搜索
AUTONOMOUS'NAVIGATIONOFPLANETARYROVERS:AFUZZYLOGICAPPROACHHomayounSerajiBruceBonJetPropulsionLaboratoryCaIiforniaInstituteofTechnologyPasadena,CA91109,USAAbstractThispaperpresentsanewstrategyforautonomousnavigationofplanetaryroversusingthefuzzylogicframeworkandanovelon-boardmeasureofterraintraversabil-ity.Thenavigationstrategyiscomprisedofthreesimple,independentbehaviorswithdifferentlevelsofresolution.Thenavigationrulesforthefirstbehavior,goal-seeking,utilizetheglobalinformationaboutthegoalpositiontogeneratethesteeringandspeedcommandsthatdrivetherovertothedesignateddestination.Thenavigationrulesforthesecondbehavior,terrain-traversing,usetheregionalinformationabouttheterrainqualitytoproducesteeringandspeedcommandsthatguidetherovertowardthesafestandthemosttraversableterrain.Theinclusionoftheregionalterraindata(suchasslopeandroughness)intherovernavigationstrategyisamajorcontributionofthispaper.Thenavigationrulesforthethirdbehavior,collision-avoidance,employthelocalinformationabouttheen-routeobstaclestodevelopsteeringandspeedcommandsthatmaneuvertheroveraroundtheencounteredobstacles.Therecommendationsofthesethreebehaviorsarethenintegratedthroughappropriateweightingfactorstogeneratethefinalcontrolactionsforthesteeringandspeedcommandsthatareexecutedbytherover.Theweightingfactorsareproducedbyfuzzyrulesthattakeintoaccountthecurrentstatusoftherover.Thecompleterovernavigationstrategyconsistsofatotalof37fuzzylogicrulesforthebehaviorsandtheirweightingfactors.Thisnavigationstrategyrequiresnoaprioriinformationabouttheenvironment,andusestheon-boardtraversabilityanalysistoenabletherovertoselecteasy-to-traversepathsautonomously.TheRoverGraphicalSimulatordevelopedatJPLfortestandvalidationofthenaviga-tionrules,aswellasforgraphicalvisualizationoftherovermotion,isdescribed.Threegraphicalsimulationcasestudiesarepresentedtodemonstratethecapabilitiesoftheproposednavigationstrategyforplanetaryrovers.Finally,thenavigationalgorithmfortheSojournerroverisdiscussedandcomparedwiththeproposedstrategy.SimulationstudiesclearlydemonstratethesuperiorperformanceofthisfuzzynavigationstrategyrelativetotheSojourneralgorithm.11IntroductionIndealingwithday-to-dayprocesses,humansmakesubjectivedecisionsbasedonqualitativeinformation.Theirperceptionofprocessesisbasedonqualitative,ratherthanquantitative,assessmentsobtainedfromlowqualityandapproximatemeasurements.Thehumancontrolstrategyforaprocesstypicallyconsistsofsimple,intuitiverulesbasedonpriorexperiencethatarebroughttobeartoaffecttheprocess.Forinstance,intheprocessofdrivingacar,thehumandriverturnsthesteeringwheeltotherightifthecarveerstoofartotheleft,andviceversa.Thedriverdeterminesthedegreeofcoursecorrectionintuitivelyandbasedonhisdrivingexperience,ratherthanrelyingonmathematicalmodelingandformulationoftheprocess.Hisactionsarebasedonhowfarfromthelanethecarhasmoved,andonhowfastthecarismoving.Similarly,thedriveradjuststhespeedofthecarbasedonhissubjectivejudgmentoftheroadconditions.Itishighlydesirabletocapturetheexpertiseofthehumandriverandtoutilizethisknowledgetodevelopautonomousnavigationstrategiesforplanetaryrovers.Fuzzylogicprovidesameanstoaccomplishthisgoal.Fuzzylogicpro-videsaformalmethodologyforrepresentingandimplementingthehumanexpert’sheuristicknowledgeandoperationalexperience.Usingthefuzzylogicframework,theattributesofthehumanreasoninganddecisionmakingcanbeformulatedbysimpleIF(antecedent)-THEN(consequent)rulesandeasily-understandablelinguisticrepresentations.Thelinguis-ticvaluesintheruleantecedentsconveytheimprecisionassociatedwithon-boardsensormeasurements;whilethoseintheruleconsequentsrepresentthevaguenessinherentinthereasoningprocesses.Theoperationalstrategiesofthehumanexpertdrivercanbetransferredviafuzzylogictoolstotherovernavigationstrategyin.theformofasetofsimpleconditionalstatementscomposedoflinguisticvariables.Theselinguisticvariablesaredefinedbyfuzzysetsinaccordancewithuser-definedmembershipfunctions.Themainadvantagesofafuzzynavigationstrategylieintheabilitytoextractheuristicrulesfromhumanexperienceandtoobviatetheneedforananalyticalmodeloftheprocess.Thenaturalappealoffuzzylogictorovernavigationhasmotivatedconsiderableresearchinthisareainrecentyears[see,e.g.,1-30].Mostofthisresearch,however,hasbeen-focusedonindoormobilerobotsthatoperateinhighlystructured,orotherwise,man-madeenvironments.NASAhasplannedanambitioussetofmissionstolaunchaseriesofspacecrafttolandontheplanetMars.Theseunmannedmissionswillcarrymobilerobots(rovers)toexploretheMartiansurfaceandtocarryoutsciencemissionsbyplacinginstrumentsonorsamplingfromsoilandrocksforanalysis.Thefirstinthisseries,theMarsPathfindermissionthatlandedonMarsinJuly1997,deployedtheSojournerroverontheMartianterrainforpositioningscienceinstrumentsagainstdesignatedrocks,asshowninFigure1.AfterthesuccessoftheSojournerrover,therehasbeenastrongmotivationtodevelopfutureplanetaryroverswithenhancedcapabilitiesthatcanexploreremoteplanetsautonomouslyandintelligentlywithminimalhumanintervention.Theautonomyandon-boardintelligenceareparticularlyimportantforMarsmissi