20Towards_Affect-sensitive_Assistive_Intervention_

20TowardsAffect-sensitiveAssistiveInterventionTechnologiesforChildrenwithAutismKarlaConn,ChangchunLiu,NilanjanSarkar,WendyStoneandZacharyWarrenVanderbiltUniversityUSA1.IntroductionInvestigationintotechnology-assistedinterventionforchildrenwithautismspectrumdisorder(ASD)hasgainedmomentuminrecentyears.TherapistsinvolvedininterventionsmustovercomethecommunicationimpairmentsgenerallyexhibitedbychildrenwithASDbyadeptlyinferringtheaffectivecuesofthechildrentoadjusttheinterventionaccordingly.Similarly,anintelligentsystem,suchasacomputerorrobot,mustalsobeabletounderstandtheaffectiveneedsofthesechildren-anabilitythatthecurrenttechnology-assistedASDinterventionsystemslack-toachieveeffectiveinteractionthataddressestheroleofaffectivestatesinhuman-computerinteraction(HCI),human-robotinteraction(HRI),andinterventionpractice.Inthischapterwepresentaphysiology-basedaffect-inferencemechanismforemotionmodeling,emotionrecognition,andemotion-sensitiveadaptiveresponseintechnology-assistedintervention.Thisworkisthefirststeptowardsdeveloping“understanding”interactivetechnologiesforuseinfutureASDintervention.Weaddresstheproblemofhowtomakecomputer-basedASDinterventiontoolsaffect-sensitivebydesigningtherapist-likeaffectivemodelsofthechildrenwithASDbasedontheirphysiologicalresponses.Byemployingthesemodels,weexplainhowarobotcandetecttheaffectivestatesofachildwithASDandadaptitsbehaviorsaccordingly.Experimentalresultswith6childrenwithASDfromcomputer-basedcognitivetasksandaproof-of-conceptexperiment(i.e.,arobot-basedbasketballgame)arepresented.ASupportVectorMachines(SVM)basedaffectivemodelyieldedapproximately82.9%successforpredictingaffectinferredfromatherapist.Therobotlearnedtheindividuallikinglevelofeachchildwithregardtothegameconfigurationandselectedappropriatebehaviorstopresentthetaskathis/herpreferredlikinglevel.Resultsshowtherobotautomaticallypredictedindividuallikinglevelinrealtimewith81.1%accuracy.Thisisthefirsttime,toourknowledge,thattheaffectivestatesofchildrenwithASDhavebeendetectedviaaphysiology-basedaffectrecognitiontechniqueinrealtime.Thisisalsothefirsttimethattheimpactofaffect-sensitiveclosed-loopinteractionbetweenarobotandachildwithASDhasbeendemonstratedexperimentally.Whilethereisatpresentnosingleacceptedintervention,treatment,orknowncureforASD,thereisgrowingconsensusthatintensivebehavioralandeducationalinterventionprogramscansignificantlyimprovelongtermoutcomesforindividualsandtheirfamilies(Rogers,AffectiveComputing,FocusonEmotionExpression,SynthesisandRecognition3661998).Despitetheurgentneedandsocietalimportofintensivetreatment(Rutter,2006),appropriateinterventionresourcesforchildrenwithASDandtheirfamiliesareoftenextremelycostlywhenaccessible(Jacobsonetal.,1998;Tarkan,2002).Therefore,animportantnewdirectionforresearchonASDistheidentificationanddevelopmentofassistiveinterventiontoolsthatcanmakeapplicationofintensivetreatmentmorereadilyaccessible.Inresponsetothisneed,agrowingnumberofstudieshavebeeninvestigatingtheapplicationofadvancedinteractivetechnologiestoaddresscoredeficitsrelatedtoautism,namelycomputertechnology(Bernard-Opitzetal.,2001),virtualrealityenvironments(Paresetal.,2005;Parsons&Mitchell,2002),androboticsystems(Dautenhahn&Werry,2004;Kozimaetal,2005;Michaud&Theberge-Turmel,2002;Pioggiaetal.,2005;Scassellati,2005).InitialresultsindicatethatsuchtechnologiesmayholdpromiseforrehabilitationofchildrenwithASD.Computerandvirtualreality(VR)basedinterventionmayprovideasimplifiedbutexploratoryinteractionenvironmentforchildrenwithASD(Mooreetal.,2000;Parsons&Mitchell,2002).VarioussoftwarepackagesandVRenvironmentshavebeendevelopedandappliedtoaddressspecificdeficitsassociatedwithautism,e.g.,understandingoffalsebelief(Swettenham,1996),attention(Trepagnieretal.,2006),expressionrecognition(Silver&Oakes,2001),andsocialcommunication(Bernard-Opitzetal.,2001;Parsonsetal.,2005).ResearchsuggestedthatrobotscanallowsimplifiedbutembodiedsocialinteractionthatislessintimidatingorconfusingforchildrenwithASD(Dautenhahn&Werry,2004).Michaud&Theberge-Turmel(2002)investigatedtheimpactofrobotdesignontheinteractionswithchildrenandemphasizedthatsystemsneedtobeversatileenoughtoadapttothevaryingneedsofdifferentchildren.Pioggiaetal.(2005)developedaninteractivelife-likefacialdisplaysystemforenhancingemotionrecognitioninindividualswithASD.Robotshavealsobeenusedtoteachbasicsocialinteractionskillsusingturn-takingandimitationgames,andtheuseofrobotsassocialmediatorsandasobjectsofsharedattentioncanencourageinteractionwithpeersandadults(Dautenhahn&Werry,2004;Kozimaetal,2005).Interactivetechnologiesposetheadvantageoffurnishingrobustsystemsthatcansupportmultimodalinteractionandprovidearepeatable,standardizedstimuluswhilequantitativelyrecordingandmonitoringtheperformanceprogressofthechildrenwithASDtoassesstheinterventionapproaches(Scassellati,2005).Byemployinghuman-computerinteraction(HCI)andhuman-robotinteraction(HRI)technologies,interactivetherapeutictoolscanpartiallyautomatethetime-consuming,routinebehavioraltherapysessionsandmayallowintensiveinterventiontobe