CognitiveandaffectivearchitectureforsocialHuman-RobotInteractions

SocialContextMostrobotsinusetodayhaveanindustrialormilitaryuse.Howeverservicerobotsfordailylifenowchallengethisprofessionalusage.In2012, about 3 millions service robots for personal and domesticusageweresold.Thesesalesmainlyincludevacuumrobotsandlawnmowingrobots.However,anewtypeofrobotsisemerging,aimingatassisting people during their daily life activities. One can think ofassistive robots takingcareofelderlypeople,orpedagogical robotsfor children. More generally these robots are called “companionrobots”becausetheirmainmissionistosupportandassistpeopleintheireveryday lifeactivitiesand tokeep themcompany.Oneof thespecificities of such robots is that they interact more and morecloselywith their humanusers, and their value ismuchmore on social than physicalinteraction [1]. Byclose,wemean that robotsmust sharenotonly thesamephysicalspacebutalsogoalsandbeliefstoachieveacommontaskthroughtheirinteractions.

ResearchchallengesDuringthepastdecades,researchinroboticshasmainlyfocusedonfundamentalskillssuch as robust perception, navigation and catching or moving things. One of thechallengesisnowtoendowourcompanionrobotswithsubtleandsmartabilitiessuchas understanding and reasoning, emotion detection and expression, empatheticbehaviour,... They should also interact intuitively and easily through speech, gestures,andfacialexpressions.

Inspiteof thenumerouscontributions inthefieldofcognitivearchitectures, (seegoodreviewsforexample[11],[12]),mostofthemaregenericandfewcanreallydealwiththe complexity of human-robot interactions (HRI). They are not tailored tomeet thespecificneedsofsocialHRI,suchashandlingemotions,language,socialnorms...

That is why developing a cognitive architecture for social robots able to take intoaccountthecomplexityofinteractionswithhumansstillremainsarealchallenge.Suchan architecture requires various features: emotions, non-verbal aspects of interaction,reactive and deliberative levels (fast emotional answer versus slower and moredeliberateanswer),explicitmanipulationofmentalstates(enablingself-explanation)...

PHDresearchobjectivesThis PHD thesis follows our previous work in three research projects funded by theFrench national research Agency (ANR): ANR CECIL project, ANR MOCA project andANRSOMBREROproject.TheseprojectshavecontributedtotheongoingdevelopmentofourCognitiveandAffectiveInteraction-OrientedarchitecturecalledCAIO(seeBestLateBreakingReportNominationpaper[10]).

ThegoalsofthisPhDthesisarethreefold:

• to fully implement this architecture and experimentation scenarios, using alightweighthumanoidrobot(forinstanceaNaorobot)

• toproposeanevaluationmethodandcriteria, andapply this toexperimentallyevaluatethearchitectureatallstagesofdevelopment

• to identify fromtheseexperimentationsthenecessaryadditionsto improvethesocialcapabilitiesofrobotsendowedwiththisarchitecture,andtorealisesomeofthese,forinstance:

o automaticactivityanalysistolearnsociallyacceptableinteractionmodels;o multimodalperceptionoftheuser,includingtheiremotions;o impactofemotionsonvariouscognitiveprocesses(memory,planning...).

University,laboratory,team• UniversityofGrenoble-Alps:http://www.univ-grenoble-alpes.fr/en/• GrenobleInformaticsLaboratory:https://www.liglab.fr• MAGMAteam:http://magma.imag.fr/

Supervisors• SylviePesty(sylvie.pesty@imag.fr)GrenobleInformaticsdeGrenoble

http://magma.imag.fr/content/sylvie-pesty

Co-Supervisors• CaroleAdam(carole.adam@imag.fr)Laboratoired’InformatiquedeGrenoble

http://magma.imag.fr/content/carole-adam • DamienPellier(damien.pellier@imag.fr)Laboratoired’InformatiquedeGrenoble

http://magma.imag.fr/content/damien-pellier

Bibliography[1] Johnson,D.O.,etal.,Sociallyassistiverobots:acomprehensiveapproachtoextending

independentliving.InternationalJournalofSocialRobotics,2013:p.195-211.

[2] Cabibihan,J.-J.,etal.,WhyRobots?Asurveyontherolesandbenefitsofsocialrobotsinthetherapyofchildrenwithautism.InternationalJournalofSocialRobotics,2013.5(4):p.593-618.

[3] Fasola, J. and M. Mataric, A socially assistive robot exercise coach for the elderly.JournalofHuman-RobotInteraction,2013.2(2):p.3-32.

[4] Mazzoleni, S., et al., Acceptability of robotic technology in neuro-rehabilitation:Preliminary results on chronic stroke patients. Computer Methods and Programs inBiomedicine,2014:p.1-7.

[5] Tencé,F.,etal.,Stablegrowingneuralgas:Atopologylearningalgorithmbasedonplayertrackinginvideogames.Appl.SoftComput.,2013.13(10):p.4174-4184.

[6] Rivière,D.,C.Adam,andS.Pesty.AreasoningmoduletoselectECA'scommunicativeintention. in International Conference on Intelligent Virtual Agents (IVA). 2012.SantaCruz,CA.p.447-454.

[7] Scherer, K.R. and H. Ellgring, Are facial expressions of emotion produced bycategoricalaffectprogramsordynamicallydrivenbyappraisal?Emotion,2007.7:p.113-130.

[8] Dulac,A.,etal.Learningusefulmacro-actionsforplanningwithN-Grams. in IEEEInternationalConferenceonToolswithArtificialIntelligence(ICTAI). 2013.Washington,DC.[9] G.Milliez,M.Warnier,A.Clodic,R.Alami.Aframeworkforendowinganinteractiverobotwith reasoning capabilities about perspective-taking and beliefmanagement. The23rd IEEE International Symposium on Robot and Human Interactive Communication,2014RO-MAN,2014.[10] W. Johal, D. Pellier, C. Adam, H. Fiorino, and S. Pesty.ACognitive andAffectiveArchitecture for Social Human-Robot Interaction. In Proceedings of the Tenth AnnualACM/IEEE International Conference on Human-Robot Interaction, Extended Abstracts(HRI'15ExtendedAbstracts).ACM,NewYork,NY,USA,71-72,2015.

[11]H.-Q.Chong,A.-H.Tan,andG.-W.Ng.Integratedcognitivearchitectures:asurvey.In:ArtificialIntelligenceReview,vol.28(2),pp.103–130,Feb.2009.

[12]Thòrisson,KristinnandHelgasson,Helgi.CognitiveArchitecturesandAutonomy:AComparativeReview.In:JournalofArtificialGeneralIntelligence,vol3,pp.1—30,2012.