Controlling-Individual-Agents-in-High-Density-Crow

Copyright©2007bytheAssociationforComputingMachinery,Inc.Permissiontomakedigitalorhardcopiesofpartorallofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthefirstpage.CopyrightsforcomponentsofthisworkownedbyothersthanACMmustbehonored.Abstractingwithcreditispermitted.Tocopyotherwise,torepublish,topostonservers,ortoredistributetolists,requirespriorspecificpermissionand/orafee.RequestpermissionsfromPermissionsDept,ACMInc.,fax+1(212)[e-mail]tion2007,SanDiego,CA,August03-04,2007.©2007ACM978-1-59593-624-4/07/0008$5.00Eurographics/ACMSIGGRAPHSymposiumonComputerAnimation(2007)D.MetaxasandJ.Popovic(Editors)ControllingIndividualAgentsinHigh-DensityCrowdSimulationN.Pelechano,J.M.AllbeckandN.I.BadlerUniversityofPennsylvania,USAAbstractSimulatingthemotionofrealistic,large,densecrowdsofautonomousagentsisstillachallengeforthecomputergraphicscommunity.Typicalapproacheseitherresembleparticlesimulations(whereagentslackorientationcontrols)orareconservativeintherangeofhumanmotionpossible(agentslackpsychologicalstateandaren’tallowedto‘push’eachother).OurHiDACsystem(forHigh-DensityAutonomousCrowds)focusesontheproblemofsimulatingthelocalmotionandglobalwayfindingbehaviorsofcrowdsmovinginanaturalmannerwithindynamicallychangingvirtualenvironments.Byapplyingacombinationofpsychologicalandgeometricalruleswithasocialandphysicalforcesmodel,HiDACexhibitsawidevarietyofemergentbehaviorsfromagentlineformationtopushingbehavioranditsconsequences;relativetothecurrentsituation,personalitiesoftheindividualsandperceivedsocialdensity.CRCategories:I.3.7[ComputerGraphics]:Three-DimensionalGraphicsandRealism—Animation;I.6.8[SimulationandModeling]:TypesofSimulation—Animation1IntroductionAnimatingmotionforlargecrowdshasbeenanimportantgoalinthecomputergraphics,movieandvideogamescommunities.Therehasbeenaconsiderableeffortonlocomotion,pathplanning,navigationinlargevirtualenvironments,andrealisticbehaviorsimulationusingcognitivemodels.Weclassifycrowdagentmotionsbythreemainapproaches:socialforcesmodels,rulebasedmodelsandcellularautomatamodels.Althoughmuchefforthasgoneintoimprovingthebehavioralrealismofeachoftheseapproaches,noneofthecurrentmodelscanrealisticallyanimatehigh-densitycrowds.Socialforcesmodelstendtocreatesimulationsthatlookmorelikeparticleanimationthanhumanmovement.Cellularautomatamodelslimitagentspatialmovementsandtendtoexposetheunderlyingcheckerboardofcellswhencrowddensityishigh.Finally,rulebasedmodelseitherdon’tconsidercollisiondetectionandrepulsionatalloradoptveryconservativeapproachesthroughtheuseofwaitingrules,whichworkfineforlowdensitiesineverydaylifesimulation,butlackrealismforhigh-densityorpanicsituations.Figure1showsataxonomyforcrowdsimulationandcomparesourmodel(HiDAC:High-DensityAutonomousCrowds)withthemainmodelsintheliteraturealongthedimensionsofanimationrealismandcrowddensity.HiDACaddressestheproblemofsimulatinghigh-densitycrowdsofautonomousagentsmovinginanaturalmannerindynamicallychangingvirtualenvironments.Oursolutiontotheproblemofrealisticallysimulatinglocalmotionunderdifferentsituationsandagentpersonalitiesusespsychological,physiologicalandgeometricalrulescombinedwithphysicalforces.SinceapplyingthesameN.Pelechano,J.M.Allbeck&N.I.Badler/ControllingIndividualAgentsinHigh-DensityCrowdSimulation©AssociationforComputingMachinery,Inc.2007.rulestoallagentsleadstohomogeneousbehavior,agentsaregivendifferentpsychological(e.g.,impatience,panic,personalityattributes)andphysiological(e.g.,locomotion,energylevel)traitsthattriggerindividualheterogeneousbehaviors.Eachagentisalsoendowedwithperceptionandreactstostaticanddynamicobjectsandotheragentswithinthenearbyspace.Figure1:Currentmodelsframeworkandourapproachforlow-levelmotion(HiDAC).Realisticmovementmaybedefinedastheemergenceofcrowdbehaviorsconsistentwithrealobservedcrowds,andappropriateindividualcollisionavoidanceandcollisionresponse.Weachievesuchrealismthroughcontextualapplicationofphysicalandgeometricalgorithms.Overlongerdistancestangentialforcesgentlysteeragentsaroundobstacles,whileovershorterdistancescollisionresponseisappliedtoavoidoverlapping.Pushingbehaviorbetweenagentsarisesfromvaryingthelong/shortpersonalspacethresholdofeachindividual.Agentsinahurrywillnotrespectothers’personalspaceandwillappeartopushtheirwaythroughthecrowd.Incontrast,more‘polite’agentswillrespectlinesandwaitforotherstomovefirst.Eachagenthasaninfluencedisk(region)infrontofthemthattriggerswaitingbehavior.Relaxedagentstemporarilystopwhenanotheragentmovesintotheirpath,whileimpatientagentsdonotrespondtothisfeedbackandtendto‘push’.Ourmodelstopsimpatientagentsfromappearingto‘vibrate’astheytrytoforcetheirwaythroughdensecrowds:weaddtemporalstoppingstatestopreventtheagentfromtryingtomoveduringashortintervaloftimealthoughitcanstillbepushedbyothers.Ouragents’behaviorisdeterminedbyahigh-levelalgorithm(including:navigationincomplexvirtualenvironments,learning,communicat