Motion Guidance Experiments with Scooter Cobot

Transcription

1 Motion Guidnce Experiments with Scooter Coot ES Boy 1, E Burdet 1,2, CL Teo 1 nd JE Colgte 3 1: COME, Dept of Mechnicl Engineering, nd 2: Division of Bioengineering, Ntionl University of Singpore e.urdet@ieee.org 3: LIMS, Dept of Mechnicl Engineering, Northwestern University Astrct Coots ssist humns y mechniclly guiding long softwre-defined pths or surfces. Coot design hs een extensively studied previously. This pper reports the first systemtic experimentl study of guidnce with coot. We investigted the movements of seven opertors with the Scooter coot in representtive environments. Anlysis of the force exerted y the opertors nd the trjectories revels significnt differences etween guided movements (GM) nd free movements (FM). While FM requires lerning for ech novel tsk, GM is optiml from the first tril: Less effort is required to move in GM thn in FM; Movements in GM re fster, smoother, nd require less ck nd forth correction thn in FM. These dvntges demonstrte the strength of the Coot concept. The results further suggest tht opertors guided y the Scooter cn hndle ojects in more open-loop wy thn with dum trolley, nd so perform fster nd concentrte on other spects of the mnipultion tsk, potentilly resulting in incresed productivity nd fewer injuries. Hndle to control scooter Force-torque sensor One of the 3 motors for steering wheels One of the 3 glidewheels for odometry 1. Introduction Plcing window or cr door into its frme is difficult opertion requiring simultneous control of six degrees of freedom (DOF). Mechnicl guides, such tht only trnsltion is needed, would mke it esy to move it into the frme using only trnsltion. Coots (or Collortive roots) developed y M Peshkin nd JE Colgte t Northwestern University [1 4] re rootic ssisting devices implementing similr strtegy. Unlike fixed mechnicl guides, coots relize softwre-defined mechnicl guidewys long desired pths or surfces. They re pssive in tht they do not generte, ut only steer the wheels to direct it to relise softwredefined pths[4]. Forces perpiculr to the wheel hedings re lnced y friction, constrining to the heding direction. ostcles pin hole Figure 1. : Scooter coot to guide plnr in position nd orienttion. : Two sujects performing pth design tests t LIMS, Northwestern University. To illustrte the coot concept nd its modes, we riefly descrie the Scooter coot with which the experiments reported in this pper were performed. The Scooter (Fig 1) is tringulr vehicle moving on plne, with steerle wheel t ech corner. In Free Mode (FM), the wheels turn like csters to lign with the force

2 exerted y the opertor, nd the coot ehves s if it hd 3 DOF (i.e., plnr position nd orienttion). This force is mesured y force-torque sensor mounted on the hndle. In Guided Mode (GM), ech wheel is steered y motor to follow guiding pth coded in softwre. Coot kinemtics, design nd control hve een investigted extensively, nd severl plnr nd sptil coots hve een relized for the utomotive industry [14]. The strtegy of mechniclly guiding is not limited to the utomotive industry. It could e used, for exmple in surgery, to fcilitte mnipultion nd keep sclpel out of risk res; to fcilitte physicl rehilittion (see the Unicycle Two-Link Arm project in [3]); or to crete n dptle wheelchir tht ssists the hndicpped ccording to their ilities [5]. While it is expected tht guidnce cn fcilitte oject hndling, we know of no systemtic study on guidnce with coot. A few previous studies on the interction with kinemtic constrints hve een relized on the multi-joint rm. [6] studied forces t sttic positions in the presence of kinemticl constrints, nd [7] studied stiffness dpttion in constrined movements. However, these works do not ddress questions such s whether nd how guidnce fcilittes ojects hndling, nd how do opertors use guidnce. The present work investigtes, for the first time, rel s performed with coot. We recorded the force exerted during movement y the opertor nd the trjectories relized in vrious environments nd conditions with the Scooter coot. We further compred the ehvior in guided nd in free modes. Our results show how opertors use the coot to fcilitte oject hndling. In prticulr, little force nd visul feedck were required in guided mode. The opertors relied on the guidewy to mneuver the Scooter without exerting much force ginst it. Tools to progrm coots sed on this pper re presented in [8].. Trining Consecutive movements performed with the Scooter in given environment my vry significntly (Fig.4). Therefore, the sujects were first trined to drive the Scooter in guided nd free modes in three typicl environments: long stright line (Fig. 2), long curved line (Fig. 2), oth drwn on the floor, nd through nrrow pssge (Fig. 2c). The sujects were required to follow the line nd plce pin fixed to the Scooter into hole t its of the pth. Corresponding guiding pths were provided to trin the sujects on guided. The sujects lso hd to mneuver the Scooter through pssge (Fig. 2c) nrrower thn the mximum dimeter of the Scooter nd to plce the pin into the hole. The ostcles were mde of Styrofom ords (Fig. 1). Agin, corresponding guiding pth ws used to trin on guided. For ll three pths the sujects were trined first in guided mode nd then in free mode, in the order of Fig. 2. They were required to perform t lest three trils in ech. For the drwn lines, lerning ws considered complete when the men distnce etween the desired nd relized pth ws less thn.2m. For the nrrow pssge, lerning ws complete when no ostcle ws hit. On verge, sujects required 5 trils (minimum 3, mximum 6) in ech environment. The trining session took n verge of 71 minutes with stndrd devition of 3.3 minutes for ech suject. c pths drwn on ground hole ostcles 2. Methods. Sujects The experiments were pproved y the Institutionl Review Bord of Northwestern University, nd performed y seven students (men ge 24, with stndrd devition 2) without known motor disility. These sujects were informed out the experiments, nd gve their consent prior to prticiption. 1 m Figure 2. Three environments the sujects used to lern to move with the Scooter coot. c. Performnce in Free nd Guided Modes Performnce fter lerning ws tested the next dy in the two environments of Figure 3. To test, sujects hd to move six times from the point to the point without colliding with ny ostcle, nd to plce the pin into hole t the point. These six pths were then lest-squre pproximted using B-splines with 1 m

3 16 control points [8], used s guiding pths. So ech free movement hd one corresponding guiding pth nd one guided. Opertors sometimes needed ck-nd-forth corrections to ccurtely position the Scooter. We counted corresponding direction reversls y the following criterion: there is reversl when the dot product etween the current (trnsltionl) velocity vector nd ny velocity vector less thn 2.5 cm wy from the current position is negtive, nd checked visully whether it did correspond to reversl (fig 7). Fst Fourier Trnsform (FFT) gve the frequency content of the force nd torque exerted y the opertor on the coot. We used the integrl of normlized FFT N ostcles Figure 3. The two environments in which free nd guided were tested. Free movements re shown in this figure. d. Dt Anlysis We wnt to compre the effort to perform movement in free nd guided modes. Oviously the energy E Fd l cnnot e used for this purpose, s it does not consider internl forces. For exmple if suject guided long pth stops t position nd exerts force norml to the guidewy, the corresponding energy (used y the muscles, though not trnsformed into work done on the oject) is not ccounted for in E. We mesured the effort T T dt (1) to rotte the Scooter, nd the trnsverse effort T F T F dt. (2) to redirect movement Similrly, [9] used the integrl of qudrtic function of the wrench F = (,F) to mesure the effort during movement. The movement durtion multiplier T (defined s the time with trnsltionl velocity exceeding.15 m/s nd rottionl velocity exceeding.3 rd/s) mkes these mesures invrint under time scling, i.e., two movements x 1 (t), t [, T], nd x 2 (t), t [, T/], with x 2 (t) x 1 (t),, hve the sme mesure. In guided mode, it would e theoreticlly possile to complete movement from the to the point with zero torque or perpiculr force, i.e., with ) = F ) =, s the Scooter needs only motive force tngent to the guidewy. v s FFT N dv 3) nd v F s FFT N F dv (4) to compre the frequency content of torque nd force etween free nd guided in the intervl [,v s ] Hz where v s = 5 Hz is the Nyquist frequency which is hlf the smpling frequency of 1kHz. As low frequencies correspond to the pth shpe nd so re similr in FM nd GM, the FFT in either FM or GM ws normlized y the lrgest mplitude to give FFT N. Hence the differences ( FM )( GM ) nd (F FM )(F GM ) mesure the difference in high frequency content etween FM nd GM. Directionl t-tests were used to investigte lerningrelted fetures nd compre GM with FM, fter Lilliefors tests on the dt to confirm tht they were T- distriuted. 3. Results. Lerning Fig. 4 shows initil trils chrcterized y lternting clockwise nd nti-clockwise torques nd corresponding oscilltions in the trjectory. Further, the torque grdully decresed with repetition of the movement, ut remined lrger thn in guided mode. We exmined this tr systemticlly using the effort mesure (defined in the Methods s the integrl of the torque). Figure 5 shows the evolution of the normlized effort mesure = mx s movement ws repeted, where mx is the mximum in tht environment for given suject.

4 free mode guided mode --- LS fit to FM dt LS fit to GM dt 1 st try 2 nd 3 rd 4 th 5 th 1 st try Figure 4. Torque in consecutive s repeted long the stright line. Clockwise torque is depicted y grey rings nd nti-clockwise torque y lck rings. The men speed ws.74 m/s in the fifth nd.88 m/s in the first guided. The (normlized) effort mesure showed significnt chnge in FM ut not in GM. The stndrd devition of FM is significntly lrger thn.1 (p<.2 for ech environment of Fig. 2), while tht of GM is significntly smller thn.4 (p<.1). For ll sujects nd environments, FM lso required significntly more effort thn GM, s min FM mx GM p<.4): see first column of Tle 1. Finlly, the dt show highly significnt lerning in the stright nd curved environments, ut not in the nrrow pssgewy (second column of Tle 1). c environment min( FM )>mx( GM ) FM, untrined > FM, trined stright pth.1.5 curved pth.34.9 nrrow pssge Tle 1. Significnce level for the difference of pplied effort etween FM nd GM (1 st column) nd for lerning in FM (2 nd column). Figure 5. Evolution of rottion effort during lerning in the three environments. Ech colour represents one of the seven sujects. The dshed lines re LS fits to FM dt of the sujects, nd the solid lines to GM dt.. Comprison etween Free nd Guided Motion After lerning, the sujects pplied significntly more torque in FM thn in GM (Fig. 6). Fig. 6c further shows significnt difference FM GM of rottion effort etween FM nd GM (p<.2). Finlly, the trnsverse

5 effort F is not significntly different in FM nd GM (p>.2). Similr results were found using the totl force Fdefined similrly to F ), s integrl of the force F. guided speed (m/s) free ed c time (s) guided d effort in mneuvering () c guided effort in pushing constrint (F ) Figure 6. The pplied torque ws lrger in free thn in guided. In nd, the grey nd lck rings represent nti-clockwise nd clockwise torque respectively. The Scooter is not shown to improve visiility. c,d: Histogrms of rottion effort mesure () nd trnsversl effort mesure (F ), etween free (grey) nd guided (lck) s done y seven humn sujects. c shows tht the rottion effort () is higher for FM thn for GM, while d shows tht the difference is not significnt for the trnsversl effort (F ). numer of corrective movements Figure 7. Differences etween free nd guided. : Speed in guided nd in the corresponding. : A movement with two direction reversls. c: Histogrm of the numer of direction reversls.

6 The sujects took less time to complete the sme pth in GM thn in FM (p<.5) (Fig. 7). As the distnce is the integrl of the speed, one my expect the men speed to e significntly lrger in GM, so tht the re under the speed curves would e similr in FM nd GM. However, the men speed in GM is lrger y only.4 m/s in men over the sujects. The longer time tken y FM is proly due to more ck-nd-forth corrective movements in FM thn in GM. A t-test confirmed tht there re more direction reversls in FM thn in GM (p<.2) (Fig 7, c). free mode curvture norml force Motions in GM showed smller high frequency content thn in FM (Fig. 8). As the FFT for either FM or GM ws similr in ll environments, we used its men over the two environments in Fig. 3. The torque contined significntly more high frequencies in FM thn in GM ((F FM ) > (F GM ) with p<.1). The force generlly contined more high frequencies in FM thn in GM, ut the difference ws not significnt ((F FM ) > (F GM ) with p>.2). guided guided mode frequency (Hz) guided Figure 9. Correltion etween norml force nd curvture. The figure shows curvture vectors (left) nd norml force (right) for movement in free mode (top, grey) nd in guided mode (down, lck). frequency (Hz) Figure 8. Typicl FFT of the pplied force () nd torque () in free (grey) nd constrined (lck) s. The mplitudes of the high frequency components in re higher thn tht in guided, in prticulr for the torque. To further compre GM with FM, we computed the correltion etween the force norml to the pth nd lso norml to the pth the curvture (Fig. 9). A high correltion coefficient indictes tht the opertor is providing centripetl force to move the Scooter long curve. Therefore high correltion is expected in FM. However, in GM, the opertor does not hve to provide centripetl force to turn, nd the correltion coefficient ecomes n indictor of the strtegy used y the opertor. A lrge (positive) correltion would indicte tht the opertor turns with the curve, while n opertor just pushing stright will produce force nticorrelted (i.e. negtively correlted) with curvture.

7 The correltion coefficient etween the FM norml force nd the curvture ws men.6 with stndrd devition.1. In GM, norml force nd curvture were uncorrelted, with men coefficient of. nd stndrd devition of Discussion From the results of the experiment, movements with the Scooter coot re fundmentlly different in guided mode (GM) versus free mode (FM): 1. For simple environments, performing optimlly in FM required severl trining trils. However, in complex environments, optiml FM movement ws hrd to chieve nd performnce did not significntly improve with prctice. In contrst, the ehvior in GM ws optiml from the first tril. 2. Significntly more effort ws required to mnoeuvre the Scooter in FM thn in GM. 3. In GM, the opertor cn rely on the constrints to guide the movement nd excessive norml forces re countercted y friction. Therefore, we might expect tht lrger norml forces to e used in GM nd in FM. However, even in highly curved pths, the totl norml force ws similr in GM to tht in FM, or smller. 4. Movements in GM were significntly fster, nd hd fewer ck-nd-forth corrective movements thn in FM. 5. The high frequency content of pplied torque ws lower in GM thn in FM, indicting smoother s in GM. These points demonstrte the dvntges of the guided mode, nd show the strength of the Coot concept. These results were otined with prticulr coot, the Scooter, nd plnr. Cn we expect similr results for generl guidnce involving six degrees of freedom nd lod support ginst grvity? The Scooter, reduces 3DOF guidnce tsk to 1D one. A 6 DOF tsk reduced to 1D tsk my yield similr or even more drmtic differences. Thus similr improvement my e expected in 6DOF mnipultion guided y coots. The ove results lso revel how the opertors ehve when using the Scooter in guided mode. The fct tht s were smoother in GM nd hd smller numer of corrective movements suggests tht opertors needed less on-line visul control for guided thn for. The opertor, relesed from the urden of mnoeuvring, cn perform fster nd concentrte on other spects of the tsk [8], e.g., precision nd sfety, perhps resulting in improved productivity nd ergonomics. How do opertors use guidnce? If they were turning the coot y themselves to minimize effort [1], their norml force would e correlted with curvture. However, the correltion coefficient ws not positive, showing tht the opertors used the guidewys to mnoeuvre the coot. It would not e surprising tht the opertors used the constrints to fcilitte nd we my expect tht the price to py for moving esily long guiding pth is higher norml force. If the opertor would push the coot stright, without knowledge of the pth, it my e expected tht the norml force would e lrge nd nticorrelted with curvture. However the norml force ws reltively smll nd not (nti)correlted with curvture, indicting tht the opertors did not just push the coot stright. Perhps the hptic informtion received through norml force helped the opertor to keep pushing minly in the correct direction. Guided y the coot, the opertor needed little force nd visul feedck to perform complex fster. Acknowledgement The experiments were performed t LIMS. We thnk Michel Peshkin, Eric Fulring, Songho Kim nd Mrk Sld for their suggestions nd contriutions nd Tim Poston for the edition. References [1] M Peshkin et l. (21), Coot Architecture, IEEE Trnsctions on Rootics nd Automtion 17(4): [2] W Wnnsuphoprsit et l. (1997), Coot Control, IEEE Interntionl Conference on Rootics nd Automtion (ICRA) 4: [3] [4] B Gillespie et l. (21), A Generl Frmework for Coot Control, IEEE Trnsctions on Rootics nd Automtion, 17(4): [5] E S Boy et l. (22), Collortive Wheelchir Assistnt, Proc. Interntionl Conference on Rootics nd Intelligent Systems (IROS). [6] T Tickel et l. (22), Kinemtic Constrints for Assisted Single-Arm Mnipultion, Proc. Interntionl Conference on Rootics nd Automtion (ICRA). [7] H Gomi (1998), Anisotropic stiffness reduction during constrined multijoint rm movement, Proc. Conference of the IEEE Engineering in Medicine nd Biology Society 2: [8] E S Boy et l. (22), Experimentl Evlution of the Lerning Coot, sum. Eurohptics 23. [9] K Lynch nd K Liu (22), Designing guides for ergonomic collortive mnipultion, Proc. Interntionl Conference on Rootics nd Automtion (ICRA). [1] E Todorov nd M I Jordn (22), Optiml feedck control s theory of motor coordintion, Nture Neuroscience 5: