin or robot in virtul environment, the suitbility of prout, its shpe n funtions n be ssesse. In the sme time, it beomes possible to settle the proess for ssembling with robot the ifferent omponents of the prout. Moreover, when simulting tsk tht shoul be performe by n opertor with virtul mnikin moel, fesibility, ess n visibility n be heke. he onitions of the performnes in terms of efforts, onstrints n omfort n lso be nlyse. Moifitions on the proess, on the prout or on the tsk itself my follow but lso better n erlier trining of the opertors to enhne their performnes in the rel environment. Moreover, suh use of the DMU les to better onformne to helth n sfety stnrs, to mximiztion of humn omfort n sfety n n optimistion of the robot bilities. With virtul relity tools suh s 3D mnipultors, it is possible to mnipulte the objet s esily s in rel to mnipulte the objet s esily s in rel environment. Some rwbks re the iffiulty to mnipulte the objet with s ese s in rel environment, ue to the lk of kinemtis onstrints n the utomti ollision voine. As mtter of ft, interferene etetion between prts is often isplye through olor hnges of prts in ollision but ollision is not voie. Another pproh onsists in integrting utomti funtionlity into the virtul environment in orer to ese the user s tsk. Mny reserh topis in the frmework of robotis eling with the efinition of ollision-free trjetories for soli objets re lso vli in the DMU. Some methoologies nee globl pereption of the environment, like (i) visibility grphs propose by Lozno-Pérez n Wesley [], (ii) geoesi grphs propose by ournssou [2], or (iii) Voronoï s igrms [3]. However, these tehniques re very CPU onsuming but le to solution if it exists. Some other methoologies onsier the moves of the objet only in its lose or lol environment. he suess of these methos is not gurntee ue to the existene of lol minim. A speifi metho ws propose by Khtib [4] n enhne by Brrqun n Ltombe [5]. In this metho, Khtib's potentils metho is ouple with n optimiztion metho tht minimizes the istne to the trget n vois ollisions. All these tehniques re limite, either by the omputtion ost, or the existene of lol minim s expline by Nmgung [6]. For these resons esigner, is require in orer to vlite one of the ifferent pths foun or to voi lol minim. he essibility n the optimum plement of n opertor to perform tsk is lso mtter of pth plnning tht we propose to solve with DMU. In orer to shorten time for trjetory serh, to voi lol minim n to suppress tiresome on-line mnipultion, we inten to settle for mixe pproh of the bove presente methoologies. hus, we use lol lgorithm bilities n globl view bility of humn opertor, with the sme pproh s [7]. Among the lol lgorithms, we present these ones ontributing to better visibility of the tsk, in term of ess but lso in term of omfort. 2. PAH PLANNING AND MULI-AGEN ARCHIECURE he bove hpter points out the lol bilities of severl pth plnners. Furthermore, humn globl vision n le to oherent prtition of the pth plnning issue. We inten to mnge simultneously these lol n globl bilities by builing n intertion between humn n lgorithms in orer to hve n effiient pth plnner [8] for mnikin or robot with respet of ergonomi onstrints or joints n mehnil limits of the robot. 2. HISORY Severl stuies bout o-opertion between lgorithm proesses n humn opertors hve shown the gret potentil of o-opertion between gents. First onepts were propose by Ferber [9]. hese stuies le to the retion of Conurrent Engineering methoology bse on network priniples, interting with ells or moules tht represent skills, rules or workgroups. Suh stuies n be linke to work one by Arn n Pelletier [0] for the esign of ognition bse multi-gent rhiteture. his work presents multi-gent rhiteture with humn n soiety behviour. It uses ognitive psyhology results within oopertive humn n omputer system. All these stuies show the importnt potentil of multi-gent systems (MAS). Consequently, we built positioning mnikin, bse on MAS, tht ombines humn intertive integrtion n lgorithms. 2.2 CHOICE OF HE MULI-AGEN ARCHIECURE Severl workgroups hve estblishe rules for the efinition of the gents n their intertions, even for ynmi rhitetures oring to the environment evolution [9, ]. From these nlyses, we keep the following points for n elementry gent efinition. An elementry gent: (i) is ble to t in ommon environment, (ii) is riven by set of tenenies (gol, stisftion funtion, et.), (iii) hs its own resoures, (iv) n see lolly its environment, (v) hs prtil representtion of the

3 hl , version 2-22 Ot 2008 environment, (vi) hs some skills n offers some servies, n (vii) hs behvior in orer to stisfy its gol, tking into ount its resoures n bilities, oring to its environment nlysis n to the informtion it reeives. he points bove show tht iret ommunitions between gents re not onsiere. In ft, our rhiteture implies tht eh gent ts on its set of vribles from the environment oring to its gol. Our Multi Agent System (MAS) will be blkbor-bse rhiteture. 2.3 PAH PLANNING AND MAS he metho use in utomti pth plnners is shemtise Figure. A humn globl vision n le to oherent prtition of the min trjetory s suggeste in [2]. Consequently, nother metho is the integrtion of n opertor to mnge the evolution of the vribles, tking into ount his or her globl pereption of the environment (Figure b). o enhne pth plnning, ouple pproh using multi-gent n istribute priniples s it is efine in [8] n be buil; this pproh mnges simultneously the two, lol n globl, bilities s suggeste Figure. he virtul site enbles grphi visuliztion of the tbse for the humn opertor, n ommunites positions of the virtul objets to externl proesses. informtion from the environment virtul site lgorithm ontribution () lgorithm lone Informtion n pereption from the environment pereption of the environment virtul site lgorithm opertor virtul site opertor (b) opertor lone ontributions ontribution () o-opertion Figure. Co-opertion priniples. As mtter of ft, this lst sheme is lerly orrelte with the blkbor bse MAS rhiteture. his priniple is esribe in [9, 3, ]. A shemti presenttion is presente on Figure 2. he only meium between gents is the ommon tbse of the virtul relity environment. he humn opertor n be onsiere s n elementry gent for the system, o-operting with some other elementry gents tht re simple lgorithms. Blkbor Shre Dt Agent Agent 2 Agent n Figure 2. Blkbor priniple with o-operting gents. 2.4 CONSIDERED APPROACH he pproh we retine is the one propose in [7] whose purpose ws to vlite new CAD/CAM solutions bse on istribute pproh using virtul relity environment. his metho hs suessfully shown its vntge by emonstrting in relisti time the ssembly tsk of severl omponents with mnikin. Suh problem ws previously solve by using rel n physil mokups. We kept the sme rhiteture n evelope some elementry gents for the mnikin (Figure 3). In ft, eh gent n be reursively ivie in elementry gents. Virtul Environment Shre t / mnipulte objet Collision Attrtion Opertor Shre t Mnikin Robot Mnikin shre t Robot shre t Repulsion Kinemtis Collision Attrtion Opertor Collision Attrtion Mehnil Opertor Shre t Ergonomis Shre t onstrints Shre t Shre t Repulsion Kinemtis Repulsion Kinemtis Altitue Vertility Vision Joint Dynmi Vision limits limits Figure 3. Co-operting gents n pth plnning tivity [7]. Eh gent i ts with speifi time smpling whih is pre-efine by speifi rte of tivity λi. When ting, the gent sens ontribution, normlize by vlue Δ i, to the environment n/or the mnipulte objet (the mnikin in our stuy). In Figure 4, we represent the Collision gent with rte of tivity equl to, the Attrtion gent hs rte of 3 n Opertor n Mnikin gents rte of 9. his perioiity of the gent tions is hrteristi of the rhiteture: it expresses priority between eh of the gols of the gents. o supervise eh gent tivity, we use n event progrmming metho where the min proess ollets gent ontributions n uptes the tbse [7]. he normliztion of the tions of the gents (the vlues Δi) inues tht the tions re reltive n not bsolute. 2.5 EXAMPLES he former metho is illustrte with two ifferent exmples. he first one (Figure 4) uses the MAS for testing the bility of mnikin for mounting n oxygen bottle insie n irplne okpit through trp. During the pth plnning proess, the opertor hs te in orer to rive the oxygen bottle towr the mile of the trp. he other gents hve te in orer to voi ollisions n to ttrt the oxygen

4 hl , version 2-22 Ot 2008 bottle towr the finl lotion. he rel time urtion is pproximtely 30s. he number of egrees of freeom is equl to 23. he seon exmple (Figure 5) is relte to the utomti mnipultion of robot whih bse is ttrte towr wll. he joints re mnge by the gents in orer to voi ollision n to solve the ssoite inverse kinemti moel. Figure 4 rjetory pth plnning of mnikin using the MAS Figure 5 rjetory pth plnning of robot using the MAS 3. VISIBILIY AND MAINAINABILIY CHECK WIH MULI-AGEN SYSEM IN VIRUAL REALIY 3. INRODUCION For the visibility hek, we hve fouse our ttention on the trunk n the he onfigurtions of mnikin (resp. the en-effetor n the film mer orienttion of robot). he joint between the he n the trunk is hrterize by three rottions α b, β b n θ b whose rnge limits re efine by ergonomi onstrints (Figure 6) (resp. the joint limits of the robot) Eyes Eyes Lens Figure 6. Exmple of joint limits n visibility pity of mnikin n of film mer. hese t n be foun using the results of ergonomi reserh [4]. o solve the problem of visibility, we efine one C whose vertex is entere between the two eyes (resp. the enter of the film mer) n whose bse is lote in the plne orthogonl to u, entere on the trget (Figure 7). he one with ε is vrible. hus, itionlly to the position n orienttion vribles of ll prts in the luttere environment (inluing the mnikin itself), we onsier in prtiulr: hree egrees of freeom for the mnikin (resp. the robot) to move it in the x-y plne: x m = (x m, y m, θ m,). It is lso possible to tke into ount egree of freeom z m if we wnt to give to the mnikin (resp. the robot) the pity to ler n obstle. hree egrees of freeom for the he joint (resp. wrist joints) to mnge the mnikin (resp. robot) vision: q b = (α b, β b, θ b ) t with their orresponing joint onstrints. he normlize ontributions from the gents re efine with two fixe prmeters: Δ pos for trnslting moves n Δ or for rotting moves. 3.2 AGENS ENSURING VISIBILIY ACCESS AND COMFOR FOR MANIKIN, AND VISIBILIY ACCESS FOR ROBO We present below ll the elementry gents use in our system to solve the ess n visibility tsk. Attrtion gent for the mnikin (resp. the robot) he gol of the ttrtion gent is to enble the mnikin (resp. the robot) to reh the trget with the best trunk posture (resp. the best bse plement of the robot), tht is: o orient the projetion of y m on the floor plne olliner to the projetion of u on the sme plne by rottion of θ m (Figure 7), o position x m n y m, oorintes of the mnikin (resp. the robot) in the environment floor, s lose s possible to the trget position (Figure 7), n for the robot. q up to q n using the inverse kinemti moel. his lst gent ts in orer to keep the robot posture, s muh s possible, in the sme spet, or posture, or onfigurtion s efine in [5].) t C R s S ε u R b θ b visul xis y trget s α b β b q 4 R m x m z m θ m y m q 5 q 6 z m q x m y m Figure 7. Mnikin skeleton n robot kinemtis; visibility one n trget efinition. his ttrtion gent only onsiers the trget n oes not tke re of the environment. his gent is similr to the ttrtion fore introue by Khtib q 2 q 3

5 hl , version 2-22 Ot 2008 [4], n gives the require ontributions x tt, y tt, n θ tt oring to the ttrtion towr trget referene s bove. hese ontributions, whih t on the mnikin (resp. the robot) leing member position n orienttion (in our se the trunk (resp. the bse of the robot n its kinemtis)), re normlize oring to Δ pos n Δ or. Repulsion gent between mnikin (resp. robot) n the luttere environment. his repulsion gent ts in orer to voi the ollisions between the mnikin (resp. the robot) n the luttere environment, whih my be stti or mobile. Severl possibilities n be use in orer to buil ollision riterion. he intersetion between two prts A n B in ollision, s shown by Figure 8, n be quntifie in severl wys. We n onsier either the volume V of ollision, or the surfe Σ of ollision, or the epth D mx of ollision (Figure 8b). he min rwbk of these pprohes omes out from the iffiulty to etermine these vlues. Moreover, 3D topologil opertions re not esy beuse mny of the virtul relity softwres use polyherl surfes to efine 3D objets. o etermine D move, the istne to voi the ollision (Figure 8b), we hve to store former positions of the mobile (mnikin or robot), so this quntifition oes not use only the tbse t given instnt but uses former informtion. his solution nnot be kept with our blkbor rhiteture tht only provies globl environment sttus t n instnt. prt A prt B () wo interseting prts. Dmx V S D move Σ (b) Volume in ollision. l 2 l S l 6 l 5 l 3 l 4 () Intersetion line. Figure 8 Collision riteri. Another quntifition of the ollision is possible with the use of the ollision line between the two prts. With this ollision line, we n etermine the mximum surfe S or the mximum length of the ollision line l = Σ l i (Figure 8). By the en, we ompute the grient of the ollision riterion oring to the Crtesin environment frme using finite ifferene pproximtion. From the grient vetor of the ollision length gr (x, y, θ ) () l, ontributions x rep, y rep, n θ rep re ompute by the repulsion gent. hese ontributions, ting on the mnikin trunk position n orienttion, re normlize oring to Δ pos n Δ or. He orienttion gent he gol of the he orienttion gent is to rotte the he of the mnikin (resp. of the film mer) in orer to observe the trget. It ensures the optimum onfigurtion tht mximizes visul omfort (resp. the visibility of the trget). Fining the optimum onfigurtion onsists in minimizing efforts on the joint oupling the he with the trunk n minimizing oulr efforts (resp. mehnil efforts or isotropy of the onfigurtion). We simplify the problem by onsiering tht the mnikin hs monoulr vision, efine by one whose prinipl xis, lle vision xis, is long y s n whose vertex is the enter of mnikin eyes (in tht se, the mnikin vision is similr to tht of film mer on robot). If the trget belongs to the vision xis, oulr efforts re onsiere null. Our purpose onsists in orienting y s suh s it beomes olliner to u by rottion of α b n θ b (Figure 7), subjet to joint limits. A joint limit verge for n ult is given in Figure 6. In the se of film mer, the orresponing vlues will be the optil hrteristis of the film mer. he lgorithm of this gent is similr to the ttrtion gent lgorithm presente there bove; ontributions α he n θ he, fter normliztion, re pplie to the joint oupling the he to the mnikin trunk (resp. the wrist joints of the robot). Visibility gent he visibility gent ensures tht the trget is visible, tht is, no interferene ours between the segment S linking the enter of mnikin eyes (resp. of the film mer) n the trget, n the luttere environment. he repulsion lgorithm is extly the sme s the one presente there bove: we etermine the ollision line length, if non equl to zero, normlize ontributions re etermine from x vis, y vis, n θ vis ompute by the visibility gent oring to the grient vetor of the ollision length, ontributions re pplie to the mnikin trunk (resp. the bse of the robot). It is to notie tht some ontributions my lso be pplie to the he orienttion (resp. the film mer orienttion). his is ue to the ft tht by turning the he, ollisions between the simplifie one with the environment my lso our. he use of simplifie one offers the vntge of ombining n ergonomi riterion with the repulsion effet. As mtter of ft, when the vision xis y s is insie the one C (Figure 7), we wien the one, respeting mximum limit. If not, we erese its vertex ngle, lso with respet of minimum limit tht orrespons to the initil onition when strting this visibility gent. he mximum limit my be expresse oring to the trget size or/n to the type of tsk to perform: proximl or istnt visul heking, globl or

6 hl , version 2-22 Ot 2008 speifi re to ontrol. Opertor gent on the mnikin (resp. on the robot) One of the ims of the stuy is to integrte humn opertor within the MAS in orer to operte in reltime. he opertor hs globl view of the luttere environment isplye by mens of the virtul relity softwre. Her or his tion must be simple n effiient. For tht purpose, we use Logiteh SpeMouse evie tht llows us to mnipulte boy with six egrees of freeom. he tion of the opertor gent only onsiers the move of the leing objet, whih is in our se the mnikin trunk (resp. the bse of the robot). Prmeters ome out from position x op n y op n orienttion θ op returne from the SpeMouse. hese ontributions re normlize, in the sme wy s with the ttrtion or repulsion gents. 3.2 RESULS AND CONCLUSIONS his metho hs been teste to hek the visul essibility of speifi elements uner trp of n irrft. he igitl moel is presente in Figure 9 n the list of elementry gents is epite in the mster gent winow in Figure 0. In this exmple, the repulsion gent for the mnikin (Repulsion), the visibility gent (Visul) n the he orienttion gent (Cone) hve speifi rte of tivity equl to, mening tht their tions hve priority but it is possible for the opertor to hnge in rel-time this tivity rte. Sine the tion of eh gent is inepenent from the other elementry gents, it is possible to intivte some of them (Puse/Work buttons). he vlues of Δ pos n Δ or, whih re use to normlize the gent ontributions, n lso be moifie in rel-time (Position n Orienttion buttons) in orer to pt the ontribution to the sle of the environment or to the tsk to perform. Our experiene shows tht the ontribution of the humn opertor is importnt in the optimiztion proess. Inee, if the utomti gent proess fils (whih n be the se when the one use in the visibility gent is in ollision with the environment), the humn opertor n: give to the MAS intermeite trgets tht will le to vli solution; move the mnikin to ple where the MAS proess oul fin solution. On the other hn, the MAS llow the humn opertor to t more quikly n more esily with the DMU. he elementry gents gurntee goo physil n visul omfort n enble to quntify n qulify it, whih woul be hr tsk for the humn opertor, even with sophistite virtul relity evies. For instne, we n evlute the rottions of the he n see how they re isperse from neutrl onfigurtion, inuing little effort. Moreover, the MAS permits us very goo lol ollision etetion n voine without ny effort of the humn opertor. he vntge of the MAS is to enble the ombintion of inepenent elementry gents to solve omplex tsks. hus, the gents prtiipting in the visibility tsk n be ouple with gents enbling essibility n mintinbility s propose by Chemil [7]. he purpose of further work onsists in globl oupling of mnikin mnipultions tking into ount visul n ergonomi onstrints n the mnipultion of moving objets s robots. he result of this work ws implemente in n inustril ontext with Snem Motors [6]. Figure 9. Digitl moel of trp of n irrft. Figure 0. Mster gent winow. 4. PAH PLANNING AND MOION CAPURE Now, we fous our work on intertive nimtion tht mkes it ble to rive n vtr in Rel ime (R) through motion pture evie, s seen on Figure [29]. In our frmework, we woul like to be ble to rive, in relisti wy, mnikin, oing shrp tsks suh s the ones tht n be one by worker, e.g. srewing srew, swing, rilling hole, or niling own nil. hese tsks ll nee shrp movements of the worker in the rel worl, n nee the sme ury in virtul worl. hus to be ble to be relisti, we will hve to emulte the rel worl s physil lws, n humn speifiities: intertion with environment, nonpenetrtion of objets, joint limits enforement, humn-like motion of the vtr. he min point in the fetures we wnt to be implemente is intertion with environment, beuse it rives the hoie of the moel we re to put into tion. If we wnt to be ble to intert in nturl wy, we hve to implement nturl behviours. ht is intertion must be one through fores. his mens kinemtil pprohes re not equte: we must use the equtions of ynmis. his perfetly fits the generl ontext of our work: our rhiteture will be ouple to portble hpti evie being evelope in [25]. he system we im t ontrolling is omplex. An effetive pproh in suh ses is to implement moulr rhiteture, so s to eouple funtionlities of our systems in ifferent moules. his will be me possible thnks to pssive pproh. Atully, pssivity will llow us to bring

7 hl , version 2-22 Ot 2008 the moulrity step further thn the simple funtionl moulrity. Inee it will llow eoupling the nlysis of the stbility of the whole system, into the nlysis of the pssivity of eh moule. he pprohes enbling tsks prioritistion suh s [2] seem interesting in se of onfliting tsks. Unfortuntely, they nnot be use in the ontext of pssivity, beuse they use projetions. As we will show, in the generl se, the use of projetions while optimising other potentils is not omptible with pssivity: in generl projetions brek pssivity. his oul seem very isturbing. Nevertheless, this limit only ppers in se of impossibility for the vtr to reproue the movements of the tor. In the ontext of engineering (t lest), one oes not relly wnt to ontrol the virtul humn in the se of unfesible movements. Inee wht we relly wnt is to know if the movement is fesible or not, this mkes big ifferene. Knowing this, we propose other ontrol moes, whih help the mnikin hieving its movements, inste of trying to ontrol it performing movements it will never suee in. hese ontrol moes re relevnt beuse of loss of informtion when we re immerse in simultions. Hpti evies pproh is interesting, but requires n hevy infrstruture. hus, if we wnte to be ble to perform preision ontrol, with light infrstruture we woul use virtul guies, bse on projetions. Nevertheless, s expline bove, we will show tht pplying projetion s is n le to the loss of the pssive nture of our ontroller, so we propose wy to buil physil projetions, whih respet pssivity, thnks to mehnil nlogies. We lso implement solution to solve for ontts. Zorn, n Hogins [23], propose solution tht hits n rets, moifying ontrol gins uring the simultion. his les to n pproh tht is known to be unsfe beuse of instbilities, n is not rel-time. Shmil, n Lin [24] use hybri they ll geometry-riven physis tht uses IK to solve for the mnikin s retion to ontt, n impulse-bse physis for the environment. hey lose the physil nture of the simultion. he metho we use oes not mke these onessions. he globl ontrol sheme we wnt is epite in Figure. We know motion pture positions re the inputs of ontroller (whih is bout to be etile), this ontroller rives physil simultion (lso to be etile); whih sens the upte worl onfigurtion to the output renerer. he mnikin we im t ontrolling is ompose of two lyers: skeleton (whih n be viewe s kinemtil hin), n rigi skin on top of it, whih will be useful for ollision etetion (of ourse, one motion is lulte, it n be sent to nie renerer whih will tkle with soft eforming skin whih is out of sope). he sheme of Figure 2, shows the whole rhiteture. vtr joint torques 6D trgets positions n veloities Control Simultion vtr joint positions rigi boies positions Figure Globl sheme of the system Controller tsk ontrol Simultion : ollision ontt points etetion ontt limits unilterl onstrints joint torques 6D trget positions, veloities projete errors internl ontrol physis joint positions joint torques guies integrtion 6D rigi boies positions R 3D Figure 2 Detile globl ontrol sheme 4. SIMULAION he bloks physis, n integrtion of the sheme Figure 2 shoul be self-motivte, s we wnt to emulte the physil lws of the rel worl. Let s just esribe some hoies we me. Physis: As stte bove, the simultion is to support ontt, n intertion with the environment, tht is fores. Hene ynmis estblishe itself s the best hoie for our moel. In first ttempt, we hve use first orer ynmil moel, s stte in [9] (though simpler it highlights the problem to be solve). Integrtion is one through n itionl joint mping term, tht is A ( q ) q & + C ( q, q &) q & + G ( q ) + B q & = Γ or B q& = Γ () with B, the mping mtrix hosen to be symmetri positive semi-efinite; q, n Γ re the joint prmeters n torques. Of ourse, these physil lws re intrinsilly pssive t port < q&, Γ >, the proof n be foun in [7]. Integrtion: We use Runge-Kutt-Munthe-Ks sheme (see [20]). his is Runge-Kutt metho eite to integrtion on Lie Groups, tht is known to be effiient. his work is left to Generlize Virtul Mehnisms (GVM) [27], librry being evelope in CEA\LIS. 4.2 CONROL In orer to unerstn the relevne of the other funtionl bloks of the igrm, we hve to woner wht mkes humn move, n hve the speifi motion he hs? In oing so, we foun three soures of movement (or influentil ftors): he first one enbles n en-effetor to reh gol in tsk spe. (Ex: I wnt my left hn to reh plug on the wll) he seon one rives onfigurtion, or gits. (Ex: it mkes the ifferene between the git of

8 hl , version 2-22 Ot 2008 fshion moel, n the git of n ol owboy) he lst one enfores physil, n biomehnil onstrints. (Ex: joint limits, non-penetrtion with environment, but lso blne ontrol, whih will be tken into ount in forthoming pper s in [28]) hese influentil ftors n be trnslte strightforwrly into ontrol iioms, being tsk spe, n null-spe ontrol uner unilterl onstrints. Bilterl n unilterl onstrints will not use ny problem. But opertionl spe, n null-spe ontrol must be stuie refully if one oes not wnt to brek pssivity. ) Externl tsk spe ontrol At first glne it oul seem interesting to bring prioritistion between externl tsks. Nevertheless, we show tht suh prioritistions n brek pssivity. his urges us to use other ontrol moes. In [2], Sentis, n Khtib introue ynmil eoupling of n externl tsks. he joint torques Γ they pply on their mnikin is ompose of the influene Γ i of n prioritise externl tsks, suh tht lower priority tsks o not isturb higher ones: Γ = n i= Γ i prev i), with Γ = Π i prev i prev( i) Γ ( ( ) i, (2) with Π prev ( i) projeting into higher priority tsks nullspe. Using suh projetions is unsfe; we show they n brek pssivity. Let s tke n exmple with two externl ports ( J, W, V, Γ ), n ( J 2, W2, V2, Γ2 ), eh port being esribe by its Jobin, the wrenh pplie, its veloity, n the torques it genertes. We give tsk the highest priority, n Π is the projetion llowing enforing priorities. he priority ppers if projets into Ker ( J ) we must enfore: B ( W V 2 + W V 2 2) t β Π. For the pssivity to be ensure, t, with β rel, W, nw2 0 (3) here lwys exists W, W 2, n J 2, suh tht: J 2W2 0 J W + J 2W2 = 0, (4) 2 J 2B Π J 2 = 0 hus in the generl se, projeting externl intertions n brek pssivity. As expline erlier, projetions re useful in se of onfliting trgets, in the se when the mnikin nnot hieve wht it is ske to o. It mens tht rel humn with the sme morphology s its virtul ounterprt, oul neither hieve the movement. In the se of engineering, we o not look for ontrolling virtul humns, in ses where they nnot hieve the trget motion. We only wnt to know if the movement is fesible or not. hus the propose ontrol is rther simple, but behves well in se of unfesible movements, n is ble to wrn in se of infesibility. In [26], they use 6D Proportionl Derivtive (PD) opertionl spe ontroller t eh point to be ontrolle on the mnikin (Figure 3): f trl = K( x x) + B ( v v). (5) Note tht if the ontrol points where linke to their trgets position, thnks to mpe spring, the fore generte woul hve the sme shpe; this mkes it ble to rw the ontroller, s its mehnil nlogy, the mpe spring. ontrol frme (frme to be ontrolle, on the mnikin) x error virtul humn to be ontrolle tsk spe ontroller (genertes wrenh on the mnikin to reue the error) x trget position (given by motion pture) Figure 3 sk spe ontrol Contenting (), n (5), we obtin B q& = J K( x x) + B ( v v), n ( ) ( B + J B J ) q = J ( K( x x) + B v ) &. his eqution mits solution if ( B J B J ) + is not singulr, tht is B must be efinite, or J must be full rnk, with B efinite, then: ( B + J B J ) J ( K( x x) B v ) q & = +. (6) 2) Constrints All the onstrints enumerte bove (joint limits, ontt ) re unilterl. hey n ll be solve through Liner Complementry Problems (LCP) lgorithms. We use GVM's unilterl onstrints solver [27, 29]. Using n pproh similr to Ruspini, n Khtib [8], we express the ontt problem in n LCP form, whih is solve for f - the ontt wrenh. his stge is pssive so long s the ynmil eqution is pssive. 3) Configurtions (internl ontrol) Null spe ontrol is usully solve s in [22], optimizing internl potentils (the hoie of these potentils is out of sope). If we o not wnt this optimiztion to isturb externl ontrol, we must work in the null spe of our externl tsk, projeting the joint torques inue from the internl potentil. Here we show this projetion n brek pssivity. Let s tke skeleton, n internl potentil U (q), to be minimize (its ssoite joint torques re Γ int ), n n externl port (efine s in ) by ( J, W, V, Γ )). Projetion Π (to be efine) is to give priority to the externl tsk: U Γint = α Π. q

9 hl , version 2-22 Ot 2008 Spee t the externl port is: U V = J q& = J B Γ = J B J W απ, (7) q If one oes not wnt the internl potentil to isturb the externl, we must tke Π s projetion in ( J ) Ker B. We must ensure this projetion keeps pssivity t ll ports. At the externl port, thnks to (7) we n write: W V = W J B J W 0. (8) So the system ompose of the externl tsk ouple to the internl one is pssive t the externl port, if the externl port ws pssive before oupling. he internl potentil s influene oes not ispper s in (8). hen, projeting n internl potentil n brek pssivity. Nevertheless, there re solutions to this problem: We n reue the internl tsk s influene through α suh tht the system remins pssive. We n use self-projetive internl potentils, Extene projetions Π prev ( i) n lso be use, they projet in ll externl ports null spe. Suh solutions re not yet implemente in our ontrol; the internl ynmi is left open loop for the time being. However, we n tune the onfigurtion through B. 4.3 PASSIVE VIRUAL GUIDES As stte before, we wnt to the possibility to guie the movements of our virtul humn (guies blok of Figure 2). he most intuitive wy to implement suh guies is to projet the error to be orrete by our opertionl spe ontroller. We showe introuing the projetion mtrix, oul brek pssivity. As in telerobotis [9], we use the virtul link onept, in orer to relize pssive projetions. Following the mehnil nlogy pproh, pssivity is ensure. trget, n tsk spe ontrol ontrol point future hole position onstrint oupling virtul humn holing rill wll to be rille virtul mehnis m: guies movement s Figure 4 Pssively guie virtul humn In Figure 4, the mnikin hols rill whih xis is ligne with the future hole xis in the wll thnks to the simple virtul mehnism in re. Any other onstrint oul be expresse thnks to virtul mehnisms. 4.4 EXPERIMENS AND RESULS he experiment onsists in rilling hole in wll thnks to rill, while lighting the future hole lotion thnks to hn light. Both tools re guie, thnks to our virtul mehnisms frmework. he rill n only move long fixe xis with fixe orienttion. his mens tht the ontroller leves only one egree of freeom to the opertor. he iretion of the spotlight is lso riven utomtilly (leving the three egrees of freeom of the light s position to the opertor). Figure 5 epits the iel xis in green () n tul xes re in re (b). Figure 5 Worker rilling hole, guie by virtul mehnisms Angle (egrees) time Figure 6 Angle between iel n tul xis of the rill, () without guie, n (b) with guie. In orer to see the effiieny of our metho, we rew the ngle between the iel xis, n the tul xis of the rill, s seen on Figure 6; in the se where the opertor is ompletely free (green), n in se where the guie is on (ornge). We lso teste the ollision engine. Figure 7 shows ntiollision in tion. Figure 7 Double, n selfollision 60 height (m) time Figure 8 Hn (), n obstle s (b) height: no penetrtion. On the urve Figure 8, we n see the height of the tble, whih must not be penetrte (ornge), n the height of the virtul humn's hn (green), while rehing, n lening on the tble. We see tht the hn never penetrtes the tble. 5. CONCLUSIONS wo pprohes hve been presente to nimte virtul mnnequins. Inee, the humn tsks n be nlyse with this both methos. he first one n be me when we hve only the DMU but, nowys, it not provies globl evlution of the humn ftigue. Some inies oming from ergonomi stuies re lrey implemente in softwre but the results re prtil. he other one n be me with humn n the prout. An b b

10 hl , version 2-22 Ot 2008 ergonomist nlyses the humn motion to qulify its ftigue by using severl methos. Suh results will be me in future works n implemente in the ontext of irrft inustry. 6. ACKNOWLEDGMENS his pper is bse on reserh me with P. Chemil (ECN), C. Le-Roy (Airbus), A. Rennuit (CEA), A. Mieilli (CEA), C. Anriot (CEA), F. Guillume (EADS/CCR), N. Chevssus (EADS/CCR). REFERENCES []. Lozno-Pérez n M. A. Wesley "An lgorithm for plnning ollision-free pths mong polyherl obstles," Comm. ACM, pp , Vol. 22, No. 0, 979. [2] P. ournssou "Géométrie et intelligene rtifiielle pour les robots," Hermès, Pris, 990. [3] J. D. Boissonnt, B. Fverjon n J. P. Merlet "Pereption et plnifition," rité es nouvelles tehnologies, Série Robotique, Hermès, Pris, 988. [4] O. Khtib "Rel time obstle voine for mnipultors n mobile robots," Interntionl Journl Robotis Reserh, pp.90-98, Vol. 5, No., 986. [5] J. Brrqun n J. C. Ltombe "Robot Motion Plnning: A Distribute Representtion Approh," Int. Journl of Robotis reserh, Vol. 0(6), pp , Deember 99. [6] I. Nmgung n J. Duffy "wo Dimensionl Collision-Free Pth Plnning Using Liner Prmetri Curve," Journl of Roboti Systems, Vol. 4(9), pp , 997. [7] P. Chemil n C. Le Roy "A istribute pproh for essibility n mintinbility hek with mnikin," Proeeings of ASME - DEC'99, Ls Vegs, September 999. [8] P. Chemil,. Dmy n C. Rouhon "Integrte esign n rtifiil relity: essibility n pth plnning," ASME Engineering System Design n Anlysis (ESDA'96), pp , Vol. 8, July 996. [9] J. Ferber "Les systèmes multi-gents, vers une intelligene olletive," InterEitions, 995. [0] J. F. Arn n S. J. Pelletier "Cognition Bse Multi-Agent Arhiteture," Intelligent Agent II, Agent heories, Arhitetures n Lnguges, August 995. [] N. R. Jennings, K. Syr n M. Woolrige, "Romp of gent reserh n evelopment," Autonomous gents n multi-gent systems, pp. 7-38, Vol., 998. [2] Y. K. Hwng, K. R. Cho, S. Lee, S. M. Prk n S. Kng "Humn Computer Coopertion in Intertive Motion Plnning," ICAR'97, pp , July 997. [3] J. Ferber n M. Ghllb "Problémtique es univers multi-gents intelligents," Ates es journées ntionles PRC-IA, pp , 988. [4] V. Riffr "Aessibilité un opérteur humin en environnement très ontrint Plement optiml et postures," PHD hesis, Nntes University, Frne, 995. [5] Ph. Wenger n P. Chemil, "Ability of robot to trvel through Crtesin free workspe in n environment with obstles," Int. Journl of Robotis Reserh, vol.0, n 3, pp , June 99 [6] B. Millé, P. Chemil, E. Rmstein, "Multi-Agent Approh n Emergene of Virtul Humn Behviour in VR Environment," IEEE Interntionl Conferene on Systems, Mn n Cybernetis, Otober [7] R. Murry, Z. Li, S. Sstry, "A Mthemtil Introution to Roboti Mnipultion", CRC Press, 994. [8] D. Ruspini, O. Khtib, "Collision/Contt for the Dynmi Simultion of Complex Environments", IEEE/RSJ Interntionl Conferene on Intelligent Robots n Systems:IROS'97, September 997, Genoble, Frne. [9] L. D. Joly, "Commne Hybrie Position/Fore pour l éléopértion: une Approhe Bsée sur les Anlogies Méniques", PhD thesis, Université Pris 6, 997. [20] E. Celleoni, B. Owren, "Lie group methos for rigi boy ynmis n time integrtion on mnifols", Computer Methos in Applie Mehnis n Engineering 92(3-4): [2] L. Sentis, O. Khtib, "sk-oriente Control of Humnoi Robots hrough Prioritiztion", IEEE- RAS/RSJ Interntionl Conferene on Humnoi Robots, Snt Moni, USA, November [22] O. Khtib, J. Wrren, V. De Spio, L. Sentis, "Humn-Like Motion From Physiologilly-Bse Potentil Energies", On Avnes in Robot Kinemtis, Springer [23] V. B. Zorn n J. K. Hogins. Motion ptureriven simultions tht hit n ret. In Proeeings of the 2002 ACM SIGGRAPH/Eurogrphis symposium on Computer nimtion, pges ACM Press, [24] H. Shmil n M. C. Lin, "Geometry-riven Intertion between Avtrs n Virtul Environments", Computer Animtion n Virtul Worls Journl, July [25] P. Grre, J.P. Mrtins, J.P. Frionneu, "Une nouvelle tehnologie 'orthèse portble", Hnip 2004, Pris, Frne,7-8 june 2004 [26] V.B. Zorn, N.C. Vner Der Horst, "Mpping optil motion pture t to skeletl motion using physil moel", Eurogrphis/SIGGRAPH Symposium on Computer Animtion, [27] X. Merlhiot, "Dossier nlyse es lgorithmes embrqués", CEA ehnil Report, Jnury 2005, to be publishe in the frmework of the LHIR RNL projet. [28] K. Hr, H. Hirukw, F Knehiro, K. Fujiwr, K. Kneko, S. Kjit, M. Nkmur, "Dynmil Blne of Humnoi Robot Grsping n Environment", Pro. of 2004 IEEE/RSJ Interntionl Conferene on Intelligent Robots n Systems IROS'2004, Otober 2004, Seni, Jpn. [29] A. Rennuit, A. Mielli, X. Merlhiot, C. Anriot, F. Guillume, N. Chevssus, D. Chblt et P. Chemil, "Pssive Control Arhiteture for Virtul Humns", Interntionl Conferene on Intelligent Robots n Systems, Emontgon, Cn, 2005.