Fisher Dept of Arti cial In telli gence Univ ersit y of Edin burgh 5 F orrest Hill Edin burgh EH1 2QL Scotland UK da vidwrbf aifh edacuk ABSTRA CT Inspired uman grasping b eha vior use of hand pre shap es has for some time b een recognized as a usef ID: 76795
Download Pdf The PPT/PDF document "Dextrous Hand Grasping Strategies Using ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
PreshapesrajectoriesvidWrenandRobertB.FisherofArti cialIngence,UnivyofEdin5ForrestHill,EdinburghEH12QL,Scotland,U.K.behaofhandshapeshasforsometimebeenrecognizedasausefulwofreducingthecomplexitofplanninggraspsforadex-troushand.ede nefourtypesoftask-speci cpreshapeandtomodesofdigitclosure,andgivecriteriafortheandpositioningofthepreshapes,oninsectingdigittrajectorieswiththegraspablesurfacesoftheobject.Graspsareplannedinsimulationusingrealrangedatafromalaserstriperandananthropomorphicmodel.Ourresultsshowthatarangeofpolyhedraledobjectscanbegraspedusingrelativelysimple,fastalgorithmswhenthehandmotsareconstrainedinthiswgrippers,objects.troushandsareessentialiftherobotistomanipulatetheobjectbeteenitsdigits.er,dextroushandshabigadverparallelwgrippersevenwhentheaimisjusttoacquiretheobjectinastablegrasp:Duetotheincreasednberofhand-objectcondextroushandgraspscanbemorestablethanpar-gripperofstabilitbecomesastheber(andarea,inthecaseofsoft- ngeredhands)ofhand-objectcontactsincreases.Adextrouscangraspobjectfromamofwristpositionstionsthancanaparallelwgripper.Thisisusefulinaclutteredentandfacilitateseasieringrationofarmmotionplanningandgraspplanning.haresub-optimal,butfast.Napier[8]introducedtheconceptofInaprecisiongrasp,theobjectisgraspedbyonlythetipofeachdigit;apoergraspwrapsaroundtheob-andpossiblythepalm|contacttheobject.Thepreci-siongraspallowsmaximasthetipsofhdigitcanrollalongtheobjectsurfaceinanydirec-po[7]furtherthesede nitionsforrobotics,yintroducingalgrinwhichtheinsidesofthedistal(ofeachdigittactstheobject.Thisliesbettheprecisionandpoergraspintermsofmanipulabilitandstabilitthetipofeachdigitcanrollalongtheob-jectsurfaceinonedirection.whenpickingobjectso the\roor,aprecisiontheonlytopoergraspsuchanobjectitmustbeliftedbyapre-pohoftheworkonrobotgraspinghasbeenconcernedwiththeofstablestaticmechanicstosynthesizeforceclosuregrasps[9]bmaximizingtheleewyincontactplacementandwtontomakethesegraspsstable[10]bymodelingtheconasvirtualsprings.er,aswithmoststabilityanal-ysis,theworkignoredthekinematicconstraintsondigitposition.robotstrained,isverycomplex.Whenmountedona6degreeoffreedomarm/wristcombination,thetomostwidelyuseddextroushands,theSalisbury[11]andtheUtah/MIT[3]respectivInrecentyearsthepreshapingparadigmhasbeenwidelyrecognizedasausefulwytoeasethecomplexitoftheof ndingfortheAhandpreshapethedigitpostureasthewristmoestoardstheobject.Thegraspisthenexecutedbyplacingthewristintoapositionthatencom-passestheobject,andthen\rexing(closing)thedigitstiltheymakecontactwiththeobject.ject.preshapedsystem,whichisusedtograsppolyhe-dralandsimplecurvedobjects,usingdatafromtoCCDcamerasandhapticexplorationtogive3Dedgesand2Dons[6],[7],giventhedesireddistancesbetusedpotentoshapethehand,butdoesnotderivthedesiredfromactualetal[1]usedpreshapingtoplanpoergraspsonobjects.Theobjectsaremodeledusingellipticalcylinders,whicareespeciallyforplanningpoergrasps|itisoneofthefewsystemswhichprovidesavisionsystemtai-loredtograspplanning.Thepreshapesareplannedusingasetofheuristicsbasedonthepropertiesoftheelliptical Whenusingthepreshapingparadigm,hoer,itisnotenoughtosimplymakeadecisionaboutwhichpreshapetrajectoriesare\rexedfromthepreshapethegrasp.ethereforede netheaspstr,whicassociatesdigittrajectorieswithhandpreshape.Thisishmoreimportantforprecisionandlateralgraspsthanforpoergrasps,becausepoergraspshaemorepossiblehand-objectconGRASPSTRAaspstrconsistsofaandasetof,fromagraspbetoftherobotwrist.Thepreshapeisaprescribedhandcon gurationandthedigittrajectoriesarethemo-tionsofthetipsofeachdigitafterthepreshapeisformedandthewristpositionhasbeen xed.possibletswhilststillallowingasucientnberofde-becopeofobjects(i.e.thoughthedigittrajectoriesarespeci ed,hdigitcanbestoppedatanypointalongitstrajectoryycontactwiththeobject).Thisapproacreducestheoftheproblemwhilstpreservingthe\rexibil-yofadextroushand.Ratherthanviewingadextroushandasacompletelygeneraldevice,itisviewedasasetof\rexibletools,whereeachgraspingstrategyisadi er-t\tool".Thedecompositionalsoallowsustoexamineumangraspingbehaviorasansuccessfulexample.canleadtousefulofhandhand)TheuseofalimitednumberprescribedtrajectoriesmeansthatthisapproachcanalsoberobotheapertheSalisburyorUtah/MITexactlyhowmhsimplerremainstobeseen.Itshouldbenotedthatthisapproachde-emphasizessta-yanalysis.Thegraspstrategiesaredevisedtomakeitelythatanygraspthusformedisstable.Thepreshapesaresymmetrical,thatallare\rexedinparallel,andthe ngerabductionsarecoupled.Inourhandmodeltherearethreedigits,sothedigit-objectcontactslieinaAssumingthereisenoughfrictiontoresistgraecanusefullylimitstabilittotheplane.perpendicularpreshapeformationthegraspwillbeinequilibriumand,ofaminimaloffriction,Asthecontactnormalsdeviatefromthisidealorasthedigitsarestoppedatdi erentpointsalongtheirtrajecto-riesbycontactwiththeobject,morefrictionisrequiredtomakethegraspstable.Itisdiculttousefullyanalyzetherelativestabilitiesofthe nalpendsobjectrobothandsoft,deformableattheendofeacdigit[12][9]thenwithsuitableforcecontrolawiderangeofdigitpositionsbeHandModelFigure1showsthehandmodelusedinourexperimenThehandisroughlythropomorphic,inthatthedigitofthecertainjointsarecoupled. 4535256045abgdd Fig.1.Handmodelshowingthelinkdistancesandthelocationofangulardegreesoffreedom.AlldistancesshowninRefertotextformoredetails.ofex-ternalforces,eachwithequalanglesof\rexion( ; ;\rtheproximal,middleanddistaljointsrespectivaboutanaxisperpendiculartothepalmandthroughtheximaljoinWithineachdigit,thedistalandmiddletsarecoupled(suchthat ),andtheanglesofabductiontscouldbehard-wiredintothehanddesign,inwhiccasetheywouldhaeastrongin\ruenceonhowtheobjectcouldbemanipulated,orjustusedasmodesofmotofacilitatetheplanningofthegraspswithcomplexhandshastheSalisburyorUtah/MIThand.GenerationoftheGraspStrategyThegraspstrategyisgeneratedfromsetsofpreshapes.Atask-speci cpreshapeisonethatisideallysuitedtoaparticulartask.Thetask-speci cpreshapeisasclosetothetask-speci cpreshapeaspossible.Thepreshapeisthenexpandedinordertoallowforerrorsinwristposi-tioningandtoaoidcollisionswiththeobject.task-speci cshapes:HOOK.Asthenamessuggest,preshapesbodythetask-requirementsofthegrasp.enthedesiredan-gleofabductionandrelativepositioningofthetipsofeacdigit,thetask-speci cpreshapesarede nedasfollo preshapebe-possible 4,i.e.itminimizesj + +\r 3 .Thisgivesacontactsurfaceatthetipsofeachdigit.preshapeperpendicularclosetoperpendicularaspossible,itminimizes .Thisgivesacontactsurfaceontheinsidesofthedistallinkofeachdigit.TheMANIPULATIONpreshapeminimizesthesumtheircentralvitminimizes(+(+(apreshapeobjectmanipulationbetthedigits.Itisusefulifthereisnopreferenceforaprecisionorlateralgrasp.TheHOOKpreshapekeepsthelinejoiningthetiptothemiddlejointofeach ngerperpendiculartothewristplane,orasclosetoperpendicularaspossible.Thisgivesapreshapewhichissuitableforgraspingyhooking. -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20PRECISION PRESHAPE -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20LATERAL PRESHAPE -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20MANIPULATION PRESHAPE -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20HOOK PRESHAPE HOOKpreshapestask-speci cpreshapeables|onecontrollingthe\rexionofthedigits,anotherabductiontask-speci cpreshapestheangleofthedistallinkiscon-strainedsothatitcannotpointayfromthe\center"ofthepreshape, .Thisisdonetomakeitlesslikelythatcollisionsoccurbeteentheobjectsurfaceandthehand.speaking,HOOKgraspspeci c,sincethedecisiontoformahookinggraspoftendependsobjectorontheundersideoftheobject.er,itbeaboutcoarseobjectgeometryThechoicebe-dependthatwpreshapesmayhaetodeformfromtheiridealsinorderto ttothespeci edgraspingpoinChoiceofpreshapethereforespeci esa,nota rmconstraint,andultimatelythetypeofgraspisalwysstronglyin\ruencedyobjectsizeandshape;itisnotpossibletoexecutealateralgrasponanobjectwhichisverysmallcomparedtothehand,norisitpossibletoexecuteaprecisiongrasponanobjectwhichislargecomparedtothehand.oranarbitraryobject,thecloserthatthe nalgraspcationiscpre,theyofthegrFigure3showsthe2categoriesofdigittrajectoryure3(a)showsthePRXIMALdigittrajectory,inwhicure3(b)showstheDISTALdigittrajectory,inwhichthedigitsare\rexedatthemiddleanddistaljoinhdigittrajectoryisde nedbyjustonevariable,sincethemiddleanddistaljointsarecoupled.Thetrajectoriesgiveadigitclosuremotionsimilartothatofthehumanhand. -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20PROXIMAL TRAJECTORY -50 0 50 -50 0 50 -120 -100 -80 -60 -40 -20 0 20DISTAL TRAJECTORY Fig.3.XIMALandDISTALtrajectoriespreshapesbeXIMALdigittrajectories6di ereningstrategies.TheHOOKpreshapecanonlybeusedinwiththeDISTALdigittrajectory,togiveatotalof7di erentgraspingstrategies.ThegraspstrategiesaregeneratedasfollopreshapeAcquisitionofobjectmodel.oftask-speci cpreshapeandwristposition.Thesearechosensuchthatthetactisassmallaspossible, inordertokeepthe nalgraspshapeasclosetothetask-speci cpreshapeaspossible.trajectorytactisassmallaspossible,andthattrajectoriesobjectrangeofwristpositions.tofpreshape,withthedigitsconstrainedtomoealongtheirtrajectories,inordertoallowforerrorsinwristpositioningandaoidcollisionstheobject.Thisamountstoplanningawholegrasp,yieldingthecon-tactpoints,thehandcon gurationandthewristposition.Detailsaregiveninthenextsection.task-speci cpreshapetrajectoryximalisagraspingstrategygeneratedfromaTERALpreshapeandPRXIMALdigittrajectoryAnalgorithmhasbeenimplementedtoplangraspsforthePRECISION,LATERALandMANIPULATIONstrate-utureworkwillextendthistoworkfortheHOOKobjectviewpointsusingalaserstriper.Theimagesaretakenataresolutionof1mminthewith256tzvalues,theresolutionofwhichdependsontheoftheobject.AtimagehasdimensionsTherangedataisthensegmeny ttingquadricsurfaces,grownfromseedsacquiredbyan(H;Kclassi cationprocedurecedure)Sur-faceswhicharetootoogoodcontactforacircular ngertiparediscarded(byero-sion/dilation),toleaeasetofgraspablefeatures[15].Aablefeisasucientlylargesurfacefeaturewithgooddigit-objectrubber-coated)pablefeaturesarethenrotatedintotheglobalcoordinateframetoprovidethevisualinputtothegraspplanner.Thegraspablefeaturesaregroupedintocandidategrasp-ingsets.Thesesetscompriseof2or3features,accordingtowhethertheto ngersonthesameorseparatefeaturerespective ndallsetswhereop-positionbeteencandidatethband ngerpatchesex-Someofthesesetswillnotbereachablebythehand;er,thesubsequentalgorithmisfastenoughnottoethisaproblem.ask-Speci cPreshapeFittingspeci cpreshapeandthewristposition.Becausethetipsofthethreedigitsformaplane,thiscanbetreatedasa2Danalysis.ede nethefollowingterms:eplaneistheplaneinwhichthetipsofthedigitslie.positionspreshapeabductionpreshapeaperturethethb-tiptothepointmidwybeteenthetdclosuristheorthogonalprojectionofthedigittrajectoriesintotheapertureplane.projectedfromthetipofeachdigittoconInchoosingthepreshapeweusethreeobservationsbasedonhumangraspingbeha,theminimumsizeofthepreshapeapertureisproportionaltotheexpecteddis-tancebeteenthegraspablefeatures(see[4]).thepreshapeapertureiskeptassmallaspossible,takingtoaccounttheuncertainyinwristandobjectposition(see[14]).,theorienoftheapertureplaneisdeterminedmorebyentalconstraintsandtheepositionsoftheobjectandrobotarmthanbytheobjectgeometry.Formanyobjectsandtaskstheaperturetodetermine,forobjectslyingonthegroundagoodapertureplanewillalmostal-ysbeonewhichisparalleltotheground,inordertoemaximumleewywithrespecttohand-groundcolli-theexperimendescribedhere,therefore,apertureplaneisalwysparalleltothegroundplane.ecombinethesewithtostabilit,wtheamountoffrictionforthecontactforcestobeinequilibrium.,wekeepthethbinaplaneperpendiculartothecontactsurface.Thisrecognizesthefactthatthethbisthesinglemostimportantdigitintermsofgraspstabilityand,assuch,shouldbekeptnor-maltothecontactsurfaceinordertoaoidslippage.oreachcandidategraspingset:ask-speci cpreshapeTERALorMANIPULATheapertureplaneisorientedparalleltothegroundpossiblepositionsapertureplaneisdeterminedsuchthatwithinthisrangetheapertureplaneintersectsallthegraspingfeaturesinthecandidategraspingset.positionoftheaperturetheaimofkthecon-tactforcesco-planar.Theanglebeteentheplaneandpotentialcontactnormalsisminimized: sumofthesquaresofthecosinesofanglesbettheplaneandtheobjectnormalsoertheapertureplane/graspingThesumisnormalized,suchthattheththegraspablefeaturesareallplanes(towithinsomethepositionoftheaperturedefaultstothecenterofitsrange.Thethb-tiptrajectoryintheprojectedclosureiseptnormaltotheintersectionofthethbconsurfacewiththeapertureplane.Thepreshapeapertureischosensuchthatthepro-jectedclosuredistanceisminimized.Ifagivenpro-jectedclosuredistancecanbeacedwitharangeofabductions,thentheabductionchosenminimizesthesumofthesquaresoftheanglesbeteenthepro-jectedclosureandthecontactnormals.positiontask-speci cpreshaperametersarethenuniquelyytheshapeaperture.ChoiceofDigitTrajectoryThedigittrajectory|PRXIMALorDISTAL|iscseninordertominimizetheclosuredistance,whilstvidingenoughleewyforexpectedwristpositioninger-ExpansionofPreshapeThetask-speci cpreshapeisexpanded,byextending(opening)thedigitsalongthechosendigittrajectoryunthereissucientleewyofpreshapeplacementoertheexpectedwristpositioningerror.Experimenobjectobjects,modelwninFigure1.Figure4showsgraspsplannedonasimplecuboidusingthePRECISION,LATERALandMANIPULATIONpre-shapes.Theexpectedpositioningerroris10mminthex,yandzdirections.Theobjectpatchboundariesaredis-Thejointsanddigit-tipsofthehandaredrawnascircles,thehandsegmentsasstraightlines.esjoin-ingdigit-tipstosurfacepatcwthedigittrajecto-riesfromthepreshapepositiontothe nalcontactpoinximal,Manipulation-Proximaltrajectoryvidesthesmallerclosuredistancebutbecausethepromaljointisclosetoitslimit,intheevtofapositioningerrorthetrajectorywouldnotreachtheobject.TheDIS-ALtrajectoryisthereforechoseninstead.Manipulation-Proximalobject,cuboid.gersabductsothattheycanboth tonthesamefeature(step#6ofalgorithm);abduction, -100 -50 0 -100 -50 0 920 940 960 980 1000 1020 1040 1060XYZ -100 -50 -100 -80 -60 -40 -20 920 940 960 980 1000 1020 1040XYZ (a)(b) -100 -50 -100 -80 -60 -40 -20 920 940 960 980 1000 1020 1040 1060XYZ -100 -100 -80 -60 -40 -20 0 920 940 960 980 1000 1020 1040XYZ Fig.4.(a)Precision-Prolstrategy,(b)Precision-Prostrategyviewedside-on,(c)Lateral-Distalstrategyviewside-on,(d)Manipulation-lstrategyviewedside-esjoiningdigit-tipstosurfacepatchesshowthedigittrajectoriesfromthepreshapepositiontothe naltactpoinouldbetoofaraparttograsptheobject.Thisdemon-stratesthatabductionisnecessaryinordertomakesomegraspskinematicallyfeasible. -50 0 50 -150 -100 960 980 1000 1020 1040 1060 1080XYZ 0 -140 -120 -100 -80 -60 960 980 1000 1020 1040 1060 1080XYZ Fig.5.(a)Manipulation-Prmalstrategy,(b)viewedside-sphere,(b)showstheplannedgrasp(Lateral-Prowsthesideviewofthegrasp|notehowthesphereisgraspedaroundit'scenter,inordertokeeptheconwsthetopviewofthegrasp|notehowthe ngersabducttogivenormalcontacts(Step#6ofalgorithm).object,withtheto ngershavingdi erentgraspingfea- -20 0 20 -120 -100 -80 890 900 910 920 930 940 -40 -20 0 20 40 -100 880 900 920 940 960 980 1000 1020XYZ (a)(b) -40 -20 0 20 40 -100 880 900 920 940 960 980 1000 1020XYZ -40 -20 0 20 -140 -130 -120 -110 -100 -90 -80 -70 -60XY Fig.6.(a)rangedata,(b)Lateral-Distalgraspstrategy,(c)edfromside,(d)viewedfromtopFigure7(a)showstherangedatafromtodi er-tviewscombined,(b)showstheplannedManipulation-(c)thefromaboorthisgraspitiscrucialtoconsiderthedigittrajecto-ries,becausetheDISTALtrajectoryfailstointersectwiththeobject(seeFigure7(d)).Thequalityofthisgrasp,intermsofbothstabilityandtask-speci cit,isloeredbthelargedeviationofthe nalthetask-speci cpreshape.Analysisofthestaticmecicsofthegraspshowsthatitrequiresaminimumcoe-toffriction=0Thealgorithmsrunonofonerangeimagetes1-2minTheplanningofagraspforasetofgraspingfea-turestakesafewseconds,thelengthoftimedependingonfeaturesizeandshape.troducedassociatingsuremodeswithpreshapes.Firmcriteria,onhmangrasping,ebeengivenfordecidingonpreshapeparameterstailoredtothetask-requiremenPreshapesarethentheaimofminimizingthedistancefrompreshapetoconThishelpsgivestablegrasps,andhelpsaoidcollisions.Theonlydrakwiththealgorithmasitstandsisthatitcanfailtoplangrasps,becauseofthelackofanyfeed-kmeckreasoningbeappliedalterthechoiceofwristpositioninthecasethatneitherdigittrajectoryintersectswiththegraspingfeatures.Ftherworkwillalsoexaminetheinitialhoiceofaperture -140 -120 -100 -80 -60 -40 -20 0 -100 -50 0 920 940 960 980 1000 1020 1040 -140 -120 -100 -80 -60 -40 -20 0 -100 -50 920 940 960 980 1000 1020 1040XYZ (a)(b) -140 -120 -100 -80 -60 -40 -20 0 -120 -100 -80 -60 -40 -20XY -140 -120 -100 -80 -60 -40 -20 0 -100 -50 920 940 960 980 1000 1020 1040XYZ Fig.7.(a)rangedata,(b)Manipulation-Plgraspstrat-,(c)viewedfromtop,(d)distaltrajectoriesdonotin-tersectwiththegraspingfeatures.planeorienIntheworkheretheplaneis xedtobetotheground.orsomeobjects,asmtherearegoodapertureplanestobeused|inthecaseofthevticallyalignedcylinder,anyplanewhichgoesthroughtheaxisofthecylinderisagoodapertureplane.,wtendtoanalyzethegraspingstrategiesforstabilit,onalargesetofrandomlygeneratedpolyhedralobjects.shouldhelpusdesigngraspingstrategieswhichmaximizethechancesofproducingstablegrasps.grasps.C.Bard,C.Bellier,C.Laugier,J.Troccaz,G.Vercelli,andB.Thievingdextrousgraspingbyintegratingplan-report,G,1993.1993.P.J.Besl.esinRangeImageUnderstanding.Springer-erlag,1988.1988.S.C.Jacobsen,E.K.Iverson,D.F.Knutti,R.T.Johnson,andK.B.Biggers.Designoftheutah/m.i.t.dextroushand.IEEEInternationalConfereonRoticsandA,volume3,pages1520{1532,1986.1986.M.Jeannerod.talcoordinationduringreacatnaturalvisualobjects.InttentionandPerformanceIXErlbaum,Hillsdale,NJ,1981.1981.S.B.KangandK.Ikeuchi.Aframeworkforrecognisinggrasps.TechnicalReportCMU-RI-TR-91-24,CarnegieMel-lonUniv,Nober1991.1991.D.Lyons.Taggedpotential elds:Anapproachtospeci ca-tionofcomplexmanipulatorcon gurations.IEEEInter-nationalConfereonRoticsandA,volume3,pages1749{1754,1986.1986.D.M.LyAsimplesetofgraspsforadextroushand.IEEEInternationalConfereonRoticsandA,pages588{593,1985.1985.J.R.Napier.Theprehensilemotofthehumanhand.BoneJointSur,38B:902{913,1956. V.-D.NguyConstructingforce-closuregraspsin3D.IEEEInternationalConfereonRoticsandA,volume1,pages240{245,1987.1987.V.-D.NguyConstructingstablegraspsin3D.InternationalConfereonRoticsandA,vume1,pages234{239,1987.1987.K.SalisburyandS.Craig.Articulatedhands:orceconandkinematicissues.InternationalJournalofRoticsR,1(1):4{17,1982.1982.K.B.ShimogaandA.A.Goldenberg.Softmaterialsforrobotic ngers.InIEEEInternationalConfereonRoticsand,volume2,pages1300{1305,1992.1992.S.A.Stans eld.Roboticgraspingobjects:10(4),August1991.1991.A.M.Wing,A.Turton,andC.FGraspsizeandaccu-racyofapproachinreacJournalofMotorBehaviour18(3),1986.1986.D.WrenandR.B.Fisher.tifyingrobot ngergraspingpointsfromrangedata.SixthIntl.Conf.onIndustrialandEngineeringApplicationsofArti cialIntelligenceandertSystems,pages522{530,1993.