协作移动机器人前因和方向外文文献翻译中英文翻译外文翻译文档格式.docx
《协作移动机器人前因和方向外文文献翻译中英文翻译外文翻译文档格式.docx》由会员分享,可在线阅读,更多相关《协作移动机器人前因和方向外文文献翻译中英文翻译外文翻译文档格式.docx(55页珍藏版)》请在冰豆网上搜索。
ComputerScienceDepartment,UniversityofCalifornia,LosAngeles,CA90024-1596
ALEXS.FUKUNAGA
JetPropulsionLaboratory,CaliforniaInstituteofTechnology,Pasadena,CA91109-8099
ANDREWB.KAHNG
Editors:
R.C.ArkinandG.A.Bekey
Abstract.Therehasbeenincreasedresearchinterestinsystemscomposedofmultipleautonomousmobilerobotsexhibitingcooperativebehavior.Groupsofmobilerobotsareconstructed,withanaimtostudyingsuchissuesasgrouparchitecture,resourceconflict,originofcooperation,learning,andgeometricproblems.Asyet,fewapplicationsofcooperativeroboticshavebeenreported,andsupportingtheoryisstillinitsformativestages.Inthispaper,wegiveacriticalsurveyofexistingworksanddiscussopenproblemsinthisfield,emphasizingthevarioustheoreticalissuesthatariseinthestudyofcooperativerobotics.Wedescribetheintellectualheritagesthathaveguidedearlyresearch,aswellaspossibleadditionstothesetofexistingmotivations.
Keywords:
cooperativerobotics,swarmintelligence,distributedrobotics,artificialintelligence,mobilerobots,multiagentsystems
1.Preliminaries
Therehasbeenmuchrecentactivitytowardachievingsystemsofmultiplemobilerobotsengagedincollectivebehavior.Suchsystemsareofinterestforseveralreasons:
•tasksmaybeinherentlytoocomplex(orim-possible)forasinglerobottoaccomplish,orperformancebenefitscanbegainedfromusingmultiplerobots;
•buildingandusingseveralsimplerobotscanbeeasier,cheaper,moreflexibleandmorefault-tolerantthanhavingasinglepowerfulrobotforeachseparatetask;
and
•theconstructive,syntheticapproachinherentincooperativemobileroboticscanpossibly
∗Thisisanexpandedversionofapaperwhichoriginallyappearedintheproceedingsofthe1995IEEE/RSJIROSconference.yieldinsightsintofundamentalproblemsinthesocialsciences(organizationtheory,economics,cognitivepsychology),andlifesciences(theoreticalbiology,animalethology).
Thestudyofmultiple-robotsystemsnaturallyextendsresearchonsingle-robotsystems,butisalsoadisciplineuntoitself:
multiple-robotsystemscanaccomplishtasksthatnosinglerobotcanaccomplish,sinceultimatelyasinglerobot,nomatterhowcapable,isspatiallylimited.Multiple-robotsystemsarealsodifferentfromotherdistributedsystemsbecauseoftheirimplicit“real-world”environment,whichispresumablymoredifficulttomodelandreasonaboutthantraditionalcomponentsofdistributedsystemenvironments(i.e.,computers,databases,networks).
Thetermcollectivebehaviorgenericallydenotesanybehaviorofagentsinasystemhavingmorethanoneagent.thesubjectofthepresentsurvey,isasubclassofcollectivebehaviorthatischaracterizedbycooperation.Webster’sdictionary[118]defines“cooperate”as“toassociatewithanotherorothersformutual,ofteneconomic,benefit”.Explicitdefinitionsofcooperationintheroboticsliterature,whilesurprisinglysparse,include:
1.“jointcollaborativebehaviorthatisdirectedtowardsomegoalinwhichthereisacommoninterestorreward”[22];
2.“aformofinteraction,usuallybasedoncommunication”[108];
3.“[joining]togetherfordoingsomethingthatcreatesaprogressiveresultsuchasincreasingperformanceorsavingtime”[137].
Thesedefinitionsshowthewiderangeofpossiblemotivatingperspectives.Forexample,definitionssuchas
(1)typicallyleadtothestudyoftaskdecomposition,taskallocation,andotherdis-tributedartificialintelligence(DAI)issues(e.g.,learning,rationality).Definitionsalongthelinesof
(2)reflectaconcernwithrequirementsforinformationorotherresources,andmaybeaccompaniedbystudiesofrelatedissuessuchascorrectnessandfault-tolerance.Finally,definition(3)reflectsaconcernwithquantifiedmeasuresofcooperation,suchasspeedupintimetocompleteatask.Thus,inthesedefinitionsweseethreefundamentalseeds:
thetask,themechanismofcooperation,andsystemperformance.
Wedefinecooperativebehaviorasfollows:
Givensometaskspecifiedbyadesigner,amultiple-robotsystemdisplayscooperativebehaviorif,duetosomeunderlyingmechanism(i.e.,the“mechanismofcooperation”),thereisanincreaseinthetotalutilityofthesystem.Intuitively,cooperativebehaviorentailssometypeofperformancegainovernaivecollectivebehavior.Themechanismofcooperationmaylieintheimpositionbythedesignerofacontrolorcommunicationstructure,inaspectsofthetaskspecification,intheinteractiondynamicsofagentbehaviors,etc.
Inthispaper,wesurveytheintellectualheritageandmajorresearchdirectionsofthefieldofcooperativerobotics.Forthissurveyofcooperativeroboticstoremaintractable,werestrictourdiscussiontoworksinvolvingmobilerobotsorsimulationsofmobilerobots,whereamobilerobotistakentobeanautonomous,physicallyindependent,mobilerobot.Inparticular,weconcentratedonfundamentaltheoreticalissuesthatimpingeoncooperativerobotics.Thus,thefollowingrelatedsubjectswereoutsidethescopeofthiswork:
•coordinationofmultiplemanipulators,articulatedarms,ormulti-fingeredhands,etc.
•human-robotcooperativesystems,anduser-interfaceissuesthatarisewithmultiple-robotsystems[184][8][124][1].
•thecompetitivesubclassofcollectivebehavior,whichincludespursuit-evasion[139],[120]andone-on-onecompetitivegames[12].Notethatacooperativeteamstrategyfor,e.g.,workontherobotsoccerleaguerecentlystartedinJapan[87]wouldliewithinourpresentscope.
•emergingtechnologiessuchasnanotechnology[48]andMicroElectro-MechanicalSystems[117]thatarelikelytobeveryimportanttoco-operativeroboticsarebeyondthescopeofthispaper.
Evenwiththeserestrictions,wefindthatoverthepast8years(1987-1995)alone,wellover200papershavebeenpublishedinthisfieldofcooperative(mobile)robotics,encompassingtheoriesfromsuchdiversedisciplinesasartificialintelligence,gametheory/economics,theoreticalbiology,distributedcomputing/control,animalethologyandartificiallife.
Weareawareoftwopreviousworksthathavesurveyedortaxonomizedtheliterature.[13]isabroad,relativelysuccinctsurveywhosescopeencompassesdistributedautonomousroboticsystems(i.e.,notrestrictedtomobilerobots).[50]focusesonseveralwell-known“swarm”architectures(e.g.,SWARMandMataric’sBehavior-basedarchitecture–seeSection2.1)andproposesataxonomytocharacterizethesearchitectures.Thescopeandintentofourworkdifferssignificantlyfromthese,inthat
(1)weextensivelysurveythefieldofco-operativemobilerobotics,and
(2)weprovideataxonomicalorganizationoftheliteraturebasedonproblemsandsolutionsthathaveariseninthefield(asopposedtoaselectedgroupofarchitectures).Inaddition,wesurveymuchnewmaterialthathasappearedsincetheseearlierworkswerepublished.
TowardsaPictureofCooperativeRobotics
Inthemid-1940’sGreyWalter,alongwithWienerandShannon,studiedturtle-likerobotsequippedwithlightandtouchsensors;
thesesimplerobotsexhibited“complexsocialbehavior”inrespondingtoeachother’smovements[46].Coordinationandinteractionsofmultipleintelligentagentshavebeenactivelystudiedinthefieldofdistributedartificialintelligence(DAI)sincetheearly1970’s[28],buttheDAIfieldconcerneditselfmainlywithproblemsinvolvingsoftwareagents.Inthelate1980’s,theroboticsresearchcommunitybe-cameveryactiveincooperativerobotics,beginningwithprojectssuchasCEBOT[59],SWARM[25],ACTRESS[16],GOFER[35],andtheworkatBrussels[151].Theseearlyprojectsweredoneprimarilyinsimulation,and,whiletheearlyworkonCEBOT,ACTRESSandGOFERhaveallhadphysicalimplementations(with≤3robots),insomesensetheseimplementationswerepresentedbywayofprovingthesimulationresults.Thus,severalmorerecentworks(cf.[91],[111],[131])aresignificantforestablishinganemphasisontheactualphysicalimplementationofcooperativeroboticsystems.Manyoftherecentcooperativeroboticsystems,incontrasttotheearlierworks,arebasedonabehavior-basedapproach(cf.[30]).Variousperspectivesonautonomyandontheconnectionbetweenintelligenceandenvironmentarestronglyassociatedwiththebehavior-basedapproach[31],butarenotintrinsictomultiple-robotsystemsandthusliebeyondourpresentscope.AlsonotethatarecentincarnationofCEBOT,whichhasbeenimplementedonphysicalrobots,isbasedonabehavior-basedcontrolarchitecture[34].
Therapidprogressofcooperativeroboticssincethelate1980’shasbeenaninterplayofsystems,theoriesandproblems:
tosolveagivenproblem,systemsareenvisioned,simulatedandbuilt;
theoriesofcooperationarebroughtfromotherfields;
andnewproblemsareidentified(promptingfurthersystemsandtheories).Sincesomuchofthisprogressisrecent,itisnoteasytodiscerndeepintellectualheritagesfromwithinthefield.Moreapparentaretheintellectualheritagesfromotherfields,aswellasthecanonicaltaskdomainswhichhavedrivenresearch.Threeexamplesofthelatterare:
•TrafficControl.Whenmultipleagentsmovewithinacommonenvironment,theytypicallyattempttoavoidcollisions.Fundamentally,thismaybeviewedasaproblemofresourceconflict,whichmayberesolvedbyintroducing,e.g.,trafficrules,priorities,orcommunicationarchitectures.Fromanotherperspective,pathplanningmustbeperformedtakingintocon-siderationotherr