USF Libraries
USF Digital Collections

Robots without faces

MISSING IMAGE

Material Information

Title:
Robots without faces non-verbal social human-robot interaction
Physical Description:
Book
Language:
English
Creator:
Bethel, Cindy L
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Robotics
Appearance-constrained robots
Affective robotics
Psychophysiology
Experimental design
Dissertations, Academic -- Computer Science and Engineering -- Doctoral -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Non-facial and non-verbal methods of affective expression are essential for naturalistic social interaction in robots that are designed to be functional and lack expressive faces (appearance-constrained)such as those used in search and rescue, law enforcement, and military applications. This research identifies five main methods of non-facial and non-verbal affective expression (body movement, posture, orientation, color, and sound). From the psychology, computer science, and robotics literature a set of prescriptive recommendations was distilled for the appropriate non-facial and non-verbal affective expression methods for each of three proximity zones of interest(intimate: contact - 0.46 m, personal: 0.46 - 1.22 m, and social: 1.22 - 3.66 m). These recommendations serve as design guidelines for adding retroactively affective expression through software with minimal or no physical modifications to a robot or designing a new robot.This benefits both the human-robot interaction (HRI) and robotics communities. A large-scale, complex human-robot study was conducted to verify these design guidelines using 128 participants, and four methods of evaluation (self-assessments, psychophysiological measures, behavioral observations, and structured interviews) for convergent validity. The study was conducted in a high-fidelity, confined-space simulated disaster site with all robot interactions performed in the dark. This research investigated whether the use of non-facial and non-verbal affective expression provided a mechanism for naturalistic social interaction between a functional, appearance-constrained robot and the human with which it interacted. As part of this research study, the valence and arousal dimensions of the Self-Assessment Manikin (SAM) were validated for use as an assessment tool for future HRI human-robot studies.Also presented is a set of practical recommendations for designing, planning, and executing a successful, large-scale complex human-robot study using appropriate sample sizes and multiple methods of evaluation for validity and reliability in HRI studies. As evidenced by the results, humans were calmer with robots that exhibited non-facial and non-verbal affective expressions for social human-robot interactions in urban search and rescue applications. The results also indicated that humans calibrated their responses to robots based on their first robot encounter.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2009.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Cindy L. Bethel.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 165 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002064177
oclc - 567688091
usfldc doi - E14-SFE0003097
usfldc handle - e14.3097
System ID:
SFS0027413:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200409Ka 4500
controlfield tag 001 002064177
005 20100323135635.0
007 cr mnu|||uuuuu
008 100323s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0003097
035
(OCoLC)567688091
040
FHM
c FHM
049
FHMM
090
TK7885 (Online)
1 100
Bethel, Cindy L.
0 245
Robots without faces :
b non-verbal social human-robot interaction
h [electronic resource] /
by Cindy L. Bethel.
260
[Tampa, Fla] :
University of South Florida,
2009.
500
Title from PDF of title page.
Document formatted into pages; contains 165 pages.
Includes vita.
502
Dissertation (Ph.D.)--University of South Florida, 2009.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
520
ABSTRACT: Non-facial and non-verbal methods of affective expression are essential for naturalistic social interaction in robots that are designed to be functional and lack expressive faces (appearance-constrained)such as those used in search and rescue, law enforcement, and military applications. This research identifies five main methods of non-facial and non-verbal affective expression (body movement, posture, orientation, color, and sound). From the psychology, computer science, and robotics literature a set of prescriptive recommendations was distilled for the appropriate non-facial and non-verbal affective expression methods for each of three proximity zones of interest(intimate: contact 0.46 m, personal: 0.46 1.22 m, and social: 1.22 3.66 m). These recommendations serve as design guidelines for adding retroactively affective expression through software with minimal or no physical modifications to a robot or designing a new robot.This benefits both the human-robot interaction (HRI) and robotics communities. A large-scale, complex human-robot study was conducted to verify these design guidelines using 128 participants, and four methods of evaluation (self-assessments, psychophysiological measures, behavioral observations, and structured interviews) for convergent validity. The study was conducted in a high-fidelity, confined-space simulated disaster site with all robot interactions performed in the dark. This research investigated whether the use of non-facial and non-verbal affective expression provided a mechanism for naturalistic social interaction between a functional, appearance-constrained robot and the human with which it interacted. As part of this research study, the valence and arousal dimensions of the Self-Assessment Manikin (SAM) were validated for use as an assessment tool for future HRI human-robot studies.Also presented is a set of practical recommendations for designing, planning, and executing a successful, large-scale complex human-robot study using appropriate sample sizes and multiple methods of evaluation for validity and reliability in HRI studies. As evidenced by the results, humans were calmer with robots that exhibited non-facial and non-verbal affective expressions for social human-robot interactions in urban search and rescue applications. The results also indicated that humans calibrated their responses to robots based on their first robot encounter.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Co-advisor: Robin R. Murphy, Ph.D.
Co-advisor: Lawrence O. Hall, Ph.D.
653
Robotics
Appearance-constrained robots
Affective robotics
Psychophysiology
Experimental design
690
Dissertations, Academic
z USF
x Computer Science and Engineering
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.3097



PAGE 1

RobotsWithoutFaces:Non-VerbalSocialHuman-RobotInteraction by CindyL.Bethel Adissertationsubmittedinpartialfulllment oftherequirementsforthedegreeof DoctorofPhilosophy Departmentof ComputerScienceandEngineering Collegeof Engineering UniversityofSouthFlorida Co-MajorProfessor: RobinR.Murphy,Ph.D. Co-MajorProfessor: LawrenceO.Hall,Ph.D. JenniferL.Burke,Ph.D. JohnFerron,Ph.D. JodiForlizzi,Ph.D. DeweyRundus,Ph.D KristenSalomon,Ph.D. DateofApproval: June17,2009 Keywords:robotics,appearance-constrainedrobots,aectiverobotics,psychophysiology, experimentaldesign c Copyright 2009 ,CindyL.Bethel

PAGE 2

Dedication Thisworkisdedicatedtomyparents,EdandLindaBethel,mybrothersChristopherand CameronBethel,JudyHyde,andtoallmyfamilyandfriendswhosecontinuedsupportandencouragementmadeitpossibleformetoaccomplishmygoals.Thankyousoverymuch!

PAGE 3

Acknowledgments ThisworkissupportedinpartunderaNationalScienceFoundationGraduateResearchFellowshipAwardNumberDGE-0135733,ARLNumberW911NF-06-2-0041,IEEERoboticsand AutomationSocietyGraduateFellowship,andtheMicrosoftRescueBuddyProject.Special thanksgotomydissertationcommitteeRobinR.Murphy,LarryO.Hall,JenniferL.Burke,John Ferron,JodiForlizzi,DeweyRundus,andKristenSalomon,andmyvolunteerresearchassistants BrianDay,ChristineBringes,MeganBrunner,AndreaVera,LeslieSalas,StephanieSmith,KimberleeFraser,CherisseBraithwaite,andCaitlinHowellfortheircontributionsandassistance.

PAGE 4

NotetoReader Theoriginalofthisdocumentcontainscolorthatisnecessaryforunderstandingthedata.The originaldissertationisonlewiththeUniversityofSouthFloridalibraryinTampa,Florida.

PAGE 5

TableofContents ListofTables vi ListofFiguresviii Abstract xi Chapter1Introduction1 1.1ResearchQuestion2 1.2WhyNon-FacialandNon-VerbalAffectiveExpression?3 1.3Non-FacialandNon-VerbalMethodsofAffectiveExpressionasAppliedtoRobot-AssistedVictimAssessment8 1.4PsychophysiologicalMeasurementsAppliedtoHuman-RobotInteraction9 1.5Contributions12 1.6OrganizationoftheDissertation13 Chapter2RelatedWork14 2.1EmotionTheoryandModels14 2.2FactorsinAffectiveExpression16 2.2.1PresentationMethods16 2.2.1.1BodyMovement,Posture,andOrientation17 2.2.1.2Color20 2.2.1.3Sound21 2.2.2Proxemics23 2.3RobotImplementations25 2.3.1Non-AnthropomorphicandAppearance-ConstrainedRobotsUsing Non-FacialandNon-VerbalAffectiveExpression27 2.3.2AnthropomorphicRobotsThatRelyHeavilyonNon-FacialandNonVerbalAffectiveExpression28 2.3.3AnthropomorphicRobotsUsingNon-FacialandNon-VerbalAffective ExpressionasaRedundantMethodofExpression30 2.4ReviewofPsychophysiology31 2.4.1BasicActivitiesandResponseTendencies31 2.4.2CommonPsychophysiologicalMeasures32 2.4.3AdvantagesofUsingPsychophysiology33 2.4.4PsychophysiologyIssues33 2.4.5HabituationandResponseFactors34 2.5ImplementationsUsingPsychophysiologicalMeasuresinHRI35 i

PAGE 6

2.5.1ImplementationsUsingPsychophysiologicalMeasuresforEmotion Detectionand/orIdentication35 2.5.2ImplementationsUsingPsychophysiologicalMeasuresforEvaluationof ParticipantReactionstoTechnology39 2.5.3ImplementationsUsingPsychophysiologicalMeasuresforReal-Time RobotControlandBehaviorModications40 2.6Summary41 Chapter3TheoryandApproach42 3.1DesignApproaches42 3.2PrescriptiveRecommendationsforNon-FacialandNon-VerbalAffectiveExpressionbyProximityZones44 3.3JusticationforthePrescriptiveRecommendations44 3.4PrescriptiveRecommendationsAppliedtoVictimAssistance47 3.5VericationofthePrescriptiveRecommendationsThroughVideoObservations UsingaPathforMedicalAssessmentofaVictim49 3.5.1MedicalAssessmentPath49 3.5.2SocialProximityZone51 3.5.3PersonalProximityZone51 3.5.4IntimateProximityZone53 3.6SocialInteractionswithAppearance-ConstrainedRobotswithSoldiers54 3.7DesignRecommendationsandtheDevelopmentofNon-FacialandNon-Verbal AffectiveExpressionsforUseinRoboticApplications55 3.8ConstraintstotheApproach56 3.9Summary56 Chapter4ExperimentalMethodsandDesign58 4.1StudyOverview58 4.2Participants60 4.3PhysiologicalRecordingApparatus65 4.4Self-AssessmentsandInterviewMeasures66 4.4.1Pre-InteractionMeasures66 4.4.2InteractionMeasures68 4.4.3Post-InteractionMeasures70 4.5SiteRequirements70 4.6ExperimentalDesign71 4.7StudyProtocol71 4.8Personnel75 4.9Summary76 Chapter5DataAnalysesandResults77 5.1DataAnalysesandResultsBackgroundInformation77 5.2Self-AssessmentManikinAnalysesandResults79 5.2.1CorrelationoftheTwoSelf-AssessmentManikinAssessments80 5.2.2MultivariateTestsforAssumptionsbyOperatingModeandRobotOrder82 ii

PAGE 7

5.2.3TheSAMValenceAnalysesandResults83 5.2.3.1DescriptiveStatisticalAnalysisforSAMValenceRecodedby OperatingModeandRobotOrder83 5.2.3.2InferentialStatisticalAnalysesforSAMValenceRecodedby OperatingModeandRobotOrder86 5.2.4TheSAMArousalAnalysesandResults88 5.2.4.1DescriptiveStatisticalAnalysisforSAMArousalbyOperatingModeandRobotOrder89 5.2.4.2InferentialStatisticalAnalysesforSAMArousalbyOperatingModeandRobotOrder91 5.3PsychophysiologyAnalysesandResults93 5.3.1HeartRateChangeAnalysesandResults94 5.3.1.1DescriptiveStatisticalAnalysisforMultivariateHeartRate ChangebyOperatingModeandRobotOrder94 5.3.1.2InferentialStatisticalAnalysesforMultivariateHeartRate ChangebyOperatingModeandRobotOrder97 5.3.1.3DescriptiveStatisticalAnalysesforUnivariateHeartRate ChangebyOperatingModeandRobotOrder99 5.3.1.4InferentialStatisticalAnalysesforUnivariateHeartRate ChangebyOperatingModeandRobotOrder100 5.3.2RespirationAmplitudeChangeAnalysesandResults100 5.3.2.1DescriptiveStatisticalAnalysisforMultivariateRespiration AmplitudeChangebyOperatingModeandRobotOrder101 5.3.2.2InferentialStatisticalAnalysesforMultivariateRespiration AmplitudeChangebyOperatingModeandRobotOrder103 5.3.2.3DescriptiveStatisticalAnalysesforUnivariateRespiration AmplitudeChangebyOperatingModeandRobotOrder103 5.3.2.4InferentialStatisticalAnalysesforUnivariateRespiration AmplitudeChangebyOperatingModeandRobotOrder104 5.3.3RespirationRateChangeAnalysesandResults104 5.3.3.1DescriptiveStatisticalAnalysisforMultivariateRespiration RateChangebyOperatingModeandRobotOrder105 5.3.3.2InferentialStatisticalAnalysesforMultivariateRespiration RateChangebyOperatingModeandRobotOrder108 5.3.3.3DescriptiveStatisticalAnalysesforUnivariateRespiration RateChangebyOperatingModeandRobotOrder111 5.3.3.4InferentialStatisticalAnalysesforUnivariateRespiration RateChangebyOperatingModeandRobotOrder112 5.3.4SkinConductanceLevelChangeAnalysesandResults113 5.3.4.1DescriptiveStatisticalAnalysisforMultivariateSkinConductanceLevelChangebyOperatingModeandRobotOrder113 5.3.4.2InferentialStatisticalAnalysesforMultivariateSkinConductanceLevelChangebyOperatingModeandRobotOrder116 5.3.4.3DescriptiveStatisticalAnalysesforUnivariateSkinConductanceLevelChangebyOperatingModeandRobotOrder116 iii

PAGE 8

5.3.4.4InferentialStatisticalAnalysesforUnivariateSkinConductanceLevelChangebyOperatingModeandRobotOrder117 5.4Self-AssessmentManikinSAMandPsychophysiologyCorrelationAnalysis andResults117 5.5Summary120 Chapter6Discussion121 6.1DiscussionoftheSelf-AssessmentManikinSAMResults121 6.1.1InterpretationandDiscussionoftheResultsforValence-RecodedResponsesbyRobot,OperatingMode,andRobotOrder122 6.1.2InterpretationandDiscussionoftheResultsforArousalResponsesby Robot,OperatingMode,andRobotOrder123 6.2DiscussionofthePsychophysiologyResults124 6.2.1InterpretationandDiscussionoftheResultsforHeartRateMeanDifferencesbyRobot,OperatingMode,andRobotOrder125 6.2.2InterpretationandDiscussionoftheResultsforRespirationAmplitude MeanDifferencesbyRobot,OperatingMode,andRobotOrder125 6.2.3InterpretationandDiscussionoftheResultsforRespirationRateMean DifferencesbyRobot,OperatingMode,andRobotOrder125 6.2.4InterpretationandDiscussionoftheResultsforSkinConductanceLevel MeanDifferencesbyRobot,OperatingMode,andRobotOrder127 6.3DiscussionoftheCorrelationAnalysisfortheSelf-AssessmentManikinand PsychophysiologyResults127 6.4DiscussionofFactorsthatmayhaveImpactedtheResults127 6.4.1InuenceofParticipantsFeelingTooSafe128 6.4.2TheEffectofParticipantsPlacedinthePronePosition129 6.4.3TheWALL-EEffect129 6.5FutureStudyImprovements131 6.5.1SuggestedStudyDesignImprovements132 6.5.2SuggestedSiteDesignImprovements133 6.6AppropriateStudyDesign134 6.7OpenResearchQuestionsResultingfromtheStudy137 6.8Summary138 Chapter7ConclusionsandFutureWork139 7.1SignicantConclusionsandContributions139 7.1.1ImportanceofNon-FacialandNon-VerbalAffectiveExpressionfor SocialHuman-RobotInteraction140 7.1.2TheCalibrationEffectforFirstRobotEncounters140 7.1.3PrescriptiveRecommendationsforNon-FacialandNon-VerbalAffectiveExpressionBasedonProximityZone141 7.1.4ValidationoftheSelf-AssessmentManikinSAMasanAssessment ToolforHuman-RobotInteractionStudies141 7.1.5RecommendationsforDesigning,Planning,andExecutingHuman-RobotInteractionStudies142 7.1.5.1ExperimentalDesignRecommendations142 iv

PAGE 9

7.1.5.2RecommendationsforStudyExecution 144 7.1.5.3OtherRecommendations 145 7.1.6ArchivalDataSet 145 7.2FutureWork 146 7.2.1ImmediateFutureResearchGoals 146 7.2.2Long-TermFutureResearchGoals 146 7.3Summary 147 References 149 Appendices 158 AppendixACopyrightPermissions 159 AppendixBIRBLettersforHumanSubjectsClearance 161 AbouttheAuthor EndPage v

PAGE 10

ListofTables Table1Meansandstandarddeviationsforparticipants'demographics.60 Table2Genderdemographicsforparticipants.61 Table3Agedemographicsforparticipants.61 Table4Educationdemographicsforparticipants.62 Table5Ethnicitydemographicsforparticipants.62 Table6Computerexperiencedemographicsforparticipants.63 Table7Robotexperiencedemographicsforparticipants.63 Table8Videogamingexperiencedemographicsforparticipants.63 Table9Robotownershipdemographicsforparticipants.64 Table10Dogownershipdemographicsforparticipants.64 Table11PearsoncorrelationmatrixforthevalencequestionsontheSelf-Assessment ManikinSAMassessments.81 Table12PearsoncorrelationmatrixforthearousalquestionsontheSelf-Assessment ManikinSAMassessments.81 Table13DescriptivestatisticsforthetwodependentvariablesforSAMvalencerecodedbyoperatingmodeandrobotorder.85 Table14DescriptivestatisticsforthetwodependentvariablesforSAMarousalby operatingmodeandrobotorder.89 Table15Descriptivestatisticsforthetwodependentvariablesforheartratechangeby operatingmodeandrobotorder.95 Table16Descriptivestatisticsforthetwodependentvariablesforheartratechangeby operatingmodeandrobotfortherstrobotinteraction.99 Table17Descriptivestatisticsforthetwodependentvariablesforrespirationamplitude changebyoperatingmodeandrobotorder.101 Table18Descriptivestatisticsforthetwodependentvariablesforrespirationamplitude changebyoperatingmodeandrobotfortherstrobotinteraction.104 Table19Descriptivestatisticsforthetwodependentvariablesforrespirationrate changebyoperatingmodeandrobotorder.107 vi

PAGE 11

Table20Descriptivestatisticsforthetwodependentvariablesforrespirationrate changebyoperatingmodeandrobotfortherstrobotinteraction.111 Table21Descriptivestatisticsforthetwodependentvariablesforskinconductance levelchangebyoperatingmodeandrobotorder.114 Table22Descriptivestatisticsforthetwodependentvariablesforskinconductance levelchangebyoperatingmodeandrobotfortherstrobotinteraction.117 Table23PearsoncorrelationmatrixfortheSelf-AssessmentManikinSAMassessmentswiththepsychophysiologydata.119 vii

PAGE 12

ListofFigures Figure1Illustrationoftwoparticpants'responsesandtheirviewsoftherobots.6 Figure2ImagesoftheiRobotPackbotScoutandtheInuktunExtreme-VGTV.7 Figure3IllustrationofArgyle'sfourprimaryproximityzoneswiththezonesofinteresthighlighted.23 Figure4Summaryofrepresentativerobotimplementationsusingnon-facialandnonverbalaffectiveexpressionforsocialhuman-robotinteraction.26 Figure5Prescriptiverecommendationsfornon-facialandnon-verbalaffectiveexpressionbasedonproximity.45 Figure6DiagramofthemedicalassessmentpathtakenbytheInuktunrobotinrelationtothesimulateddisastervictim.49 Figure7ObservationsatpointT4fromtherobot'sandvictim'sperspectivesinthe intimateproximityzone.51 Figure8Psychophysiologysensorplacementdiagrams.65 Figure9Conceptualdrawingoftheexperimentalsiteplanusingthespaceallocatedin ENB223.72 Figure10Actualhigh-delity,simulateddisastersitewithconned-spaceboxand Inuktunrobot.73 Figure11PackBotinarubblepileandInuktuninthesimulateddisastersite.73 Figure12BoxplotforSAM Valence Recoded-Inuktunbyoperatingmodeandrobot order.84 Figure13BoxplotforSAM Valence Recoded-Packbotbyoperatingmodeandrobot order.84 Figure14Statisticallysignicantthree-wayinteractionforSAM Valence Recodedby robot,operatingmode,androbotorder;separatedbyrobotorder.87 Figure15Statisticallysignicantthree-wayinteractionforSAM Valence Recodedby robot,operatingmode,androbotorder;separatedbyoperatingmode.88 Figure16BoxplotforSAM Arousal-Inuktunbyoperatingmodeandrobotorder.90 Figure17BoxplotforSAM Arousal-Packbotbyoperatingmodeandrobotorder.90 Figure18ArousalbyoperatingmodemaineffectfromtheSAMassessments.92 viii

PAGE 13

Figure19ArousalbyrobotordermaineffectfromtheSAMassessments.93 Figure20Boxplotforheartratechange-Inuktunbyoperatingmodeandrobotorder.96 Figure21Boxplotforheartratechange-Packbotbyoperatingmodeandrobotorder.96 Figure22Heartratechangebyrobotandordertwo-wayinteraction.98 Figure23Boxplotforheartratechange-Robot1byoperatingmodeandrobotorder.100 Figure24Boxplotforrespirationamplitudechange-Inuktunbyoperatingmodeand robotorder.102 Figure25Boxplotforrespirationamplitudechange-Packbotbyoperatingmodeand robotorder.102 Figure26Boxplotforrespirationamplitudechange-Robot1byoperatingmodeand robotorder.105 Figure27Boxplotforrespirationratechange-Inuktunbyoperatingmodeandrobot order.106 Figure28Boxplotforrespirationratechange-Packbotbyoperatingmodeandrobot order.106 Figure29Respirationratechangebyoperatingmodeandrobotordertwo-wayinteraction.109 Figure30Respirationratechangebyrobotandrobotordertwo-wayinteraction.110 Figure31Boxplotforrespirationratechange-Robot1byoperatingmodeandrobot order.112 Figure32Boxplotforrespirationratechange-Robot1byoperatingmodeandrobot order.113 Figure33Boxplotforskinconductancelevelchange-Inuktunbyoperatingmodeand robotorder.115 Figure34Boxplotforskinconductancelevelchange-Packbotbyoperatingmodeand robotorder.115 Figure35Boxplotforskinconductancelevelchange-Robot1byoperatingmodeand robotorder.118 Figure36InuktunandPackbotintheinitialsimulateddisastersite.122 Figure37AngledviewoftheInuktunExtreme-VGTVrobot.130 Figure38LetterofpermissionforuseofcopyrightedmaterialfromHalMyers,Ph.D., PresidentofThoughtTechnologyLtd.159 Figure39Electronice-mailpermissionforuseofcopyrightedmaterialfromLawrence Klein,VicePresidentofThoughtTechnologyLtd.160 ix

PAGE 14

Figure40InstitutionalReviewBoardapprovalletterforinitialreviewforIRB106255 page1of2.162 Figure41InstitutionalReviewBoardapprovalletterforinitialreviewforIRB106255 page2of2.163 Figure42InstitutionalReviewBoardapprovalletterforcontinuingreviewforIRB 106255page1of2.164 Figure43InstitutionalReviewBoardapprovalletterforcontinuingreviewforIRB 106255page2of2.165 x

PAGE 15

RobotsWithoutFaces:Non-VerbalSocialHuman-RobotInteraction CindyL.Bethel ABSTRACT Non-facialandnon-verbalmethodsofaffectiveexpressionareessentialfornaturalisticsocialinteractioninrobotsthataredesignedtobefunctionalandlackexpressivefacesappearanceconstrainedsuchasthoseusedinsearchandrescue,lawenforcement,andmilitaryapplications. Thisresearchidentiesvemainmethodsofnon-facialandnon-verbalaffectiveexpressionbody movement,posture,orientation,color,andsound.Fromthepsychology,computerscience,and roboticsliteratureasetofprescriptiverecommendationswasdistilledfortheappropriatenonfacialandnon-verbalaffectiveexpressionmethodsforeachofthreeproximityzonesofinterest intimate:contact0.46m,personal:0.461.22m,andsocial:1.223.66m.Theserecommendationsserveasdesignguidelinesforaddingretroactivelyaffectiveexpressionthroughsoftware withminimalornophysicalmodicationstoarobotordesigninganewrobot.Thisbenetsboth thehuman-robotinteractionHRIandroboticscommunities. Alarge-scale,complexhuman-robotstudywasconductedtoverifythesedesignguidelines using128participants,andfourmethodsofevaluationself-assessments,psychophysiological measures,behavioralobservations,andstructuredinterviewsforconvergentvalidity.Thestudy wasconductedinahigh-delity,conned-spacesimulateddisastersitewithallrobotinteractions performedinthedark.Thisresearchinvestigatedwhethertheuseofnon-facialandnon-verbal affectiveexpressionprovidedamechanismfornaturalisticsocialinteractionbetweenafunctional, appearance-constrainedrobotandthehumanwithwhichitinteracted. Aspartofthisresearchstudy,thevalenceandarousaldimensionsoftheSelf-Assessment ManikinSAMwerevalidatedforuseasanassessmenttoolforfutureHRIhuman-robotstudies.Alsopresentedisasetofpracticalrecommendationsfordesigning,planning,andexecuting xi

PAGE 16

asuccessful,large-scalecomplexhuman-robotstudyusingappropriatesamplesizesandmultiple methodsofevaluationforvalidityandreliabilityinHRIstudies. Asevidencedbytheresults,humanswerecalmerwithrobotsthatexhibitednon-facialand non-verbalaffectiveexpressionsforsocialhuman-robotinteractionsinurbansearchandrescue applications.Theresultsalsoindicatedthathumanscalibratedtheirresponsestorobotsbasedon theirrstrobotencounter. xii

PAGE 17

Chapter1 Introduction Theworldisconstantlyevolvingandchangingandasaresult,advancesintechnologyare essential.Itisinevitablethatrobotswillbecomeapartofourdailylives;thereforeitisimperativethatroboticistsdeterminehowtheserobotswillinteractinasociallyappropriatemannerwith humans.Therehasbeenastrongfocusonbuildingbetterandmorereliablerobots,developing morecomplexbehaviors,andimprovingtheintelligencecapabilitiesoftheserobots;however therealsoneedstobeconcertedeffortputforthtodeterminetheimpactofthesedevelopments onthehumanswhowilleitherbeoperatingorinteractingwiththeserobots.Itisessentialthatthe roboticscommunitydevoteeffortandresourcestoensuresafeandappropriatesocialinteractions withtheserobots.Thiscanbeaccomplishedthroughhuman-robotstudiesusingproperdesignand evaluationmethodstorevealwherethecurrentstateofthetechnologyexcelsandwhereitneeds modicationorimprovement.Itisexpectedthatsoontherewillbearobotineveryhome. Thischapterbeginswiththeprimaryandsecondaryresearchquestionsbeinginvestigatedas partofthisthesis,inSection1.1.ItwilldiscussinSection1.2,theimportanceofusingnon-facial andnon-verbalaffectiveexpression.Section1.3describeshownon-facialandnon-verbalaffectiveexpressioncanbeusedforrobot-assistedvictimassessmentintheurbansearchandrescue domain.InSection1.4detailsareprovidedonhowpsychophysiologymeasurescanbeusedto evaluatehumans'responsestorobots.ContributionsofthisresearcharepresentedinSection1.5. AnoutlineoftheorganizationofthisthesisispresentedinSection1.6. 1

PAGE 18

1.1ResearchQuestion Theprimaryresearchquestionthatthisworkaddressesis: Cantheuseofnon-facialandnon-verbalaffectiveexpressionprovideamechanismfornaturalisticsocialinteractionbetweenafunctional,appearance-constrainedrobotandthehumanwith whichitisinteracting? Appearance-constrainedrobotsareusedindifferentapplications,suchassearchandrescue, lawenforcement,andmilitary.Becauseofthetypesofusesfortheserobots,theyofteninteract withhumansandthereforethedevelopmentofanaturalisticsocialinterfacewouldallowtherobotstocompletetheirtasksinanimprovedmanner.Insearchandrescueapplications,anaturalisticsocialinteractionthroughnon-facialandnon-verbalaffectiveexpressionwillreducestress levelsandkeepvictimscalmuntilassistancecanarrive.Inthecaseofmilitaryandlawenforcementapplications,therobotstendtoattracttheattentionofthosenearby,whichcouldimpedethe operationsoftherobot.Inthosecases,therobotwouldneedtoexhibitanaggressivebehaviorto keeppeopleatadistancesothatitcanbeutilizedinamannerrequiredtoaccomplishthetaskat hand.ThishasbeentruefortherobotsusedinIraqandAfghanistanwherechildrenarefascinated bytherobotsandwanttoapproachthem.Inthatsituationtherobotwouldneedtodisplayanaggressiveaffectiveexpressiontokeepthechildrenawayfromtherobotsotheywillbesafeandthe robotcanaccomplishitstask. Thisresearchquestionbringsupadditionalquestionsthatneedtobeaddressed.Thosequestionsareasfollows: Whatmethodsofnon-facialandnon-verbalaffectiveexpressioncanbeusedforanaturalistic socialinteractioninappearance-constrainedrobots?ThisisaddressedinChapter3through thedevelopmentofasetofprescriptiverecommendationsinwhichbodymovements,postures,orientation,illuminatedcolor,andsoundaretherecommendedmethods. Doesthedistancebetweenthehumanandtherobotwithwhichitisinteractingimpactthe typesandappropriatenessofthenon-facialandnon-verbalmethodofaffectiveexpression utilized?FromtheliteratureandobservationsdiscussedinChapter3,distancecaninuence whichnon-facialandnon-verbalmethodsaremostappropriatetouse.Theseguidelinesare presentedinasetofprescriptiverecommendations. 2

PAGE 19

Isitpossibletoimplementthesenon-facialandnon-verbalmethodsofaffectiveexpressions onanappearance-constrainedrobotwithoutdesigninganewrobotormakingphysicalmodicationsontherobot?TheapproachforthisisdiscussedinChapter3.Throughtheuseof theprescriptiverecommendationsdiscussedinChapter3itispossibletoimplementnon-facialandnon-verbalaffectiveexpressionsthroughsoftwareadaptationondifferenttypesof robots.Forthepurposesofthisresearch,theimplementationswereprogrammedintoanInuktunExtreme-VGTVInuktunandaniRobotPackBotScoutPackbot.Theseconceptscanbe appliedtoothertypesofappearance-constrained,non-anthropomorphicandanthropomorphic robots. Isitpossibletoevaluatethroughsoundexperimentalmethodstheeffectivenessofthenonfacialandnon-verbalmethodsofaffectiveexpressioninappearance-constrainedrobots?A majorissueinHRIresearchstudiesisthelackofmetricsandmeasurementstoevaluatethe impactsofthesetypesofsocialinteractionsusinganappropriatesamplesizetorepresentthe populationbeingstudied.Agoalofthisresearchistoutilizemultiplemethodsofevaluation forconvergentvalidity,incorporatepsychophysiologicalmeasurestoassessparticipantarousal responses,useanappropriatenumberofparticipantstoeffectivelyevaluatethemethodssee Chapter4,performin-depthstatisticalanalysis,andpresentanystatisticallysignicantresults seeChapter5. Thisresearchisacomprehensiveinvestigationintotheuseofnon-facialandnon-verbalaffectiveexpressionasamethodofnaturalisticsocialinteractioninnon-anthropomorphicandappearance-constrainedrobots.Theapplicationdomainusedforthisinvestigationisrobot-assisted searchandrescue;howevertheresultswillbeapplicabletootherdomainssuchasmilitary,lawenforcement,education,service,andentertainmentapplications.Itisexpectedthattheresultsfrom thisstudywillbenetnotonlytheHRIcommunitybutalsotheroboticscommunityasawhole. 1.2WhyNon-FacialandNon-VerbalAffectiveExpression? Theuseofaffectiveexpressionandsocialinteractionisanemergingareaofimportancein robotics,withthefocushistoricallyonfacialexpressionsand/oranimalmimicryBethelandMurphy,2006a,Breazeal,2002,Ca nameroandFredslund,2000,Fong,Nourbakhsh,andDauten3

PAGE 20

hahn,2003,Mizoguchietal.,1997,Scheeffetal.,2000.However,alargenumberofmobile robotscurrentlyinuseforapplicationssuchassearchandrescue,lawenforcement,andmilitary arenotanthropomorphic,donothaveanymethodofprojectingfacialexpressions,andcannotbe re-engineeredtoexplicitlysupportaffectiveexpression.Thisposessignicantchallengesasto howtheseappearance-constrainedrobotswillsupportnaturalistichuman-robotinteractionBethel andMurphy,2006a,BethelandMurphy,2008. Appearance-constrainedrobotsarenotengineeredtobeanthropomorphicanddonothavethe abilitytoexhibitfacialexpressionsormakeeyecontact.Appearanceconstraintsstemfromeither thelimitationsoftheapplicationorfromeconomics.Mobilityisamajorlimitation;forexample, uneventerrainmaydrivetheuseoftracksinsteadofusinganthropomorphiclegs.Powerandplatformsizearetwootherlimitations;robotssuchasthoseforoperationinhighlyconnedspaces maynothaveenoughspaceoron-boardpowertoaddfacialfeatures.Extraeffectorsmayinterfere withthemissione.g.,snagonwiresoroverhangsordecreasereliabilitye.g.,dustbreakingthe effectors.Theenvironmentitselfmaylimitaffect,forexample,loworunconstrainedlightingmay preventtheviewingofavatarsonscreens.Intermsofeconomics,inmanycasesmanufacturers andorganizationshavealreadydevelopedandinvestedlargeamountsofmoneyintotheirrobot designsanditwouldnotbepracticaltoaltertherobot'sphysicalappearancetoproduceamore naturalisticsocialinteraction. Whilefacialdisplayshavebeenthemostcommonmechanismforexpressingaffectforrobotsingeneral,someroboticistshaveusedbodymovementBreazeal,2002,BreazealandFitzpatrick,2000,Breazeal,2004,Maeda,1999,MoshkinaandArkin,2005,Shimokawaand Sawaragi,2001,postureBreazeal,2002,Scheeffetal.,2000,BreazealandFitzpatrick,2000, Breazeal,2004,MoshkinaandArkin,2005,orientationBreazeal,2002,BreazealandFitzpatrick,2000,Breazeal,2004,Bruce,Nourbakhsh,andSimmons,2002,Nourbakhsh,Kunz, andWilleke,2003,colorMoshkinaandArkin,2005,SuganoandOgata,1996,andsound BreazealandFitzpatrick,2000,Breazeal,2004,Bruce,Nourbakhsh,andSimmons,2002, Nourbakhsh,Kunz,andWilleke,2003,aseithertheprimarymethodofexpressionortoprovide affectiveexpressionredundancy.Theseapplicationsweredevelopedmostlyforeducationaland/or entertainmentpurposes.Althoughfacialexpressionhasbeenshowntobequiteeffectiveintheexpressionofaffect,someresearchersfeelthatbodymovementandposturemayrevealunderlying 4

PAGE 21

emotionsthatmightbehiddenotherwiseBreazealandFitzpatrick,2000,Argyle,1975,Bull, 1987,Fast,1970,Picard,1997. Theimpactofspatialdistancesonbodymovement,posture,orientation,color,andsoundin roboticshasbeenacknowledgedbyseveralroboticistsbuthasnotbeencodiedMizoguchiet al.,1997,BickmoreandPicard,2004,Dautenhahn,Ogden,andQuick,2002,Fincannonet al.,2004.Inthepsychologycommunity,spatialdistancesbetweenindividualssociallyinteractingproxemicsisdividedintofourmaincategories:intimatecontact0.46m,personal.46 1.22m,social.223.66mandpublic.66mandbeyond.ArgyleandFast describepersonalspacetobetheareathatindividualsmusthavesurroundingthemtofeelcomfortable,safe,andprotected.BasedonexperimentsconductedbyArgyleandFast, ifanindividualorobjectintrudesintothepersonalspaceofanotheritwillproduceadiscomfort responsefromthepersonwhosepersonalspacewasinvaded. Thisresearchfocusesontheuseofaffectiveexpressioninnon-anthropomorphicandappearance-constrainedrobotsforhuman-robotinteractionsoccurringwithinthreemetersofeachother BethelandMurphy,2006a.Mostsocialinteractionsoccurwithinthisdistancerange.ItisparticularlymotivatedbyastudyconductedbyMurphy,Riddle,andRasmusseninusingmanpackablerobotstoactasasurrogatepresencefordoctorstendingtotrappedvictims,butitisexpectedtobeapplicabletoanyclosehuman-robotinteractionBethelandMurphy,2006a,Bethel andMurphy,2008.Thestudyidentieshowtherobotwillinteractwiththevictimasoneofthe fourmajoropenresearchissuesMurphy,Riddle,andRasmussen,2004.Theynotedthattherobotsoperatingwithinthreemetersofthesimulatedvictimswereperceivedascreepyandnot reassuringseeFigure1BethelandMurphy,2006a,BethelandMurphy,2008. Ineachofthesecases,therobotswereoperatedinhighlyconnedspacesandtheaddition offacesorotherdevicesmightinterferewiththecriticalattributeofmobility.Anothersource ofmotivationisFincannonetal.,whererescueworkersworkingwithinthreemetersofa smalltank-likerobotduringbreachingoperationsexpectedittofollowsocialconventionsdespite therobot'snon-anthropomorphicappearance.Therefore,threemetersappearstobeareasonable, thoughempirical,radiusforconsideringhuman-robotinteractioninsearchandrescueapplications. 5

PAGE 22

Figure1.Illustrationoftwoparticpants'responsesandtheirviewsoftherobots.Theirresponses visiblyshowtheywerenotcomfortablewiththerobotswithwhichtheywereinteracting.Thetop leftisaparticipant'sresponsetotheInuktunpicturedinthetopright.Thebottomleftimageisa participant'sresponsetothePackbotpicturedonthebottomright. 6

PAGE 23

TherobotsusedforthisresearchweretheInuktunExtreme-VGTVandtheiRobotPackbot ScoutdisplayedinFigure2,whicharerobotsusedinactualurbansearchandrescueoperations. Inrescueresponsestheserobotsaretypicallytele-operatedforbettercontrolandtobeableto navigateinuncertainenvironmentsasfoundindisastersituations.TheInuktunhasatetherfor powerandforrobotretrieval;howeverthePackbotiscontrolledthroughwirelesstechnology. TheInuktunrobotislessthanhalfthesizeofthePackbotandismuchmoremaneuverableespeciallyinconnedspaces.IncomparisonthePackbot,ismorepowerful,anditslargersizeaffords ittheabilitytoovercomelargerobstacles.Bothrobotsareconsideredpolymorphicbecausethey havetheabilitytochangetheirshape.TheInuktuncantransformfromaatposition,lowtothe ground,toatriangularshapethatcanbeusedtoovercomeobstacles.Thecamerafacerotatesso thattherobotcanbeusedtoinspectsurroundingareasfordamageandforvictimassessment.The Packbotcanraisefromitsatpositiontowhatistermedacrabpositionwhereitraisesontoits ipperstopullitselfoverobstacles.Additionally,thePackbotcanreverseitsippersandraisethe backofitsbodysothatitscamerafacecanscandownwardstolookingratesorlocationsbelowits cameralevelforinspectionpurposes.Overalltheyservedifferentpurposes,butbothcanbevery usefulinsearchandrescueresponses;hencethereasonforselectingthemforuseinthisresearch. Figure2.ImagesoftheiRobotPackbotScoutandtheInuktunExtreme-VGTV.ThePackbotis ontheleftandInuktunisontheright.Bothrobotsareexamplesofshapeshiftingorpolymorphic robots. Militaryapplicationsareanotherdomaininwhichtherobotsareappearance-constrained.As partofthoseapplicationstherobotscanbeusedtolocateandinteractwithsoldiersinjuredinbattle.Additionally,thereareapplicationsinwhichthemilitaryisrequestedtoassistwithsearchand 7

PAGE 24

rescueinnaturaland/orman-madedisasters.Inthosesituationstherobotswouldstillneedtointeractwithvictimsastheyarelocatedintraumatizedstates.Anin-depthdiscussionofmilitary applicationsisbeyondthescopeofthisparticularresearch.Thefocusofthisworkistheareaof searchandrescueapplications;howevertheprinciplescanapplytotheseotherapplications. Thisresearchsynthesizesthecognitiveliteratureonaffectiveexpressiontheoryandthelessonslearnedfromtherobotimplementationspracticetodatetogenerateaprescriptivemapping ofwhichnon-facialandnon-verbalexpressionsareappropriateincloseoperation.Althoughnonanthropomorphicaffectivefeaturesarebeingusedbyroboticists,thechoiceoffeaturesforaparticularapplicationisadhocandsothismappingwillcontributetotheformalizationoftheuse ofaffectinroboticsBethelandMurphy,2006a,BethelandMurphy,2008.Theserecommendationsformamulti-modalsystemofnon-facialandnon-verbalaffectiveexpressionthatcanbe usefulinanytypeofsocialroboticapplicationanthropomorphicandnon-anthropomorphic.In addition,theprescriptivemappingcanguidetheadditionofappropriate,andeconomical,affectiveexpressionsimplythroughsoftwarecontrol,withminimalornophysicalchangestotherobot BethelandMurphy,2006a,BethelandMurphy,2008. 1.3Non-FacialandNon-VerbalMethodsofAffectiveExpressionasAppliedtoRobot-AssistedVictimAssessment Inaworldwheredisastersappeartobeprevalent,eitherfromterrorists'activitiesornatural causes,thereisanincreasingneedfortheuseofsearchandrescuerobotstosociallyinteractwith survivorsandtoactasavirtualpresenceforemergencyrespondersandmedicalpersonnel.There aretwomaincapacitiesinwhichtheserobotshavebeenutilized:tolocatevictims,andtoassess structuraldamageFincannonetal.,2004.Theserobotsprovideavirtualpresenceforsupport teamsandmedicalpersonnellocatedoutsideofthedisastersiteFincannonetal.,2004.Whena victimislocatedinadisastersite,therobotwillbeusedasasurrogatepresenceforphysiciansand medicalpersonneltoperformaninitialmedicalassessmentofthevictim'sconditionandtocontinuallymonitorthevictim'smedicalstatusuntilassistancecanarrivetypically4-10hoursafterthe victimislocatedMurphy,Riddle,andRasmussen,2004,Riddle,Murphy,andBurke,2005.In thistypeofapplication,therobotmustinteractwithboththevictimandtheoperators.Theoperatorsareabletocaptureimagesofthevictimthroughon-boardcamerasaswellascommunicate 8

PAGE 25

withvictimsusingon-boardtwo-wayaudio.Additionally,theoperatorsworkwiththemedical professionalssothattheymayinteractwiththevictimsthroughtheuseoftherobot.Underthese conditions,therobotisbothateammemberoftheemergencyresponseteaminadditiontoasocialagentthatmustinteractwithvictimsandkeepthemcalmuntilassistancecanreachthemfor extrication. Aftervictimsarelocated,theprimarygoalofsocialinteractionbetweentherobotandthese victimsistokeepthemcalmthroughtheappropriateuseofnon-facialandnon-verbalmethods ofaffectiveexpression.Asapartofthissocialinteractionitisnecessarytodetermineavictim's medicalstatusthroughrobot-assistedmedicalreachbackassessments.Socialinteractionbetween thesearchandrescuerobotandvictimsshouldincludeamulti-modalapproachforaffectiveexpression,assuggestedinBethelandMurphy,2006a,BethelandMurphy,2008. AsetofprescriptiverecommendationswereinitiallydevelopedandpresentedinBetheland Murphy,2006a,renedinBethelandMurphy,2008asurveyofrelatedliterature,andfurther researchBethelandMurphy,2007conrmedtheappropriatenessoftherenedrecommendations.Itwasdeterminedfromvideoobservationsthattheserecommendationswereperceptually signicantatthepredicteddistances.Theprescriptiverecommendationsaddresstheappropriatenessofeachofthevemethodsofnon-facialandnon-verbalaffectiveexpressionsbyproximity zone.Asimulatedvictimmedicalassessmentscenariowasdevelopedandutilizedforthisresearch studyfromobservationsoftwodifferentmedicalassessmentscenariosconductedbymedicalprofessionalsinastudybyRiddle,Murphy,andBurke.Conrmationoftheprescriptiverecommendationswasobtainedthroughvideoobservationsofthemedicalassessmentpathfromthree viewpointsrobot-eye,victim-eye,andoverhead-view.Resultsfromthisvericationstudyindicatethatfuturehuman-robotinteractionHRIstudiesusingtheserecommendationsareviable.As aresultofthelarge-scalehuman-robotstudyperformedaspartofthisresearch,theprescriptive recommendationswerefurtherrenedandthenalversionisdisplayedinFigure5onpage45. 1.4PsychophysiologicalMeasurementsAppliedtoHuman-RobotInteraction Human-robotinteractionisanemergingeldofresearch;howeverthedevelopmentofmethodstoevaluatetheeffectivenessoftheseinteractionsislacking.Intheearlyphasesofthiseldthe focuswasonthedevelopmentofspecicroboticsystemsandapplications.Methodsfortesting 9

PAGE 26

andevaluationhavebeenadoptedandmodiedfromsucheldsashuman-computerinteraction, psychology,andsocialsciencesKiddandBreazeal,2005.Themannerinwhichahumaninteractswitharobotissimilarbutnotidenticaltointeractionsbetweenahumanandacomputeror ahumaninteractingwithanotherhuman.Asrobotsbecomemoreprevalentinday-to-daylife, itwillbeincreasinglyimportanttohaveaccuratemethodsofevaluatinghowhumansfeelabout theirinteractionswithrobotsandhowtheyinterprettheactionsoftherobotsBetheletal.,2007a, Betheletal.,2007b. Therearevemainmethodsofevaluationusedforhuman-robotinteractionHRIstudies: self-reportmeasures,behavioralmeasures,psychophysiologicalmeasures,interviews structuredandunstructured,andtaskperformancemetricsKiddandBreazeal,2005,Burke etal.,2004,Betheletal.,2007a,Betheletal.,2007b.Themostcommonmethodsutilizedin mostHRIstudiesareself-reportandbehavioralmeasures.ThereareatleastveHRIresearch studiesusingrobotsthatincludetheuseofpsychophysiologicalmeasuresandevenfewerHRI studiesusingtaskperformancemetrics.Eachmethodhasitsadvantagesanddisadvantages;howeversomeofthedisadvantagescanbeovercomebyusingmorethanonemethodofevaluation KiddandBreazeal,2005. Theuseofparticipants'self-reportsisoneofthemostcommonlyusedmethodsofevaluationinHRIstudiesandoftenincludedaspartofapsychophysiologicalevaluation.Self-report measuresincludepencil-and-paperorcomputerizedpsychometricscales,questionnaires,and/or surveys.Participantsprovideapersonalreportoftheirmotivesandfeelingsaboutanobject,situation,orinteractions.Self-reportsprovidevaluableinformationbutthereareproblemswithvalidity andcorroboration.Participantsmaynotanswerexactlyhowtheyarefeelingbutratheranswer questionsastheyfeelotherswouldanswerthemorinawaytheythinktheresearcherwantsthem toanswer.Anotherissuewithself-reportingmeasuresistheinabilityforobserverstocorroborateimmediatelyanddirectlytheinformationprovidedbyparticipantsElmes,Kantowitz,and RoedigerIII,2006.Participantsmaynotbeintouchwithwhattheyarefeelingabouttheobject orsituationandthereforemaynotreporttheirtruefeelings.Theresponsescouldbedependenton participants'moodandstateofmindonthedayofthestudyElmes,Kantowitz,andRoedigerIII, 2006,JohnsonandChristensen,2004.Forthesereasons,itisimportanttoperformpsychophys 10

PAGE 27

iologicalmeasurestoaddanotherdimensionofunderstandingofparticipants'responsesandphysiologicalreactionsinHRIstudiesBetheletal.,2007b. Behavioralmeasuresareprobablythesecondmostcommonmethodofevaluationinhumanrobotinteractionstudiesandoftenincludedinpsychophysiologicalevaluationsforconvergent validityofparticipants'self-reportresponsesandmeasuredphysiologicalreactions.Johnsonand Christensendeneobservationasthewatchingofbehavioralpatternsofpeopleincertain situationstoobtaininformationaboutthephenomenonofinterest.TheHawthorneeffectisan areaofconcernwithobservations.Ifparticipantsknowthattheyarebeingobserved,itwillimpact theirbehaviorsElmes,Kantowitz,andRoedigerIII,2006,JohnsonandChristensen,2004.For thisreason,psychophysiologicalmeasurescanassistwithobtainingabetterunderstandingofthe participants'underlyingresponsesexpressedatthetimeoftheobservations.Thebenetofusing behavioralmeasuresisthatresearchersareabletorecordtheactualbehaviorsofparticipantsand donothavetorelyonparticipantstoreportaccuratelytheirintendedbehaviorsorpreferencesin additiontoobtainingpsychophysiologicalmeasuresforconvergentvalidityElmes,Kantowitz, andRoedigerIII,2006,Steinfeldetal.,2006,Betheletal.,2007b. ThedesignofaqualityresearchstudyforuseinHRIapplicationsthatproducesresultsthat areveriable,reliable,andreproducibleisamajorchallengeBetheletal.,2007a,Betheletal., 2007b.Psychophysiologicalmeasurementscancomplicatethisprocessbecausetheresultsare notalwaysstraightforwardandconfoundscanleadtomisinterpretationofdata.Thereisatendencytoattributemoremeaningtoresultsbecauseofthetangiblenatureoftherecordings.Informationneedstobeobtainedfromparticipantspriortobeginningastudytohelpreducethese confoundse.g.,healthinformation,stateofmind.Multiplephysiologicalsignalsshouldbeused inordertondcorrelationsintheresults.Steinfeldetal.describetheneedforthedevelopmentofcommonmetricsasanopenresearchissueinHRI.TheydiscussanapproachofdevelopingcommonmetricsforHRI;howeverthisapproachisorientedmoretowardanengineering perspectiveanddoesnotcompletelyaddressthesocialinteractionperspective.Bothperspectives havevaluebutrequirefurtherinvestigation.Inordertoobtaincredibilityintheresearchcommunity,HRIstudiesneedtobesupportedbyqualityexperimentaldesignswithadequatesamplesizes andmulti-facetedmethodsofmeasurementtoprovideconvergentvalidity.Noneofthesemeasures alonearesufcienttointerpretaccuratelytheresponsesofparticipantstoarobotwithwhichthey 11

PAGE 28

areinteracting.Inorderforastudytohavecorroborationandconsistencyinitsevaluations,at leasttwomethodsofmeasurementshouldbeusedKiddandBreazeal,2005,Elmes,Kantowitz, andRoedigerIII,2006,JohnsonandChristensen,2004,Betheletal.,2007a,Betheletal., 2007b. 1.5Contributions Thefocusofthisresearchprovidescontributionstotheroboticsandhuman-robotinteraction communitiesintheareasoffundamentalscience,economicbenets,measurements,metrics,and socialbenets.Thecontributionsareasfollows: FundamentalScienceThisworkistherstcomprehensiveinvestigationofnon-facialand non-verbalaffectiveexpressionsforuseinappearance-constrainedandanthropomorphicrobots. EconomicBenetDevelopedasetofprescriptiverecommendationstoaddretroactivelyaffectiveexpressionbasedonproximityzonetobothappearance-constrainedandanthropomorphicrobotsthroughsoftwaremodication.Thiswillminimizetheneedforphysicalrobot modicationandnewrobotdevelopment. ImmediateSocialBenetThisresearchprovidesameansofsocialinteractionwithvictims locatedinsearchandrescueoperationstokeepthemcalmandtoreassurethemuntilassistancecanarrivewhichistypically410hoursafterthevictimislocated.Theresultsare applicabletotheurbansearchandrescueenvironmentandtointeractionsinwhichthegoal wouldbetoelicitacalmingresponse.Furtherresearchneedstobeexploredtodeterminethe implicationstootherrobotapplications. MetricsandMeasurementsThisworkvalidatesthevalenceandarousaldimensionsofthe Self-AssessmentManikinSAMasanassessmenttoolforuseinHRIstudies.Thedominancedimensionwasnotvalidatedsinceapproximately25%oftheparticipantsverballyreportedconfusionwiththewordingoftheSAMassessmentquestionsrelatedtothedominance dimension;thereforefutureworkwillberequiredtovalidatethisportionoftheassessment. Throughtheuseofpsychophysiologicalmeasuresonasignicantsamplesizethisresearch 12

PAGE 29

demonstratesthattheremaybevalueinusingthistoolinHRIstudiestoevaluateparticipants' responsestotherobotswithwhichtheyareinteracting.Additionally,asetofpracticalrecommendationsareprovidedfordesigning,planning,andexecutingsuccessful,complexhumanrobotstudiesusingappropriatesamplesizesandmultiplemethodsofevalationforusebythe HRIcommunity. 1.6OrganizationoftheDissertation Thisworkisorganizedintosevenchapters.Chapter2coversrelatedworkintheareasofemotiontheoriesandmodels,factorsinaffectiveexpressionpresentationmethodsandproxemics, robotimplementations,reviewofpsychophysiology,andimplementationsusingpsychophysiologicalmeasuresinHRIstudies.Adiscussionoftheprescriptiverecommendationsdevelopedforthe appropriateuseofnon-facialandnon-verbalmethodsofaffectiveexpressionbyproximityzoneis coveredinChapter3.Additionally,coverageisgiventotheapproachtakentodevelopnaturalistic socialinteractioninappearance-constrainedrobotsusingthedevelopedprescriptiverecommendations.TheexperimentalmethodsanddesigndetailsarepresentedinChapter4withthestatistical dataanalysesandresultsdiscussedinChapter5.Adiscussionofthestatisticallysignicantndings,factorsthatmayhaveimpactedtheoutcomes,suggestedstudyimprovements,appropriate studydesign,andopenquestionsarepresentedinChapter6.Chapter7detailstheconclusionsand contributionsresultingfromthiswork,aswellasabriefdiscussionoffuturework,whichrequires furtherexploration. 13

PAGE 30

Chapter2 RelatedWork Thepurposeofthischapteristoprovideabackgroundandhistoryoffactorsrelatedtothe expressionordisplayofaffectoremotioninhuman-robotinteractionresearch.Emotiontheory frompsychologyandsocialsciencesprovidesthefoundationforaffectiveexpressioninrobotsystems.Non-facialandnon-verbalpresentationmethodsofaffectandproxemicsmustbeconsidered whendevelopingasocialroboticsystemwhethertherobotisappearance-constrainedoranthropomorphic.Previousroboticimplementationscanassistinclarifyingopenresearchissuesrelatedto affect,whatissuesposeachallenge,andthedirectionoffutureresearchefforts.Finally,psychophysiologicalmeasuresusedinHRIstudiescanprovideusefulinformationonthedevelopmentof systemstoidentifyanddetectarousallevelsinparticipantsinteractingwithrobots,evaluateparticipants'responsestotherobotswithwhichtheyareinteracting,andtocontrolrobotbehaviorsand interactions. 2.1EmotionTheoryandModels Althoughemotiontheoryandmodelinghasbeenthesubjectofsignicantresearchinelds suchaspsychology,neurology,computerscience,androbotics;emotiontheoryandmodelingare independentofthedevelopmentofnon-facialandnon-verbalmethodsofaffectiveexpression.In theroboticscommunity,theselectionofwhichemotionstoexpresshasfollowedprimarilytwo paths:basicemotionsfear,anger,sadness,joy,anddisgust,andemotionsascognitiveprocesses asseenintheOrtony,Clore,andCollinsmodel,whichincludesbipolarpairingssuchas joy-distress.Roboticistshavefocusedprimarilyontheuseoffacialexpressionsfordisplaying emotionseventhoughithasbeenshownthattheuseoffacialexpressionsaloneisinadequateto expressdistinctemotionsRussellandLemay,2000.Thisreinforcestheneedforredundantor 14

PAGE 31

differentmethodsofaffectiveexpressionsuchasnon-facialandnon-verbalmethodsofaffective expression. Emotionshavebeenthesubjectofcontroversyandscienticinvestigationinawidevariety ofeldssuchas:psychology,physiology,neurology,andmorerecentlycomputerscienceandrobotics.Thecontroversyassociatedwithemotionresearchbeginsattheveryfoundation,theterminologyanddenition.Therehasbeennoconsensusamongresearchersonwhatanemotionisor howitisdenedFrijda,2000.RussellandLemaydescribevedifferentwaystoclassify emotionalepisodestocreatewhathetermsbasiccategories.Thesecategoriesare: Classifyemotionalepisodesbasedonfacialexpressionswhichtheybelieveisinadequatefor tworeasons:rst,thereareemotionsthatdonothaveacorrespondingfacialexpression;and second,someemotionscouldsharethesamefacialexpression; Emotionalepisodescouldbeclassiedbyautonomicnervoussystemactivitywhichhasbeen showntobeeffective; EmotionalepisodescouldbeclassiedbythecognitiveprocessesinvolvedsimilartotheOCC modeldevelopedbyOrtony,Clore,andCollins; Theseepisodescouldalsobeclassiedbytheactioninvolved;however,thisismoreofaspeculativeapproach; Emotionalepisodescouldbeclassiedbythebrainstructuresandneurotransmittersinvolved. Althoughtheseclassicationsprovideavarietyofapproachesthatcanbetakeninthetheoryandmodelingofemotionresearch,roboticimplementationshavefocusedontherstbullet point,whichreliesheavilyonfacialexpressionsastheprimarymethodofconveyingaffectBreazeal,2002,Bruce,Nourbakhsh,andSimmons,2002,Ca nameroandFredslund,2000,Lisetti, 2002,Lisettietal.,2004,Scheeffetal.,2000,Michaudetal.,2001,Vel asquez,1997.As indicatedintherstbulletpoint,therearepotentialissueswithrelyingprimarilyontheuseof facialexpressionstoclassifyandexpressemotions,whichreinforcestheneedtodevelopother methodsofaffectiveexpression.Althoughextensiveresearchhasbeendevotedtothetheoryof emotionandemotionalmodeldevelopment,thefocusofthisresearchwasonconceptsandde15

PAGE 32

velopmentsinnon-facialandnon-verbalmethodsofaffectiveexpressioninrelationtoproxemics whichareindependentoftheemotionalmodelsupportingthoseexpressions. 2.2FactorsinAffectiveExpression Therearetwomainfactorsrelatedtoaffectiveexpressionidentiedinthepsychologyliteratureandsupportedbyobservationsfromcomputerscienceandrobotics:apresentationmethodsforaffectiveexpression,andbtheimportanceandimpactofproxemics,therelativespatial distancebetweenagentsinsocialinteractionsBethelandMurphy,2006a,BethelandMurphy, 2008,BethelandMurphy,2006b.Thesefactorsprovideakeyunderstandingintothedevelopmentofamulti-modalsystemofaffectiveexpression,whichisessentialfornon-anthropomorphic andappearance-constrainedrobotsandprovidesaffectiveexpressionredundancyinrobotsthat haveananthropomorphicappearance.Ingeneral,researchersintheeldsofpsychology,computerscience,androboticshavefocusedmostoftheirattentiononfacialexpressionasthemain methodforexpressingaffectandhaveneglectedotherimportantmethodsofpresentationsuchas bodymovement,posture,orientation,color,andsoundCa nameroandFredslund,2000,Argyle, 1975,Bull,1987. 2.2.1PresentationMethods Non-facialandnon-verbalpresentationmethodsofaffectcanbeseparatedintovedifferent cues:bodymovements,posture,orientation,color,andsoundBethelandMurphy,2006a,Bethel andMurphy,2008,BethelandMurphy,2006b,BethelandMurphy,2007.Argyleand Bartneckdescribesomeaffectiveexpressionsusingbodymovements:depressionslow andhesitatingmovements,andelationfast,expansive,andemphaticmovements.Robotorientationtowardapersonitisinteractingwithisindicativeofitsperceivedattentivenessandcaring forthatpersonFong,Nourbakhsh,andDautenhahn,2003,Bruce,Nourbakhsh,andSimmons, 2002,BickmoreandPicard,2004,Fincannonetal.,2004.Argylediscussestheuseof colortoproduceanaffectiveresponsesuchas:blueelicitspleasant,calm,secure,andtenderresponses;yellowelicitscheerful,joyful,andjovialresponses;andredelicitsanger,affection, andlove.Normandiscussesthatvocalpatternsandtonecanexpressaffecteveniftheliteralmeaningisnotunderstood;howeverScheeffetal.describesituationswheretonescan 16

PAGE 33

beoffensivetoindividualsinteractingwiththeirrobot.Additionally,Bartneckdescribes theuseofmusictoexpressaffectsuchas:tendernessexpressedwithslowtempo,softtimbre, intensevibrato,andlowtomoderatesoundlevels;whereasangerhasafasttempo,highsound level,sharptimbreanddistortedtones.ResearchconductedbyBalkwill,Thompson,andMatsunagaindicatesthatlistenersaresensitivetoemotionexpressedinbothfamiliarandunfamiliarmusicbasedontheperceptionofacousticcuesandtranscendingculturalboundariesfor affectiveexpression. 2.2.1.1BodyMovement,Posture,andOrientation Thestudyofnon-verbalcommunicationbypsychologistsdoesnotofferacompleteunderstandingofbodymovement,posture,andorientation.Bullnotesthatpsychologistsinvestigatingnon-verbalcommunicationhaveneglectedtheareasofgestureandposture,insteadfocusingonfacialexpression,gaze,pupildilation,andinterpersonaldistance.SpiegelandMachotka, 1974offeratleasttworeasonsforthislackoffocus.First,itisdifculttoseparatebodycommunicationfromverbalmethodsofcommunication.Second,thereisnosystemformeasuringand interpretingbodymovementcommunicationthatisindependentoflanguage.Despitethesebarriers,researcherssuchasBuck,Bull,Fast,andSpiegelandMachotka opinethatbodymovementsandposturecanrevealmoreabouttheactualaffectivestateofanindividualthanfacialexpressionsorevenverbalcommunication.ResearchconductedbyFast, suggeststhatbodymovementandposturearehumans'mostprimitiveandbasicmethodsofconveyingaffect.Buckarguesthatfurtherinvestigationintotheinuenceofbodilyfeedback throughtheuseofbodymovement,posture,andorientationisnecessaryforunderstandingaffectiveresponses. Movement,posture,andorientationarerelatedtoeachotherandoftenusedinterchangeably. Argyle,Bull,andFastindicatethatpostureandorientationcanindicateliking,interest,andlevelofcomfortbetweenindividuals.Basedontheirqualitativeandquantitativestudies,interestandlikingcanbeexpressedthroughforwardleaning,closerproximity,raised head,anddirectorientation.Insomesituations,theydiscussthatorientationindicatinglikingcan alsobeinasidebysidepositionespeciallynotedinclosefriends.Fastdescribesthatdifferentposturesarerelatedtodifferentemotionalstatesandthroughobservationofanindividual, 17

PAGE 34

theiremotionalstatecanbedecodedbytheirposture.Additionally,hementionsthatbychanging orientationawayfromapersoncanbeperceivedasboredomandadesiretodiscontinuecommunicationandinteraction. Asnotedearlier,bodymovementdominatesnon-facialandnon-verbalaffect.Argyle, asocialpsychologist,whoseworkinnon-verbalaffectisseminal,identiedfouraffectiveexpressionsthatcouldbedisplayedthroughbodymovement: Extremeinhibitionwithdrawingorunnecessarymovements Depressionslowandhesitatingmovements Elationfast,expansive,andemphaticmovements Anxietydgetingorhidingmovements Thoughthiscategorizationisextractedprimarilyfrommovementsofthehandsandfeet,the movementdescriptioncanclearlybeappliedtoarobotplatformoreffectorsonaplatform.His studyshowedthatinsomecasesbodymovementsbetterreectedaparticipant'strueemotional stateeventhoughitcontradictedtheirfacialexpressionsandverbalcommunicationoftheiremotionalstate.ThissupportsthepostulationsofBuck,SpiegelandMachotka,Bull ,andFast. Fromthecomputerscienceandroboticsperspective,motionandbodymovementhavebeen usedbyBreazeal,BreazealandFitzpatrick,Scheeffetal.,Bruce,Nourbakhsh,andSimmons,Nourbakhsh,Kunz,andWilleke,Lisettietal.,Thrun ,andhavebeenpositedbyMizoguchietal.,BreazealandFitzpatrick,Picard ,BickmoreandPicard,tobeessentialelementsinpresentingaffectiveexpressionto individualswhoareinteractingsociallywithrobots.Thereappeartobetwomotivationsforusing affectivemovement,posture,andorientation:tocreateattractionandtocreatetrust. EightrobotsystemsusedeithermovementMizoguchietal.,1997,Scheeffetal.,2000, BreazealandFitzpatrick,2000,MoshkinaandArkin,2005,Nourbakhsh,Kunz,andWilleke, 2003,Lisettietal.,2004,Thrun,1998,postureMizoguchietal.,1997,Scheeffetal.,2000, MoshkinaandArkin,2005,orientationMizoguchietal.,1997,Scheeffetal.,2000,BreazealandFitzpatrick,2000,MoshkinaandArkin,2005,Bruce,Nourbakhsh,andSimmons, 18

PAGE 35

2002,Nourbakhsh,Kunz,andWilleke,2003,Lisettietal.,2004,Thrun,1998,orallthree Mizoguchietal.,1997,Scheeffetal.,2000,MoshkinaandArkin,2005,toattracthumanattentionandinteraction.Basedonaseriesofexperimentsandthreerobotsspanningaveyear period,Nourbakhsh,Kunz,andWillekestatethemostsuccessfulwayforarobottoattract attentionisbydeliberatelyorientingtherobottowardapersonandaddressingthem.Additionally, theydescribethatindividualsaremorelikelytointeractwitharobotthatisinmotion,thanarobot thatisinaxedposition.Unfortunately,intheroboticsystemsthathaveimplementedthesemethodsofaffectiveexpression,bodymovement,orientation,andpostureweresecondaryandcoupled tosometypeoffacialexpressioneitheronaphysicalrobotoranavatarconnectedtoarobotand viewedonamonitorMizoguchietal.,1997,Scheeffetal.,2000,BreazealandFitzpatrick, 2000,MoshkinaandArkin,2005,Bruce,Nourbakhsh,andSimmons,2002,Nourbakhsh, Kunz,andWilleke,2003,Lisettietal.,2004,Thrun,1998.Therefore,thereisnoquantitative measurementoftherelativecontributionsofthesedifferentexpressionsofaffect.Therehasbeen somediscussionontheimportanceofpostures;howeveritappearsthatthisisnotaseffectivein presentingspecicaffectiveexpressionsBull,1987,Fast,1970,EkmanandFriesen,1967. AccordingtoArgyle,Bull,andSpiegelandMachotkapostureisindicative ofoverallliking,interest,andopennessbetweenindividuals. Movements,posture,and/ororientationhavebeenidentiedascreatingtrustbetweentherobotandhumaninthreestudies.BickmoreandPicarddiscussthatpeoplearemorelikely totrustsomethingthattheyperceivecaresforthem.Whilecaringisoftendemonstratedthrough verbalcommunicationandfacialexpressions;theyalsobelievethatitcanbeexpressedinarobot oravatarthroughcloseconversationaldistance,noddingmovementofaperceivedheadorthrough bodymovementandgesturesBickmoreandPicard,2004.Theirquantitativeresultsindicatethat byapplyingthesenon-verbalbehaviors,suchasbodyandfacialorientationoftherobotoravatar towardtheindividualincreasestheperceptionofcaringforthatindividualBickmoreandPicard, 2004. Fincannonetal.describethatrescueworkersfeelmorecomfortableintheirsocialinteractionifthesearchandrescuerobotwasorientedtowardtherescueworkerduringallverbal andnon-verbalcommunication.Byorientingtherobottowardtherescueworker,itwasperceived byrescuersthattheywerethefocusofattentionoftherobotandthattherobotisintentlylistening 19

PAGE 36

totherescuer'scommunication.Theirexperimentalresultsindicatedthathumansinteractingwith arobotusedthesamesocialconventionsforeyecontactandpersonaldistanceastheywouldin humantohumancontact. Dautenhahnetal.conductedexperimentsinalabsettingtodeterminepreferencesfor robotapproachorientationusingthePeopleBotTMrobot.Fromtheseexperimentsitwasdeterminedthatparticipantsinbothaconferenceandalabsettingweremostcomfortablewiththerobotapproachingfromtherightandweremostuncomfortablewitharobotapproachingthemfrom thefront.Participantsreportedthatthefrontaloramoredirectapproachwasmorethreatening,aggressive,andtheyhadconcernsabouttherobotnotstoppingandrunningintothem.Therefore,it appearsthatparticipantstrustedtherobotsthatapproachedfromeithertheleftortherightmoreso thanthosethatapproachedthemfromthefront. 2.2.1.2Color Colorasanexpressionofaffecthasnotbeendeeplyresearchedbyeitherthepsychologyor roboticscommunities.Withinthepsychologycommunity,Argylediscussesthatcolors areoftenassociatedwithaffectandinsomecasescolorcanproduceanaffectiveresponse.The followingcolorsareassociatedwiththeseaffectiveresponses: Redanger,affection,love Bluepleasant,calm,secure,tender Yellowcheerful,joyful,jovial Orangedisturbed,hostile,upset Purplesad,depressed,dignied,stately Greenpleasant,leisurely,incontrol Blacksadness,anxiety,fear Whitejoy,lightness,neutral AlthoughArgylediscussesacolormappingforaffect,thecolorshavemultipleemotionsassociatedwitheachcolor.Theuseofcolorwillrequireadditionalinvestigationtodevelop 20

PAGE 37

astrongmappingofacolortoaspecicemotion.Thecolorofblueorlightbluehasbeenmore commonlymappedtopleasantandcalmenergy,whichcouldproveusefulinsearchandrescue applications. Intheroboticscommunity,theonlyknownworktoutilizecolorforaffectiveexpressionisthe WAMOEBA-1RrobotfromresearchconductedbySuganoandOgata.Theirnon-anthropomorphicrobothassevenlampscontainingthethreeprimarycolorsinthehead-likeportionof therobot.Therobotexpressedfearusingthecolorblue;angerwasdisplayedbythecolorred;and yellowindicatedexpectationandpleasure.Itappearsthatthecolor-affectassociationswereempiricalratherthanextractedfromthepsychologicalliterature.Theaffectivemeaningassociatedwith redandyellowaresimilartoArgyle'sassociations;howevertheircorrespondencesdepartfrom thoseassociationswhenitcomestotheaffectforblue. 2.2.1.3Sound Theuseofnon-verbalsoundasamethodofaffectiveexpressionhasnotbeenthoroughlyinvestigatedbythepsychologyorroboticscommunities.Inthepsychologycommunitytheresearch relatedtonon-verbaluseofsoundhasbeenassociatedwiththestudyofanimalcommunication sinceitisaprimarymethodofconveyingaffectbetweenanimalsArgyle,1975,Buck,1984. Anexamplewouldbeinprimates,distressisindicatedbyascreechingsound,growlingisusedto conveyanger,andsoftgruntsareusedtokeeptouchwiththoseincloseproximitywhiletraveling Argyle,1975,BethelandMurphy,2006b,BethelandMurphy,2008.Theworkthathasbeen conductedonnon-verbalsoundsrelatedtoaffectiveexpressioninhumansisprimarilyfocusedon vocalpatternsandtoneofvoiceArgyle,1975,Buck,1984,Norman,2004.Norman indicatesthatvocalpatternsandtonecanbeusedtoexpressaffectandindicatesthatthesepatterns andtonescanbeinterpretedeveniftheliteralmeaningisnotunderstood.Healsoobservesthat petscaninterpretaperson'saffectivestatebasedontheirvocalpatternsandtone.Scherer indicatesthatrecognitionofaperson'semotionincontentandcontext-freejudgmentsituations, vocalcueshavebeenasindicativeofsomeemotionsasfacialcues. IntheroboticscommunitysoundhasbeenusedasasupplementalmethodofaffectiveexpressionBreazeal,2002,Scheeffetal.,2000,BreazealandFitzpatrick,2000.Scheeffetal., 2000foundthatwhenthesoundsproducedarenotvocalizedinwordsthatareunderstandable 21

PAGE 38

bytheindividualinteractingwiththerobot,thereisatendencytodislikethesounds.Kismet,the robotdevelopedbyBreazealusesavocalresponsetoreinforceitsemotionaldisplayBreazeal, 2002,BreazealandFitzpatrick,2000;howeveritisnottheprimarymodeofaffectiveexpression.El-NasrandSkubicutilizesoundtoindicatetheemotionsofanger,fear,andpainona mobilerobot. Bartneckdescribestheuseofmusicasamethodofaffectiveexpression;howeverthis methodismoredifculttoimplementbecauseitisculturallydependent.Hediscussesmusicparametersforvariousaffectiveexpressionsas: HappinessFasttempo,moderatevariationsintiming,moderatetoloudsoundlevel,mostly staccatoarticulation,fasttoneattacks,brighttimbre,lightornovibrato. SadnessSlowtempo,lowsoundlevel,legatoarticulation,slowtoneattacks,slowanddeep vibrato,softtimbre,nalricard,atintonation. AngerFasttempo,highsoundlevel,nonalricard,mostlynon-legatoarticulation,very sharptoneattacks,sharptimbre,distortedtones. FearLargetempovariations,largedeviationintiming,verylowsoundlevel,largedynamic variation,mostlystaccatoarticulation,fastandirregularvibrato,pausesbetweenphrases,and softspectrum. TendernessSlowtempo,lowtomoderatesoundlevel,legatoarticulation,slowtoneattacks, softtimbre,andintensevibrato. Althoughhisfocusisonmusicparameters,theacousticcuesassociatedwitheachemotionare consistentwithrelatedresearchbyArgyle,1975,NassandBrave,2005,Balkwill,Thompson, andMatsunaga,2004,Juslin,1997,Scherer,1979,SchererandOshinsky,1977,Schubert, 2004.Balkwill,Thompson,andMatsunagadiscussesthatthereisastrongassociation betweenmusicandemotions.Hisquantitativeresultsindicatethatitispossibletodetectemotions suchasjoy,anger,andsadnesspresentedthroughtheperceptionofacousticcuesinfamiliarand unfamiliarmusictranscendingculturalboundaries.Itappearsthatifsoundandmusicarecorrectly usedandimplementedinrobots,itcouldbeavaluabletoolforaffectiveexpression. 22

PAGE 39

2.2.2Proxemics Theunderlyingthesisofproxemicsresearchisthatthespatialdistancebetweenhumanshas asignicantimpactonthequalityandcomfortlevelofinteractions;thishasbeenpredictedto extendtohumans'interactionswithrobotsNassandMoonandrecentlyconrmedby Mizoguchietal.,BickmoreandPicard,Dautenhahn,Ogden,andQuick, Fincannonetal..Althoughtherehavebeensomedifferencesinthedivisionofsocialdistances,theconsensusappearstobefourprimaryzonesSeeFigure3:intimatecontact0.46 meters,personal.461.22meters,social.223.66meters,andpublic.66metersandbeyondMizoguchietal.,1997,Argyle,1975,Hall,1966,Waltersetal.,2005.Studieshave indicatedthatindividualsaremostcomfortableinteractingwithrobotsinthesocialdistancezone Mizoguchietal.,1997,Waltersetal.,2005.Itisimportanttodeterminethebestmethodof affectiveexpressiontouseineachofthesocialdistancezonestoensurethecomfortlevelofthe individualwiththerobotinsocialinteractions. Figure3.IllustrationofArgyle'sfourprimaryproximityzoneswiththezonesofinteresthighlighted. 23

PAGE 40

Therearethreecompetingclassicationsofsocialdistance.Theearliestwasrstdenedby anthropologistHall,thecreatorofproxemics,asfourdistancezonesbetweenpairsofindividualsasintimate,personal,social,andpublicwitheachzoneconsistingofacloseandfarphase. Argyleeliminatesthecloseandfarsubdivisionsanddenesboundariesoneachofthefour proximityzones.SpiegelandMachotkapositanindividual-centric,ratherthanpair-centric, partitioningofspace:internal,proximal,axial,distal,andlimbic,whereinternalwhichliterally meansinsidethebodytolimbicwhichisdetailedasbeingbeyondthebordersofsensoryprocessing.SpiegelandMachotka'spartitioningappearstoberelativelyabstractanddifculttoincorporateintoahuman-robotinteractionapplication. Argyle'scategoriesappearmorerelevanttothestudyofhuman-robotinteraction.Hisfour proximityzonesare: Intimatefromcontactto0.46metersThiszoneisforindividualswhoareinvolvedinan intimaterelationship,theycantouch,smell,feelbodyheat,talkinawhisper,butcannotsee theotherpersonverywell. Personal.461.22metersThiszoneisadistanceatwhichdiscomfortcouldbefeltifthat spaceispenetratedbysomeonewhomtheindividualisnotfamiliarwith,eachpersoncanbe clearlyseen,andtheycantoucheachotherbyreaching. Social.223.66metersThiszoneisusedforformalbusinesspurposessuchassitting acrossadeskfromoneanother,bodymovementsareclearlyvisible,andspeechneedstobe louder. Public.66andbeyondThiszoneisusedforimportantpublicgures,facialexpressions aredifculttosee,loudervoiceisneededtocommunicate,andbodymovementsneedtobe exaggeratedtobevisible. Forthepurposesofthesearchandrescueapplicationsmotivatingthisresearch,onlytherst threezoneswillbediscussed,sincetheycoverthethreemeterdistancebetweenhumanandrobot SeeFigure3.Inmostsearchandrescueoperationsthespacearoundavictimwouldbelessthan threemeters;thereforethatwillbethezoneofinterestforthisresearch. Fromtheroboticsperspective,Mizoguchietal.reportempiricalresultsindicatingthat individualsinteractingwiththeirrobotweremostcomfortableiftherobotinteractedatasocial 24

PAGE 41

distanceconsistentwithHall;howeverthiswasdependentonthespeedatwhichthe robotwasmoving.Theindividualsinteractingweremorecomfortableallowingarobottocome withinacloserdistanceifitmovedslowly.Additionally,theresultsspecifythatnodifferencewas foundintheacceptedproximitydistancebasedonwhethertheindividualinteractingwasstanding orsittingduringtheexperimentswiththerobotMizoguchietal.,1997. Waltersetal.utilizesimilarproximityzonespresentedbyArgylewiththeexceptionthattheintimatezoneisbrokenintocloseintimateandintimatezonesintheirexperiments withthePeopleBotTMrobot.Insixtypercent%oftheparticipantsstudied,theinteraction andapproachdistancesbetweentherobotandthehumanparticipantwereconsistentwithhumanto-humansocialdistancesfallingwithinthepersonalandthesocialdistancezones.Theremaining 40%oftheparticipantswerecomfortablewiththerobotinteractingintheintimatezoneatadistanceof0.45morless.Inthesecases,theparticipantsdidnottreattherobotasasocialentityand alsostatedthattheywerenotthreatenedoruncomfortableinteractingwiththerobot. 2.3RobotImplementations Robotimplementationsusingaffectiveexpressionscanbedividedintothreebasiccategories: non-anthropomorphicandappearance-constrainedrobotsusingnon-facialandnon-verbalaffectiveexpression;anthropomorphicrobotsrelyingheavilyonnon-facialandnon-verbalaffective expression;andtraditionalanthropomorphicrobotsusingnon-facialandnon-verbalaffectiveexpressionasaredundantmethodofexpressioninatightly-coupledconjunctionwithconventional facialexpressionsofaffectRefertoFigure4forasummaryofsomerepresentativeimplementationsfoundintheliterature.Anin-depthdiscussionofeachoftheseimplementationsispresented inthefollowingsubsections.Thesebasiccategorieswerederivedbasedonthetypeofrobotthe non-facialandnon-verbalmethodsofaffectiveexpressionwereimplementedon,andtowhatdegreethesemethodswereutilizedintheimplementationforsocialhuman-robotinteractions.Six oftheimplementationsreviewedinthissectionrelyheavilyonnon-verbalandnon-facialmethodsofaffectiveexpressionMizoguchietal.,1997,Maeda,1999,MoshkinaandArkin,2005, ShimokawaandSawaragi,2001,Bruce,Nourbakhsh,andSimmons,2002,SuganoandOgata, 1996.Threeofthesiximplementationsdiscusseddisplayaffectiveexpressionusingnon-anthropomorphicmobilerobotsMaeda,1999,ShimokawaandSawaragi,2001,SuganoandOgata, 25

PAGE 42

1996.Thisreviewofrepresentativeimplementationsconcentratesondeterminingwhataffective expressionwasusedforwhichproximityzone,anyjusticationforthatchoice,andmeasurement ofsuccess.Althoughitispossibletoexpressthesixbasicemotionsinadditiontootheremotions withtheuseofnon-facialandnon-verbalaffectiveexpressions,someapplicationsdonotrequire alloftheseemotionstobeimplementedBethelandMurphy,2008. Figure4.Summaryofrepresentativerobotimplementationsusingnon-facialandnon-verbalaffectiveexpressionforsocialhuman-robotinteraction. 26

PAGE 43

2.3.1Non-AnthropomorphicandAppearance-ConstrainedRobotsUsingNon-Facialand Non-VerbalAffectiveExpression Threeoftheroboticimplementationsrelysolelyonnon-facialandnon-verbalmethodsofaffectiveexpression.SuganoandOgatadevelopedarobot,theWAMOEBA-1R,thatdisplayedemotionusingcolor.Thecolorreectedtheinternalstateoftherobotbasedonbatterystatus,location,movement,andsensordata.TheWAMOEBA-1Rwasdevelopedtoevaluatelearning behaviorbasedonself-preservationandexpresseditsinnerstatethroughtheuseofcolor.Therobotoperatedprimarilyinthepersonalandsocialdistancezones.Colorforexpressionwasvisible tohumansinboththepersonalandsocialzones.Labexperimentswereperformed,theyreported empiricalresultsfortheuseofcolorasamethodofaffectiveexpressiontosuccessfullyconvey theinternalstateoftherobottohumanobservers;howevertheeffectivenesswasnotcompletelyor objectivelyevaluatedduringthesestudiesSuganoandOgata,1996. Maedaimplementedaffectiveexpressionusingorientationandspeedofmovementon aminiatureKheperarobot.Thepurposeofthisresearchwastoevaluatetheefciencyofasimulatordevelopedtogenerateemotionsusingfuzzylogic.Emotionwasexpressedthroughreactionstoalightsourceusedinthelaboratoryexperiment.Ifscaledforthesizeoftherobot,itappearedtofunctionintheintimate,personal,andsocialdistancezonesinrelationshiptothelight source.Iftherobotexperiencedjoythenitcheerfullywanderedaroundthelightsource;however ifthevalueforjoygeneratedbythesimulatorincreased,thentherobotmotionincreasedandbecamecyclic.Fortheaffectiveexpressionofanger,therobotrushedtowardthelightsourceandif thevalueforangerincreasedthentherobotmovedfasterandwouldoverrunthelightsourceand returnedtoitagain.Inthecaseofsadnesstherobotmovedawayfromthelightsource,ifitwas present,andwenttoadarklocationinthelab;howeverifthesadnessvalueincreased,therobot's motionwouldbecomeslowandifalightsourcewaspresentitmovedbackwardsandvibratedits body.MaedaconductedexperimentsinalabenvironmentusinghumanobservationsandCCD cameratracesforevaluation;howevertheoutcomeswerenotclearandthereisnostatisticaldata presentedtosupportthestatedsuccessfulresults.Maedanotethatfromtheexperiments conductedthatitwasdifculttorecognizethespecicemotionwhentheintensityofangerwas equaltothevalueforsadness. 27

PAGE 44

ShimokawaandSawaragidevelopedanagent-robotsystemusingaradio-controlled modeltankwithavideocamerawithpan,tilt,andzoommountedontopwithahead-likeappearance.Thecamerawascapableofproducinganoddingorshakingappearancetousers.The mainmethodofemotiondisplayforthissystemwastheuseofmotion,velocity,anddirection throughtheuseofathrottleandruddersystem.Emotionswerevariedusingcontinuousandinterruptedmovements;forward,backward,andturningdirections;andthechangesinvelocityofthe movements.Theirexperimentswereconductedinthesocialdistancerange.Theyimplemented theemotionsofjoy,fear,sadness,surprise,anger,anddisgust;howevertheydeterminedthatthe movementsforjoywerethemostreliablyinterpretedbyobservers.Thefocusofthisresearchwas thedevelopmentofnaturalandintuitivesocialinteractionsbetweentherobotandthehumans withwhichitisinteracting.Therobotwasdesignedsothatfeedbackfromhumanobserverswas recordedbythesoftwareagentandtheagentwouldinturnmodifytherobot'saffectiveexpressionswithgeneticalgorithmswhichadaptedtothefeedbackoftheobserver.Thepaperdiscussed theempiricalresultsfromoneobserver,whointeractedwiththerobotfor163iterations.Thepaperdidnotdiscusshoweachemotionwasexpressedthroughmotion,direction,andvelocity. 2.3.2AnthropomorphicRobotsThatRelyHeavilyonNon-FacialandNon-VerbalAffective Expression Generally,non-facialandnon-verbalaffectiveexpressionisusedasasecondarymethodto relayaffect.HoweverinthreeofthesiximplementationscoveredinthisreviewMizoguchiet al.,1997,MoshkinaandArkin,2005,Bruce,Nourbakhsh,andSimmons,2002,eitherbody movement,posturesororientationwereasignicantmeansofaffectiveexpressiontothosewho interactedwiththeirrobots. Vikia,amuseumrobotdevelopedbyBruce,Nourbakhsh,andSimmonswasofinterest becauseitorientedtowardamuseumvisitor.Itwasanavatarthatappearedonamonitorattached toaxedroboticbaseandinteractedinthepersonalandsocialdistancezones.Ifamuseumvisitorwasdetectedwithinthesocialzone,theywereacknowledgedthroughfacialexpression,verbal interaction,andorientation,whichindicatedfocusofattention.Bruce,Nourbakhsh,andSimmons determinedthataddingtheabilitytoorienttherobottowardthevisitordisplayedrobotinterestinthevisitorandincreasedthelikelihoodthatvisitorswouldinteractwithVikia.Theywere 28

PAGE 45

abletosupportthisclaimwithdatacollectedoverfourdaysandtrialsperformedtwicedaily.The resultsindicatedthattheuseoforientationandnofacialexpression p -valueof0.002with99% condencehadahigherprobabilityofcompellingvisitorstostopandinteractwiththerobotthan usingnofacialexpressionandnoorientation p -valuenotpresentedforthisscenario,orjustusing facialexpressionandnoorientation p -valueof0.042with95%condence.Therewasanadditiveeffectifbothfacialexpressionandorientationwereusedtoencouragevisitorstointeractwith therobot.AlthoughVikiadoesutilizetheanthropomorphicfeaturesoftheavatarface,theuseof orientationdisplayedrobotinterestandplayedasignicantroleinachievingthegoalofattracting peopleforinteractionregardlessofwhetherafacewaspresentornot. Mizoguchietal.developedanexpressivemobilerobotthatwasanimal-likeinappearance.Therobotincludedahead,twomoveablearms,andwheelsformobility.Theheadcontained largeroundeyesthatwerenotexpressive;howevertheheaddidhavetheabilitytoorienttoward auser.Theentirepurposeofthisrobotwastointeractwithhumans,morespecicallytheelderly, andtodisplayaffectusingbodymovement,poses,orientation,andproxemics.Itusedgestures patternedafterballetposestocommunicatewithhumans,mobilitytoexpressspatialdistancefrom theindividualwithwhichitwasinteracting,andtrajectoryofmotionasmethodsofaffectiveexpression.Therobotoperatedinthepersonalandsocialdistancezones.Whentherobotwasinthe personalzonetheindividualwithwhichitwasinteractingappearedtoshowgreaterinterestinthe robot.Thedistanceofinteractionwasdependentonthespeedofmovementoftherobot.When therobotwasmovingataslowspeeditcouldinteractineitherzone;howeverwhenitwasmovingatahighspeed,participantswerecomfortablewiththerobotinthesocialzone,butbecame uncomfortablewithitinthepersonalzone.Expressionwasgeneratedbythechangingofposes gestures,thespeedofmovementandposturechanges,rotationofthebody,andthepatternsof movements.Experimentswereconductedinalaboratorysetting.Participantsobserved11differentrobotposesandcompletedpsychologicalquestionnaires.Theparticipantsreportedtheircomfortwiththerobotandtheirimpressionthattherobottookinterestinthem.Orientationtowarda testerwascrucialintheimpressionofrobotinterest.Asignicantproblemwiththisapproachwas thattherewasnoassociationofamovementwithaparticularaffectiveexpression.Statisticaldata tosupporttheirconclusionswasnotpresented. 29

PAGE 46

MoshkinaandArkinutilizedspeedofbodymovement,tailandearpositions,andthe redilluminationofanLEDscreentoexpressaffectonaSonyroboticdogAIBO.ThemotivationsforthisworkweretodeterminewhethertheAIBOwasabletodisplayemotionsandifthe displayofemotionsmadeinteractionsbetweenparticipantsandtheAIBOmoreenjoyable.The emotionsexpressedintheirexperimentswerealertinterestcrawlinggait,earsandtailraised, friendlyinterestcrawlinggait,earsraised,andtailwagging,alertjoyfastwalk,earsandtail raised,friendlyjoyfastwalk,earsraised,tailwagging,angerfastwalk,earsat,tailraised,and redLEDscreen,andfearrobotbackedupusingacrawlinggait,tailandheaddown,redLED lightsoff.Thedogreactedtocommandsandstimuluspresentedinthepersonalandsocialproximityzones.Thelaboratoryexperimentswereconductedwithtwogroups,onethatinteractedextensivelywithastandardSonyAIBOandasecondgroupinteractedextensivelywiththeemotionequippedSonyAIBO.Laterthegroupswereabletointeractwiththeoppositetypeofdogand giveimpressions.Theirresearchpresentedadetailedexperimentaldesignandwassupportedwith athoroughstatisticalanalysisofresults.Theresultsindicatedthatthoseparticipantswhobelieved theAIBOdisplayedemotionsand/orpersonalityoutof10ineachgroupatotalof20participantsfoundthatemotionand/orpersonalitymadetheinteractionmoreenjoyable.Theyalso foundthatwomenweremorelikelytoattributeaffecttothedogthanmenmeanvaluewomen= 3.8,meanvaluemen=2.5,F=4.829, p < 0.043MoshkinaandArkin,2005. 2.3.3AnthropomorphicRobotsUsingNon-FacialandNon-VerbalAffectiveExpressionas aRedundantMethodofExpression Intraditionalsocialrobotimplementations,bodymovement,posture,orientation,color,and/or soundareusedforaffectiveexpressionredundancyasamulti-modalmethodofexpression.These robotsareperceivedtobecuteinappearanceandhaveexpressiveroboticfacesandaredevelopedexplicitlyforsocialinteractionBreazeal,2002,Ca nameroandFredslund,2000,Scheeff etal.,2000.Althoughtheyutilizenon-facialandnon-verbalaffectiveexpressionsforredundancy, theyarenottheprimaryfocusofthissurveyofimplementations.Therobotsutilizedinsearchand rescue,lawenforcement,andmilitaryapplicationsdonothaveanexpressivefaceforaffectiveexpressionandmustrelyonothermethodsofexpression. 30

PAGE 47

TherobotKismetdevelopedbyBreazealisanexampleofananthropomorphicrobot thatutilizesnon-facialandnon-verbalaffectiveexpressionmethodsforredundancyinsocialinteractions.WhenobjectsappeartooclosetoKismet,itdisplaystheexpressionofsurprisewith astartledfacialexpressionbutalsoitrapidlywithdrawsbymovingitsheadandneckawayfrom thestimulusandproducesavocalresponseBreazealandFitzpatrick,2000.Kismettypically interactswithcaregiverswhoareinthepersonalandsocialdistancezonesandexhibitsthepreviouslymentionedreactionwhenapersonorobjectinvadesitsintimateandpersonaldistancezones ofsafety.Additionally,Kismetdisplaysawithdrawingmovementforredundancypurposesfor theaffectiveexpressionsofdisgustandfear.Experimentswereperformedinalaboratorysetting; howevertheeffectivenessofKismet'snon-facialandnon-verbalaffectiveexpressionswerenot specicallyisolatedandevaluatedinthesesocialinteractions. 2.4ReviewofPsychophysiology PsychophysiologyfocusesontheinteractionbetweenthemindandbodyStern,Ray,and Quigley,2001.Stern,Ray,andQuigleydenedpsychophysiologyasanyresearchin whichthedependentvariablethesubject'sresponseisaphysiologicalmeasureandtheindependentvariablethefactormanipulatedbytheexperimenterisabehavioralone.PsychophysiologicalmeasuresareusefulevaluationtoolsforHRIstudiesifusedappropriately.Thereisa tendencyintheresearchcommunitytoattributetheresultsofpsychophysiologicalmeasuresto speciccausesand/oremotions;howeverthereadingsmaybeattributedtomultiplefactorsor confoundsmakingitdifculttoisolatethespecicfactorsItohetal.,2006,KiddandBreazeal, 2005,Kuli candCroft,2006,Liu,Rani,andSarkar,2006,Picard,Vyzas,andHealey,2001, Ranietal.,2004.Becausetheresultsarevisibleastangibleoutput,researchershaveatendency tomakestrongerassumptionsthanmaybeaccurateStern,Ray,andQuigley,2001. 2.4.1BasicActivitiesandResponseTendencies Psychophysiologicalmeasurescanbeanalyzedintermsofthreebasictypesofactivities: spontaneous,tonic,andphasic.Aspontaneousresponseisameasurableresponsethatoccurs whenthereisnoknownstimuluspresentedStern,Ray,andQuigley,2001.Tonicresponsesare thebaselineorrestinglevelresponsesofactivityforaparticularphysiologicalmeasure.Thislevel 31

PAGE 48

occurswhenparticipantsbeingmeasuredarenotrespondingtoaknownorunknownstimulus. Thismeasureistypicallytakenattheendofarestingperiod,typicallyinthelastthreetoveminutesofatenminuterestingperiod.Thephasicorevokedresponseoccurswhenparticipantshave discreteresponsestoaspecicorknownstimulus.Itisimportantduringthistypeofmeasurementtoaccountforinternalandexternalstimulithatmayimpactparticipants'responsestothe presentedstimuli.Thiscanbeaccomplishedthroughself-reportsorinterviewstomakesureother factorse.g.,stateofmind,mood,healtharenotcontributingtothemeasuredresponses. Therearetwotypesofpsychophysiologicalresponsetendencies,stimulus-responsespecicity andindividual-responsestereotypythatcommonlyoccurinpsychophysiologicalstudies;however theyarenotmutuallyexclusive.Stimulus-responsespecicityiswhenastimulusorstressorproducesasimilarpatternofphysiologicalresponsesamongmostsubjectsorparticipantsstudied. Typically,morethanonetypeofresponseisinvolvedbutthepatternofresponseswouldbeconsistentamongmostparticipantssubjectedtothesamestimulusorstressor.Individual-response stereotypyoccurswhenafewindividualsexhibitapatternofresponsesdifferentthanexpected toaspecicstimulusorstressor.Also,individualsmayhavethesameidiosyncraticresponseto differentstressors,nomatterwhatthestressorsmaybeStern,Ray,andQuigley,2001. 2.4.2CommonPsychophysiologicalMeasures Therearenumeroustypesofpsychophysiologicalmeasuresavailabletoresearchers;howevertheytendtobeapplicationandenvironmentspecic.Ifaparticipantisinalaboratorysetting,inaxedlocationconnecteddirectlytostationaryequipment,theavailablemethodsofmeasurementarenumerous.Themostcommonmeasuresusedinacontrolledlaboratorysettingare: cardiovascularsystemheartratevariabilityHRV,respiratorysinusarrhythmiaRSA,cardiac outputCO,interbeatintervalIBI,bloodpressureBP;electrodermalactivityskinconductanceactivitySCA,skinconductanceresponseSCR;respiratorysystembreathsperminute, respirationvolume;muscularsystemelectromyographyEMG;andbrainactivityelectroencephalographyEEGandimagingBrownley,Hurwitz,andSchneiderman,2000,Dawson, Schell,andFilion,2000,Stern,Ray,andQuigley,2001,TassinaryandCacioppo,2000.Some ofthesemeasuressuchasEEGandimagingarenotconducivetoHRIstudiesbecausethepartic 32

PAGE 49

ipantmustbeinaspecictestinglocation,withlittleornomovement,anddirectlyconnectedto thetestinganddatacollectionequipmentforaccuraterecordingofdata. ThemostcommonmeasuresusedinHRIstudiesinclude:HRV,RSA,IBI,BVP,SCR,and EMGItohetal.,2006,Kuli candCroft,2005,Kuli candCroft,2006,Liu,Rani,andSarkar, 2006,Ranietal.,2004.Thesepsychophysiologicalmeasuresareavailableinambulatoryrecordingunits,whichallowsparticipantstobeplacedinaeldlocationorevenallowsthemtobe mobile.However,inmostcasesthemeasuresmustbeadjustedformovementartifactsorsignal noise. 2.4.3AdvantagesofUsingPsychophysiology ThereareadvantagesofusingpsychophysiologicalmeasuresinHRIapplicationsandexperiments.Theprimaryadvantageisthatparticipantscannotconsciouslymanipulatetheactivitiesof theirautonomicnervoussystemANS;thereforethereadingsreectparticipants'stateduringthe timeofevaluationItohetal.,2006,KiddandBreazeal,2005,Kuli candCroft,2006,Liu, Rani,andSarkar,2006,Picard,Vyzas,andHealey,2001,Ranietal.,2004.Additionally,psychophysiologicalmeasuresofferanon-invasivemethodthatcandeterminethestresslevelsand reactionsofparticipantsinteractingwithtechnologyItohetal.,2006,Kuli candCroft,2006, Liu,Rani,andSarkar,2006,Picard,Vyzas,andHealey,2001,Ranietal.,2004. 2.4.4PsychophysiologyIssues Theuseofpsychophysiologymeasurescanposesignicantchallenges.Theabilitytogather reliabledatafromparticipantsinreal-worldHRIscenarioscanbedifcultKiddandBreazeal, 2005.Properpreparationoftheareawhereelectrodesareplaced,locationofelectrodeplacement, andmakingsureappropriateamountsofconductinggelorpasteareusedarefactorswhichimpact thequalityofrecordings. Itisimportantandsometimescomplicatedtodeterminebaselinevalues;andthelawofinitial valuescanmakethisissueevenmoreproblematicBrownley,Hurwitz,andSchneiderman,2000, Dawson,Schell,andFilion,2000,Stern,Ray,andQuigley,2001,TassinaryandCacioppo, 2000.TheLawofInitialValuesindicatesthattheinitialstateofaparticipantdeterminesthe levelofpossiblechangesinthatstatethatcanoccurStern,Ray,andQuigley,2001.Ifpartici33

PAGE 50

pantsarerecordedathighinitialstates,thenfurtherincreasesinphysiologicalresponselevelsare limited,similarlyifparticipantsstartatlowerinitialstates,itwilllimitfurtherdecreasesinresponselevels. 2.4.5HabituationandResponseFactors Habituationisafactorthatreducesparticipants'responsesduetorepetitivepresentationof thesameorsimilarstimuliinpsychophysiologicalstudies.Therearetwoprimarytypesofhabituation:short-termoccursduringasingleevaluationsessionandlong-termoccursovermultiplesettingsoveraperiodofdaysorweeks.Habituationoccursmorerapidlywhenastimulus ispresentedfrequently.Onemethodtoreducetheeffectsofhabituationistoaskparticipantsto completearatingquestionnairebetweenthepresentationofeachstimulustoinduceabehavioral response.HabituationhasitsstrongesteffectstowardtheendofanystudyandneedstobeconsideredintheevaluationofdataduringpsychophysiologicalstudiesStern,Ray,andQuigley,2001. Threeresponsefactorsmayneedtobeconsideredinanypsychophysiologicalstudy:orienting response,startleresponse,anddefensiveresponse.Theorientingresponserelatestohowaparticipantrespondstonovelstimuli.Itcausestheparticipanttoorienttowardthenovelstimulito identifywhatitisanditslocation.Oncetheparticipantdeterminesthestimulusisnotathreator concerning,theeffectsoftheorientingresponseareinhibited.Therefore,dependingonthetypeof test,therstfewsecondsfollowingthepresentationofanovelstimulusshouldinsomecasesbe disregardedwhenevaluatingthedatadependingontheapplication.Therearesomecaseswhere researchersmaywanttoevaluateormeasuretheorientingresponsetowardarobotpresented toparticipantsaspartoftheirstudy.Thestartleresponseoccursduetoasuddenonsetofanintensetypeofstimuluse.g.,doorslamorlightningstrike.Datacollectedafterastartleresponse wouldbehandledsimilartoanorientingresponsebydisregardingthedatafortherstfewsecondsfollowingthepresentationofthestimulus;howeverthiswouldbedependentonthefocusof theresearchstudy.Thedefensiveresponseoccursasaresultofintense,threatening,dangerous,or painfulstimuli.Thistypeofresponsepreparesthebodyforghtorightactivation.TheinclusionofthisdatawoulddependonthetypeofstudyconductedStern,Ray,andQuigley,2001. 34

PAGE 51

2.5ImplementationsUsingPsychophysiologicalMeasuresinHRI ImplementationsusingpsychophysiologicalmeasurementsinHRIfallintothreeprimarycategories:participantemotiondetectionand/oridenticationbasedonphysiologicalmeasuresPicard,Vyzas,andHealey,2001,Ranietal.,2002,Kuli candCroft,2003,Ranietal.,2003, Sarkar,2002;evaluationofparticipants'physiologicalresponsestotechnologyKuli candCroft, 2006,Kuli candCroft,2005;andreal-timerobotcontrolandbehaviormodicationsbasedon physiologicalresponsesfromparticipantsRanietal.,2004,Itohetal.,2006,Liu,Rani,and Sarkar,2006.Thereareatleasttenimplementationsrelatedtotheuseofpsychophysiological measuresintheroboticscommunity.Ofthetenimplementationsdiscussedinthissection,only veinvolvepsychophysiologicalmeasurementsofparticipantsindirectinteractionwithsometype ofrobotRanietal.,2004,Itohetal.,2006,Kuli candCroft,2006,Liu,Rani,andSarkar, 2006,Kuli candCroft,2005.Ofthetenstudiespresented,onlytheKuli candCroftstudyincludedasignicantnumberofparticipantswith36individualsKuli candCroft,2006.Inthe otherstudiespresented,ifthenumberofparticipantswasgiven,therangewasfromoneto14 whichmakesitdifculttovalidatethereliabilityoftheresultspresented.ThishasbeenaconsistentissuewithreliabilityandvalidityofmetricsandmeasurementsinHRIresearchstudies. 2.5.1ImplementationsUsingPsychophysiologicalMeasuresforEmotionDetectionand/or Identication Thereareveprimaryresearchstudiesthatfocusonusingpsychophysiologicalmeasurements todetectand/oridentifyspecicemotionsexpressedbyparticipants.Inmostcases,thesestudies arepreliminaryinvestigationstoformthebasisoffurtherhuman-robotinteractionstudiesand/or thedevelopmentofacontrolarchitectureorbehaviorsystem. Picard,Vyzas,andHealeyfocustheirstudyonthedevelopmentofamachinethatcan accuratelyrecognizeeightdistincthumanemotionsgivenfourphysiologicalsignals.Theyindicatethatmachineintelligencemustalsoincludeemotionalintelligence.Thisstudyinvestigates acommonissueinmultiplesessionpsychophysiologicalmeasurementswhichistheproblemof day-to-dayvariationsinaparticipant'semotionalresponses.Foraccurateaffectrecognitionthey feelitisimportanttoincludemultipletypesofsignalsfromtheparticipant,andtoobtaininformationrelatedtotheparticipant'scontext,situation,goals,andpreferencesPicard,Vyzas,and 35

PAGE 52

Healey,2001,BickmoreandPicard,2004.Theyusedasingle-participantmultiple-daydatacollectionmethod.Theparticipantthatwasusedwasanactresswhoexpressedeightdifferentemotions:noemotionneutral,anger,hate,grief,platoniclove,romanticlove,joy,andreverence.The actressnotonlyexpressedeachemotionexternally,butfocusedonfeelingeachemotioninternally. Theexperimentsincluded25-minutedailysessionsacross20days.Thevephysiologicalsignals recordedwereelectromyographyEMGofthemasseterfacialmuscles,bloodvolumepressure BVP,heartrateHR,skinconductanceSCR,andrespiration.Thephysiologicalsignalswere processedusingSequentialFloatingForwardSearchSFFS,FisherProjectionFP,andtheirown HybridSFFSwithFisherProjectionSFFS-FP.TheresultsusingtheSFFS-FPalgorithmsindicatedthattheyobtainedan81%recognitionaccuracyfortheeightcategoriesofemotionswhich ishigherthanmachinerecognitionofaffectfromspeech-70%andalmostasaccurateasautomatedrecognitionofaffectfromfacialexpressions-98%Picard,Vyzas,andHealey,2001, BickmoreandPicard,2004.Theseresultsweresignicantbecausendingsdiscussedinthepsychophysiologyliteratureindicatedthatonlyarousallevelscouldbedetectedthroughtheuseof psychophysiologicalmeasuresPicard,Vyzas,andHealey,2001,BickmoreandPicard,2004. Ranietal.focusedtheirinitialstudyontheideathatifarobotcandetectstressquickly, thenitcanrespondtothehumaninreal-time.Arobotwasnotusedintheirinitialstudybutinsteadtheyhadparticipantsplayvideogamesandassessedtheirstresslevelsthroughself-report, heartratevariabilityandinterbeatintervalIBI.Afrequencydomainanalysiswasperformedof theIBIsignaltodetectwhetherparticipantswereexperiencingstress.Theyanalyzedtheresultsof participants'stressanddevelopedaroboticarchitecturetocontrolarobotbasedonpsychophysiologicalinputsreceivedfromthehumanwithwhichtherobotisinteracting.Thedevelopedrobot architectureincludedacollectionofelectrocardiographyECGsignalsfromparticipantsandthe calculationoftheIBI.Next,theyperformedawavelettransformation.Theauthorsthencalculated standarddeviationsofthesympatheticandparasympatheticfrequencybands.Thesestandarddeviationsbecameinputvariablesintoafuzzylogicsystemandtheoutputwasastressindexvalue. Ifthestressindexvaluewasgreaterthanathresholdvalue,therobotwouldreceiveanalarmsignal andwouldtakeactiontoassistparticipants.Someproblemsencounteredweresimulatingstressfulsituationsthatelicitedtheappropriateresponse,dayvariabilityinparticipants,andvariability betweenparticipants. 36

PAGE 53

Sarkarproposedanapproachandperformedsomeinitialexperimentsforasystemthat wouldenablearobottorecognizethepsychologicalstateofahumanwithwhichitisinteracting andmodifyitsactionsorbehaviorstomakethehumanmorecomfortableinteractingwiththerobot.Thisstudywasbasedontheassumptionsthattheaffectivestateoftheparticipantwasdirectly relatedtotheinteractionwiththerobotandthatonlypsychophysiologicalmeasureswereusedto recognizeaffect.Theseassumptionsarelimitingandnotrealisticbutnecessaryforthetractability ofthestudy.Agoalofthisstudywastorecognizehumanaffectthroughtheuseofpsychophysiologicalmeasures.Thenextgoalwastoidentifytheroboticactionsassociatedwiththemeasured affectivestateandmodifytherobot'sactionstoaltertheaffectivestateofthehumanwithwhich itwasinteracting.Thenalgoalwastodesigncontrolrulesfortherobottoassociateactionswith theresultingaffectexpressedbythehumanwithwhichitwasinteracting.ThestudyutilizedHRV, EMGofthecervicaltrapeziusmuscle,temperatureanalysis,SCR,andECGsignals.Theycreated anonlinestressdetectionalgorithmthatwasbasedonECGsignalsandthepowerspectrumof theIBIderivedfromtheECGsignaltoobtainfrequencybandsforthesympatheticandparasympatheticactivityoftheANS.Thisdatawasprocessedusingfuzzylogictoformthebasisofthe controlarchitecturedescribedinRanietal.,2002. Ranietal.focusedtheircontinuedresearchstudiesonaffectrecognitionbasedon physiologicalmeasuresobtainedfromawearablebiofeedbacksensorsystem.Thestudyincluded sixparticipantsandwasafullywithin-subjectsdesign.Theparticipantsweregiventwoversions ofthreeproblemsolvingtaskssolvinganagrams,mathproblemsolving,andsounddiscriminationofvaryingdifcultyacrosssixexperimentalsessionstoinduceparticipantanxiety.They measuredECG,SCR,EMGofthecorrugatorsuperciliileftbrowandmasseterjawmuscles, skintemperature,andrelativepulsevolume.Self-reportswerealsoutilizedtocorroboratephysiologicaldatacollectedwithparticipantanxietylevelsreported.Thephysiologicalsignalswere processedusingfuzzylogicalongwithdecisiontreelearningforaffectdetection.Thedatawas dividedintotwosets,onefortrainingthesystemandtheotherfortestingthesystem.Theresults indicatedtheywereabletodetectanxietyreliablyinparticipantsinvolvedintheproblemsolving sessions.Theyfoundthatthedecision-treelearningclassicationsystemwasmorereliablethan thefuzzylogicsystemofclassication. 37

PAGE 54

Kuli candCroftbegantheirseriesofresearchstudiesbyestimatingparticipantintent usingphysiologicalsignalsandperformedsomepreliminarytests.Theyfeltthatbydetermining participants'intentthroughphysiologicalmeasures,therobotcouldgainabetterunderstandingof participants'ratingofitsperformancewithouthavingtopollparticipantsrepeatedlyforexplicit feedback.Theydiscussedtheimportanceofusingmorethanonephysiologicalsignalfordeterminingparticipantintentaccurately.Forthepurposeofthisstudytheyusedavalenceandarousal systemofevaluatingintent.Theymeasuredbloodvolumepressure,SCR,chestcavityexpansion andcontraction,andEMGofthecorrugatorsuperciliieyebrowmuscle.Theyprocessedthesignalsusingafuzzyinferenceenginewithvesetsofrules.TherstsetofrulesevaluatedtherelationshipbetweenSCRandarousal.Thesecondsetofruleslookedattherelationshipbetween EMGandvalence.Thethirdsetofrulescorrelatedcardiacactivitytovalenceandarousal.The fourthsetofrulesrelatedvasomotorresponsestoarousal.ThefthsetofrulescorrelatedrespiratoryactivitywithemotionalstateKuli candCroft,2003.Theexperimentsinthisstudyuseda picture-basedsystemandfollowedthepsychophysiologicaltestingandmeasurementprocedures developedbyLang,Bradley,andCuthbert.Theproceduresconsistedofabaselinemeasurementtakenfromparticipantsandthentheywereshownanemotionallyarousingimageforten secondsandthenwereaskedtoratetheemotionalcontentoftheimageusingvalenceandarousal scalesKuli candCroft,2003,Lang,Bradley,andCuthbert,1993.Thestudywasperformed usingfourparticipantsandinonecasetheEMGelectrodeswerenotproperlyattachedandthe datawasnotusable.Onaverage,arousalwascorrectlydetected94%ofthetime.Thechangeof valencewascorrectlydetectedonaverage80%ofthetime.Whenthevalencewascorrectlydetected,75%ofthetimethedirectionofthevalencewascorrectlydetected.Theyusedtheresults ofthisstudytodeveloparobotplanningandcontrolstrategywhichwouldinteractwithahuman andrespondtothehuman'semotionalarousalinreal-time.Theyusedthecollecteddatatotrain theirfuzzylogicplanningandcontrolsystem.Theyfoundthatrespirationrateswerenotusefulin determiningparticipants'arousalresponsesbecausetheresponsetimewastoolongforreal-time application.Additionally,theydeterminedthatchangesinheartrateweredifculttoassociate withaspeciceventorcontextKuli candCroft,2003.Skinconductanceresponseshowedalinearcorrelationtoarousalandwasshowntobeaneffectivemeasure.ResultsindicatedarelationshipbetweentheEMGmeasuresofthecorrugatorsuperciliimusclewithvalanceinparticipants. 38

PAGE 55

2.5.2ImplementationsUsingPsychophysiologicalMeasuresforEvaluationofParticipant ReactionstoTechnology TherearetwoprimarystudiesthatutilizepsychophysiologicalmeasurestoevaluatehowparticipantsrespondtoroboticimplementationsandbehaviorsKuli candCroft,2006,Kuli cand Croft,2005.Bothstudieswereconductedtodeterminehowparticipantsreactedtotheirrobotic manipulatorarm.Theyutilizedarobotmanipulatorarmandevaluatedparticipantsfortheiranxietylevelswhileexperiencingvariousmovementsoftheroboticarm.Therobotperformedtwo setsofmovements:pickandplace,andreachandretract.Therewerealsotwoscenariosforeach movementtype;asetofclassicpotentialeldsplannedmotionsandasetofsafeplannedmotions. Thereweretwogoalsassociatedwiththerststudy,participants'subjectiveandphysiological responsestotherobotmotions,anddetermineifaparticularsetofrobotmotionscouldreduceparticipants'anxietylevelsKuli candCroft,2005.Thegoalsofthesecondstudyweretovalidatea previouslydevelopedinferenceengineKuli candCroft,2003withastatisticallysignicantsamplesizeparticipants;todevelopandtestareliablesystemfordeterminingtheparticipants' responsestotherobotmotions;andtodeterminewhethertheperceptionofsafemotionsrelatedto thetypeofmotionpathplanningusedKuli candCroft,2006. Inbothstudiestheymeasuredheartrate,SCR,andEMGofthecorrugatorsuperciliieyebrowmuscle.Theauthorsdeterminedthatparticipants'arousalresponsescouldbemostreliably detectedwithSCR,butheartratehadacontributoryimpact,althoughlessreliable.Psychophysiologicalresponseswerecomparedwithparticipants'self-reports.EMGofthecorrugatorsupercilii musclewasnotareliablepredictorofparticipants'valencepositiveornegativeandarousallevel intheinteractionsbetweentherobotandtheparticipant.Inmostparticipantsnochangeswere noted.Theresultsindicatedthatparticipantshadlowerarousalresponseswiththesafeplanned motionsoftheroboticmanipulatorarmandfeltcalmerwhentherobotmotionswereslower.Participantstendedtoshowstrong,measurablephysiologicalresponsestofastroboticarmmovements.Theresultsalsoindicatedthatphysiologicalsignalsprovidedusefulinformationandadded alevelofperceivedsafetyforhumansinteractingwithrobots. 39

PAGE 56

2.5.3ImplementationsUsingPsychophysiologicalMeasuresforReal-TimeRobotControl andBehaviorModications Threeprimarystudieshavebeenconductedrelatedtotheuseofpsychophysiologicalmeasuresforthedevelopmentandimplementationofreal-timerobotcontrolarchitecturesandadaptationofrobotbehaviorsRanietal.,2004,Itohetal.,2006,Liu,Rani,andSarkar,2006. AstudybyRanietal.involvedthedevelopmentofaroboticsystemthatmonitoreda participant'sanxietylevelandwouldrespondappropriatelytoassisttheparticipant.Theyuseda subsumptionarchitectureinwhichtherobotwouldnormallyoperateinthewanderingmode;howeveriftherobotreceivedahighanxietylevelsignalfromtheparticipantitwouldstopthewanderingbehaviorandeitherrushtotheaidoftheparticipant.Iftherobotencounteredanobstacle orsomethingthatthreateneditssurvivalitwouldceaseallotherbehaviorstoattendtoitssurvivalthenreturnrsttoanyaffectsignalsdetectedandthentoawanderingmode.Theparticipant playedvideogamesofdifferingdifcultytoinducedifferentaffectlevels.Theyusedself-report questionnairesandperformedmeasurementsforheartratevariabilityHRV,IBI,skinconductanceresponseSCR,andelectromyographyEMGofthecorrugatorsuperciliieyebrowand masseterjawmuscles.Thestudyresultsindicatedthatcardiacactivity,SCR,andEMGwereall goodindicatorsofanxietyandcorrelatedwiththeparticipant'sself-report.Onelimitationofthe studywasthatonlyoneparticipantwasusedintheexperimentsconducted. Itohetal.,developedabioinstrumentationsystemtomeasurehumanstresswheninteractingwithaxedhumanoidrobotwithonlyanupperbody.Theirwearablesystemmeasured ECG,respiration,EDAchangesinskinresistance,pulsewavetransittime,bloodpressure,and upperbodymovements.TheexperimentsreliedheavilyonIBIderivedfromECGtomeasurethe activityofthesympatheticLF-HRVandparasympatheticHF-HRVorRSAdivisionsoftheautonomicnervoussystemANS.Ifparticipants'stresslevelincreasedpastacertainthresholdthen therobotwouldshaketheparticipants'handtodecreasestresslevels.Thephysiologicalresponses indicatedareductioninparticipants'stressaftertherobotshooktheirhand.Theirsystemmodied therobot'sbehaviorsinreal-timeinresponsetothephysiologicaldatacollectedfromparticipants. Resultsindicatedthatbloodpressureandpulsewavetransittimeweredisruptedduetomovement artifactsandthedatawasnotuseful;howevertheratioofsympatheticandparasympatheticactivitywasusefultodetectparticipants'stresslevelsduringtheirrobotinteractionsItohetal.,2006. 40

PAGE 57

Liu,Rani,andSarkarperformedastudyinwhicharobotmodieditsbehaviorbased onthepsychophysiologicalresponsesofthepersonwithwhichitwasinteracting.Inthisstudy14 participantsperformedtwodifferentversionsofrobot-basedbasketballRBB,counterbalanced. Inoneversion,thegamedifcultywasbasedonparticipants'performanceandintheotherversion thegamedifcultywasbasedonparticipants'physiologicalreadingsforanxiety.Asparticipants' anxietylevelincreasedthedifcultylevelwoulddecreaseandviceversa.Themodicationofthe gameoccurredinreal-timeinresponsetoparticipants'anxietylevelsobtainedfromphysiologicaldatacollected.Thestudyusedself-reportsofanxietyinadditiontomeasuringcardiovascular activityIBI,relativepulsevolume,pulsetransittime,andpre-ejectionperiod,SCRtonicand phasic,andEMGactivityfromthecorrugatorsuperciliieyebrow,zygomaticuscornerofthe mouth,anduppertrapeziusshouldermuscles.Theresultsindicatedthat11outof14participantshadloweranxietylevelsplayingthepsychophysiological-basedversionofRBBthatadjusteddifcultybyparticipants'measuredanxietylevels.Additionally,nineparticipantsof14had improvedperformancescoreswiththepsychophysiological-basedversionofRBBLiu,Rani,and Sarkar,2006. 2.6Summary Thischapterbeginswithaverybriefdiscussionoftheliteratureassociatedwiththetheoryof emotionandemotionalmodelswhichwasnecessarytoexplaintheimpactofemotiononhumanrobotinteractionandalsotoprovidesupportfortheneedfornon-facialandnon-verbalmethods ofaffectiveexpressionasaprimarymethodofinteractionorforredundancypurposes.Thereis adiscussionofrelatedliteratureforfactorsinaffectiveexpressionincludingpresentationmethodsbodymovement,posture,orientation,color,andsoundandproxemicsdistancebetweentwo agents.Detailsareprovidedonthreecategoriesofrobotimplementationsthateitherusenon-facialandnon-verbalmethodsofaffectiveexpressionasaprimarymechanismofsocialinteraction orforredundancypurposesbasedonwhethertherobottypewasanthropomorphicornon-anthropomorphic.ThereisasummaryoftheseimplementationspresentedinFigure4onpage26.There isareviewofpsychophysiologyliterature,includingfoundationalknowledge,andHRIimplementationsusingpsychophysiologymeasureseitherforemotiondetectionand/oridentication,evaluationofparticipants'reactionstotechnology,real-timerobotcontrol,andbehaviormodications. 41

PAGE 58

Chapter3 TheoryandApproach Theobjectiveofthischapteristhepresentationofthetheoryandapproachtotheincorporationofnon-facialandnon-verbalaffectiveexpressionsbyproximityzoneintorobotimplementations.Thisincludesadiscussionofdesignapproaches,thedevelopmentofprescriptivedesign recommendationssynthesizedfromanextensivereviewofliterature,justicationfortheseprescriptiverecommendations,andanexampleofhowtheseprescriptiverecommendationscouldbe appliedtovictimassistanceintheurbansearchandrescuedomain.Additionally,detailsarepresentedofavericationoftheprescriptiverecommendationsconductedthroughvideoobservations ofapathusedformedicalassessmentofavictim.Thereiscoverageofsocialinteractionsbetweenappearance-constrainedrobotsandsoldiers,designrecommendationandthedevelopment ofnon-facialandnon-verbalaffectiveexpressionsinroboticimplementationsandconstraintsthat mayimpactthisapproach. 3.1DesignApproaches Therearedifferentapproachestodesign;howeverthissectionwilldiscusstwoprimaryapproaches,theiterativeapproachandrecommendationsbasedonestablishedknowledgeand/or literature.Theiterativeapproachiscommonlyusedtoaddressdifferentresearchproblemssuch asuser-interfacedesign,softwaredevelopment,andbuildingphysicalobjects.TheiterativeapproachtypicallyentailsaprocessapproachbyphaseZimmerman,Forlizzi,andEvenson,2004. IntheapproachdiscussedbyZimmerman,Forlizzi,andEvenson,therearesixphasesin thedesignprocess,whichencompasses:denetechnicalassessmentoftheproblem,teambuilding,hypothesize,discovercontexts,benchmarks,determininguserneeds,synthesizeprocess maps,opportunitymap,frameworks,personas,scenarios,constructdesignoffeaturesandfunctions,behaviors,interactionandowmodels,collaborativedesign,reneevaluation,scoping, 42

PAGE 59

interaction,specications,andreectpostmortem,opportunitymap,marketacceptance,benchmarking.Thisisarepetitivedesignprocessinwhichdesignerslearnthroughtrialanderrortestinganditerationsofdevelopmenttodeterminethebestdesigntomeettheusers'needs.Thistype ofdesignprocessisutilizedinuserinterfacedesign,todeterminethroughuserinvolvementwhat isneededandhowtodesigntomeetthoseneeds.Oftenitisdifculttodeterminewhatusersmay wantuntiltheyactuallyusetheproductandcandeterminewhatisworkingandwhereshortfalls mayexist.Thesoftwaredesignprocessoftengoesthroughasimilariterativeprocessofbuilding preliminarysoftwaretotesttheproofofconceptandthendesignimprovementsindifferentiterationsandthroughusertesting.Similarly,engineerswillgothroughaniterativeprocesswhen designingandbuildingphysicalobjectssuchascars.Aprototypeisoftendesignedandbuilt,then basedonfeedbackfromusersthedesignsaremodiedandimprovedbeforetheactualproductionitemisnalized.Thisisagoodapproach;howeveritcanbetimeconsumingandcostlywhen multiplegenerationsoriterationsaredesigned.Theendresultisoftenthedevelopmentofadesign heuristicwhichcanbeusedasafoundationorstartingpointtoreducethenumberofiterations infuturedesigns.Thedesignheuristicisasetofguidelinesorrecommendationsdevelopedasa resultoftrialanderrormethodsandexperimentation. Anotherapproachisthedevelopmentofasetofprescriptiverecommendationsbasedonlongstandingcustomorbasedonestablishedknowledge.Thatistheapproachtakeninthisresearch. Extensiveresearchintoliteratureassociatedwithpsychology,sociology,anthropology,animalbehavior,robotics,andcomputersciencewassynthesizedintoasetofprescriptiverecommendations forappropriatebehaviorsormethodsofnon-facialandnon-verbalaffectiveexpressionsbasedon thedistancebetweenagentsinteracting.Fromtheliterature,theseagentscouldbehumans,animals,avatars,and/orrobots;howeverthedevelopedprescriptiverecommendationsarebasedon establishedknowledgeassociatedwithdifferentmethodsofnon-facialandnon-verbalaffective expressionbyinter-agentproximity.Theseapproachesarenotmutuallyexclusiveandcanbeused together.Byusingestablishedknowledgeitcancomplementtheiterativeapproachandalsoreducethenumberofiterationsrequiredinthedesignprocess. 43

PAGE 60

3.2PrescriptiveRecommendationsforNon-FacialandNon-VerbalAffectiveExpressionby ProximityZones Asetofpresciptiverecommendationsfortheappropriatenessofthenon-facialandnon-verbalaffectiveexpressionsbyproximityzoneisneededtoguidetheapplicationofaffectivecuesfor particularrobotsandscenariosBethelandMurphy,2006a.Figure5proposessuchrecommendations.Theprescriptiverecommendationsprovideguidelinessuchastheuseofbodymovements andposturesareeffectiveinthepersonalandsocialdistancezones;howevertheyaredifcultto interpretintheintimatezonebecauseofsightconstraintsatsuchcloseproximityArgyle,1975. Bodyorientationisperceivableinanyoftheproximityzones.Theuseofcolorforaffectiveexpressionisappropriateinanyofthedistancezonesdependingontheintensityofthelight.Inthe caseofsound,itismostappropriatewhenexpressedintheintimateandpossiblythepersonaland socialzones;howeverinrelationtosoundinthepersonalandsocialzones,backgroundnoiseand environmentalconditionswillbeakeyfactorintheappropriatenessofthismethodBetheland Murphy,2008. 3.3JusticationforthePrescriptiveRecommendations TheprescriptiveaffectiveexpressionrecommendationsdescribedinFigure5isasynthesisof thecognitiveliteratureandroboticexperiencereectingthepreferrednon-facialandnon-verbal affectiveexpressionforthethreerelevantproximityzonesArgyle,1975,Hall,1966,Bethel andMurphy,2008.FollowingArgyleandHall,thepreferrednon-facialandnonverbalaffectiveexpressionsinthesocialandpersonalproximityzonesarebodymovementand posture;theyalsonotethatsoundcanbeheardatamediumlevelinthepersonalzoneandata loudlevelinthesocialzone.Bodymovementandposturearevisibleinbothofthesetwoproximityzonesandthereforeeasytointerpretwhethertheyareconstantasinpostureorchangingas inbodymovement;howeversoundisdependentonthevolumeandbackgroundnoiseineachof thesezones.Unlessthevolumecaneasilybeadjustedtoaccommodatedifferentdistanceranges, itisnotlikelythatsoundwillbeeffectiveinthesocialzone.Maedapreferredbodymovementinthepersonalandsocialproximityzonesastheprimarymethodofaffectiveexpressionon 44

PAGE 61

Figure5.Prescriptiverecommendationsfornon-facialandnon-verbalaffectiveexpressionbased onproximity.Theredcellsindicatethesemethodsarenotappropriateinthisproximityzone,the yellowcellsindicatethismethodmaybeappropriateintheseproximityzones,andthegreencells indicatethatthesemethodsareappropriateforuseintheseproximityzones. 45

PAGE 62

theKheperarobot.MoshkinaandArkinreliedonbodymovement,posture,andorientation inthesocialandpersonalzonesforconveyingaffectintheSonyAIBOusedintheirexperiments. FollowingtheworkfromBruce,Nourbakhsh,andSimmonsandNourbakhsh,Kunz, andWilleke,orientationisthepreferrednon-facialandnon-verbalmethodofaffectiveexpressioninthepersonalandsocialproximityzones.Bruce,Nourbakhsh,andSimmons discussthatusingtrackingandorientationalonewasmoreeffectivethanusingafacealoneorno faceandnotrackingforattractingvisitorinteractionsandindicatingattentiveness.Fincannonet al.discussthatorientationtowardtheworkerwaspreferredintheintimateproximityzone especiallyduringcommunicationinteractions.ArgyleandHalldonotdirectlydiscussorientation;howeverbasedondiscussionsofvisibilityinthesezones,orientationwouldbe observable. BasedonresearchfromSuganoandOgataandMoshkinaandArkin,illuminatedcolorisvisibleandaneffectivenon-facialandnon-verbalaffectiveexpressioninthepersonalandsocialproximityzones.SuganoandOgatapreferredtheuseofcolorintheformofilluminatedlightsontheWAMOEBA-1Rrobottoexpressaffectinboththepersonalandsocial proximityzones;howevertheWAMOEBA-1Rmustbeorientedtowardanobserverfortheaffectiveexpressiontobevisiblesincethelightsaretheonlymeansithasforexpressingaffect.The SonyAIBOusedbyMoshkinaandArkinutilizedaredilluminatedLEDscreentosupport affectiveexpressioninthepersonalandsocialzones.Theliteratureislackingontheuseofilluminatedcolorintheintimateproximityzone.BasedonthedescriptiongivenbyHallofwhat isvisibleineachproximityzoneanilluminatedcoloredlightislikelyvisible;howeveritwouldbe dependentonintensityasindicatedwithagreenentryinFigure5. ArgyleandHallsupporttheeffectivenessofsoundinthesocialzoneataloud volume;howevertheeffectivenessisunclearintheroboticsliterature.Bruce,Nourbakhsh,and SimmonsusessoundeffectivelyatthepersonalandsocialdistancesinVikiathemuseum robotoperatinginarelativelyquietenvironment;howeverFincannonetal.discussthatdue toexternalnoisefoundinthesearchandrescueenvironment,closerproximityimproveshearing. BasedontheFincannonetal.example,ear-to-robotsoundcommunicationwasmost effectiveintheintimatezone;howevertheappropriatenessofsoundinthepersonalandsocial proximityzonesisdependentonthebackgroundenvironmentalnoiselevelsandwouldneedto 46

PAGE 63

beeithermediumtoloudinvolumeasindicatedbyayellowentryinFigure5.Duethefactthat soundlevelswouldneedtobeadjustabletoaccommodatethevolumeneededtobeeffectivein thesocialzone,andinmostsituationsenvironmentalnoisesmayimpedesoundtransmissionat thesocialdistance.TheentryinFigure5forthesocialzoneismarkedyellowindicatingmaybe appropriateatthisdistance. Thepreferredaffectiveexpressionsintheintimatezoneareorientationandsound,because visionisconstricted,whichpreventstheperceptionofbodymovementandpostureasindicated withredentriesinFigure5Argyle,1975,Hall,1966.Fincannonetal.discussthatboth orientationandsoundarehighlyeffectiveintheintimatedistance;howevertheeffectivenessofilluminatedcolorinthiszoneisnotclearbasedontheliterature.Theresultsfromtheresearchstudy performedforthisdissertationindicatedthattheuseofcolorintheintimatezoneisappropriate dependingontheintensityofthelight. 3.4PrescriptiveRecommendationsAppliedtoVictimAssistance InordertoillustrateoneofthepotentialusesoftheprescriptiveaffectiveexpressionrecommendationsinFigure5,itishelpfultoconsiderhowitcouldbeappliedtoasearchandrescue robotwhichlocatesvictimsandstayswiththemuntilassistancecanarriveusually4-10hours Murphy,Riddle,andRasmussen,2004.Thehuman-robotinteractioninthiscasewouldspan allthreeoftheproximityzones:intimate,personal,andsocial.Therobotwouldrstapproacha socialdistancefromlocatedvictims,andinthissituationtherobot'sbodymovementsshouldbe sufcientlylargeorexaggeratedenoughtobevisiblebyvictimsatthesocialdistance.Slowand smoothmovementswouldhelptoreassurethevictims.Abruptandjerkymovementswouldgive theimpressiontherobotisangryBartneck,2000.Therobotshouldbeorientedtowardvictims toindicateconcernandattentivenessFong,Nourbakhsh,andDautenhahn,2003,Bruce,Nourbakhsh,andSimmons,2002,Nourbakhsh,Kunz,andWilleke,2003,BickmoreandPicard, 2004,Fincannonetal.,2004;howevertheapproachwouldbelessthreateningifmadefrom therightorleftsideiftheenvironmentpermitsWaltersetal.,2005.Therobotcouldalsobegin displayingblueilluminatedlighttoindicateacalmpresenceapproaching.Soundatthisdistance wouldnotlikelybeaudibleduetoenvironmentalnoiseoccurringinthedisastersitethatwould drownoutanysoundemittedbytherobotFincannonetal.,2004,BethelandMurphy,2008. 47

PAGE 64

Therobotwouldquicklymovefromthesocialzoneintothepersonalzoneinordertoperform apreliminarymedicalassessmentofvictims.Atthisdistancetherobotshouldbeslightlyraised andorientedtowardvictimstoindicateitsinterestandattentiontothem.Bodymovementsofthe robotatthisdistanceshouldbeslowanddeliberatetonotalarmvictimsandtokeepthemcalm. Atthepersonaldistancetheuseofanilluminatedbluelightingfeaturewouldbemorereadily visiblebythevictimandasdescribedbyArgylebluewouldexpresscalm,secure,tender, andpleasantaffectbytherobottowardvictims.Additionally,theuseofsoothingtonesNorman, 2004ormusicBartneck,2000couldbeutilizedtoexpresscalmnessandtendernesstoelicita calmingresponseinthevictims.Tonescouldbeusedtoindicateapositiveornegativeresponse toacknowledgecommunicationreceivedbytherobotfromthevictims.Additionally,tonesand/or musiccanbeusedbytherobottoindicateunderstandingtothevictimregardingtheirsituation. Therobotwouldalsooperateinbriefintervalsintheintimatezonewhenassessingthestatus ofvictims,andpossiblytoadministerairand/orwaterMurphy,Riddle,andRasmussen,2004. Actualbodymovementsandpostureswouldnotbevisiblebyvictimsattheintimatedistanceand thereforeitmustrelyonorientation,color,andsoundasmethodsofaffectiveexpression;however insomesituationsorientationcouldbeimpededduetospaceconstraintsand/orthelocationofthe robot.Therobotwouldneedtocontinuetomoveslowlyandshouldremainorientedtowardvictimsifatallpossibletoshowattentivenessandcaringsimilartothepersonalzone.Soundwould beeasilyaudibleatthisdistanceandwouldbeoneofthepreferredmethodsofexpressiontoreducethestresslevelsofvictimsduringinteractionsintheintimatezone.Itwouldbeimportant fortherobottoemitsoothingtonesormusictokeepvictimscalmduringmedicalevaluationsand robotcontact.Thiswouldbethemoststressfulinteractionbetweentherobotandvictims;thereforesoothingsoundswouldbeavaluabletoolduringthistypeofinteraction.Theuseoftheblue illuminatedlightwouldpossiblybeeffectiveintheintimatezoneaddingtothecalmingeffect. Duringvictimrecovery,therobotwouldoperateprimarilyinthepersonalproximityzone .461.22meters,givingtherespondersenoughstandoffdistancetomonitorthestatusofthe victimsuntilassistancecanarrive,whilenotbeingtoointrusive.Murphy,Riddle,andRasmussen surveyed28medicalprovidersandseveralrespondentsdiscussedtheneedforasearchand rescuerobottobecomfortingtovictims.Additionally,theycommentedonthecreepyappear 48

PAGE 65

anceoftherobot;thereforetheuseofthesenon-facialandnon-verbalaffectiveexpressionscould proveinvaluabletovictimsthatareexperiencinghighlevelsofstressandtrauma. 3.5VericationofthePrescriptiveRecommendationsThroughVideoObservationsUsinga PathforMedicalAssessmentofaVictim Thedetailsofarobot-assistedmedicalassessmentreachbacktaskillustratedinFigure6 arethefocusofavericationoftheprescriptiverecommendations.Theseincluderobot-eyeand victim-eyeviewsatsevenpointsofinterestforthetask;andthesuitabilityoffourofthetheve non-facialandnon-verbalmethodsofaffectiveexpressionineachofthethreeproximityzonestraversedbytherobot.Theappropriatenessoftheuseofilluminatedcolorbyproximityzonewasnot includedinthispreliminaryresearchinvestigationforthevericationofthisprescriptiverecommendationBethelandMurphy,2007. Figure6.DiagramofthemedicalassessmentpathtakenbytheInuktunrobotinrelationtothe simulateddisastervictim. 3.5.1MedicalAssessmentPath Therobot-assistedmedicalassessmentpathdepictedinFigure6,robotmovements,postures, andorientationwasderivedfromobservationsoftwovideotapedstudiesconductedbyMurphy, Riddle,andRasmussenandRiddle,Murphy,andBurke.Duringthosetwostudies, 49

PAGE 66

medicalpersonneldirectedtherobotoperatorsregardingthemovements,locations,andorientation oftherobotwithrespecttothevictim.ThepositionsshowninFigure6wereliteralintermsofthe robot'sdistancefromthevictim.Thesepositionswerekeypositionsformedicalassessment;howevertheynaturallycoincidedwiththeproximityzonesdisplayedinFigure6BethelandMurphy, 2007. Figure6containssevenkeypointsofinterestinwhichtherobotmustperformspecictasks relatedtovictimlocationandmedicalassessmentBethelandMurphy,2007. T1socialzoneRobotlocatesthevictim,raisestofullheight,rotatesside-to-side,andtilts thecamerafaceupanddown,tosurveytheareasurroundingthevictim. T2personalzoneRobotperformsapreliminarymedicalassessmentofthevictimandsurroundingenvironment.Therobotperformsthesamesequenceofmovementsthatoccurred atT1.Aftercompletionofthepreliminarymedicalandenvironmentalassessment,therobot lowerstoaatpositionandmovestopointT3ofFigure6. T3intimatezoneRobotgentlypushesthevictiminthefootregiontodetermineifthevictimisreactiveandconscious. T4intimatezoneRobotmovestotheheadregionofthevictimtoobtainmedicalsensor readingsandtodetermineifthevictimisbreathing.Thisisthemostintimidatingposition therobotcanbelocatedinrelationtothevictimSeeFigure7.Therobotonlystaysinthis positionlongenoughtoobtainthenecessarydata. T5personalzoneRobotmovesbackwardintothepersonalzonetosurveytheoverallconditionofthevictimandthesurroundingenvironment.Therobotperformsthesamemotion sequencesperformedatpointT1. T6intimatezoneRobotmovesintotheintimatezonetothevictim'schestandupperabdominalregion.Therobotfocusesonthechestareatodetermineifthevictimhasanyvisible injuriesanddeterminewhetherthevictimhaschestmovementindicativeofbreathing. T7personalzoneRobotmovesbackwardintothepersonalzonetopointT7,whereitperformsanotheroverallvictimstatusassessmentsimilartopointT5.PointsT5andT7arecomfortablelocationsforcontinuedsocialinteractionandvictimmonitoring. 50

PAGE 67

Figure7.ObservationsatpointT4fromtherobot'sandvictim'sperspectivesintheintimateproximityzone. 3.5.2SocialProximityZone Robotactivitiesinthesocialzonemainlyinvolvesearching,possiblevictimlocation,andgeneralstructuralassessment.Interactionwithavictimatthisdistanceislimited;howeverthevictim maybeabletoseetherobotandneedstofeelcomfortablewithitapproachingforassessment.ObservationsatpointT1onFigure6indicatethatthevictimisabletoseetherobot,whatposition itisin,largebackandforthrotationalmovements,translationalmotion,andorientationwhichis consistentwiththeprescriptiverecommendationsdescribedinthesocialsectionofFigure5. Smallermovementsoftherobotsuchasthetiltingmotionofthecamerafacearelessvisible atthisdistance.Additionally,fromtherobotperspectiveatpointT1,therobotisabletodetectthe victim;howeveritisnotabletodeterminethemedicalconditionofthevictimatthisdistance. 3.5.3PersonalProximityZone Thepersonalproximityzoneisappropriateforhuman-robotsocialinteraction,assessingand monitoringtheoverallconditionofthevictim'sbody,inadditiontodeterminingthestructural integrityoftheenvironmentsurroundingthevictim.Therobotiscapableofviewingtheentire lengthofthevictim'sbody,performingapreliminaryassessmentofinjuries,anddeterminethe stabilityofthestructuressurroundingthevictimatpointT2fromFigure6. 51

PAGE 68

Consistentwiththeprescriptiverecommendationsdescribedinthepersonalzoneentriesin Figure5thevictimiscapableofviewingallthemovementsoftherobotincludingthesmallmovementsofthecameraface.Additionally,atpointT2thevictimcandeterminetheorientationof therobot.Atthisdistance,sounds,tones,and/ormusiccouldpossiblybeheardifitismediumto loudvolumeanddependentonthebackgroundenvironmentalnoiseBethelandMurphy,2006a, BethelandMurphy,2006b,BethelandMurphy,2008. AtpointT5asshowninFigure6,therobotcanperformanoverallscanofthevictim'sbody andthesurroundingenvironment.Thevictim'suppertorsoandfaceareclearlyvisiblewhenthe robotisorientedtowardthevictim'sface.Fromthevictim'sperspective,therobot'stranslational androtationalmovements,orientation,posture,andsmallmovementsofthecamerafacearevisible.ConsistentwiththeentriesinthepersonalzoneofFigure5,therobotisabletocommunicate affecttothevictimthroughtheuseofbodymovements,postures,andorientation.Mediumtoloud sounds,tones,and/ormusiccouldpossiblybeusedforaffectredundancy,iftheenvironmental conditionsareamenableBethelandMurphy,2006b. Therobotisabletoperformanotheroverallevaluationofthevictim'sstatusandsurveythe areasurroundingthevictimatpointT7fromFigure6.Thislocationisalsoasafestandoffdistancetocontinuallymonitorthevictim'sstatus.SocialinteractionsatpointT7betweenthevictimandtherobotwouldbelessstressfulforthevictimforlong-terminteractionsuntilassistance canarriveforextricationBethelandMurphy,2006a,BethelandMurphy,2008.AtpointT7 mostofthevictim'sbodyisvisibletotherobotwhenitisorientedtowardthefaceofthevictim.Orientationtowardthevictim'sfacecanindicateliking,interest,andcaringinadditionto thedevelopmentoftrustbetweentherobotandthevictimBickmoreandPicard,2004.From thevictim'sperspectiveatpointT7,therobot'smovements,posture,andorientationarevisible; consistentwiththeprescriptiverecommendationsdisplayedinthepersonalzoneentriesinFigure 5.Sounds,tones,and/ormusicmaybeappropriateforuseatthislocation;howeveritwoulddependonbackgroundenvironmentalnoiselevelsBethelandMurphy,2006a,BethelandMurphy, 2006b,BethelandMurphy,2008. 52

PAGE 69

3.5.4IntimateProximityZone Theintimateproximityzoneisusefulforclose-rangevictimmedicalassessment.FromobservationsoftheoriginalmedicalreachbackstudiesdiscussedinMurphy,Riddle,andRasmussen, 2004,Riddle,Murphy,andBurke,2005therobotonlyremainsatthislocationT4inFigure 6longenoughtoobtainmedicalsensorreadingsfromthevictim.Atthisdistanceparticipantsin thevictimrolereportedtherobotascreepyinappearance,especiallywhentherobotisnearthe face.Atthislocation,itisimportantthatrobotmovementsareminimalandareperformedslowly toavoidstartlingthevictimBethelandMurphy,2008. AtpointT3fromFigure6,therobotisusedtogentlypushthevictim'slegtodetermineif thereisanyresponse.Atthislocation,thechestandfaceofthevictimisnotvisibletodetermine ifthevictimisbreathingoraware.Ifthevictimmovesinresponsetothispush,medicalpersonnel candeterminethatthevictimisdisplayingawarenessMurphy,Riddle,andRasmussen,2004, Riddle,Murphy,andBurke,2005. ConsistentwiththeprescriptiverecommendationsdescribedintheentriesintheintimatesectionofFigure5,therobotisnotvisibletothevictimintheintimatezoneatpointT3.Theuseof bodymovementandpostureatthislocationisnotappropriate;howeversounds,tones,and/ormusicwouldbeanappropriatemethodofaffectiveexpressionBethelandMurphy,2006a,Bethel andMurphy,2006b,BethelandMurphy,2008. Therobotcanbeusedtoobtainmedicalsensorreadingsfromthevictim'sfacialareawhile activeatpointT4ofFigure6.Atthislocation,medicalpersonnelcanassessthevictim'slevelof awarenessandinteractionthroughobservationoffacialresponsesseeFigure7Murphy,Riddle, andRasmussen,2004,Riddle,Murphy,andBurke,2005.Fromthevictim'sperspective,the robotiscloseandreportedascreepyinappearanceatthisdistanceasshowninFigure7.Asper theentriesintheintimatesectionofFigure5orientationoftherobotisvisible.Consistentwith theintimateentriesinFigure5thefullbodyoftherobotisnotvisibleatpointT4;thereforethe useofbodymovementandposturewouldnotbeanappropriatemethodofaffectiveexpressionat thislocation.Soundwouldbeamoreappropriatemethodofexpressingaffectatthislocationin theintimatezoneBethelandMurphy,2006a,BethelandMurphy,2006b,BethelandMurphy, 2008. 53

PAGE 70

Therobotcanbeusedtoevaluateavictim'sbreathingandabdomenatpointT6ofFigure6. Therobotisnotvisibletothevictimatthislocation.ConsistentwiththeprescriptiverecommendationsdescribedintheintimateentriesofFigure5,bodymovementandposturewouldnotbe anappropriatemethodofexpressingaffecttothevictimforsocialinteraction.Inthiscase,orientationoftherobotisnotvisible.ThemostappropriatemethodofexpressingaffectatpointT6 wouldbesounds,tones,and/ormusicasindicatedintheentriesfortheintimatedistancezoneof Figure5. Thisinitialevaluationthroughvideoobservationsofthemedicalassessmentpathprovided vericationofbodymovements,postures,orientation,andsomeofthesoundentriesBetheland Murphy,2007.Illuminatedcolorwasnotevaluatedinthisscenario.Thisinitialstudywasused toverifytheprescriptiverecommendationsandshowtheviabilityoffuturemorecomprehensive studieswithhumansubjects.Theprescriptiverecommendationshavebeendevelopedandapreliminaryvericationprovidethefoundationforapproachingtheuseofnon-facialandnon-verbal affectiveexpressioninappearance-constrainedandanthropomorphicrobotsfornaturalisticsocial interaction.Human-robotinteractionstudieshavefocusedontheuseoffacialexpressionstoprovidethatnaturalisticinterface;howeveritisoftenthecasethattherobotdoesnothaveafaceor theabilitytoutilizefacialexpressionsforaffectandsocialinteraction. 3.6SocialInteractionswithAppearance-ConstrainedRobotswithSoldiers Accordingtoanarticleonwashingtonpost.comGarreau,2007soldiersusingiRobot'sPackBotScoutrobotshavebondedwithandsociallyinteractedwiththeirrobotseventhoughtherobots displayednoaffectiveexpression.Thesoldiershavegonesofarastogivetheirrobotspromotions andcommendationsbasedontherobot'sperformanceintheeldandthefactthattherobotsaved thelivesofsoldiers.Thesoldiersviewtheserobotsasteammates.Whenacentipederobotwas downtoonelegafterdetonatingseveralimprovisedexplosivedevicesanditwasdraggingitsbody withitsremainingleg,aColonelwatchingtheactivitystoppedtherestoftherobot'smissionbecausehefeltitwasinhumaneGarreau,2007.Iftherobotsbecomeincapacitatedthesoldierswill taketheminforrepairbuttheywanttheirrobotreturnedtothem,notareplacement.Theywill insistthattheirrobotberepairedbecauseoftheirattachmenttotherobot,evenwhenitisbeyond 54

PAGE 71

repair.ThisreinforcestheconceptrstpresentedbyReevesandNass,thatifitmoves,humanswilltreatitasasocialentityandwillinnatelyinteractwiththeobjectinasocialmanner. 3.7DesignRecommendationsandtheDevelopmentofNon-FacialandNon-VerbalAffectiveExpressionsforUseinRoboticApplications Basedontheprescriptiverecommendationsordesignguidelines,non-facialandnon-verbal affectiveexpressionsneedtobedevelopedfortheroboticapplicationinwhichtherobotwillbe interactingwithahuman.Itispossibletodevelopmovements,postures,orientation,color,and soundprimitivesthatcanbeappliedtoanyrobot;howevermoreintricatemovementswillbelimitedtoaspecicfunctionalityofarobote.g.,polymorphicrobotsvs.xedplatformrobots. Movementpatternsthatdisplaynon-facialandnon-verbalaffectiveexpressions,robotorientationtowardthehumanswithwhichtheyareinteracting,androbotposturescanbeincorporatedintoroboticsystemsthroughsoftwareimplementation.Thiswouldeliminatetheneedfor physicalmodicationoftherobotortheneedtobuildnewrobotstoexhibitnon-facialandnonverbalaffectiveexpression.Itisimportant,whendesigningtoproduceacalmingeffect,thatthe robotmovesincontrolledandslowermovements,remainslowertothegroundifatallpossible toreduceanytypeofloomingeffect,andlimitstheuseofmovementwhenincloseproximityto humans.Todisplaycaringandattentivenesstowardhumans,therobotsshouldremainoriented towardthehumanasmuchaspossible.Ifthegoalistokeephumansatadistancefromtherobot oraparticularareasuchasrequiredincrowdcontrol,military,andlawenforcementapplications, therobotshouldmoveinquickanderraticmovementpatterns,possiblychargingtowardthehumanstokeepthematadistance.Fromthisresearchandtheliterature,movementsandbehaviors commonlyexhibitedinhuman-humaninteractionsareapplicableinhuman-robotinteractionsas well;howevermoreextensiveresearchisrequiredinallvemethodsofnon-facialandnon-verbal affectiveexpressiontodeterminethemosteffectiveuseofthesemethodsandspecicapplications. Forthepurposeofthisresearch,subjectmatterexpertsSMEs,JeffConoverapuppeteerwith DisneyandDanielDeutschaDisneyconsultant,wereutilizedtoassistinthedevelopmentoftypes ofmovements,andpostures,thatwouldbecalmingandlessthreateningtobeusedinsearchand rescueapplicationsorinanyapplicationwherethereisadesireforthehumaninteractingwiththe robottobecalmandexperiencelessstress.Additionally,theSMEsassistedinidentifyingmove55

PAGE 72

mentsandposturesthatwouldbethreateningoraggressivetokeephumansawayfromrobots, suchasneededinmilitaryandlawenforcementoperations. Fortheinitialimplementationassociatedwiththisresearch,lightblueneonlightingtapewas usedtoproduceacalmingeffectwhenusedinsearchandrescueapplications.Additionally,this lightingeffectilluminatedthebodyoftherobotwhichwasutilizedtodecreasethevictims'stress levelstomaketherobotmorevisiblewhenapproachingthem.Lightbluelightingwasselected becausebasedontheresearchliterature,ithasbeenrecognizedtoelicitacalmingandpleasant responseArgyle,1975.Moreextensiveresearchneedstobeconductedtodetermineamapping ofcolortospecicemotions. Afterthedevelopmentofthebodymovements,postures,andorientation,thesenon-facialand non-verbalmethodsofaffectiveexpressionwereimplementedinsoftwareontheactualrobot.The lightblueneonlighttapewasphysicallymountedontheunder-carriageofeachrobot.Thenext stepintheprocesswastotestthenon-facialandnon-verbalaffectiveexpressionswithhumanparticipantstodeterminetheeffectivenessofthesocialinteractions,thevalidityoftheprescriptive recommendations,andtoestablishandvalidateresearchmethods,metrics,andmeasurements. 3.8ConstraintstotheApproach Therearecertainlimitationstoimplementingthefullsetofnon-facialandnon-verbalaffectiveexpressionbasedontheprescriptiverecommendationsdevelopedinthisresearch.Depending onthedesignoftherobot,notallofthemethodscanbeimplemented.Forexample,theiRobot PackBotinitscurrentcongurationdoesnothavetwo-wayaudiocommunicationmakingthedevelopmentofsoundsdifculttoimplementwithoutdesignmodication. 3.9Summary Thischapterfocusesonthetheoryandapproachesnecessarytoincorporatenon-facialand non-verbalaffectiveexpressionintoroboticimplementations.Therearetwodesignapproaches discussedtheiterativedesignapproachandprescriptiverecommendations.Theseapproaches areoftencomplementaryandarenotmutuallyexclusive.Thereisadiscussionofthedevelopmentofasetofprescriptiverecommendationsfortheappropriateuseofvemethodsofnon-facial andnon-verbalaffectiveexpressionbodymovements,postures,orientation,color,andsound 56

PAGE 73

basedontheinter-agentdistanceorproximitydistancebetweentwoagents.Adetailedexample isprovidedonhowtheserecommendationscouldbeimplementedinanurbansearchandrescue applicationassociatedwithvictimmedicalassessment.Detailsareprovidedofavideoobservation usedtoverifytheprescriptiverecommendationsbasedonapathusedforthemedicalassessment ofavictim.Thereisabriefdiscussionofhowsoldierssociallyconnectwiththeappearance-constrainedrobotstheymustuseinmilitaryoperations.Designrecommendationsandthedevelopmentofnon-facialandnon-verbalmethodsofaffectiveexpressionforuseinroboticsystems,with examplesofmovementpatternsareprovidedinadditiontosomeconstraintstothisapproachthat wererelatedtotherobotsusedinthisresearch. 57

PAGE 74

Chapter4 ExperimentalMethodsandDesign Theprimaryobjectiveofthisresearchwastodeterminewhethertheuseofnon-facialand non-verbalaffectiveexpressionprovidedamechanismfornaturalisticsocialinteractionbetweena functional,appearance-constrainedrobotandthehumanwithwhichitwasinteracting.Theexperimentalobjectivesforthisresearchweretodetermineiftheuseofnon-facialandnon-verbalaffectiveexpressionswithtwodifferentappearance-constrainedsearchandrescuerobotsreducedparticipants'arousallevels,toallowthemtofeellessinhibitedandmoreincontrol,andcausethem toviewtheinteractionasmorepositive.Thechapterdiscussestheuseofpoweranalysistodeterminetheappropriatesamplesize,participanteligibilityrequirements,andsourcesofrecruitment. Detailsofthepsychophysiologyequipmentusedforthestudyareprovided.Thereisanin-depth discussionoftheself-assessmentmeasurementtoolsandthestructuredinterviewquestionsused inthisstudy.Designrequirementsforthesimulateddisastersiteareprovidedinadditiontotheexperimentaldesignforthestudy.Additionally,thereisadetaileddescriptionofthestudyprotocol, aswellaspersonnelrequirementsandresponsibilities. 4.1StudyOverview Atotalof128participantswererandomlyassignedtooneoftwogroupsthatinteractedwith twodifferentappearance-constrainedrobots.Therobotswerepre-programmedtooperateineitherastandardmodeoranemotivemodesothatparticipantswouldallhavesimilarexperiences. Thebehaviorsoftherobotsprogrammedtooperateinthestandardmodeweredevelopedbased onvideoobservationsofrobotoperatorsduringsearchandrescuetrainingexercises.Therobotsin thestandardmodeexhibitedfastanderraticmovements,wouldrisetofullheightcreatingaloomingeffect,andoftenturnedawayfromtheparticipantstosurveythesurroundingenvironment.The behaviorsfortherobotsprogrammedtooperateintheemotivemodeweredevelopedbasedonlit58

PAGE 75

eraturefrompsychologyandanimalbehavior,andfromconsultationswithsubjectmatterexperts, JeffConoveraDisneypuppeteerandDanielDeutschaDisneyconsultant.Therobotsprogrammed tooperateintheemotivemodedisplayedslowandcontrolledmovementstoexpressinterestand curiosityastheyapproachedparticipants,stayedlowertothegroundtoreducetheloomingeffect, remainedorientedtowardtheparticipantstoexhibitcaringandattentiveness,anddisplayedalight bluelightingeffectontheunder-carriageoftherobottoelicitacalmingresponseandilluminate therobotforvisibilitypurposes. Participantswereprovidedeightself-assessmentspriortoanyrobotinteractions,sixassessmentsaftereachrobotinteraction,andafollow-upquestionnaireattheendoftheirparticipation inthestudythataddressedtheirfeelingsregardingtheentirestudyexperience.Psychophysiologicalmeasureswereutilizedtomeasureparticipants'arousallevelstodeterminewhethertherewere differencesbetweentheirphysiologicalmeasuresduringarestingperiodobtainedaftertheparticipantwasplacedintheconned-spacesimulateddisastersitecomparedtotheirresponsesto eachrobotwithwhichtheyinteracted.Videodatawascollectedfromfourdifferentcameraperspectivesoverheadviewoverallsiteimage,faceviewfaceanduppertorso,participantview frominsidetheboxjustbehindandabovetheparticipant,andtherobotviewimagesfromthe robots.Followingtheexperiments,participantswereinterviewedtodeterminewhattheywere feelingabouttheirinteractionswiththerobotsandaboutthestudyoverall.Throughtheuseofall fourmethodsofevaluation,convergentvaliditymaybeobtainedtodeterminetheeffectiveness oftheuseofnon-facialandnon-verbalaffectiveexpressionfornaturalisticsocialinteractionina simulateddisasterapplication.Forthepurposesofthisdissertation,theitemsanalyzedwerethe resultsfromthevalenceandarousaldimensionsoftheSelf-AssessmentManikinSAM,andthe psychophysiologymeasuresofheartrate,respirationamplitude,respirationrate,andskinconductancelevel.Additionally,acorrelationanalysiswasconductedtodetermineiftherewerecorrelationsbetweentheSAMassessmentsandthepsychophysiologydata.Thedominancedimensionof theSAMassessmentwasremovedfromtheanalysisbecausenumerousparticipantsverballyexpresseddifcultyunderstandingthequestionsapproximately25%ofparticipants;thereforethis dimensionwasremovedfromconsiderationinthisstudy.Thiswillneedtobeevaluatedfurther atalaterdatewitheitheradifferentmeasurementtoolorbymodicationofthequestionsonthe SAMassessment. 59

PAGE 76

4.2Participants Atotalof128participantswererecruited,basedonapoweranalysisfor80%power,amediumeffectsizeof0.35,twogroupsbeingevaluated,andan =.05.Participantswererandomly assignedbetweentwogroupsthoseinteractingwithrobotspre-programmedtooperateineitherastandardmodeoranemotivemode.Participantsincludedbothmales.3%andfemales .7%,whichwasproportionatetotheoverallpopulationontheUSFTampacampus.Partcipantswerebetweentheagesof18and62,fromdifferenteducationalbackgrounds,andethnicities.ParticipantswererecruitedfrombothwithinandoutsideoftheUniversityofSouthFlorida. Themeansandstandarddeviationsforparticipants'demographicsaredisplayedinTable1.The frequenciesandpercentagesfortheparticipants'demographicsarepresentedasfollows:gender Table2,ageTable3,educationTable4,ethnicityTable5,computerexperienceTable6, robotexperienceTable7,videogamingexperienceTable8,robotownershipTable9,anddog ownershipTable10. Table1.Meansandstandarddeviationsforparticipants'demographics.Thisincludesgenderfemaleor1-male,age-62,educationalbackground-somehighschoolto8-postdoctorate, ethnicity-caucasianto7-mixed/other,computerexperience-noexperienceto5-expert,robot experience-noexperienceto5-expert,videogamingexperience-noexperienceto5-expert, whetheraparticipantownsarobot-noor1-yesand/oradog-noor1-yes. VariableName MeanM Std.Dev.S Gender .38 .49 Age 22.84 9.13 Education 2.55 1.07 Ethnicity 1.60 2.51 ComputerExperience 3.19 1.00 RobotExperience .52 .99 VideoGamingExperience 2.79 1.45 OwnaRobot .04 .20 OwnaDog .42 .50 Participantswereinterviewedorcompletedanexclusionquestionnaireonlinepriortoinvolvementinthestudytodetermineeligibility.Participantswererequiredtobeofgoodhealthwithno knownhistoryofhighbloodpressure,arrhythmias,cardiovasculardisease,asthma,mentalillness specicallypanicdisorders,anxietydisorders,and/orclaustrophobia,ortakinganymedication 60

PAGE 77

Table2.Genderdemographicsforparticipants.Theresponsesforthisdemographicare0-female and1-male. Gender Frequency Percentage 0-female 79 61.7% 1-male 49 38.3% Total 128 100.0% Table3.Agedemographicsforparticipants.Theresponsesforthisdemographicarefrom18to 62yearsofage. Age Frequency Percentage 18 39 30.5% 19 24 18.8% 20 13 10.2% 21 10 7.8% 22 9 7.0% 23 8 6.3% 24 6 4.7% 25 3 2.3% 26 1 .8% 27 2 1.6% 28 1 .8% 34 1 .8% 37 1 .8% 39 1 .8% 43 1 .8% 46 2 1.6% 53 1 .8% 54 2 1.6% 57 1 .8% 61 1 .8% 62 1 .8% Total 128 100.0% 61

PAGE 78

Table4.Educationdemographicsforparticipants.Theresponsesforthisdemographicare0-some highschool,1-highschoolgraduate,2-somecollege,3-AA AS,4-Bachelordegree,5-somegraduateschool,6-Mastersdegree,7-Ph.D./M.D./J.D.and8-postdoctorate. EducationalBackground Frequency Percentage 0-somehighschool 0 0.0% 1-highschoolgraduate 8 6.3% 2-somecollege 76 59.4% 3-AA AS 20 15.6% 4-Bachelordegree 16 12.5% 5-somegraduateschool 5 3.9% 6-Mastersdegree 3 2.3% 7-Ph.D./M.D./J.D. 0 0.0% 8-postdoctorate 0 0.0% Total 128 100.0% Table5.Ethnicitydemographicsforparticipants.Theresponsesforthisdemographicare0-Caucasian,1-BlackorAfricanAmerican,2-AsianorAsianAmerican,3-AmericanIndianorAlaska Native,4-AraborMiddleEastern,5-HispanicorLatino,6-NativeHawaiianorOtherPacicIslander,and7-Other/MixedorNon-Specied. Ethnicity Frequency Percentage 0-Caucasian 80 62.5% 1-BlackorAfricanAmerican 9 7.0% 2-AsianorAsianAmerican 9 7.0% 3-AmericanIndianorAlaskaNative 0 0.0% 4-AraborMiddleEastern 1 .8% 5-HispanicorLatino 13 10.2% 6-NativeHawaiianorOtherPacicIslander 3 2.3% 7-Mixed/OtherorNon-Specied 13 10.2% Total 128 100.0% 62

PAGE 79

Table6.Computerexperiencedemographicsforparticipants.Theresponsesforthisdemographic are0-noexperience,1-novice,2-someexperience,3-moderateexperience,4-strongexperience, and5-expert. ComputerExperience Frequency Percentage 0-noexperience 2 1.6% 1-novice 5 3.9% 2-someexperience 15 11.7% 3-moderateexperience 63 49.2% 4-strongexperience 31 24.2% 5-expert 12 9.4% Total 128 100.0% Table7.Robotexperiencedemographicsforparticipants.Theresponsesforthisdemographicare 0-noexperience,1-novice,2-someexperience,3-moderateexperience,4-strongexperience,and 5-expert. RobotExperience Frequency Percentage 0-noexperience 90 70.3% 1-novice 21 16.4% 2-someexperience 11 8.6% 3-moderateexperience 3 2.3% 4-strongexperience 1 .8% 5-expert 2 1.6% Total 128 100.0% Table8.Videogamingexperiencedemographicsforparticipants.Theresponsesforthisdemographicare0-noexperience,1-novice,2-someexperience,3-moderateexperience,4-strongexperience,and5-expert. VideoGamingExperience Frequency Percentage 0-noexperience 8 6.3% 1-novice 19 14.8% 2-someexperience 28 21.9% 3-moderateexperience 26 20.3% 4-strongexperience 31 24.2% 5-expert 16 12.5% Total 128 100.0% 63

PAGE 80

Table9.Robotownershipdemographicsforparticipants.Theresponsesforthisdemographicare 0-noand1-yes. OwnaRobot? Frequency Percentage 0-no 123 96.1% 1-yes 5 3.9% Total 128 100.0% Table10.Dogownershipdemographicsforparticipants.Theresponsesforthisdemographicare 0-noand1-yes. OwnaDog? Frequency Percentage 0-no 74 57.8% 1-yes 54 42.2% Total 128 100.0% thatcouldimpactthecardiovascularsystem.Additionally,participantsshouldhavenohistoryof drugoralcoholabuseinthepastyearandhavenotreceivedprofessionalpsychiatriccounseling forthepastyear.Theserequirementswerenecessaryforobtainingaccuratepsychophysiology measurementsfromeachparticipant. ParticipantswererecruitedviathePsychologyDepartmentSONASystem,yersoncampus, announcementstoclassesindifferentdepartmentsoncampus,andpersonalcontacts.Participation inthestudywasvoluntary.Adoorprizeticketwasprovidedtoeachparticipantandinsomecases classorextracreditwasofferede.g.extracreditiscommonlygiventoPsychologyDepartment studentsforparticipationinresearchstudiesasaddedincentiveforincreasedparticipation.Appointmentswerescheduledforeacheligibleparticipanttoavoidschedulingconictsandoptimize participants'andresearchers'time.Thedoorprizeticketwasprovidedtoeachparticipantregardlessofwhethertheyactuallycompletedthestudy.Thedrawingwasheldaftertheconclusionof alldatacollectionandthetwowinningparticipantswerecontactedviae-mailandphone,andthen giventheir$100cashprizes. 64

PAGE 81

4.3PhysiologicalRecordingApparatus AProComp5InnitiphysiologicalrecordingunitwasprovidedbyThoughtTechnology,Inc. http://www.thoughttechnology.com/foruseinthisresearchstudy.ThoughtTechnology,Inc.also providedtheBiographInnitisoftwarefordatacollectionandsignalprocessing,inadditiontothe followingsensors:oneelectrocardiographECGwiththreeleadelectrodes,onebloodvolume pulseBVP,oneskinconductancelevelSCL,andtworespirationstraingaugesforthoracicand abdominalmeasures.Additionally,Uni-GelECGdisposablesilver-silverchlorideelectrodeswere usedforECGrecordings. TheelectrocardiogramECGsignalwasusedtomeasureheartrateHR.TheECGsignal wastransducedusingonegroundandtwoactiveelectrodesinamodiedleadIIconguration SeeFigure8Stern,Ray,andQuigley,2001.ThesignalswereprocessedusingtheBiograph InnitisoftwaredesignedforusewiththeProComp5Innitiphysiologicalrecordingunit. Figure8.Psychophysiologysensorplacementdiagrams.Theimageontheleftillustratesthe modiedleadIIcongurationforECGmeasures.Thecenterdiagramdisplaysthesensorplacementformeasuringskinconductancelevels,andthediagramontherightshowsthesensorcongurationformeasuringthoracicandabdominalrespiration.TheseimageswereusedwithpermissionfromThoughtTechnology,Ltd. SkinconductancelevelSCLwasmeasured.TheSCLwastransducedwithtwoelectrodes placedontherstandlastvolarsurfacesofthedistalphalangesofthenon-dominanthandasdisplayedinFigure8Dawson,Schell,andFilion,2000.ThesensorsprovidedbyThoughtTechnologies,Inc.didnotrequireanyspecialelectrodesorskinpreparation,paste,orconductinggel. ThesignalswereprocessedusingtheBiographInnitisoftware. 65

PAGE 82

HeartrateHRandbloodvolumepressureBVPsignalswerealsomeasuredusingtheBVP sensor.TheHR/BVPwastransducedwithasingleelectrodeattachedtothemiddlengerofthe non-dominanthand.Thissensordidnotrequirespecialelectrodesorskinpreparation.Thesignals wereprocessedusingtheBiographInnitisoftware.Duetothemeasurementsfromthissensor beingredundantwiththeheartratedataobtainedfromtheECGsensoraswellastheanalysisof theBVPsensordatarevealednumerousproblemswithsignalnoiseandmovementartifacts,the datawasnotconsideredreliablefromthissensor;thereforethisdatawasnotusedforanalysispurposes. Respirationsignalsweremeasuredwiththesignalstransducedwithtwostraingaugesensors atthethoracicandabdominalregionsRefertoFigure8.Itwasimportanttoobtainbothmeasuresbecausesomeparticipantsmaybreathefromtheirabdomenandothersfromtheirupper chest.TherespirationmeasurementswillbeusedinconjunctionwithECGsignalstodetermine RSAatalaterdate.ThesignalswereprocessedusingtheBiographInnitisoftware. Participantdatawastransmittedthroughberopticcabletoalaptopcomputer.Followingthe datacollectionofeachparticipantthedatacollectedwasbackeduptothreedifferentexternalhard drivesthatweregeographicallydispersedforsafestorage.Auniqueidentiferwasassignedtoeach participantforcondentialityreasonstotrackalldatacollected. 4.4Self-AssessmentsandInterviewMeasures Participantswereprovidedaseriesofquestionnairesthatwereincludedinabookletthatthey receivedafteritwasdeterminedthattheywereeligibleforthestudy.Thebookletincludedpreinteraction,interaction,andpost-interactionassessmentsandquestionnaires,witheachsectionof thebookletseparatedbydividers.Participantswereprovidedtheassessmentsonesectionatatime tocomplete,andthentheassessmentswerescannedandreassembledintoabookletformforsafe storageinalockedcabinet. 4.4.1Pre-InteractionMeasures Afteritwasdeterminedthatparticipantswereeligibleforthestudy,theycompletedthepreinteractionassessmentsandquestionnaires.Thefollowingpre-interactionassessmentswereprovidedtoeachparticipant: 66

PAGE 83

DemographicsThisincludedquestionsregardinggender,age,occupation,educationlevel, majoreldofstudywhereapplicable,ethnicity/race,priorcomputerexperience,priorrobot experience,priorvideogamingexperience,robotownership,anddogownership.Thisquestionwasdevelopedbytheresearcherforthepurposesofthisstudy. HealthInformationThisquestionnairerequestedinformationregardingthecurrenttime,age, gender,allprescriptionandnon-prescriptionmedicationsandsupplementstakenwithinthe past48hours.Additionallyitrequiredparticipantstostatewhentheylasthadsomethingto eatandwhattheyate,whentheylastdrankcaffeineandhowmuchtheyconsumed,andwhen theylastsmokedcigarettes,howmanytheysmoked,andhowmanycigarettestheyregularly smokedinaday.Thisquestionnairewasdevelopedbytheresearcherforthepurposesofthis study. PhysicalActivityQuestionnairePaffenbarger,Wing,andHyde,1978Thisquestionnaire measuredtheparticipants'physicalactivitylevelsincludinghowmanyightsofstairsthey walkedupdaily,howmanycityblockstheywalkeddaily,anysportsorrecreationalactivities theyparticipatedinoverthepastweek,andatableofactivitiesandfrequencyofinvolvement. SleepQualityAssessmentBuysseetal.,1989Thiswasamulti-pageassessmentthatasked multiplequestionsonthequalityandamountofsleepparticipantsoverthepreviousmonth. Itincludedquestionsonsleeppatterns,bedtimehabits,lengthofsleep,factorsthatdisrupted theirsleep,andmanyotherrelatedfactors. PerceivedStressScalePSSCohen,Kamarck,andMermelstein,1983Thisassessment measuredtheparticipantslevelsofstressoverthepreviousmonth.Itincludedquestionsregardinghowtheyhandledstress,howoftentheywouldgetupset,howcondentweretheyin theirabilitytohandledifcultsituationandrelatedconcepts. SingleItemSocialSupportSIMSSBlakeandMcKay,1986Thisconsistedofonequestion:Howmanypeopledoyouhavenearthatyoucanreadilycountonforrealhelpintimes oftroubleordifculty,suchastowatchoverchildrenorpets,giveridestothehospitalor store,orhelpifyouaresick?Ithadascaleof0,1,2-5,6-9,10ormore. 67

PAGE 84

StateTraitAnxietyInventorySTAISpielberger,Gorsuch,andLushene,1970Onlythe stateportionofthisstandardizedpsychologicalassessmentwasused.Thisassessmentallowed participantstorespondtoquestionsregardinghowtheyfeltpriortointeractingwiththerobots.Thisprovidedabaselineoftheircurrentfeelingsandstate.Additionally,thisassessment wasgivenaftereachrobotinteractionwiththenalassessmentcomparedtothepre-interactionversiontomakesureparticipantsreturnedtotheirpreviousstateonceexperimentswere concluded.Thiswascheckedforethicalreasons. PositiveandNegativeAffectSchedulePANASWatson,Clark,andTellegen,1988This isastandardizedpsychologicalassessmentthatallowedparticipantstoratetheirfeelingsand emotionsonave-pointLikertscale.Thisinstrumentidentiedthestrengthoffeelingsand emotionspriortoandfollowinganyrobotinteractions. Thedatafromthesequestionnaireswasprocessedandsavedforarchivalpurposesandforfutureexploration.Analysesoftheseassessmentsarebeyondthescopeofthisthesis;howeverfuture dataanalyseswillprovideanopportunityforfurtherinsightsandapplications. 4.4.2InteractionMeasures Afterthepresentationofeachrobot,participantswererequestedtocompletesixdifferentselfassessments.Thisgaveeachparticipantabehavioraltasktoperformbetweeninteractionstoassist inreducinganypossibleeffectsofhabituation.Thefollowingself-assessmentswerecompleted afterthepresentationofeachrobot: Self-AssessmentManikinSAMBradleyandLang,1994Thisassessmentratesparticipants'experienceswiththerobotsonaninepointLikertscaleofvalence,arousal,anddominance.Thisisatwopartassessmentwhichposestheevaluationterminologyintwodifferent waysforincreasedvalidity.Thequestionsassociatedwiththedominancedimensionwere misunderstoodandappearedtoconfuseapproximately25%oftheparticipantsbasedonverbal accounts;thereforethisdatawasremovedfromtheanalyses.Thedominancedimensionwill needtobevalidatedatalaterdateusingadifferentassessmenttoolorthroughmodifyingthe questionsontheSAMassessmentforthisdimension.Twoversionsofthisassessmentwere 68

PAGE 85

developedtodetermineifbothversionselicitedsimilarresponsesfromparticipants.AcorrelationanalyseswasconductedtodeterminethevalidityoftheSAMassessmentforuseinHRI studies.Furtherinformationonthisvalidationprocessandtheresultsofthecorrelationare availableinChapter5andChapter6commencingonpages77and121respectively. RobotAssessmentThisassessmentwasdevelopedtodetermineparticipants'feelingsregardingfourdifferentaspectsoftherobottheyexperienced.Theassessmentevaluatedthe appearance,sounds,movements,andspeedoftherobot.Eachcriteriawasevaluatedbasedon aninepointLikertscaleforvalence,arousal,anddominance.Numerousparticipantsverbally expressedconfusionandquestionedthewordingassociatedwiththedominancedimension approximately25%ofparticipants;thereforethisdimensionwasnotutilized.Thisquestionnairewasdevelopedforthepurposesofthisstudyasadistractorsothatparticipantswouldnot inferthetruepurposeofthestudy. AffectGridScaleRussell,Weiss,andMendelsohn,1989,Mutluetal.,2006Modied fromtheoriginalversionforuseintheHRIdomain.Thisassessmentwassetupinagrid andrequestedparticipantstoratetwoquestions:yourperceptionoftherobotinteractionlow arousaltohigharousalANDlowpleasuretohighpleasure,andyourperceptionoftherobot's attitudehostiletofriendlyANDsubmissivetodominant.Thisassessmentwassetupasgrid orgraphdesignandapproximately25%oftheparticipantsexpressedconfusiononhowtoplot thepointsandcompletethisgraph;thereforeitwaswithdrawnfromanalyses. ObservationsoftheRobotMutluetal.,2006Thisquestionnairewasmodiedfromthe originalversiontocorrespondwiththishuman-robotstudy.Severalquestionswereremoved duetothefactthattheoriginalassessmentwasgearedtowardahumanoidandwerenotapplicabletothishuman-robotstudy. StateTraitAnxietyInventorySTAISpielberger,Gorsuch,andLushene,1970Onlythe stateportionofthisstandardizedpsychologicalassessmentwasused.Thisassessmentallowed participantstorespondtoquestionsregardinghowtheyfeltduringtheirinteractionswitheach robot. 69

PAGE 86

PositiveandNegativeAffectSchedulePANASWatson,Clark,andTellegen,1988This standardizedpsychologicalassessmentallowedparticipantstorateonavepointLikertscale thefeelingsandemotionstheyexperiencedduringeachrobotinteraction. 4.4.3Post-InteractionMeasures Onceparticipantscompletedtheassessmentsimmediatelyfollowingtheirlastrobotinteractiontheyparticipatedinanaudiorecordedstructuredinterview.Participantswererequestedto answerthefollowingquestions: Whatwereyoufeelingduringtheexperiments? Werethereanyfeelingsthataroseduringtheexperimentsthatimpactedyouinapositiveway? Weretheyanyfeelingsthataroseduringtheexperimentsthatimpactedyouinanegativeway? Wasthereanythingthatoccurredduringtheexperimentthatwasproblematicforyouinany way? Doyouhaveanysuggestionsforimprovingtheexperimentalprocess? Doyouhaveanyothercommentsorsuggestionsaboutthisexperience? Followingtheaudiorecordedstructuredinterview,participantswereprovidedwithonelast self-assessmentthatexaminedtheirfeelingaboutthestudyoverall.Thisquestionnairewasderived fromahuman-robotstudyconductedatCarnegieMellonUniversityMutlu,Hodgins,andForlizzi,2006.Aftercompletionofthislastassessmentparticipantsweredebriefedonthestudyand wererequestednottodiscussthedetailsofthestudywithothers. 4.5SiteRequirements Thesitewasdesignedtobeinanindoorlocationthatallowedforthecontroloftemperature andsoundforqualitypsychophysiologicaldatacollection.Ahigh-delity,simulateddisastersettingwasneededwithaconnedspaceforparticipantstohaveasenseofbeinginanactualdisastersetting,andtoinduceamildtomediumstressconditionduringthecourseoftheexperiments. Therewasaneedtoprovideadequatespacetotestthesocialimpactofinteractionsineachofthe 70

PAGE 87

threeproximityzonesofinteresttoverifythedevelopedprescriptiverecommendations.Thesite wasdesignedwithawoodenchutethattherobotswouldcomethroughtoenterthesimulateddisastersitetopreventparticipantsfromviewingtherobotspriortotheactualstudyinteractions.The siterequiredenoughspacetosetupavideocamcordertoobtainanuppertorsoandfacialperspectiveofparticipantsduringtheirinteractionswiththerobots.Therewerechairsandasixfoot tableforthelaptop,psychophysiologicalrecordingequipmentandaseparateareaforcompleting questionnairesandoperatingtherobotandvideorecordingequipment.TheexperimentswereconductedinUniversityofSouthFlorida,EngineeringBuildingII,Room223.Allrobotinteractions wereconductedinthedark;thereforeallvideoequipmenthadinfraredrecordingcapability.Additionallyinfraredlightingwasusedtoprovideilluminationforthevideorecording.Videocameraswerealsoinstalledinsidetheconned-spaceboxtorecordtheparticipant'sperspective.Each robotwasequippedwithcamerastoobtaintherobot'sperspective,andacamerawasmounted atceilingheighttoobtainanoverviewperspectiveoftheentiresite.Theconceptualdrawingof theexperimentalsiteplanisshowninFigure9.AphotoofthenalsimulateddisastersiteisdisplayedinFigure10. 4.6ExperimentalDesign Eachparticipantinteractedwithtwodifferentrobotswithin-subjectsfactorusedinactual urbansearchandrescueresponsesSeeFigure11:aniRobotPackBotScoutandanInuktunExtreme-VGTV.Participantswererandomlyassignedtooneoftwoconditionsbetween-subjects factor:robotsprogrammedtooperateinthestandardmode,orrobotsprogrammedtooperatein theemotivemode.Thiswasa2x2mixedmodeldesignrobotsx2operatingconditions.The ordertherobotswerepresentedwascounterbalanced,abetween-subjectsfactor.Additionally,age andgenderwerebalanced. 4.7StudyProtocol Uponarrivalattheexperimentsite,participantswereinterviewedtoconrmstudyeligibility, giveninstructions,briefedontheactivitiesofthestudy,andwererequestedtosigntheprovided informedconsentsforstudyparticipationandaudio/video-recordingbeforebeginningtheproto71

PAGE 88

Figure9.ConceptualdrawingoftheexperimentalsiteplanusingthespaceallocatedinENB223. 72

PAGE 89

Figure10.Actualhigh-delity,simulateddisastersitewithconned-spaceboxandInuktunrobot. Figure11.PackBotinarubblepileandInuktuninthesimulateddisastersite. 73

PAGE 90

col.Thenparticipantsreceivedandwereinstructedtocompletethepre-interactionassessments andquestionnairesasoutlinedinsubsection4.4.1. Next,theambulatoryphysiologicaldatarecorderandsensorswereconnectedtotheparticipantandtestedtoverifythatallconnectionsweresecureandworkingproperly.Participantswere thenplacedonapaddedcotinthepronepositionontheirrightsideandlistenedtocalmingmusic toobtainbaselinephysiologicalmeasurespriortobeingplacedintheconned-spacesimulated disastersitefortherobotinteractions.Baselinemeasurementswererecordedinthelastthreeminutesofatenminuterestingperiod. Afterbaselinemeasurementsweretaken,participantswereplacedinthepronepositionon theirrightsideintheconned-spacewithinthesimulateddisastersitedesignedforthisstudy. Thesitewasbuiltinanindoorlocationsothattheenvironmentcouldbecontrolledfornoiseand temperaturetoobtainanaccuraterecordingofthephysiologicalmeasures.Participantswererequestedtorestfortenminutespriortostartingtherobotscenarios.Duringthelastthreeminutes ofthisrestperiodbaselinereadingsfromallsensorswererecorded. Participantswererandomlyassignedtooneoftwogroupsstandardoperatingrobotsvs.emotiveoperatedrobotsandwerepresentedtherobotsInuktunExtreme-VGTVorPackBotina counterbalancedorder.Eachparticipantexperiencedtworobotinteractionsthatlastedapproximatelyvetosevenminuteseach.Therobotpathandmovementswerepre-programmedforconsistencyintherobotinteractions.Thepaththerobotsfollowedwasbasedonobservationsofa robot-assistedvictimassessmenttrainingexercise,usinginformationobtainedfromactualemergencymedicalpersonnelregardinghowtheywouldhavetherobotsoperatedtoperformthistype ofmedicalassessmentRiddle,Murphy,andBurke,2005,Murphy,Riddle,andRasmussen, 2004.Psychophysiologicalmeasureswerecollectedcontinuouslyduringtherobotinteractions. Followingeachinteraction,participantswereinstructedtocompleteinteractionassessments,as outlinedinsubsection4.4.2toevaluatetheirexperiencesandfeelingsduringtherobotinteraction.Byhavingparticipantscompleteassessmentsfollowingtheinteractionsitwasexpectedtoreducetheeffectsofhabituation;howeverbasedontheresults,thismaynothavehadthedesiredeffect.Participantsweregivenfteenminutestocompletetheseassessmentswiththegoalthatthey wouldreturntobaselinelevelsbeforeexperiencingthenextrobot.Participantswerepermittedto situptocompletetheassessmentforms;howevertheyreturnedtothepronepositionontheirright 74

PAGE 91

sideforthenextsetofbaselinephysiologicalreadingsandwhiletheyinteractedwiththerobotsin ordertohaveconsistentphysiologicalreadings. Alltaskswerevideotapedfromoverhead,faceviewuppertorsoandface,participant,and robotperspectivesusinginfraredvideocameras.Allrobotinteractionswereperformedinthedark tosimulateactualdisasterconditions.Therecordingssynchronizedbytheoverheadlightsbeing turnedoffandonfortherobotinteractionsfordatacollectionconsistency. Followingthetworobotinteractionsandcompletionoftheinteractionassessments,participantswereremovedfromthesimulateddisastersiteandallphysiologicalequipmentwasremoved. Datawascollectedviaber-opticcabletransmissiontotheonsitelaptopandbackeduptothree externalharddrivesthatweregeographicallydispersedeachday.Thedatawassecuredinalocked cabinetinasecuredbuildingforsecurityandcondentialitypurposes.Condentialdatawasonly sharedamongtheresearchgroup.Informedconsentformswerestoredinalockedlocationseparatefromtheotherdatacollected. Participantswereinterviewedusingthequestionsinsubsection4.4.3.Theinterviewwasaudiorecordedforaccuracyindatacollection.Participantswerethendebriefedonthepurposeofthe study.Followingtheinterview,participantswereinstructedtocompletethepost-interactionfollow-upassessmentregardingtheirfeelingsaboutthestudyasawhole.Videotapeswereremoved, labeled,andstoredforbehavioralcodingatalaterdate.Allvideotapeswerealsobebackedupto threeexternalharddrivesthatweregeographicallydispersed.Theoriginaltapeswerekeptina lockedofceinasecuredbuilding. 4.8Personnel Thisstudyrequiredaminimumofoneassistantforrecordingpsychophysiologymeasurementsandoperatingthefaceviewvideocamera.Thisassistantalsoplacedtherobotsinthestartingpositionforeachinteractionandmadesuretherobotswerepoweredonandthebatterieswere charged.Theywereresponsibleforallvideooperationsandmaintenance.Thesedutiesincluded: verifyingthatallcameraswereloadedwithvideotapes,batterieswerecharged,A/Vcablingwas attached,allrecordingsweretimesynched,allvideotapeswereproperlylabeledandbackedup. 75

PAGE 92

Forsafetyoftheparticipants,assistants,investigator,androbotstherewasalwaysaminimum oftwotrainedpersonnelonsiteduringthestudy.Theinvestigatorandallexperimentpersonnel hadcellphonesforemergencypurposes. 4.9Summary Thischapterbeginswithanoverviewoftheoverallstudydetails,followedbyadiscussion ofhowtodeterminetheappropriatesamplesizeandmethodsofrecruitingparticipantsforhumanstudies.Thedemographicsoftheparticipantsinvolvedinthisstudyarepresentedforcompletenessandalsoforreproducibilityofthestudy.Detailsareprovidedonthepsychophysiology equipmentusedforthisstudy,whatmeasurementswererecorded,andproperelectrodeplacement. Thereisdetailedcoverageontheself-assessmenttoolsutilizedindifferentaspectsofthisstudy pre-interaction,interactionandpost-interactiontasks.Otherimportantaspectspresentedinthis chapterarethesiterequirements,adescriptionoftheexperimentaldesign,andthestudyprotocol utilizedaspartofthisresearch.Additionally,thereiscoverageonthepersonnelrequirementsand responsibilitiesneededforthisstudy. 76

PAGE 93

Chapter5 DataAnalysesandResults Thischapterpresentsthedetailsofthedataanalysesconductedandresultsfromdatacollected aspartofthiscomplex,large-scale,comprehensivestudyevaluatingtheeffectivenessofnon-facial andnon-verbalaffectiveexpressionforuseasamechanismfornaturalistichuman-robotsocialinteraction.Onlyasmallportionofthedatacollectedaspartofthisstudywillbeanalyzedaspartof thisdissertation.Thefocusforthisresearchisassociatedwithparticipants'valenceandarousalresponsestothetwourbansearchandrescuerobotsusedinthisstudy,theInuktunExtreme-VGTV InuktunandtheiRobotPackbotScoutPackbot.Thechapterbeginswithpreliminaryinformationregardingthetypesofanalysesconducted,theequationusedtocalculateeffectsizewiththe associatedscale,andthenullhypothesesevaluated.Theanalysesperformedaspartofthisstudy fallintothreecategories:Self-AssessmentManikinSAMdataanalyses,psychophysiological dataanalyses,andcorrelationsbetweentheSAMandpsychophysiologicaldata.Withineachtype ofanalyses,descriptiveandinferentialstatisticsarepresented. 5.1DataAnalysesandResultsBackgroundInformation DataanalysesforthestudyconductedincludedanalysesofthedatarelatedtotheSelf-AssessmentManikinSAMBradleyandLang,1994onthedimensionsofvalenceandarousal, psychophysiologicaldatawasanalyzedforarousal,andacorrelationanalysiswasperformedon theSAMassessmentandpsychophysiologicaldata.Additionally,twoversionsoftheSAMassessmentweredevelopedforthisstudyandaspartoftheanalyses,acorrelationwasperformedto determineiftheresultsfromthetwoversionsoftheSAMassessmentscouldbecombinedforease ofanalysisandtovalidatetheSAMassessmentasatoolfortheHRIcommunity.Alltestswere conductedusingan =.05.EffectsizeswerecalculatedforthestatisticallysignicantF-testsusingCohen's f effectCohen,1988. 77

PAGE 94

TheeffectsizesforallF-testswerecalculatedusingthefollowingequationforCohen's f effect: f = q df F N ThescaleusedforinterpretingCohen's f effectisasfollows: .00-.09nonotableeffect .10-.24smalleffect .25-.39mediumeffect .40+largeeffect Thenullhypothesesassociatedwiththisresearchfocusedonthreedimensions,whichwere valencepositiveversusnegative,arousalexcitedversuscalm,anddominancein-controlversuscontrolled.ValencewasevaluatedusingtheSAMassessments,arousalwasmeasuredthrough boththeSAMassessmentsandpsychophysiologicaldata;howeverthedominancedimensionwas notsuccessfullyevaluatedinthisstudyduetonumerousparticipantsapproximately25%ofthe participantsverballyreportingconfusionwiththewordingofthedominancequestionsonthe SAMassessments.ThiswillneedtobeinvestigatedfurtherinafuturestudywithadifferentassessmenttoolorrevisionstotheSAMassessments.Therefore,thenullhypothesesthatwereevaluatedforthisthesiswerethearousalandvalencedimensionsasfollows: Within-SubjectsFactorRobotInuktunExtreme-VGTVInuktunversusiRobotPackbot ScoutPackbot ParticipantswillexperiencenodifferenceinstressandarousallevelswiththeInuktun comparedtothePackBot. ParticipantswillviewinteractionswiththeInuktunExtreme-VGTVasnodifferentthan interactionswiththeiRobotPackBotScout. Between-SubjectsFactorOperatingModeStandardversusEmotive Participantswillexperiencenodifferenceinstressorarousallevelswiththerobotsoperatedintheemotivemodecomparedtothoseoperatedinthestandardmode. 78

PAGE 95

Participantswillviewinteractionswiththerobotsoperatedintheemotivemodeasthe samewhencomparedtothestandardmoderobots. Between-SubjectsFactorRobotOrderInuktunFirstversusPackbotFirst Participantswillexperiencenodifferenceinstressandarousallevelswhethertheysee theInuktunFirstorthePackbotFirst. ParticipantswillviewinteractionswiththeInuktunasnodifferentthaninteractionswith thePackbotregardlessofwhichrobotispresentedrst. 5.2Self-AssessmentManikinAnalysesandResults TheSelf-AssessmentManikinSAMdatawasanalyzedontwoofthethreedimensions,valencepositiveversusnegativeandarousalexcitedversuscalmBradleyandLang,1994.The orginalSAMdocumentwasmodiedtocorrespondwiththeHuman-RobotInteractionHRIdomainandtwoversionsoftheassessmentweredevelopedinordertovalidatethisassessmenttool foruseintheHRIcommunity.TheSAMassessmenthasalreadybeenvalidatedforuseinthepsychologycommunityBradleyandLang,1994.Thissectiondiscussesthedataanalysesperformed usingtheSAMassessmentdataincludingthecorrelationperformedbetweenthetwoversionsof thisdevelopedassessmenttool,themultivariatetestsforassumptionsbyoperatingmodeandorder,thedescriptiveandinferentialstatisticalresultsfromtheSAMdatacollectedonthevalence andarousaldimensions. ThedataanalysesfortheSAMdataincludedbothdescriptiveandinferentialstatisticalanalyses.TheanalyseswereperformedusingSAS9.2andSPSS17.0.Forthedescriptivestatistics, analysisofcentraltendencieswasconductedwhichincludedthemean,standarddeviation,skewness,kurtosisanddistributioncharacteristicsforthevalenceandarousalvariablesbyoperating modeandbyorder.Assumptionsformultivariatenormalityandhomogeneityofthecovariance matrixBox'sMTestweretestedandanalyzedbyoperatingmodeandbyorder. Theinferentialstatisticalanalysesincludedrepeatedmeasuresmultivariateanalyses.There weretwolevelsforthewithin-subjectsfactorofrobotsInuktunandPackbot.Thereweretwobetween-subjectsfactorseachhavingtwolevels:operatingmodestandardversusemotive,androbotorderInuktunFirstversusPackbotFirst.TherewerefourdependentvariablesArousal-Inuk79

PAGE 96

tun,Arousal-Packbot,Valence-Inuktun,andValence-Packbot,whichwereevaluatedforstatistical signicanceusinganF-testwithTypeIIISumofSquaresSS.Repeatedmeasuresmultivariate analysisofvarianceMANOVAtestswereconductedforvalenceandarousalbythefollowing factors:operatingmode,robot,robotorder,operatingmodebyrobotorder,robotbyrobotorder, robotbyoperatingmode,androbotbyoperatingmodebyrobotorder.StatisticallysignicantresultsarepresentedusingF-testvaluesandeffectsizegivenbyCohen's f Cohen,1988.Further analyseswasconductedtoprovideamoredetailedexplanationofanyinteractionsthatwerestatisticallysignicant. 5.2.1CorrelationoftheTwoSelf-AssessmentManikinAssessments TheSAMassessmentisavalidatedassessmenttoolthathasbeenutilizedbythepsychology community;howeveritrequiredsomemodicationofwordingtobeusefulintheHRIdomain. InordertovalidatethistoolforuseintheHRIcommunity,thewordingwasmodiedandtwo formsoftheassessmentweredevelopedthatwereintendedtoelicitsimilartypesofresponses fromparticipantsbutphrasedinaslightlydifferentmanner.Thesemodicationswereconsistent withtheoriginalformatoftheassessmenttool.EachparticipantreceivedbothSAMassessments foreachoftherobotinteractions.Thedatawascompiledandstoredinaspreadsheetforanalyses. TherstpartofthedataanalysesprocessrequiredtestingtodetermineiftherewerestatisticallysignicantpositivecorrelationsbetweentheresponsestothetwoversionsoftheSAMassessmentssothattheresponsescouldbecombinedandaveragedforeaseofuseinthedataanalysisprocess.ThecorrelationmatricesfortheSAMvalencequestionsandtheSAMarousalquestionsaredisplayedinTable11andTable12respectively.Althoughallthecorrelationsarestatisticallysignicant,thecellsthatarehighlightedinyellowarethecorrelationsofinterestforcombiningthequestionsforeaseofanalysesandvalidation. CorrelationswereperformedforthetwoversionsoftheSelf-AssessmentManikinforthe valenceresponsesSAMval-11andSAMval-14totherstrobotandtherewasastrongstatisticallysignicantpositivecorrelation[r=.54, p < .0001]betweenSAMval-11andSAMval-14 withSAMval-14accountingfor29.16%ofthevariabilityinSAMval-11.Inthecaseofthetwo versionsofthevalenceresponsestothesecondrobotinteractionSAMval-21andSAMval-24, 80

PAGE 97

therewasastrongstatisticallysignicantpositivecorrelation[r=.74, p < .0001]withSAMval-24 questionaccountingfor54.76%ofthevariabilityintheSAMval-21question. Table11.PearsoncorrelationmatrixforthevalencequestionsontheSelf-AssessmentManikin SAMassessments. SAMval-11 SAMval-14 SAMval-21 SAMval-24 SAMval-11 r=1.00 SAMval-14 r=.54 r=1.00 p < .0001 SAMval-21 r=.21 r=.31 r=1.00 p =.020 p < .0001 SAMval-24 r=.17 r=.28 r=.74 r=1.00 p =.050 p =.001 p < .0001 Correlationissignicantatthe0.05level-tailed. APearsoncorrelationwasperformedforthetwoversionsoftheSAMarousalresponses SAMarous-12andSAMarous-15totherstrobotinteractionandtherewasastatisticallysignicantpositivecorrelation[r=.37, p < .0001]betweentheresponsesusingtheSAMarous-12questionandtheSAMarous-15questionwiththeSAMarous-15accountingfor13.69%ofthevariabilityintheSAMarous-12question.Therewasastatisticallysignicantpositivecorrelation[r=.36, p < .0001]betweenSAMarous-22andSAMarous-25responsestothesecondrobotexperienced bytheparticipantswithSAMarous-25accountingfor12.96%ofthevariabilityinSAMarous-22. Table12.PearsoncorrelationmatrixforthearousalquestionsontheSelf-AssessmentManikin SAMassessments. SAMarous-12 SAMarous-15 SAMarous-22 SAMarous-25 SAMarous-12 r=1.00 SAMarous-15 r=.37 r=1.00 p < .0001 SAMarous-22 r=.47 r=.34 r=1.00 p < .0001 p < .0001 SAMarous-25 r=.32 r=.30 r=.36 r=1.00 p < .0001 p =.001 p < .0001 Correlationissignicantatthe0.05level-tailed. 81

PAGE 98

Duetothesepositivecorrelations,itwasdeterminedthattheresponsestothetwodifferent versionsoftheSAMassessmentscouldbecombinedforfurtherdataanalyses.Thevaluesfor theresponsestothetwoversionswereaveragedandthosenumberswereusedfortherestofthe analysesdiscussedinSection5.2.ValidationofthevalenceandarousaldimensionsoftheSAM assessmentforuseinHRIwasobtainedbydevelopingtwoversionsoftheSAMassessmentwith statisticallysignicantpositivecorrelationsonthesetwodimensionsusingdatafrom127participants.Thestudywasconductedwith128participants;howeveroneparticipantdidnotcomplete oneportionoftheSAMassessmentandthereforetheSAMassessmentdataforthisparticipant wasremovedfromthedataset.TheSAMassessmenthasalreadybeenvalidatedforuseinpsychologyBradleyandLang,1994andthroughthedevelopmentofthetwoversionsoftheSAM assessmentbasedonHRIterminology,thepositivecorrelationsindicatethattherevisedassessmentkeepswiththeoriginalintentoftheassessmenttoolinthisnewdomain;thereforethevalenceandarousaldimensionsoftheSAMassessmenthasbeenmodiedandvalidatedforuseby theHRIcommunityforfuturestudiesevaluatingvalenceandarousal. 5.2.2MultivariateTestsforAssumptionsbyOperatingModeandRobotOrder Theindependenceassumptionwasmetbythemethodusedtocollectthedatainwhichparticipantscompletedtheirassessmentsindependentlywithoutconsultationwithothers.Testswere conductedtodeterminemultivariatenormality.Themultivariateskewnessoftheresidualsfrom theRepeatedMeasuresMANOVAdidnotappeartobestatisticallysignicantforskewness[ b 1 ;p = .85, 2 ,N=127=18.48, p =.560].Themultivariatekurtosisdidnotshowstatisticallysignificantkurtoticbehaviorbyfallinginarangeof-1.96to1.96[ b 2 ;p =22.94, z upper =-.86, z lower =1.63].Thetestsindicatedthatthemultivariatenormalityassumptionhasbeensatised.Byscreeningformultivariateoutliersorextremevalues,themaximumMahalanobisDistancewas13.10 [F,122=3.57, p =.010]usingan =.05thereappearedtobeanindicationofmultivariateoutliersorextremevalues.Furtherexaminationofthedatadidnotindicateanyvaluesthatwerenot plausibleorwerelargelydifferentfromothervaluespresented;thereforetheanalysesproceeded withtheopinionthatthisassumptionwasrobusttoanyconcerns.BasedontheresultfromBox's MTest,therewasastatisticallysignicantdifferenceinthecovariancematrices[ 2 ,N=127 82

PAGE 99

=55.79, p < .0001].Basedonthesamplesizethetestwasrobusttotheviolationofhomogeneity ofthecovariancematrices. 5.2.3TheSAMValenceAnalysesandResults Therewasathree-wayinteractionthatwasstatisticallysignicantinthevalencedimension oftheSelf-AssessmentManikinanditwasdiscoveredinboththewithin-subjectsandbetweensubjectsfactors.Therewerestatisticallysignicantresultsforthemaineffectsofvalencebyrobot andvalencebyrobotorder;howeverthesewillnotbediscussedfurtherbecausethethree-way interactiontakesprecedenceintheseresults.Therewasastatisticallysignicantresultforathreewayinteractionforvalenceresponsebyrobot,operatingmode,androbotorder.DescriptivestatisticsforthevariablesassociatedwiththeSAMvalencedimensionarepresented,followedbya discussionofthestatisticallysignicantresultsfromtheinferentialstatisticsforthethree-wayinteractionforvalencebyrobot,operatingmode,androbotorder.Theresultsfromthethree-way interactionwillbepresentedintwodifferentformatstobetterdecomposeandexplaintheresults. Aspartoftheanalyses,thescaleforvalencewasrecodedinreverseorderforeaseofinterpretation.Therevisedscaleis0-negativeto8-positive. 5.2.3.1DescriptiveStatisticalAnalysisforSAMValenceRecodedbyOperatingModeand RobotOrder Therearetwomaindependentvariablesfortherecodedvalencedimension-negativeto8positive:ValenceR-Inuktun,andValenceR-Packbot.Thesetwovariableswereanalyzedbasedon bothoperatingmodeandrobotorderusingthefollowingcodingscheme:00StandardMode-InuktunFirst,01StandardMode-PackbotFirst,10EmotiveMode-InuktunFirst,11EmotiveModePackbotFirst.ThedescriptivestatisticsforeachoperatingmodeandrobotorderscenarioforValenceR-InuktunandValenceR-PackbotareprovidedinTable13.TheboxplotsforValenceR-InuktunandValenceR-PackbotbyoperatingmodeandrobotorderaredisplayedinFigures12and13 respectively. ThevariableValenceR-Inuktun-00wasrelativelynormallydistributedwithaslightnegative skewness Sk =-.17andmoderatelyplatykurtic Ku =-.62hadameanvalueof5.17.ValenceRInuktun-01hadameanvalueof6.22andappearedtobenormallydistributedwithamoderateneg83

PAGE 100

Figure12.BoxplotforSAM Valence Recoded-Inuktunbyoperatingmodeandrobotorder. Figure13.BoxplotforSAM Valence Recoded-Packbotbyoperatingmodeandrobotorder. 84

PAGE 101

Table13.DescriptivestatisticsforthetwodependentvariablesforSAMvalencerecodedbyoperatingmodeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01Standard Mode-PackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-PackbotFirst. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu ValenceR-Inuktun-00 5.17 1.70 -.17 -.62 ValenceR-Inuktun-01 6.22 1.33 -.54 -.41 ValenceR-Inuktun-10 5.72 1.41 -.15 -.20 ValenceR-Inuktun-11 6.76 1.52 -1.41 1.13 ValenceR-Packbot-00 4.41 1.65 -.26 -.18 ValenceR-Packbot-01 5.28 1.53 -.18 -.65 ValenceR-Packbot-10 3.72 1.93 .17 -.68 ValenceR-Packbot-11 5.95 1.49 -.33 -.72 ativeskewness Sk =-.54andslightlyplatykurtic Ku =-.41.ValenceR-Inuktun-10hadamean valueof5.72,seemednormallydistributedwithaslightnegativeskewness Sk =-.15,andslightly platykurtic Ku =-.20.ValenceR-Inuktun-11hadameanvalueof6.76andwasrelativelynormally distributedwithastrongnegativeskewness Sk =-1.41andwassignicantlyleptokurtic Ku = 1.13.Overall,participantsreportedfeelingthemostpositivewheninteractingwiththeInuktun robotintheEmotiveModewhenviewedafterinteractingwiththePackbotintheEmotiveMode. ItappearedfromthedatathatparticipantsweremorepositiveingeneraltotheInuktunrobotwhen itwasviewedafterthePackbotineitheroperatingmode. ValenceR-Packbot-00wasrelativelynormallydistributed;howevertherewasaveryslight negativeskewness Sk =-.26andslightlyplatykurtic Ku =-.18withameanvalueof4.41.For thevariableValenceR-Packbot-01themeanvaluewas5.28anditappearedtoberelativelynormallydistributedwithaslightnegativeskewness Sk =-.18andmoderatelyplatykurtic Ku =-.65. ValenceR-Packbot-10hadameanvalueof3.72andseemedrelativelynormallydistributedwith aslightpositiveskewness Sk =.17andmoderatelyplatykurtic Ku =-.68.ValenceR-Packbot11hadameanvalueof5.95,appearedtobenormallydistributedwithaslightnegativeskewness Sk =-.33andwasmoderatelyplatykurtic Ku =-.72.Whencomparingthemeanvaluesforthe Packbotconsideringoperatingmodeandrobotorder,participantsreportedviewingtheinteractions withthePackbotFirstintheemotivemodeasmorepositive.Ingeneral,participantsreportedthat theyfeltthemostnegativeaboutinteractionswiththePackbotwhenviewedaftertheInuktunin theemotivemode. 85

PAGE 102

5.2.3.2InferentialStatisticalAnalysesforSAMValenceRecodedbyOperatingModeand RobotOrder TheanalysesforevaluatingthedependentvariablevalencerecodedfromtheSelf-Assessment Manikinassessmentsincludedawithin-subjectsfactorofrobotInuktunandPackbot,withtwo between-subjectsfactorsoperatingmodestandardoremotiveandrobotorderInuktunFirstor PackbotFirst.Thestudyevaluatedwhethertherewereanystatisticallysignicantdifferences inthevalenceresponsesbasedonrobot,operatingmode,and/orrobotorder.Ifanystatistically signicantdifferenceswereobtainedthentheassociatednullhypthesiswasrejected. BasedontheresultsfromtheMANOVAforvalencebyoperatingmodeandrobotorder,there wasastatisticallysignicantthree-wayinteractionforSAMvalenceresponsesbyrobot,operatingmode,androbotorder,[F,123=4.50, p =.036, =.05].Theeffectsizeforthisthree-way interactionwas f =.19,asmalleffect.TheresultsaredisplayedinFigures14and15. Furtheranalysiswasconductedtodecomposethisthree-wayinteraction,bysplittingtheresultsusingtherobotorder.ForparticipantsintheInuktunFirstorder,therewasasignicantrobot byoperatingmodeinteraction,[F,62=5.64, p =.021, =.05], f =.30amediumeffect.The interactionwasdecomposedusingpairwisecomparisonsRefertoFigure14.TherewasastatisticallysignicantdifferencebetweentheresponsestotheInuktunandthePackbotforbothoperatingmodes.TheresultsfromthepairwisecomparisonswiththeInuktunFirstforbothoperating modesindicatedthatthedifferencebetweentheInuktunandthePackbotwasgreaterintheemotiveoperatingmodeM Inuktun -M Packbot =2.00,[F,123=38.95, p < .0001, =.05], f =.55, alargeeffect,thaninthestandardmodeM Inuktun -M Packbot =0.77,[F,123=5.71, p =.018, =.05], f =.21,asmalleffect.Theseresultssignifythatparticipantsrespondedmorepositivelyto theInuktuncomparedtothePackbotinbothoperatingmodeswithagreaterdifferenceobserved fortherobotsintheemotivemode.However,thisinteractiondidnotresultinstatisticallysignificantdifferencesbetweenrobotsineachoperatingmode[Fvalues < 2.74, p values > .100].The differenceinmeansfortheInuktunemotive-standardwas.55andforthePackbotemotivestandardwas-.69. PairwisecomparisonsforthePackbotFirstindicatedastatisticallysignicantdifferencein responsestotheInuktuncomparedwiththePackbotinbothoperatingmodes.ParticipantsrespondedmorepositivelytotheInuktunthanthePackbotinthestandardmode[F,123=8.56, 86

PAGE 103

Figure14.Statisticallysignicantthree-wayinteractionforSAM Valence Recodedbyrobot, operatingmode,androbotorder;separatedbyrobotorder. p =.004, =.05], f =.26amediumeffect,withadifferenceinmeansInuktun-Packbotof.94, andintheemotivemode[F,123=6.14, p =.015, =.05], f =.22asmalleffect,withadifferenceinmeansInuktun-Packbotof.81.However,therewasnosignicantinteractionforthe PackbotFirstorder,[F,61=.12 p =.729, =.05],suggestingthatthedifferencebetweenrobotsdidnotdifferbyoperatingmode.RefertoFigure14.Alsotherewasnotastatisticallysignicantdifferencebetweenoperatingmodes[Fvalues < 2.74, p values > .100].Thedifferencein meansforInuktunemotive-standardwas.54andforthePackbotemotive-standardwere.67. Afurtheranalysiswasconductedtodecomposethethree-wayinteraction,byconductinganalysesseparatelyforeachoperatingmodeandthisrevealedadditionalinformation.Inthestandard mode,therewasnotastatisticallysignicantrobotbyorderinteraction,[F,62=.14, p =.706, =.05].Therewasastatisticallysignicantdifferenceonhowpositivelyparticipantsrespondedto eachrobotbasedonthesequenceofinteractionRefertoFigure15.Pairwisecomparisonsindicatedthat,inthestandardmode,participantsrespondedmorepositivelytotheInuktunwhenitwas viewedafterthePackbot[F,123=7.85, p =.006, =.05], f =.25amediumeffect,withamean differencePackbotFirst-InuktunFirstof1.05,comparedtowhentheInuktunwasviewedrst followedbythePackbot[F,123=4.44, p =.037, =.05], f =.19asmalleffect,withamean differencePackbotFirst-InuktunFirstof.88.Thispatternwasalsoobservedinthepairwise comparisonsconductedforrobotsintheemotivemodewiththeparticipantsrespondingmorepositivelyoveralltotheInuktunwhenitwasviewedafterthePackbot[F,123=7.62, p =.007, = .05], f =.24amediumeffect,withameandifferencePackbotFirst-InuktunFirstof1.04,comparedtotheInuktunviewedrstfollowedbythePackbot[F,123=28.46, p < .0001, =.05], 87

PAGE 104

f =.47alargeeffect,withameandifferencePackbotFirst-InuktunFirstof2.23.However,the robotbyorderinteractionwasstatisticallysignicant,F,61=6.81, p =.011, =.05], f =.33a mediumeffect,indicatingthatthedifferenceforthePackbotwasgreaterthanthedifferenceforthe Inuktun. Overall,participantsweremorepositivetowardtheInuktunthantheyweretowardthePackbot.However,thisdifferencewasmorepronouncedforparticipantswhosawtheInuktunFirstin theemotivemode.Further,whenparticipantsviewedtheInuktunFirstfollowedbythePackbot,in eitheroperatingmode,theoverallvalencescoresforbothrobotswerelower.Thisdifferencewas mostpronouncedforthePackbotwhenitwasseenintheemotivemode.Therewasnosignicant differenceinthevalenceresponsesbasedontheoperatingmodeoftherobots. Figure15.Statisticallysignicantthree-wayinteractionforSAM Valence Recodedbyrobot, operatingmode,androbotorder;separatedbyoperatingmode. Otherresultsforwithinandbetween-subjectsfactorswerenotstatisticallysignicant.Further analyseswereconductedtodetermineiftherewereanystatisticallysignicantresults,takinginto considerationgenderandsourceofrecruitmentwithnostatisticallysignicantresultsdiscovered. Theresultsforthemaineffectofoperatingmodewasnotstatisticallysignicant,[F,123=1.35, p =.247].Therewerenootherstatisticallysignicantresultsfortwo-wayorthree-wayinteractions forvalence,with[Fvalues < 2.94,and p values > .089]. 5.2.4TheSAMArousalAnalysesandResults Thereweretwomaineffectsthatwerestatisticallysignicantinthearousaldimensionofthe Self-AssessmentManikinandthosewerediscoveredinthebetween-subjectsfactorsofoperating modestandardversusemotiveandrobotorderInuktunFirstversusPackbotFirst.Descriptive 88

PAGE 105

statisticsforthevariablesassociatedwiththeSAMarousaldimensionarepresented,followedby adiscussionofthestatisticallysignicantresultsfromtheinferentialstatisticsforthearousalby operatingmodeandarousalbyrobotordermaineffects. 5.2.4.1DescriptiveStatisticalAnalysisforSAMArousalbyOperatingModeandRobot Order Therearetwomaindependentvariablesforthearousaldimension-excitedto8-calm: Arousal-Inuktun,andArousal-Packbot.Thesetwovariableswereanalyzed,basedonbothoperatingmodeandrobotorderusingthefollowingcodingscheme:00StandardMode-InuktunFirst,01 StandardMode-PackbotFirst,10EmotiveMode-InuktunFirst,11EmotiveMode-PackbotFirst. ThedescriptivestatisticsforeachoperatingmodeandrobotorderscenarioforArousal-Inuktun andArousal-PackbotareprovidedinTable14.TheboxplotsforArousal-InuktunandArousalPackbotbyoperatingmodeandrobotorderaredisplayedinFigures16and17respectively. Table14.DescriptivestatisticsforthetwodependentvariablesforSAMarousalbyoperating modeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01StandardModePackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-PackbotFirst. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu Arousal-Inuktun-00 3.70 1.85 .51 -.94 Arousal-Inuktun-01 4.36 1.44 .37 .14 Arousal-Inuktun-10 4.02 1.27 .06 -.05 Arousal-Inuktun-11 5.87 2.09 -.55 -.99 Arousal-Packbot-00 3.52 1.55 .09 -.85 Arousal-Packbot-01 4.20 1.71 .54 -.18 Arousal-Packbot-10 4.05 1.57 .01 -.20 Arousal-Packbot-11 5.32 1.66 -.05 -1.03 ThevariableArousal-Inuktun-00wasrelativelynormallydistributedwithaslightpositive skewness Sk =.51andmoderatelyplatykurtic Ku =-.94hadameanvalueof3.70.ArousalInuktun-01hadameanvalueof4.36andappearedtobenormallydistributedwithaslightpositiveskewness Sk =.37andslightlyleptokurtic Ku =.14.Arousal-Inuktun-10hadameanvalue of4.02andseemednormallydistributedwithaveryslightpositiveskewness Sk =.06andvery slightlyplatykurtic Ku =-.05.Arousal-Inuktun-11hadameanvalueof5.87andwasrelatively normallydistributedwithaslightlynegativeskewness Sk =-.55andwasmoderatelyplatykurtic 89

PAGE 106

Figure16.BoxplotforSAM Arousal-Inuktunbyoperatingmodeandrobotorder. Figure17.BoxplotforSAM Arousal-Packbotbyoperatingmodeandrobotorder. 90

PAGE 107

Ku =-.99.Overall,participantsreportedfeelingthecalmestwheninteractingwiththeInuktun robotintheEmotiveModewhenviewedafterinteractingwiththePackbotintheEmotiveMode. ItappearedfromthedatathatparticipantswerecalmeringeneraltotheInuktunrobotwhenitwas viewedafterthePackbotineitheroperatingmode. Arousal-Packbot-00wasrelativelynormallydistributed;howevertherewasaveryslightpositiveskewness Sk =.09andmoderatelyplatykurtic Ku =-.85withameanvalueof3.52.Forthe variableArousal-Packbot-01themeanvaluewas4.20anditappearedtoberelativelynormally distributedwithapositiveskewness Sk =.54andslightlyplatykurtic Ku =-.18.Arousal-Packbot-10hadameanvalueof4.05andseemedrelativelynormallydistributedwithanegliblepositiveskewness Sk =.01andslightlyplatykurtic Ku =-.20.Arousal-Packbot-11hadameanvalue of5.32,appearedtobenormallydistributedwithaveryslightnegativeskewness Sk =-.05and wasmoderatelyplatykurtic Ku =-1.03.WhencomparingthemeanvaluesforthePackbotconsideringoperatingmodeandrobotorder,participantsreportedbeingthecalmestwheninteracting withthePackbotrtintheemotivemode.Participantresponsedmostcalmlytobothrobotswhen interactingwiththePackbotFirstfollowedbytheInuktunwithbothrobotsoperatedintheemotivemode. 5.2.4.2InferentialStatisticalAnalysesforSAMArousalbyOperatingModeandRobot Order TheanalysesforevaluatingthedependentvariablearousalfromtheSelf-AssessmentManikinassessmentsincludedawithin-subjectsfactorofrobotInuktunandPackbot,withtwobetween-subjectsfactorsoperatingmodestandardoremotiveandrobotorderInuktunFirstor PackbotFirst.Thestudyevaluatedwhethertherewereanystatisticallysignicantdifferences inthearousalresponsesbasedonrobot,operatingmode,and/orrobotorder.Ifanystatistically signicantdifferenceswereobtainedthentheassociatednullhypthesiswasrejected. ThereweretwostatisticallysignicantmaineffectsobtainedfromtherepeatedmeasuresmultivariateanalysisofvarianceMANOVAbyoperatingmodeandrobotorderthatwereconducted. Therewasanarousalbyoperatingmodeandanarousalbyrobotordermaineffectdiscovered. TheSAMarousalresponsebyoperatingmodehadastatisticallysignicantresult[F,123= 12.05, p =.001, =.05].Cohen's f effectformeasuringeffectsizewasusedwith f =.31which 91

PAGE 108

Figure18.ArousalbyoperatingmodemaineffectfromtheSAMassessments. wasamediumeffectSeeFigure18Cohen,1988.Theresultsindicatethatbasedonparticipants'responsestotheSAMassessmentsthattheparticipantswhointeractedwiththeemotive operatedrobotsM=4.81,basedon0-excitedto8-calmwerecalmertobothrobotsthanthose thatinteractedwiththestandardoperatedrobotsM=3.95. ArousalresponsebyrobotordermaineffectfromtheSAMassessmentswasstatisticallysignicant.Theresultwas[F,123=19.98, p < .0001, =.05],andexhibitedalargeeffectsizeof f =.40.ThegraphofthisresultisshowninFigure19.Theresultsindicatethatparticipantsoverall werecalmerwhentheyinteractedwiththePackbotFirstfollowedbytheInuktunM=4.94,based on0-excitedto8-calmcomparedtothoseparticipantswhoviewedtheInuktunFirstfollowedby thePackbotM=3.82. Theresultsfromtheotherwithinandbetween-subjectsanalyseswerenotstatisticallysignificant.Theresultsobtainedforarousalbyrobotwere,[F,119=1.34, p =.249].Therewereno statisticallysignicantresultsobtainedforanyofthetwo-wayorthreewayinteractionsevaluated,with[Fvalues < 3.19,and p values > .077].Thereforenofurtheranalysiswasconductedfor arousalresponsesbasedontheSAMassessments. 92

PAGE 109

Figure19.ArousalbyrobotordermaineffectfromtheSAMassessments. 5.3PsychophysiologyAnalysesandResults Therewerefourdifferenttypesofpsychophysiologicalmeasuresusedtoevaluateparticipants' arousallevels:heartrateHRchange,respirationamplitudeRAchange,respirationrateRR change,andskinconductancelevelSCLchange.Thesephysiologicalchangeswerecalculated bysubtractingthesecondrestingratemeanvaluesfromindividualrobotinteractionmeanvalues e.g.,HR-change-Inuktun=HR-Inuktun-HR-Resting2,HR-change-Packbot=HR-Packbot-HRResting2,....Descriptiveandinferentialstatisticswerecalculatedforeachpsychophysiology measurebasedonoperatingmodeandrobotorder.Thedescriptivestatisticsincludeananalysisof centraltendenciesmean,standarddeviation,skewness,kurtosis,anddistributioncharacteristics foreachpsychophysiologicalmeasurebyoperatingmodeandrobotorder. TheinferentialstatisticalanalysesincludedarepeatedmeasuresmultivariateanalysisofvarianceMANOVA.Thereweretwolevelsforthewithin-subjectsfactor,whichisrobotInuktun andPackbot.Thereweretwolevelsforeachofthetwobetween-subjectsfactors:operatingmode standardoremotiveandrobotorderInuktunFirstorPackbotFirst.Therewereeightdependent variablesHRchange-Inuktun,HRchange-Packbot,RAchange-Inuktun,RAchange-Packbot, 93

PAGE 110

RRchange-Inuktun,RRchange-Packbot,SCLchange-Inuktun,SCLchange-Packbotthatwere evaluatedforstatisticalsignicanceusingF-testswithTypeIIISumofSquaresandeffectsizewas givenbyCohen's f Cohen,1988.One-wayanalysisofvarianceANOVAtestswererunforany statisticallysignicantinteractionsusinganF-testvalueandeffectsizegivenbyCohen's f Cohen,1988.Analyseswereconductedgenderusedasacovariancefactor;howevertherewasno statisticallysignicantdifferencesobservedwithgender. Additionally,aunivariateanalysiswasconductedusingeachofthepsychophysiologicalresponsesmeasuredHR,RA,RR,SCLfromtherstrobotinteractiontoseeiftherewereanystatisticallysignicantdifferencesfromtherstrobotinteraction.Thephysiologicalchangevariables werecalculatedbysubtractingthesecondrestingratefromthevaluesobtainedfromtherstinteractione.g.,HRchange-robot1=HR-robot1-HR-resting2,....Therewerefourdependent variablesusedintheunivariateanalysisHRchange-Robot1,RAchange-Robot1,RRchange-Robot1,SCLchange-Robot1withstatisticallysignicantresultspresentedusingtheresultsfrom F-testswithTypeIIISumofSquaresandeffectsizewasgivenbyCohen's f Cohen,1988. 5.3.1HeartRateChangeAnalysesandResults Therewasatwo-wayinteractionthatwasstatisticallysignicantintheheartrateHRchange analysis,robotbyrobotorderinteractioneffect.Descriptivestatisticsforthevariablesassociated withtheHRchangearepresented,followedbyadiscussionofthestatisticallysignicantresult fromtheinferentialstatisticalanalysisforthetwo-wayinteractionforHRchangebyrobot,and robotorder.Additionally,aunivariateanalysisofvariancewasconductedforheartratechangeusingdatafromtherstrobotinteractiononly;howevertherewerenostatisticallysignicantresults observed. 5.3.1.1DescriptiveStatisticalAnalysisforMultivariateHeartRateChangebyOperating ModeandRobotOrder ThereweretwomaindependentvariablesforHRchange:HR-Inuktun,andHR-Packbot. Thesetwovariableswereanalyzedbasedonbothoperatingmodeandrobotorderusingthefollowingcodingscheme:00StandardMode-InuktunFirst,01StandardMode-PackbotFirst,10 EmotiveMode-InuktunFirst,11EmotiveMode-PackbotFirst.Thedescriptivestatisticsforeach 94

PAGE 111

operatingmodeandrobotorderscenarioforHR-InuktunandHR-PackbotareprovidedinTable 15.TheboxplotsforHR-InuktunandHR-PackbotbyoperatingmodeandrobotorderaredisplayedinFigures20and21respectively. Table15.Descriptivestatisticsforthetwodependentvariablesforheartratechangebyoperating modeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01StandardModePackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-PackbotFirst. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu HR-Inuktun-00 .84 3.94 .71 .15 HR-Inuktun-01 -2.78 3.31 -.35 .12 HR-Inuktun-10 -.01 2.48 -.42 .92 HR-Inuktun-11 -3.34 2.97 .11 -.77 HR-Packbot-00 -2.70 4.22 -.65 2.15 HR-Packbot-01 1.24 3.90 2.29 7.54 HR-Packbot-10 -3.65 4.07 -1.30 2.35 HR-Packbot-11 .25 3.19 -.26 .13 ThevariableHR-Inuktun-00wasrelativelynormallydistributedwithamoderatepositive skewness Sk =.71andmoderatelyleptokurtic Ku =.15hadameanvalueof.84.HR-Inuktun-01 hadameanvalueof-2.78andappearedtobenormallydistributedwithaslightnegativeskewness Sk =-.35andveryslightlyleptokurtic Ku =.12.HR-Inuktun-10hadameanvalueof-.01and seemedrelativelynormallydistributedwithanegativeskewness Sk =-.42andslightlyleptokurtic Ku =.92.HR-Inuktun-11hadameanvalueof-3.34andwasrelativelynormallydistributedwith averyslightpositiveskewness Sk =.11andwasmoderatelyplatykurtic Ku =-.77.Participants whointeractedwiththeInuktunFirstinthestandardoperatingmodeshowedaslightincreasein heartrateoflessthanonebeatperminute.Therewasessentiallynochangeinheartratewhen viewingtheInuktunFirstintheemotivemodeM=-.01versusheartratesrecordedduringthe secondrestingperiod.WhentheInuktunwasviewedsecondineithermode,theheartratechange exhibitedalargedecreasefromthevaluesrecordedduringthesecondrestingperiod. HR-Packbot-00wasrelativelynormallydistributed;howevertherewasamoderatenegative skewness Sk =-.65andstronglyleptokurtic Ku =2.15withameanvalueof-2.70.ForthevariableHR-Packbot-01themeanvaluewas1.24anditappearedtoberelativelynormallydistributed withalargepositiveskewness Sk =2.29andstronglyleptokurtic Ku =7.54.HR-Packbot-10 hadameanvalueof-3.65andseemedrelativelynormallydistributedwithastrongnegativeskew95

PAGE 112

Figure20.Boxplotforheartratechange-Inuktunbyoperatingmodeandrobotorder. Figure21.Boxplotforheartratechange-Packbotbyoperatingmodeandrobotorder. 96

PAGE 113

ness Sk =-1.30andstronglyleptokurtic Ku =2.35.HR-Packbot-11hadameanvalueof.25,appearedtobenormallydistributedwithaslightnegativeskewness Sk =-.26andwasveryslightly leptokurtic Ku =.13.InthestandardoperatingmodewiththePackbotFirst,participants'heart rateshowedaslightincreasefromtheirratesinthesecondrestingperiodM=1.24.Theheart ratechangeforthePackbotFirstintheemotivemodealsosawaveryslightincreaseovertheir ratesduringthesecondrestingperiodM=.25.Theheartratewasnoticeablylowerwhenthe Packbotwasviewedsecondineitheroperatingmode.Thedropinheartrateforthesecondrobot interactionregardlessofoperatingmodeorrobotindicatedahabituationeffectmayhaveoccurred duringthesecondrobotinteraction.Duetothisresult,itwasdeterminedthatfurtherinferential analysesshouldbeconductedtodetermineiftherewereanystatisticallysignicantresultsobservedwhenconsideringonlytherstrobotinteraction. 5.3.1.2InferentialStatisticalAnalysesforMultivariateHeartRateChangebyOperating ModeandRobotOrder Theanalysesforevaluatingthedependentvariableheartratechangeincludedawithin-subjectsfactorofrobotInuktunandPackbot,andtwobetween-subjectsfactorsoperatingmode standardoremotiveandrobotorderInuktunFirstorPackbotFirst.Thestudyevaluated whethertherewereanystatisticallysignicantdifferencesintheheartratechangebasedonrobot, operatingmode,and/orrobotorder.Ifanystatisticallysignicantdifferenceswereobtainedthen theassociatednullhypthesiswasrejected. Theresultsshowedthattherewasastatisticallysignicanttwo-wayinteractionforHRchange byrobot,androbotorder,[F,117=99.43, p < .0001, =.05].Theeffectsizeforthistwo-way interactionwas f =.91,alargeeffect.TheresultsaredisplayedinFigure22. Furtheranalysesusingpairwisecomparisonswereconductedtodecomposethistwo-wayinteraction.Therewasastatisticallysignicantdifferenceinheartratechangeobservedbetween therobotsdependingonthesequencetherobotswereviewed.WhentheInuktunwasviewedrst thedifferencebetweenrobotsforHRchangeM Inuktun -M Packbot =3.59,[F,117=47.22, p < .0001, =.05], f =.62,alargeeffect.WhenthePackbotwasviewedrstthemeandifferencebetweenrobotsforHRchangewasalsosignicantM Inuktun -M Packbot =-3.81,[F, 117=52.25, p < .0001, =.05], f =.66,alargeeffect.Therewasalsoastatisticallysignicant 97

PAGE 114

Figure22.Heartratechangebyrobotandordertwo-wayinteraction. differenceinmeansbyrobotorderforHRchangefortheInuktuncomparedwiththePackbot M InuktunFirst -M PackbotFirst =3.48,[F,117=35.05, p < .0001, =.05], f =.54,alargeeffect.TherewasastatisticallysignicantdifferenceinHRchangeforthePackbotwhenthePackbotwasviewedrstM InuktunFirst -M PackbotFirst =-3.92,[F,117=31.09, p < .0001, = .05], f =.51,alargeeffect. UsingtheFisher'sLeastSignicantDifferenceLSDcalculation,differenceswereexaminedbetweentheInuktunandPackbotwheneachoftheserobotswasviewedrst,andbetween therobotswheneachwasviewedsecond.BasedontheFisher'sLSD,thedifferenceinthemeans forHRchangemustbegreaterthan1.04atan =.05toindicateasignicantdifference.There wasnostatisticallysignicantdifferenceinHRchangebetweentheInuktunwhenitwasviewed rstcomparedtothePackbotwhenitwasviewedrst.Similarly,therewasnodifferenceinHR changebetweentheInuktunandPackbotwhenbothwereviewedsecond.Heartrateincreased totherstrobot,regardlessofrobottype.Heartratedecreasedtothesecondrobot,regardlessof robottype,likelybecauseparticipantshadhabituatedtotheappearanceoftherobotsduringthe study,asindicatedbyalowerheartrateduringthesecondrobotinteractionthanduringresting baseline Otherresultsforwithinandbetween-subjectsfactorswerenotstatisticallysignicant.Further analyseswereconductedtodetermineiftherewereanystatisticallysignicantresultstakinginto 98

PAGE 115

considerationgenderwithnostatisticallysignicantresultsdiscovered.Therewerenostatistically signicantresultsforthemaineffectsofrobot,operatingmode,androbotorder,oranyadditional interactions,with[Fvalues < 2.49,and p values > .117]. 5.3.1.3DescriptiveStatisticalAnalysesforUnivariateHeartRateChangebyOperating ModeandRobotOrder TherewasonedependentvariablefortheHRchangevaluesfortherstrobotinteraction,HRRobot1.Thisvariablewasanalyzedbasedonbothoperatingmodeandrobotusingthefollowing codingscheme:00StandardMode-Inuktun,01StandardMode-Packbot,10EmotiveMode-Inuktun,11EmotiveMode-Packbot.ThedescriptivestatisticsforeachoperatingmodeandrobotscenarioforHR-Robot1areprovidedinTable16.TheboxplotforHR-Robot1byoperatingmode androbotisdisplayedinFigure23. Table16.Descriptivestatisticsforthetwodependentvariablesforheartratechangebyoperating modeandrobotfortherstrobotinteraction.Thecodingschemeis00StandardMode-Inuktun, 01StandardMode-Packbot,10EmotiveMode-Inuktun,and11EmotiveMode-Packbot. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu HR-Robot1-00 .84 3.94 .71 .15 HR-Robot1-01 1.24 3.90 2.29 7.54 HR-Robot1-10 -.01 2.48 -.42 .92 HR-Robot1-11 .25 3.19 -.26 .13 ThevariableHR-Robot1-00wasrelativelynormallydistributedwithamoderatepositive skewness Sk =.71andmoderatelyleptokurtic Ku =.15hadameanvalueof.84.HR-Robot101hadameanvalueof1.24andappearedtobenormallydistributedwithastrongpositiveskewness Sk =2.29andstronglyleptokurtic Ku =7.54.HR-Robot1-10hadameanvalueof-.01and seemedrelativelynormallydistributedwithanegativeskewness Sk =-.42andslightlyleptokurtic Ku =.92.HR-Robot1-11hadameanvalueof.25andwasrelativelynormallydistributedwith aslightnegativeskewness Sk =-.26andwasmoderatelyleptokurtic Ku =.13.Inthestandard modeparticipantheartratesexhibitedasmallincreasetobothrobots,InuktunM=.84andPackbotM=1.24.Intheemotivemodetherewasrelativelylittlechangeinheartratefromtherates recordedfromthesecondrestingperiod,InuktunM=-.01andPackbotM=.25.Theseval99

PAGE 116

Figure23.Boxplotforheartratechange-Robot1byoperatingmodeandrobotorder. uesindicatedthatparticipantsmayhavebeenslightlymorearousedtobothrobotsoperatedinthe standardmodecomparedtoparticipantsinteractingwithrobotsintheemotivemodeduringthe rstrobotinteraction. 5.3.1.4InferentialStatisticalAnalysesforUnivariateHeartRateChangebyOperating ModeandRobotOrder Theunivariateanalysisforheartratechangefortherstrobotinteractiondidnotindicateany statisticallysignicantmaineffectsorinteractions.Theresultsindicated[Fvalues < 2.17,and p values > .144].TherewasaslighttrendtowardamaineffectforHR-Robot1foroperatingmode with[F,117=2.17, p =.144, =.05],thoughnotstatisticallysignicant. 5.3.2RespirationAmplitudeChangeAnalysesandResults TherewerenostatisticallysignicantresultsobservedintherespirationamplitudeRA changeanalysis.DescriptivestatisticsforthevariablesassociatedwiththeRAchangearepresented,followedbyadiscussionoftheinferentialstatisticalanalysisforRAchange.Additionally, 100

PAGE 117

aunivariateanalysisofvariancewasconductedforrespirationamplitudechangeusingdatafrom therstrobotinteractiononly;howevertherewerenostatisticallysignicantresultsobserved. 5.3.2.1DescriptiveStatisticalAnalysisforMultivariateRespirationAmplitudeChangeby OperatingModeandRobotOrder Thereweretwomaindependentvariablesforrespirationamplitudechange:RA-Inuktun,and RA-Packbot.Thesetwovariableswereanalyzedbasedonbothoperatingmodeandrobotorder usingthefollowingcodingscheme:00StandardMode-InuktunFirst,01StandardMode-Packbot First,10EmotiveMode-InuktunFirst,11EmotiveMode-PackbotFirst.Thedescriptivestatistics foreachoperatingmodeandrobotorderscenarioforRA-InuktunandRA-Packbotareprovidedin Table17.TheboxplotsforRA-InuktunandRA-Packbotbyoperatingmodeandrobotorderare displayedinFigures24and25respectively. Table17.Descriptivestatisticsforthetwodependentvariablesforrespirationamplitudechange byoperatingmodeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01 StandardMode-PackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-Packbot First. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu RA-Inuktun-00 -.10 .77 -.33 -.17 RA-Inuktun-01 .03 .69 0.23 -.11 RA-Inuktun-10 -.08 .95 -.11 2.95 RA-Inuktun-11 -.12 1.02 -.83 1.04 RA-Packbot-00 -.18 .79 -1.17 1.78 RA-Packbot-01 .17 .77 1.68 3.51 RA-Packbot-10 .12 .79 .75 2.70 RA-Packbot-11 -.07 .94 -.78 .58 ThevariableRA-Inuktun-00wasrelativelynormallydistributedwithamoderatepositive skewness Sk =.77andveryslightlyplatykurtic Ku =-.17hadameanvalueof-.10.RA-Inuktun01hadameanvalueof.03andappearedtobenormallydistributedwithaslightpositiveskewness Sk =.23andveryslightlyplatykurtic Ku =-.11.RA-Inuktun-10hadameanvalueof-.08 andseemedrelativelynormallydistributedwithaveryslightnegativeskewness Sk =-.11and stronglyleptokurtic Ku =2.95.RA-Inuktun-11hadameanvalueof-.12andwasrelativelynormallydistributedwithanegativeskewness Sk =-.83andwasmoderatelyleptokurtic Ku =1.04. 101

PAGE 118

Figure24.Boxplotforrespirationamplitudechange-Inuktunbyoperatingmodeandrobotorder. Figure25.Boxplotforrespirationamplitudechange-Packbotbyoperatingmodeandrobotorder. 102

PAGE 119

Therewereverylittledifferencesinparticipants'respirationamplitudechangefortheInuktunobservedregardlessofoperatingmodeandrobotorder. RA-Packbot-00wasrelativelynormallydistributed;howevertherewasamoderatenegative skewness Sk =-1.17andstronglyleptokurtic Ku =1.78withameanvalueof-.18.ForthevariableRA-Packbot-01themeanvaluewas.17anditappearedtoberelativelynormallydistributed withalargepositiveskewness Sk =1.68andstronglyleptokurtic Ku =3.51.RA-Packbot-10had ameanvalueof.12andseemedrelativelynormallydistributedwithapositiveskewness Sk =.75 andstronglyleptokurtic Ku =2.70.RA-Packbot-11hadameanvalueof-.07,appearedtobenormallydistributedwithanegativeskewness Sk =-.78andwasslightlyleptokurtic Ku =.58.The resultsforrespirationamplitudechangeindicatedverylittledifferenceforthePackbotregardless ofoperatingmodeandrobotorder. 5.3.2.2InferentialStatisticalAnalysesforMultivariateRespirationAmplitudeChangeby OperatingModeandRobotOrder Resultsforthewithinandbetween-subjectsfactorsforrespirationamplitudewerenotstatisticallysignicant.Furtheranalyseswereconductedtodetermineiftherewereanystatistically signicantresultstakingintoconsiderationgenderwithnostatisticallysignicantresultsdiscovered.Therewerenostatisticallysignicantresultsforrespirationamplitudechangeformainor interactioneffectsbyrobot,operatingmode,androbotorder,with[Fvalues < 2.01,and p values > .159]. 5.3.2.3DescriptiveStatisticalAnalysesforUnivariateRespirationAmplitudeChangeby OperatingModeandRobotOrder TherewasonedependentvariablefortheRAchangevaluesfortherstrobotinteraction,RARobot1.Thisvariablewasanalyzedbasedonbothoperatingmodeandrobotusingthefollowing codingscheme:00StandardMode-Inuktun,01StandardMode-Packbot,10EmotiveMode-Inuktun,11EmotiveMode-Packbot.ThedescriptivestatisticsforeachoperatingmodeandrobotscenarioforRA-Robot1areprovidedinTable18.TheboxplotforRA-Robot1byoperatingmode androbotisdisplayedinFigure26. 103

PAGE 120

Table18.Descriptivestatisticsforthetwodependentvariablesforrespirationamplitudechange byoperatingmodeandrobotfortherstrobotinteraction.Thecodingschemeis00Standard Mode-Inuktun,01StandardMode-Packbot,10EmotiveMode-Inuktun,and11EmotiveModePackbot. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu RA-Robot1-00 -.10 .77 -.33 -.17 RA-Robot1-01 .16 .77 1.68 3.52 RA-Robot1-10 -.08 .95 -.11 2.95 RA-Robot1-11 -.07 .94 -.78 .58 ThevariableRA-Robot1-00wasrelativelynormallydistributedwithaslightnegativeskewness Sk =-.33andveryslightlyplatykurtic Ku =-.17hadameanvalueof-.10.RA-Robot101hadameanvalueof.16andappearedtobenormallydistributedwithastrongpositiveskewness Sk =1.68andstronglyleptokurtic Ku =3.52.RA-Robot1-10hadameanvalueof-.08 andseemedrelativelynormallydistributedwithaveryslightnegativeskewness Sk =-.11and stronglyleptokurtic Ku =2.95.RA-Robot1-11hadameanvalueof-.11andwasrelativelynormallydistributedwithanegativeskewness Sk =-.78andwasslightlyleptokurtic Ku =.58.The resultsindicatetherewasverylittledifferenceinrespirationamplitudechangeintherstrobot interactionregardlessofrobotandoperatingmode. 5.3.2.4InferentialStatisticalAnalysesforUnivariateRespirationAmplitudeChangeby OperatingModeandRobotOrder Theunivariateanalysisforrespirationamplitudechangefortherstrobotinteractiondidnot indicateanystatisticallysignicantmaineffectsorinteractions,[Fvalues < .82,and p values > .366]. 5.3.3RespirationRateChangeAnalysesandResults Therewasatwo-wayinteractionthatwasstatisticallysignicantintherespirationrateRR changeanalysis,robotbyrobotorderinteractioneffect,aswellasastatisticallysignicanttwowayinteractionforrespirationratechangebyoperatingmode,androbotorder.DescriptivestatisticsforthevariablesassociatedwiththeRRchangearepresented,followedbyadiscussionofthe statisticallysignicantresultsfromtheinferentialstatisticalanalysisforthetwo-wayinteractions 104

PAGE 121

Figure26.Boxplotforrespirationamplitudechange-Robot1byoperatingmodeandrobotorder. forRRchangebyrobot,androbotorderandalsoforoperatingmodeandrobotorder.Additionally,aunivariateanalysisofvariancewasconductedforrespirationratechangeusingdatafrom therstrobotinteractiononly,resultinginastatisticallysignicanttwo-wayinteractionforrespirationratechangebyrobot,andoperatingmode. 5.3.3.1DescriptiveStatisticalAnalysisforMultivariateRespirationRateChangebyOperatingModeandRobotOrder ThereweretwomaindependentvariablesforRRchange:RR-Inuktun,andRR-Packbot. Thesetwovariableswereanalyzedbasedonbothoperatingmodeandrobotorderusingthefollowingcodingscheme:00StandardMode-InuktunFirst,01StandardMode-PackbotFirst,10 EmotiveMode-InuktunFirst,11EmotiveMode-PackbotFirst.Thedescriptivestatisticsforeach operatingmodeandrobotorderscenarioforRR-InuktunandRR-PackbotareprovidedinTable 19.TheboxplotsforRR-InuktunandRR-PackbotbyoperatingmodeandrobotorderaredisplayedinFigures27and28respectively. 105

PAGE 122

Figure27.Boxplotforrespirationratechange-Inuktunbyoperatingmodeandrobotorder. Figure28.Boxplotforrespirationratechange-Packbotbyoperatingmodeandrobotorder. 106

PAGE 123

Table19.Descriptivestatisticsforthetwodependentvariablesforrespirationratechangebyoperatingmodeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01Standard Mode-PackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-PackbotFirst. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu RR-Inuktun-00 .76 1.62 .30 -.01 RR-Inuktun-01 .41 1.40 .37 .36 RR-Inuktun-10 -.19 1.27 -.17 -.71 RR-Inuktun-11 .78 1.46 .65 2.38 RR-Packbot-00 1.18 2.09 .98 .69 RR-Packbot-01 .13 1.42 -.71 .85 RR-Packbot-10 .19 1.72 .52 .78 RR-Packbot-11 .25 1.49 -.74 1.77 ThevariableRR-Inuktun-00wasrelativelynormallydistributedwithaslightpositiveskewness Sk =.30andveryslightlyplatykurtic Ku =-.01hadameanvalueof.76.RR-Inuktun-01 hadameanvalueof.41andappearedtobenormallydistributedwithaslightpositiveskewness Sk =.37andslightlyleptokurtic Ku =.36.RR-Inuktun-10hadameanvalueof-.19andseemed relativelynormallydistributedwithaveryslightnegativeskewness Sk =-.17andplatykurtic Ku =-.71.RR-Inuktun-11hadameanvalueof.78andwasrelativelynormallydistributedwith apositiveskewness Sk =.65andwasstronglyleptokurtic Ku =2.38.Participantswhointeracted withtheInuktunFirstinthestandardoperatingmodeshowedaslightincreaseinrespirationrate oflessthanonebreathperminuteM=.76,whichwasevenlesswhentheInuktunwasviewed secondinthestandardmodeM=.41.Therewasaveryslightdecreaseinrespirationratewhen viewingtheInuktunFirstintheemotivemodeM=-.19;howeverwhentheInuktunwasviewed secondintheemotivemodetherewasanincreaseinrespirationrateM=.78. RR-Packbot-00wasrelativelynormallydistributed;howevertherewasamoderatepositive skewness Sk =.98andleptokurtic Ku =.69withameanvalueof1.18.ForthevariableRRPackbot-01themeanvaluewas.13anditappearedtoberelativelynormallydistributedwitha negativeskewness Sk =-.71andmoderatelyleptokurtic Ku =.85.RR-Packbot-10hadamean valueof.19andseemedrelativelynormallydistributedwithapositiveskewness Sk =.52and moderatelyleptokurtic Ku =.78.RR-Packbot-11hadameanvalueof.25,appearedtobenormallydistributedwithanegativeskewness Sk =-.74andwasstronglyleptokurtic Ku =1.77.In thestandardoperatingmodewiththePackbotviewedsecond,participants'respirationrateshowed 107

PAGE 124

aslightincreasefromtheirratesinthesecondrestingperiodM=1.18comparedtothePackbot viewedrstinthestandardmodeM=.13.TherespirationratechangeforthePackbotrstin theemotivemodealsosawaveryslightincreaseovertheirratesduringthesecondrestingperiod M=.25,witharespirationratechangethatwasjustslightlylesswhenthePackbotwasviewed secondintheemotivemodeM=.19. 5.3.3.2InferentialStatisticalAnalysesforMultivariateRespirationRateChangebyOperatingModeandRobotOrder TheanalysesforevaluatingthedependentvariablerespirationratechangeincludedawithinsubjectsfactorofrobotInuktunandPackbot,andtwobetween-subjectsfactorsoperatingmode standardoremotiveandrobotorderInuktunFirstorPackbotFirst.Thestudyassessedwhether therewereanystatisticallysignicantdifferencesintherespirationratechangebasedonrobot, operatingmode,and/orrobotorder.Ifanystatisticallysignicantdifferenceswereobtainedthen theassociatednullhypthesiswasrejected. Theresultsshowedthattherewasastatisticallysignicanttwo-wayinteractionforRRchange byoperatingmode,androbotorder,[F,122=5.76, p =.018, =.05].Theeffectsizeforthis two-wayinteractionwas f =.21,asmalleffect.TheresultsaredisplayedinFigure29. Pairwisecomparisonswereconductedtodecomposethistwo-wayinteraction.Therewasa statisticallysignicantdifferenceobservedinthestandardmode[F,124=3.92, p =.050, = .05], f =.18,asmalleffect.ThedifferenceforRRchangeM InuktunFirst -M PackbotFirst =.70, withtherespirationratesignicantlyhigher,regardlessofrobottype,intheInuktunFirstconditionthaninthePackbotFirstcondition.Therewasnostatisticallysignicantdifferenceobserved intheemotivemode[F,124=2.11, p =.149],forRRchangeM InuktunFirst -M PackbotFirst = -.52.Thisndingwouldindicatethattherewasasignicantdifferenceinthestandardmodedependingonthesequenceofrobotsexperienced,thoughthiseffectwasnotevidentintheemotive mode.Whenrobotswereoperatedinstandardmode,RRchangetobothrobotswassignicantly greaterforparticipantsgiventheInuktunFirstorderthanthosegiventheinthePackbotFirstorder.However,whenrobotswereoperatedintheemotivemode,RRchangetobothrobotsdidnot differbetweenparticipantsgivendifferentorders. 108

PAGE 125

Figure29.Respirationratechangebyoperatingmodeandrobotordertwo-wayinteraction. Fromadifferentpairwisecomparison,therewasastatisticallysignicantdifferenceinthe meansforparticipantswhoviewedtheInuktunFirstM standard -M emotive =.97,[F,124= 7.63, p =.007, =.05], f =.24,asmalleffect.Thisindicatedthattherewasasignicantincrease inRRchangetobothrobotsfortheparticipantswhoviewedtheInuktunFirstinthestandard mode,whichindicatedaslightarousalresponseofapproximatelyonebreathperminuteovertheir secondrestingrespirationrate.ConverselytherewasnochangeintherespirationratetobothrobotswhentheInuktunwasviewedFirstfollowedbythePackbotintheemotivemodeoverthe meanrespirationratesrecordedduringthesecondrestingperiod.Additionally,therewerenostatisticallysignicantdifferencesbetweenoperatingmodesforparticipantswhoviewedthePackbot FirstfollowedbytheInuktunM standard -M emotive =-.25,[F,124=.47, p =.493, =.05]. WhentheInuktunwasviewedrst,theRRresponsetobothrobotswasgreaterinthestandard modethanintheemotivemode.However,whenthePackbotwasviewedrst,therewasnodifferenceintheRRresponsetobothrobotsbetweenoperatingmodes.Thisindicatedthatparticipants respondedinasimilarmannertobothoperatingmodeswhenthePackbotwasviewedrst. Theresultsshowedthattherewasastatisticallysignicanttwo-wayinteractionforRRchange byrobot,androbotorder,[F,124=10.91, p =.001, =.05].Theeffectsizeforthistwo-way interactionwas f =.29,amediumeffect.TheresultsaredisplayedinFigure30. 109

PAGE 126

Figure30.Respirationratechangebyrobotandrobotordertwo-wayinteraction. Pairwisecomparisonswereconductedtodecomposethistwo-wayinteraction.Therewasa statisticallysignicantdifferencebetweenrobotswhenintheInuktunFirstconditionM Inuktun M Packbot =.39,[F,124=5.39, p =.022, =.05], f =.21,asmalleffect.Participantsresponded withanincreaseinrespirationratetothePackbotafterviewingtheInuktun.Additionally,there wasastatisticallysignicantdifferenceinRRchangetobothrobotswhenthePackbotwasviewed rstfollowedbytheInuktunM Inuktun -M Packbot =.41,[F,124=5.51, p =.020, =.05], f = .21,asmalleffect.Inthiscase,therewasanincreaseintheRRwhentheInuktunwasviewedfollowingthePackbot.Itappearedinthiscasethatitwasnotrelevantwhichrobotwasviewedrst, theRRchangewaslessthantheRRchangeforthesecondrobot.Apossibleexplanationforthis phenomenonisthatparticipantsexperiencedasomewhatlowerRRtotherstrobot,regardlessof robottypebecausetheywereanticipatingwhatwouldhappenandwerebeingmoreattentiveresultinginslightlyslowerbreathing;whereaswiththesecondrobot,regardlessofrobottypethey weremorecomfortableandbreathingataslightlyfasterrate.Thiswouldbeindicativeofahabituationeffect.Thoughstatisticallysignicant,thisdifferencewasonlyaboutonehalfofabreath perminute,whichwasnotalargedifference.Usingpairwisecomparisons,therewasnostatisticallysignicantdifferenceintheRRchangebetweentheInuktunandthePackbot,regardlessof presentationorder[Fvalues < 2.68, p values > .104]. 110

PAGE 127

Otherresultsforwithinandbetween-subjectsfactorsforrespirationratechangewerenotstatisticallysignicant.Furtheranalyseswereconductedtodetermineiftherewereanystatistically signicantresultstakingintoconsiderationgenderwithnostatisticallysignicantresultsdiscovered.Therewerenostatisticallysignicantresultsforrespirationratechangemainorinteraction effectsbyrobot,operatingmode,orrobotorder,with[Fvalues < 2.10,and p values > .150]. 5.3.3.3DescriptiveStatisticalAnalysesforUnivariateRespirationRateChangebyOperatingModeandRobotOrder Therewasonedependentvariablefortherespirationratechangevaluesfortherstrobotinteraction,RR-Robot1.Thisvariablewasanalyzedbasedonbothoperatingmodeandrobotusing thefollowingcodingscheme:00StandardMode-Inuktun,01StandardMode-Packbot,10Emotive Mode-Inuktun,11EmotiveMode-Packbot.Thedescriptivestatisticsforeachoperatingmodeand robotscenarioforRR-Robot1areprovidedinTable20.TheboxplotforRR-Robot1byoperating modeandrobotisdisplayedinFigure31. Table20.Descriptivestatisticsforthetwodependentvariablesforrespirationratechangebyoperatingmodeandrobotfortherstrobotinteraction.Thecodingschemeis00StandardModeInuktun,01StandardMode-Packbot,10EmotiveMode-Inuktun,and11EmotiveMode-Packbot. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu RR-Robot1-00 .77 1.62 .30 -.01 RR-Robot1-01 .13 1.43 -.70 .85 RR-Robot1-10 -.19 1.27 -.17 -.71 RR-Robot1-11 .25 1.49 -.74 1.77 ThevariableRR-Robot1-00wasrelativelynormallydistributedwithaslightpositiveskewness Sk =.30andveryslightlyplatykurtic Ku =-.01hadameanvalueof.77.RR-Robot1-01 hadameanvalueof.13andappearedtobenormallydistributedwithamoderatenegativeskewness Sk =-.70andmoderatelyleptokurtic Ku =.85.RR-Robot1-10hadameanvalueof-.19and seemedrelativelynormallydistributedwithaveryslightnegativeskewness Sk =-.17andmoderatelyplatykurtic Ku =-.71.RR-Robot1-11hadameanvalueof.25andwasrelativelynormally distributedwithamoderatenegativeskewness Sk =-.74andwasstronglyleptokurtic Ku =1.77. 111

PAGE 128

Figure31.Boxplotforrespirationratechange-Robot1byoperatingmodeandrobotorder. 5.3.3.4InferentialStatisticalAnalysesforUnivariateRespirationRateChangebyOperatingModeandRobotOrder Theunivariateanalysisforrespirationratechangefortherstrobotinteractiondidreveala two-wayinteractionforrobot,andoperatingmode[F,124=4.30, p =.040, =.05],withan effectsizeof f =.18,asmalleffectSeeFigure32.Pairwisecomparisonswereconductedtodecomposethistwo-wayinteractionfurther.Theresultsindicatedtherewasastatisticallysignicant differenceinRRchangewhenparticipantsviewedtheInuktunintheirrstinteractionM standard -M emotive =.96,[F,124=6.93, p =.010, =.05], f =.23,asmalleffect.Thisindicatedthat participantswhointeractedwiththeInuktunintheemotivemodefortheirrstinteractionhada slightlylowerRRchangeovertheirsecondrestingrate,whichwouldindicateaslightcalmingeffect.Conversely,participantswhointeractedwiththeInuktunintheirrstinteraction,operatedin thestandardmodeexhibitedaslightincreaseinRRchangeovertheirsecondrestingrate,which wouldindicateaslightarousalresponse.TherewasnostatisticallysignicantdifferenceinRR changeforparticipantsviewingthePackbotfortheirrstinteraction,regardlessoftheoperating mode[Fvalues < 3.015, p values > .085]. 112

PAGE 129

Figure32.Boxplotforrespirationratechange-Robot1byoperatingmodeandrobotorder. TherewerenoadditionalstatisticallysignicantmaineffectsorinteractionsforRR-Robot1, [Fvalues < 2.62,and p values > .108].TherewasaslighttrendtowardamaineffectforRR-Robot1foroperatingmodewith[F,124=2.62, p =.108, =.05],thoughnotstatisticallysignicant. 5.3.4SkinConductanceLevelChangeAnalysesandResults TherewerenostatisticallysignicantresultsobservedintheskinconductancelevelSCL changeanalyses.DescriptivestatisticsforthevariablesassociatedwiththeSCLchangearepresented,followedbyadiscussionoftheinferentialstatisticalanalysisforSCLchange.Additionally,aunivariateanalysisofvariancewasconductedforskinconductancelevelchangeusing datafromtherstrobotinteractiononly;howevertherewerenostatisticallysignicantresults observed. 5.3.4.1DescriptiveStatisticalAnalysisforMultivariateSkinConductanceLevelChangeby OperatingModeandRobotOrder ThereweretwomaindependentvariablesforSCLchange:SCL-Inuktun,andSCL-Packbot. Thesetwovariableswereanalyzedbasedonbothoperatingmodeandrobotorderusingthefol113

PAGE 130

lowingcodingscheme:00StandardMode-InuktunFirst,01StandardMode-PackbotFirst,10 EmotiveMode-InuktunFirst,11EmotiveMode-PackbotFirst.Thedescriptivestatisticsforeach operatingmodeandrobotorderscenarioforSCL-InuktunandSCL-PackbotareprovidedinTable21.TheboxplotsforSCL-InuktunandSCL-Packbotbyoperatingmodeandrobotorderare displayedinFigures33and34respectively. Table21.Descriptivestatisticsforthetwodependentvariablesforskinconductancelevelchange byoperatingmodeandrobotorder.Thecodingschemeis00StandardMode-InuktunFirst,01 StandardMode-PackbotFirst,10EmotiveMode-InuktunFirst,and11EmotiveMode-Packbot First. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu SCL-Inuktun-00 .83 .98 .59 .81 SCL-Inuktun-01 .52 1.37 -.28 1.01 SCL-Inuktun-10 .50 1.08 .82 .93 SCL-Inuktun-11 .51 1.05 1.05 2.16 SCL-Packbot-00 .72 1.12 1.48 2.73 SCL-Packbot-01 .58 .93 .16 -.13 SCL-Packbot-10 .32 1.01 -.24 .59 SCL-Packbot-11 .52 .79 .05 .65 ThevariableSCL-Inuktun-00wasrelativelynormallydistributedwithapositiveskewness Sk =.59andmoderatelyleptokurtic Ku =.81hadameanvalueof.83.SCL-Inuktun-01hada meanvalueof.52andappearedtobenormallydistributedwithaslightnegativeskewness Sk = -.28andleptokurtic Ku =1.01.SCL-Inuktun-10hadameanvalueof.50andseemedrelatively normallydistributedwithamoderatepositiveskewness Sk =.82andleptokurtic Ku =.93.SCLInuktun-11hadameanvalueof.51andisrelativelynormallydistributedwithapositiveskewness Sk =1.05andwasstronglyleptokurtic Ku =2.16.Thelargestincreaseinskinconductancelevel wasobservedwhentheInuktunwasviewedrstinthestandardoperatingmodeM=.83.There wereverylittledifferenceswiththeotherskinconductancelevelchangesfortheInuktunregardlessofoperatingmodeorrobotorder. SCL-Packbot-00wasrelativelynormallydistributed;howevertherewasapositiveskewness Sk =1.48andstronglyleptokurtic Ku =2.73withameanvalueof.72.ForthevariableSCLPackbot-01themeanvalueis.58anditappearedtoberelativelynormallydistributedwithavery slightpositiveskewness Sk =.16andveryslightlyplatykurtic Ku =-.13.SCL-Packbot-10had 114

PAGE 131

Figure33.Boxplotforskinconductancelevelchange-Inuktunbyoperatingmodeandrobotorder. Figure34.Boxplotforskinconductancelevelchange-Packbotbyoperatingmodeandrobotorder. 115

PAGE 132

ameanvalueof.32andseemedrelativelynormallydistributedwithaveryslightnegativeskewness Sk =-.24andslightlyleptokurtic Ku =.59.SCL-Packbot-11hadameanvalueof.52,appearedtobenormallydistributedwithaveryslightpositiveskewness Sk =.05andwasslightly leptokurtic Ku =.65.Theresultsforskinconductancelevelchangeindicatedthelargestincrease inskinconductancelevelchangeoccurredwhenthePackbotwassecondinthestandardmode M=.72.ThesmallestincreaseinskinconductancelevelchangewasobservedwhenthePackbotwasviewedsecondintheemotivemodeM=.32.Therewereverylittledifferencesinskin conductancelevelchangeforthePackbotFirstinthestandardmodeM=.58comparedtothe PackbotFirstintheemotivemodeM=.52. 5.3.4.2InferentialStatisticalAnalysesforMultivariateSkinConductanceLevelChangeby OperatingModeandRobotOrder Resultsforthewithinandbetween-subjectsfactorsforskinconductancelevelchangewere notstatisticallysignicant.Furtheranalyseswereconductedtodetermineiftherewereanystatisticallysignicantresultstakingintoconsiderationgenderwithnostatisticallysignicantresults discovered.Therewerenostatisticallysignicantmainorinteractioneffectsforskinconductance levelchange,[Fvalues < 1.54,and p values > .218]. 5.3.4.3DescriptiveStatisticalAnalysesforUnivariateSkinConductanceLevelChangeby OperatingModeandRobotOrder TherewasonedependentvariablefortheSCLchangevaluesfortherstrobotinteraction, SCL-Robot1.Thisvariablewasanalyzedbasedonbothoperatingmodeandrobotusingthefollowingcodingscheme:00StandardMode-Inuktun,01StandardMode-Packbot,10Emotive Mode-Inuktun,11EmotiveMode-Packbot.Thedescriptivestatisticsforeachoperatingmodeand robotscenarioforSCL-Robot1areprovidedinTable22.TheboxplotforSCL-Robot1byoperatingmodeandrobotisdisplayedinFigure35. ThevariableSCL-Robot1-00wasrelativelynormallydistributedwithapositiveskewness Sk =.59andmoderatelyleptokurtic Ku =.81hadameanvalueof.83.SCL-Robot1-01hada meanvalueof.58andappearedtobenormallydistributedwithaveryslightpositiveskewness Sk =.16andveryslightlyplatykurtic Ku =-.13.SCL-Robot1-10hadameanvalueof.50and 116

PAGE 133

Table22.Descriptivestatisticsforthetwodependentvariablesforskinconductancelevelchange byoperatingmodeandrobotfortherstrobotinteraction.Thecodingschemeis00Standard Mode-Inuktun,01StandardMode-Packbot,10EmotiveMode-Inuktun,and11EmotiveModePackbot. VariableName MeanM Std.Dev.S SkewnessSk KurtosisKu SCL-Robot1-00 .83 .98 .59 .81 SCL-Robot1-01 .58 .93 .16 -.13 SCL-Robot1-10 .50 1.08 .82 .93 SCL-Robot1-11 .52 .79 .05 .65 seemedrelativelynormallydistributedwithamoderatepositiveskewness Sk =.82andleptokurtic Ku =.93.SCL-Robot1-11hadameanvalueof.52andwasrelativelynormallydistributed withaveryslightpositiveskewness Sk =.05andwasleptokurtic Ku =.65.Theresultsindicated thelargestincreaseinskinconductancelevelchangeoccurredwhentheInuktunwasviewedinthe standardmodeM=.83.Therewereverylittledifferencesobservedintherestoftheskinconductancelevelchangeregardlessofrobotoroperatingmode. 5.3.4.4InferentialStatisticalAnalysesforUnivariateSkinConductanceLevelChangeby OperatingModeandRobotOrder Theunivariateanalysisforskinconductancelevelchangefortherstrobotinteractiondidnot indicateanystatisticallysignicantmaineffectsorinteractions.Theresultsindicated[Fvalues < 1.37,and p values > .245]. 5.4Self-AssessmentManikinSAMandPsychophysiologyCorrelationAnalysisandResults ThelastphaseoftheanalysisprocessincludeddeterminingiftherewereanystatisticallysignicantcorrelationsbetweentheSelf-AssessmentManikinSAMandthepsychophysiologydata collected.TherewerefourvariablesfromtheSAMassessmentsValR-Inuktun,ValR-Packbot, Arous-Inuktun,Arous-Packbotwereevaluatedforcorrelationswiththeeightpsychophysiology variablesHR-Inuktun,HR-Packbot,RA-Inuktun,RA-Packbot,RR-Inuktun,RR-Packbot,SCLInuktun,SCL-Packbot.Thecorrelationmatrixwiththestatisticallysignicantresultshighlighted 117

PAGE 134

Figure35.Boxplotforskinconductancelevelchange-Robot1byoperatingmodeandrobotorder. inyellowisdisplayedinTable23.Therewerethreestatisticallysignicantcorrelations:ArousalInuktunwaspositivelycorrelatedwithHR-Packbot[r=.21, p =.021, =.05]withHR-Packbot accountingfor4.41%ofthevariabilityintheArousal-Inuktun,asmalleffect;ValenceR-Packbot waspositivelycorrelatedwithHR-Packbot[r=.24, p =.010, =.05]withHR-Packbotaccounting for5.76%ofthevariabilityinValenceR-Packbot,whichwasasmalleffect;andValenceR-Packbot waspositivelycorrelatedwithRR-Inuktun[r=.20, p =.021, =.05]withRR-Inuktunaccountin for4.00%ofthevariabilityinValenceR-Packbot,whichwasasmalleffect.Althoughthesecorrelationswerestatisticallysignicantthereeffectsweresmall.Thestrongestcorrelationisbetween theValenceR-PackbotandHR-Packbot,andadditionallythiswastheonlystatisticallysignicant correlationwithinthesamerobot. 118

PAGE 135

Table23.PearsoncorrelationmatrixfortheSelf-AssessmentManikin(SAM)assessmentswiththepsychophysiologydata. HR-Inuktun HR-Packbot RA-Inuktun RA-Packbot RR-Inuktun RR-Packbot SCL-Inuktun SCL-Packbot ValR-Inuktun r=-.10 r=.01 r=-.06 r=.02 r=-.04 r=-.04 r=.05 r=.06 p=.299 p=.950 p=.503 p=.832 p=.699 p=.665 p=.579 p=.590 ValR-Packbot r=-.17 r=.24 r=.05 r=.04 r=.20 r=.08 r=.04 r=.05 p=.071 p=.010 p=.614 p=.664 p=.021 p=.399 p=.622 p=.590 Arous-Inuktun r=-.10 r=.21 r=.04 r=.06 r=.15 r=.08 r=.04 r=.12 p=.257 p=.021 p=.685 p=.503 p=.097 p=.378 p=.699 p=.178 Arous-Packbot r=-.17 r=.05 r=.15 r=.13 r=-.03 r=-.08 r=-.02 r=.01 p=.067 p=.575 p=.104 p=.152 p=.721 p=.383 p=.869 p=.882 Correlationissignicantatthe0.05level(2-tailed). 119

PAGE 136

5.5Summary ThemostsignicantndingassociatedwiththeresearchquestionwasthestatisticallysignicantmaineffectforSAMarousalbyoperatingmode.Thisresultindicatedthatparticipants reportedbeingcalmerwheninteractingwithboththerobotsoperatingintheemotivemodecomparedtothoseoperatedinthestandardmode.Therewasastatisticallysignicantthree-wayinteractionforSAMvalencerecodedforrobot,operatingmode,androbotorder.Theresultsindicated thattherewasagreaterdifferenceexhibitedinhowpositivelyparticipantsviewedtheirinteractionswiththePackbotcomparedtotheInuktunintheemotivemode.Overall,participantsviewed theirinteractionswiththeInuktunmorepositivethanthePackbot,andingeneralweremorepositivelyandcalmertobothrobotswhenthePackbotwasviewedrst.Theheartratechangehada statisticallysignicanttwo-wayinteraction;howevertheprimaryndingwasthatthereappeared tobeahabituationeffectthatoccurredduringthesecondrobotinteractionregardlessofthetype ofrobotandoperatingmode.Therewasastatisticallysignicanttwo-wayinteractionfortherespirationratechangebyoperatingmodeandorder,inadditiontoastatisticallysignicanttwo-way interactionforrobotbyrobotorder.Participantsexhibitedastatisticallysignicantdifferencein respirationratechangefortheemotivemodecomparedtothestandardmodewhentheInuktun wasviewedrst.ThiswasnotapparentwhenthePackbotwasviewedrst.Theresultsalsoindicatedthatparticipantshadanincreaseintheirrespirationratechangetothesecondrobotthey experiencedcomparedtotherstrobotregardlessoftherobottype.Thereappearedtobeahabituationeffectthatoccurred;thereforefurtheranalyseswereconductedtoevaluatethepsychophysiologicalresponsesintherstinteractiononly.Theonlystatisticallysignicantndingwas thatparticipantshadanincreaseinrespirationratechangefortheInuktuninthestandardmode andaslightdecreaseintherespirationratechangeintheemotivemode.Acorrelationanalysis wasconductedevaluatingpossiblecorrelationsbetweentheSAMassessmentdataandthepsychophysiologydata.Therewasonlyoneimportantndingthatwasstatisticallysignicantforthis correlationanalysisandthatwasbetweenvalencerecodedPackbotandtheheartratechangefor Packbot.Acorrelationanalysiswasperformedtovalidatethevalenceandarousaldimensionsof theSAMassessmentforuseasanevaluationtoolbytheHRIcommunity. 120

PAGE 137

Chapter6 Discussion Theresultsfromthisresearchrevealedseveraltrendsandinsightsassociatedwithnon-facial andnon-verbalsocialhuman-robotinteraction,suchashumansreportedfeelingcalmerwheninteractingwithrobotsthatwereprogrammedtooperateinanemotivemode,andtheytendtocalibratetheirresponsestorobotsbasedontheirrstrobotencounter.Robotswillplaymoresignicantrolesinourdailylivesanditisimportanttogainabetterunderstandingofhowhumanswill respondtotheserobotswithwhichtheywillinteract.ThischapterdiscussesandinterpretstheresultsobtainedfromthehumanstudyassociatedwiththisresearchincludingcoverageoftheSelfAssessmentManikinSAM,psychophysiologymeasurements,andthecorrelationbetweenthe SAMandpsychophysiologyresults.Additionally,thereisadiscussionoffactorsthatmayhave impactedtheoutcomes,suggestedstudyimprovements,appropriatestudydesign,andopenquestionsdiscoveredasaresultofthestudy. 6.1DiscussionoftheSelf-AssessmentManikinSAMResults TheSelf-AssessmentManikinSAMresultswerethemostrevealingandsignicanttothis study.Asevidencedbythendings,itisapparentthatnon-facialandnon-verbalaffectiveexpressiondoesimpactsocialhuman-robotinteractionsasrelatedtotheurbansearchandrescueapplicationandalsoapplicationswhereitisimportanttokeephumanscalmduringsocialhuman-robot interaction.Thiswasobservedstronglyinarousalresponsesandalsotosomedegreeinthevalenceresponsesofparticipantsintheresearchstudy.Thissectiondiscussesandinterpretsboththe valenceandarousalresultsfromtheSAMassessments. 121

PAGE 138

6.1.1InterpretationandDiscussionoftheResultsforValence-RecodedResponsesbyRobot,OperatingMode,andRobotOrder Themostsignicantresultassociatedwithvalenceisthethree-wayinteractionbetweenrobotInuktunandPackbot,operatingmodestandardoremotive,androbotorderInuktunrst orPackbotrst.Inallscenarios,theInuktunwasviewedmorepositivelythanthePackbotregardlessofoperatingmodestandardoremotiveandrobotorderInuktunFirstorPackbotFirst. ParticipantsoverallrespondedmorepositivelytotheInuktuncomparedwiththePackbot,which wasexpected.Whilecompletingtheirassessmentsaftertherobotinteractions,approximately25% oftheparticipantscommentedverballytheyfelttheInuktunwascute.Additionally,theymade commentsabouttheInuktunhavingeyesvisibleinthedark,andtheyfeltmorecomfortablewith theInuktunSeeFigure36. Figure36.InuktunandPackbotintheinitialsimulateddisastersite. Therewasastatisticallysignicantdifferencebetweenhowpositivelyparticipantsviewed theInuktuncomparedtothePackbotinbothoperatingmodes;howevertherewasamorenotable differenceobservedintheInuktuncomparedtothePackbotwhenprogrammedtooperateinthe emotivemodeandtheInuktunwasviewedrstfollowedbythePackbot.Anexplanationforthis mightbethatparticipantsfoundtheinteractionswiththeInuktunintheemotivemodeaspositive;howeverwheninteractingwiththePackbotintheemotivemodefollowingtheInuktunthe largersizeandlackofvisiblefacialfeatureswhenthePackbotwasoperatedinthedarkmayhave 122

PAGE 139

comeacrossasnoticeablydifferent,impactingparticipants'responses.Thisdifferencemayhave beenmoreapparentintheemotivemodebecausebothrobotsmovedmoreslowlyatsimilarspeeds andmakingiteasiertodistinguishthedifferencesbetweentherobots.Inthestandardmode,both robotsaremovingquicklyandthemovementswereerraticinnature,whichmayhavedistracted participantsfrompayingattentiontotheactualdifferencesintherobots. Thesequenceinwhichtherobotsappearedsignicantlyimpactedhowpositivelyparticipants respondedtobothrobots.IfthePackbotwasviewedrstfollowedbytheInuktun,participants ratedtheInuktunasmorepositivelythanthePackbotinbothmodes.Specically,ifparticipants viewedthePackbotrstfollowedbytheInuktun,thescoressignicantlyincreasedfortheInuktunovertheirscoresforthePackbot.Conversely,whentheInuktunwasviewedrstfollowedby thePackbot,participantsrespondedmorepositivelytotheInuktunbutthescoressignicantlydecreasedwhenthePackbotwasviewednext.Thiseffectheldtrueinbothoperatingmodes.This wouldindicatethatparticipantshada calibrationeffect totherstrobotwithwhichtheyinteracted.Asaresultitisadvisablethatiftworobotsmustbesentintoadisastersituation,thatthe larger,lessmaneuverable,andpossiblymoreintimidatingrobotbesentinrstandthenparticipantswillreactlessstronglytothefollowingrobots. Therewerenostatisticallysignicantdifferencesobservedbetweentheoperatingmodeswhen evaluatingeachindividualrobot.Inotherwords,theresponseswerethesameoveralltotheInuktunprogrammedtooperateinthestandardandtheemotivemode;thesamewastrueoftheresponsestothePackbotprogrammedtooperateinthestandardmodeoremotivemode. Theseresultsindicatethatallthreefactors,operatingmode,robotorder,androbottype,have signicanteffectsonhowpositiveornegativehumansfeelabouttheirinteractionswiththeseparticularrobotsanditmaytranslatetoothertypesofrobots,thoughmoreresearchwouldneedtobe conductedtomakeadenitivedetermination.Thesefactorsappeartobeinter-relatedandshould beconsideredwhendevelopingrobotsystemsandevaluatingsocialhuman-robotinteractions. 6.1.2InterpretationandDiscussionoftheResultsforArousalResponsesbyRobot,OperatingMode,andRobotOrder Oneofthemostsignicantndingsinthisstudythatsupportstheprimaryresearchquestion Cantheuseofnon-facialandnon-verbalaffectiveexpressionprovideamechanismfornatural123

PAGE 140

isticsocialinteractionbetweenafunctional,appearance-constrainedrobotandthehumanwith whichitisinteracting?wasdiscoveredintheSAMarousalresults.Participantsreportedfeeling calmerwheninteractingwithbothrobotsintheemotivemodeM=4.81comparedtointeracting withbothrobotsinthestandardmodeM=3.95onascaleof0-excitedto8-calm.Thisindicates thatthemovements,posture,orientation,andilluminatedcolorcanmakeadifferencetosocialhuman-robotinteractionwhenattemptingtoelicitacalmingresponsefromhumanstowardtherobots withwhichtheymayinteract. TheresultsfortheSAMarousalresponsesalsoindicatedthattherewasasignicantmaineffectforrobotorder.Asevidencedbytheresults,participantswerecalmeroveralltobothrobots whentheyinteractedwiththePackbotrstregardlessofoperatingmode.Thisreinforcestheidea thathumanscalibratetheirresponsestorobotsbasedontheirrstinteraction.IfthePackbotis viewedrsttheoverallresponsestobothrobotswasM=4.94,whichwassignicantlycalmer comparedtotheresponsestobothrobotswhentheInuktunwasviewedrstM=3.82. Onepurposeofthisstudywastodetermineiftherewasabetterwaytooperaterobotsusedin urbansearchandrescueUS&Roperationstokeepvictimscalmeruntilassistancecouldarrive toextractthemfromthedisastersite.Theresultssupportthatappropriaterobotmovements,posture,orientation,andilluminatedcolorwillaidinkeepingvictimscalmeruntilhelpcanarrivein comparisontohowrobotshavebeentypicallyoperatedinUS&Rtrainingexercises. 6.2DiscussionofthePsychophysiologyResults Thepsychophysiologyresultswerelessforthcominginrelevancetotheprimaryresearch questionasameasurementofarousalinthisrobot-humanstudy.Thereweresomesmalleffects thatwerediscovered,butfurtherstudiesneedtobeconductedandmoredetailedanalysesare neededtodetermineifthereareanyadditionalsignicantresponsesmeasuredthroughpsychophysiology.ThediscussionwillcoverresultsobtainedfromtheheartrateHR,respirationamplitudeRA,respirationrateRR,andskinconductancelevelSCLmeandifferences. 124

PAGE 141

6.2.1InterpretationandDiscussionoftheResultsforHeartRateMeanDifferencesbyRobot,OperatingMode,andRobotOrder Therewasastatisticallysignicanttwo-wayinteractioneffectdiscoveredforHRmeandifferencebyrobotandrobotorder.Theresultsindicatethatwhicheverrobotisviewedrst,theheart ratewilldisplayanincrease,indicatingarousal.Aspartofthisinteractioneffect,aninteresting ndingisthatthereisasignicantdropinheartrateforthesecondinteractionregardlessofrobot andorderInuktunsecondM=-3.18andPackbotsecondM=-3.06.Thedropisverysimilarfor bothrobots.Thisresultwouldindicatethatparticipantsmayhavehabituatedtothesecondrobot interactionandtheimpactoftheinteractionwasnolongersignicant.Itappearsthatparticipants becamemorecomfortablewiththesecondrobotfromaphysiologyperspectiveregardlessofthe particularrobot.Thisresultledtoperforminganothersetofanalysesforpsychophysiologymeasurestakingintoconsiderationonlyparticipants'responsestotherstrobot.Unfortunately,there werenostatisticallysignicantndingsdiscoveredforHRmeandifferencefromthisanalysis. TherewasatrendthatappearedthatdidapproachsignicanceforHRmeandifferencebyoperatingmode,inwhichtheheartratemeandifferenceforparticipantsinteractingwiththerobotinthe standardmodeM=.45wasslightlyhigherthanparticipantswhointeractedwiththerobotinthe emotivemodeM=.03. 6.2.2InterpretationandDiscussionoftheResultsforRespirationAmplitudeMeanDifferencesbyRobot,OperatingMode,andRobotOrder TherewerenostatisticallysignicantresultsuncoveredforrespirationamplitudeRAmean difference.Fromtheseresultsitwouldappearthattherewasnosignicantdifferenceinhow deeplyparticipantswerebreathingduringtherobotinteractions.Itwouldbeexpectedthatduringhigharousalparticipantswouldtendtobreathemoreshallowly;howeverthiswasnotobserved inthisstudy. 6.2.3InterpretationandDiscussionoftheResultsforRespirationRateMeanDifferences byRobot,OperatingMode,andRobotOrder Themostsignicantndingsforpsychophysiologywerediscoveredintherespirationrate RRmeandifferences.TherewasanoperatingmodebyrobotordereffectforRRmeandiffer125

PAGE 142

encethatwasstatisticallysignicant.IftheInuktunwasviewedrstinthestandardmodethere wasanincreaseintheRRmeandifferenceM=.97indicatingapproximatelyonebreathper minuteincreaseovertherespirationrateobservedduringthesecondrestingperiodcomparedto theRRmeandifferenceoftheInuktunviewedrstintheemotivemodeM=.00,whichindicatesnochangeinbreathsperminuteovertheirsecondrestingperiod.Thiswouldindicatethat participantswerecalmertotheInuktunintheemotivemodecomparedtothestandardmodewhen theInuktunwasviewedrst.TherewasnosignicantdifferenceintherespirationratemeandifferencewhenthePackbotisviewedrstregardlessofoperatingmode.Participantsappearedto onlyexhibitdifferentphysiologicalresponsepatternstotheInuktuninthedifferentoperating modesandnotthePackbot.Furtherstudiesneedtobeconductedtodetermineifthisisaconsistentpattern. Additionally,therewasastatisticallysignicanttwo-wayinteractioneffectobservedforRR meandifferencebyrobot,androbotorder.Theseresultsforthisinteractionwereconfusing.Regardlessoftheoperatingmodeorrobot,whicheverrobotwasviewedsecond,therewasanincreaseinrespirationrate,withM=.68whenInuktunwasviewedsecondandM=.60whenthe Packbotwasviewedsecond.Alternatively,therewasalowerrespirationrateforwhicheverrobot wasviewedrstduringtheinteractionsM=.29InuktunandM=.19Packbot.Onepossibleexplanationforthistrendwouldbethatparticipantswerehabituatingtothesecondrobotinteraction, andbecameirritatedwithhavingtoparticipateinanotherinteractioncausinganincreaseinthe respirationrate.Furtherresearchwouldneedtobeconductedtodetermineifthispatternofphysiologicalresponsesisconsistent. Furtheranalysiswasconductedevaluatingrespirationratemeandifferenceforonlytherst robotinteractiontodetermineifanyadditionalstatisticallysignicantresultswerediscovered. Therewasatwo-wayinteractionbetweenrobotandoperatingmoderevealed.Participantswho experiencedtheInuktunintheemotivemodesawaslightdecreaseinrespirationratesovertheir respirationratesduringthesecondrestingperiodM=-.19comparedtoparticipantswhoexperiencedtheInuktuninthestandardmodesawanincreaseinrespirationratesduringtheirinteractionoverthesecondrestingperiodM=.77.Thoughthedifferenceisonlyaboutonebreathper minute,itisstatisticallysignicant.Thistrendwouldshowthatparticipantswhoexperiencedthe Inuktunintheemotivemodemayhavebeenslightlycalmerthantheparticipantswhoviewedthe 126

PAGE 143

Inuktuninthestandardmode.Unfortunatelythereisnostatisticallysignicantdifferenceinthe respirationratesforthoseparticipantswhoexperiencedthePackbotfortheirrstinteractionregardlessofoperatingmodeM=.25emotivemodeandM=.13instandardmode.Thereisno denitiveexplanationforthistrendatthistime,andfurtherstudieswouldberequiredtodetermine ifthisisacommontrend. 6.2.4InterpretationandDiscussionoftheResultsforSkinConductanceLevelMeanDifferencesbyRobot,OperatingMode,andRobotOrder TherewerenostatisticallysignicantresultsuncoveredforskinconductancelevelSCL meandifference.Fromtheseresultsitwouldappearthattherewasnosignicantdifferenceinparticipants'sweatingresponseduringtherobotinteractions.Itwouldbeexpectedthatduringhigh arousalthatparticipantswouldtendtohaveanincreaseinsweatproductionontheirhandsand feet;howeverthiswasnotobservedinthisstudy. 6.3DiscussionoftheCorrelationAnalysisfortheSelf-AssessmentManikinandPsychophysiologyResults AlthoughtherewerethreestatisticallysignicantcorrelationsdiscoveredbetweentheSAM andpsychophysiologydata,onlyoneofthecorrelationsappearstobemeaningful.ThestatisticallysignicantcorrelationforSAMValence-RecodedPackbotM=5.96waspositivelycorrelatedwiththeHRmeandifferenceforPackbotM=-1.23.ThiscorrelationsuggeststhatifparticipantsviewthePackbotmorepositivelytheyexperienceanoveralldecreaseintheirheartrateto thePackbotwithoutconsideringoperatingmodeorrobotorder.OverallthereisnotastrongcorrelationobservedbetweentheresponsesreportedbyparticipantstotheSAMassessmentsandtheir physiologicalresponsestotherobots.Moreresearchneedstobeperformedintheuseofpsychophysiologymeasuresinthehuman-robotinteractiondomain. 6.4DiscussionofFactorsthatmayhaveImpactedtheResults Therewerethreeprimaryfactorsthatmayhaveinuencedtheresultsofthisstudy.Therst isthatparticipantsappearedtofeeltoosafeinthesimulateddisastersiteduringtheirinteractions 127

PAGE 144

withtherobots.Thesecondfactoristhatparticipantswereplacedinthepronepositionduring theirinteractionswiththerobots.ThethirdfactorwasthereleaseoftheDisneymovieWALL-E justpriortothestartofthestudy.Itissuspectedthatthesethreefactorshadastronginuenceon howparticipantsrespondedtotherobotsduringthetwointeractions. 6.4.1InuenceofParticipantsFeelingTooSafe Psychophysiologymeasureslevelsofarousalandcanonlydetectarousalwhenitisgenuinely present.Arousalresponsesthebodyfeelsarenotsomethingthatcanbesimulatedorconsciously manipulated.Thatisoneofthejusticationsforusingpsychophysiologymeasurementsinresearchstudiesbecauseparticipantscannotmanipulatethesemeasurements.Theproblemwiththis studywasthatitappearedthatmanyparticipantswerefeelingtoosafeinthesimulateddisastersiteduringtherobotinteractions.Althoughthesitewasmadetolookasrealisticaspossible, safetyfactorswereputinplacetomakesureparticipantswerenotharmedincaseofaproblem withtherobotsorifrubbleshouldbecomedislodged.Participantswereplacedinaverysturdy conned-spaceboxthathadawoodpanelatthebottomedgeoftheboxandwiremeshonthe frontsideoftheconned-spaceboxtopreventtherobotsfrommakingcontactwithparticipants. Themeshalsoassistedwithkeepingparticipantssafeifrubbleshouldbecomelooseandfallfrom therubblepile.Additionally,theconned-spaceboxwasbuilttomeetthe95thpercentileformale height,whichresultedintheboxnotfeelingasconnedasitwouldbeinarealdisasterexperience.Participantswereplacedonaclosedcellfoammatinsidetheconned-spaceboxwhich mayhavebeentoocomfortabletosimulatewhatmightbeexperiencedinarealdisaster.From thevideoobservations,itwouldappearthemanyoftheparticipantswereveryrelaxedduringthe interactionswhichwouldhaveasignicantinuenceontheirphysiologicalresponses.Therewere severalparticipantswhodidshowvisiblereactionstotherobots;howeverthoseparticipantsmay nothavebeenalargeenoughportionoftheoverallsamplesizetohaveasignicantimpactonthe results.Analysisofthevideodatawasnotconductedaspartofthisdissertation. Aninterestingphenomenonwasobservedduringthepreparationandtheactualperformance ofthestudy.Inthedevelopmentphaseofthestudy,approximatelyeighttotenactualroboticsstudentsandaroboticsfacultymemberwereusedtotesttheeffectsoftheconned-spacesiteandthe interactionswiththerobots.Theresponsesfromtheroboticsstudentsandroboticsfacultymember 128

PAGE 145

wereactuallystrongerthanthereactionsofmanyoftheparticipantsinthestudy.Thisobservation wassurprising.Apossiblereasonforthisdifferenceinresponseswasthattheroboticsstudents andtheroboticsfacultymemberwerefamiliarwithboththeInuktunandPackbotrobots.They hadanunderstandingthatsituationscanoccurwhererobotscanexhibitunexpectedbehaviorsdue tooperatorerrors,programming,and/orsystemproblems.Theroboticsstudentsandtherobotics facultymemberwerealsoawareofthepoweroftheserobotsandthefactthatifsomethingwould happentogowrong,thattheserobotscouldcausephysicalharm.Thisisespeciallytrueofthe Packbotrobotduetoitssize,weight,andpower.Thestudyparticipantswouldnothavethistype ofknowledgeandautomaticallyassumewhentheyparticipateinastudythattheirsafetyisinsuredbecauseoftheInstitutionalReviewBoardIRBprocessrequiredbeforeanyhumanstudy canbeperformed.Therefore,participantsdidnotnecessarilyfeelastheymightfeelinanactual disastersituation,whichlikelyinuencedtheirpsychophysiologyresponses.Itappearsthatthe participantsofthestudyhadaninnateleveloftrustintherobots;howeverfurtherexplorationof thispossiblephenomenonisrequired.Suggestedimprovementstoovercometheseissueswillbe providedlaterinthischapterSeeSection6.5. 6.4.2TheEffectofParticipantsPlacedinthePronePosition Anotherpossiblefactorthatmayhaveinuencedthephysiologicalresponsesoftheparticipantswashavingthemplacedintheproneposition.Intypicalpsychophysiologicalstudies,participantsareplacedinaseatedpositionandusuallyaregivencomplicatedmentaltaskstoperform. Thereisnodataontheeffectsofhavingparticipantsplacedinthepronepositioninpsychophysiologicalstudies;thereforeitisdifculttoknowifthisisanappropriateplacementforobtaining accuratepsychophysiologicalmeasurements.Additionally,participantswerenotgivenanymentaltaskstoperformandwereinanobservationalmodeofinteractionwhentherecordingswere obtained.Theimpactofparticipantplacementonphysiologicalresponsesrequiresfurtherexploration. 6.4.3TheWALL-EEffect Anotherfactorthatmayhaveinuencedparticipants'responseswasthereleaseoftheDisney lmWALL-Ejustpriortothestartofthisresearchstudy.Thisanimatedmoviewasastoryabout 129

PAGE 146

afunctionaltrashcompactorrobot.TheWALL-Erobothadtriangularshapedtracksandaboxlikeappearance.Itdidhaveexpressiveeyesthatappearedasaheadontopoftherobotbody.The WALL-ErobothasfeaturesthataresimilartotheInuktunrobote.g.,triangularshapedtracks, boxstylebody,polymorphicorshape-changingfeatures,....AnimageoftheInuktunrobotis displayedinFigure37.ImagesoftheWALL-Erobotcanbeviewedatthewebsite: http://adisney.go.com/disneyvideos/animatedlms/wall-e/. Figure37.AngledviewoftheInuktunExtreme-VGTVrobot. TheInuktun,especiallywhenoperatedinthestandardmodesoundedverysimilartothe WALL-Erobotmovingoverrubbleorgravel.Thehardplastictracksmadeasimilarclackety clacksoundwhenmovingacrossthehardsurfaceoorsinthesimulateddisastersite.Additionally,theInuktunintheemotivemodemaderapid,erraticmovementswhichcouldbecompared tothemovementsoftheWALL-Erobot.Thiswasnotintentionalinthedesign;howeveritdidreectasimilaritytothemovierobot.Thedesignofthestandardmovementsweredevelopedfrom 130

PAGE 147

observationsofrobotoperatorsrunningrobotsinsearchandrescuetrainingexercises.ItwascoincidentalthatthemovementscloselyresembledthemovementsoftheWALL-Erobotinthemovie. Approximately25%oftheparticipants,especiallythosewhointeractedwiththeInuktunrobotoperatedinthestandardmodecommentedonhowcutetherobotlookedandthatitremindedthem ofWALL-E.Hadtherebeenarealizationthatthiswouldbeanissue,thenscreeningwouldhave beendonetodetermineifparticipantshadviewedthemoviepriortotheirinteractions;howeverit isexpectedthatthiswouldhavedramaticallyreducedtheparticipantpoolanditwasnotadvisable tobeginscreeningoncethestudywasunderway.Itissuspectedthatparticipantsdidnotreactas stronglytotheInuktunrobotinthestandardmodebecauseoftheirassociationwiththeWALLErobot.InpriorinteractionswiththeInuktunrobotwhenitwasapproachingatfacelevelwitha personlayinginthepronepositionontheground,theInuktunwasdescribedatcreepyinappearance.ThereactiontotheInuktuninthisresearchstudywasastrongdeparturefromthispreviousreactioninmanycases.ItdoesappearthattheWALL-Emoviemayhavebeenasignicant factortheresponsesofparticipantstotheInuktunrobot. Thisfactordoesbringupaninterestingquestion.Wouldvictimsbecalmeriftheycouldassociatethesearchandrescuerobotwitharobotthatisknowntothem,suchasWALL-EorR2D2? Shouldthesetypesofrobotsbedesignedtoresembleknownfriendlyrobots?Cantheybedesignedandbuilttoappearlikeaknownrobotandstillperformtheirfunctionalduties?Further researchneedstobeperformedtodeterminetheanswerstothis,butitdoesappearfromthisstudy thatitcouldmakeadifferenceinvictimresponses. 6.5FutureStudyImprovements Ifthisstudyweretobeconductedagainatafuturedatethereareseveralimprovementsthat mightmaketheresultsandtheirinterpretationmoreevident.Theseimprovementswouldbemade intheareasofthestudyandsitedesign.Itisexpectedthatmakingthesesuggestedimprovements wouldmakeiteasiertoidentifywhichofthestudyfactorshaveasignicantimpactontheresults, andimproveparticipantresponsesbyhavinganevenmorerealisticenvironment. 131

PAGE 148

6.5.1SuggestedStudyDesignImprovements Afutureresearchstudywouldbeimprovedbylimitingthenumberoffactorsbeingevaluated. Inthecurrentstudy,therewasawithin-subjectsfactorofrobot,withtwolevelsInuktunandPackbot.Byselectingonlyonetypeofrobottobeusedinthestudyitwouldlimitthenumberoffactorsevaluatedandmakeiteasiertoevaluatetheresults.Theuseofonlyonerobotwouldremove theinuenceofrobotandrobotorderandwouldisolatetheinuenceofoperatingmode,which wastheprimaryfactorofinterest.Thestudywasoriginallyconductedwithtworobotstodetermineiftheresultswouldextendtomorethanonerobotinsteadofbeingrobotspecic;howeverin thiscase,itseemstohavecomplicatedtheanalysesandresults.Someoftheresultsdidextendto bothrobots,thoughitmadeitmuchmoredifculttoisolatetheimpactsinsomepartsoftheanalyses.Additionally,havingparticipantsinteractwithtworobotsresultedinapracticeand/orhabituationeffect.Aftertheinitialinteraction,thenoveltyappearedtodissipateandmayhaveinuenced participants'responses.Thiswasespeciallyevidentinthepsychophysiologyresults. Anotherstudydesignissueisrelatedtotherealismoftheparticipants'responses.Theresults wouldbemoreevidentifthelevelofrealismwasincreased.Thismightbeaccomplishedbyhavingparticipantsviewvideofootagefromactualdisastersituationsjustpriortotheinteractions withtherobots.Ifrealvideoimageswereusedjustpriortointeractions,itwouldinuencethe frameofmindoftheparticipantsandmakeiteasierforthemtovisualizewhatitmightbeliketo beinanactualdisastersituation.Additionally,iftherewassomewaytoinstillsomelevelofphysicaldiscomfortthatwouldbesomewhatcomparabletohowtheymightfeelinanactualdisaster situation,thentheresponseswouldbemoresimilartohowarealvictimmightrespond.Though gettingapprovalfromtheIRBforthistypeofphysicaldiscomfortmightbedifculttoobtain,it wouldbeinvaluabletothestudyandtheresultswouldbemoreapplicabletohowactualvictims wouldrespond. Fromtheknowledgegainedfromthisstudyitismoreevidentwhattypesofassessmentquestionswouldbevaluabletoaskinfuturestudies.Forexample,itwouldbeinterestingtoexplorethe areaoftrustfurtherandhowthatimpactstheresponsesofparticipantstotherobots.Thisisaresultofthedifferencesinresponsesbetweenpeopleexperiencedwithrobotsandtheparticipantsof thisstudy. 132

PAGE 149

Additionally,removingthelightbluelightingeffectfortherobotsoperatedintheemotive modewouldallowthestudyfocustobeontherobotmovements,postures,andorientation.Althoughtherewaslittlementionofthelightbluelightingeffect,itisdifculttodetermineifand/or howmuchofaninuencethelightingeffecthadontheresults.Bynotincludingthelightblue lightingeffectinafuturestudyitwouldallowresearcherstoisolatetheeffectsoftherobotmovements,postures,andorientation.Aseparatestudywouldneedtobeconductedtodeterminethe inuenceofcolorandcoloredlightingeffects.Thisisanopenareaofresearchacrossdisciplines anddoesrequirefurtherinvestigation.Theresultsfromthisstudydidnotindicatehowmuchofan impactthelightbluelightingeffectmayhavehadonparticipants'responses. 6.5.2SuggestedSiteDesignImprovements Thereareseveralsuggestionsthatmightimprovetherealismofthesimulateddisastersite, whichcouldhaveasignicantimpactonthestudyresults.Ifthestudywererepeated,therealism ofthesiteshouldbeimproved.Thecloserthatasitecanbetoareal-worldenvironmentthebetter theresponseswouldreecttheactualsituation.Thereareseveraldesignimprovementsthatcould bemadetothesimulateddisastersite.Thoseincludesoundeffects,moreconningspace,and increasingthediscomfortlevelofparticipantsinthestudy. Thestudywouldbemorerealisticandwouldmakeiteasierforparticipantstoimaginewhat theymightfeellikeinanactualdisastersituationifsoundeffectsfromanactualdisastereventhad beenplayedduringtherobotinteractions.Itwasoftendifcultfortheparticipantstoimaginehow theymightfeel,whentheycouldhearthesoundsofinstructorstalkinginthenextroom,doors openingandclosing,andstudentsinteractingintheotherpartofthelaboratory. Ideally,itwouldbemostrealistictohavethestudyconductedinanoutdoorsettinginasimulateddisastersettinglikeDisasterCity,availableatTexasA&MUniversity;howeverduetothe useofpsychophysiologymeasuresinthisstudy,itwasessentialtoconductthestudyindoorsin atemperature-controlledenvironment.Thereisatrade-offbetweenusingmultiplemeasuresof evaluation,suchaspsychophysiologymeasuresandtherealismofanoutdoorsetting.Itisunclear ifanindoor,temperature-controlledenvironmentcouldbedesignedtoresemblemorecloselyan actualdisastersetting. 133

PAGE 150

Theconned-spaceboxthatwasusedwouldneedtobedesignedtobemoreconningand lesscomfortableinfuturestudies.Themoreconnedapersonfeelsthemorelikelytheywillbe abletoputthemselvesintowhattheymightfeelinanactualdisastersituation.Theconned-space boxusedinthisstudywasjustnotconningenoughtoprovidethatlevelofrealism.Italsoprovidedwhatparticipantsmayhaveperceivedasasafespacetobeplaced;thereforeresultingin diminishedresponsestotherobotswithwhichtheywereinteracting.Thefeelingofconnement mighthavealsobeenenhancedbyplacingaheavyfeelingobjectontopofthepersoninsidethe conned-spaceboxsuchasaleadapronusedinX-rayfacilities.Theleadapronwouldbeheavy enoughthatparticipantswouldfeelasiftheywerenotabletomoveandthatwouldimprovethe realismofthesimulateddisastersite. Therealismmightalsobeimprovedbymakingthesurfaceonwhichtheparticipantisplaced lesscomfortable.Inthisstudyparticipantswereplacedonaclosedcellfoampadthathadasomewhatbumpytextureonit.Thepadwasrmbutstillprovidedparticipantswithacertainlevelof comfortfrombeingplaceddirectlyontheoororwoodsurfaceunderneath.Itwouldhaveprovidedsomeminordiscomfortifthesurfacewasmorerealisticandhadsomerocksorsmallrubble intheareainwhichtheywerelying.Thismighthaveincreasedparticipantsabilitytoimaginehow theywouldfeelinanactualdisastersettinginvokingmorerealisticornaturalresponsesthatwould bedetectablethroughtheuseofpsychophysiologymeasures.Thetrickyparttothiswouldbedeterminingifparticipantsarereactingtothisenvironmentortheactualrobot.Furtherresearchin thisareaneedstobeperformed. 6.6AppropriateStudyDesign Human-RobotInteractionHRIisarapidlyadvancingareaofresearch,andassuchthereisa growingneedforstrongexperimentaldesignsandmethodsofevaluation.Thisbringscredibility andvaliditytoscienticresearchthatinvolveshumansassubjectsasrecognizedinthepsychology andsocialscienceelds.TwoprimaryissuesobservedinHRIstudiesisthelackofsignicantsizedparticipantpoolsthatcloselyrepresentthepopulationsbeingstudiedandthelackofmultiplemethodsofassessmentusedtoobtainconvergentvalidityinHRIstudiesKiddandBreazeal, 2005,Elmes,Kantowitz,andRoedigerIII,2006,JohnsonandChristensen,2004. 134

PAGE 151

DeterminingtheappropriatesamplesizeappearstobeachallengeinhumanstudiesinHRI. Anaprioripoweranalysiscanbeconductedtoestimatethenumberofparticipantsnecessaryfora study.Apoweranalysisisastatisticalcalculationthatcanbeperformedtodeterminetheappropriatenumberofparticipantsneededforobtainingaccurateandreliableresultsbasedonthenumber ofgroupsinthestudy,alphalevel,expectedeffectsize,andacertainlevelofstatisticalpower. Therearetypicallytablesintheappendicesofmoststatisticalbooksthatwillprovidepoweranalysisvalues.Additionally,thereissoftwarefreelyavailableonlinethatwillassistwiththistypeof calculatione.g.,G*Power3http://www.psycho.uni-duesseldorf.de/abteilungen/aap/gpower3/. Itwasdiscoveredintheresearchstudyassociatedwiththisdissertationthatdeterminingthe appropriatesamplesizewasimportanttothesuccessofthestudy.Manyofthestatisticallysignicantresultsdiscoveredinthisstudyhadsmalleffectsandthoseeffectsmaynothavebeendiscoveredhadthesamplesizenotbeentheappropriatesize.Thisisthelargestknownsamplesize reportedintheliteratureforahuman-robotstudyinacontrolledenvironment.Additionally,the samplesizewaslargeenoughthatthedataforthesecondinteractioncouldberemovedandthere wasstillenoughstatisticalpowertorununivariateanalysesforthepsychophysiologicalmeasures. Thiswouldnothavebeenpossibleifapoweranalysiswasnotperformedtoestimatetheappropriate-sizedparticipantpool.TherehavebeenstudiessuchasthoseconductedatCarnegieMellon UniversitywiththeroboceptionistthathadalargenumberofparticipantsMakatchev,Lee,and Simmons,2009;howeverthesewerelongitudinalobservationalstudiesandnotcontrolledexperimentsthatmanipulatedspecicfactors. ThefocusuntilrecentlyinHRIwasonthedevelopmentofspecicroboticsystemsandapplicationsinsteadofmethodsofevaluationandmetrics.Somemethodsoftestingandevaluation havebeenadoptedand/ormodiedfromsucheldsashuman-computerinteraction,psychology, andsocialsciencesKiddandBreazeal,2005;however,themannerinwhichahumaninteracts witharobotissimilarbutnotidenticaltointeractionsbetweenahumanandacomputerorahumaninteractingwithanotherhuman.Asrobotsbecomemoreprevalentindailylife,itwillbeincreasinglyimportanttohaveaccuratemethodsofevaluatinghowhumansfeelabouttheirinteractionswithrobotsandhowtheyinterprettheactionsoftherobotsBetheletal.,2007b. Anotherresultoftheresearchstudyassociatedwiththisdissertationisthevalidationofthe arousalandvalencedimensionsoftheSelf-AssessmentManikin,anassessmenttoolvalidatedand 135

PAGE 152

usedinthepsychologycommunity.Byvalidatingthismeasurementtool,itprovidestheHRIcommunitywithanothertoolfordatacollectioninfuturehuman-robotstudies.Thiswaspossibledue tothelargesamplesize,andpre-planninginthestudydesignphase,whichallowedforthedevelopmentoftwoversionsoftheSAMassessments.Eachquestionintheassessmentwaswrittenin aslightlydifferentmannersothatitwouldelicitsimilarresponsesfromparticipants.Acorrelationanalysisrevealedpositivecorrelationsbetweenthetwoversionsoftheassessment,allowing fortheresponsestobecombinedandaveragedforeaseofanalysesandtovalidatetheSAMassessmentforuseintheHRIcommunity.ThequestionsdevelopedforthetwoversionsoftheSAM assessmentwereconsistentwiththeoriginalintentofthepsychologyversionofthisassessment; howevertheyweremodiedtobeapplicabletotheHRIdomain. ThereareveprimarymethodsofevaluationusedforhumanstudiesinHRI:self-assessments, behavioralmeasures,psychophysiologicalmeasures,interviewsstructuredandunstructured,and taskperformancemetricsKiddandBreazeal,2005,Betheletal.,2007b,Betheletal.,2007a. FromthereviewofHRIliteratureitappearsthemostcommonmethodsutilizedinHRIstudiesare self-assessment,behavioralmeasures,orinterviews.Thereislimitedresearchintheuseofpsychophysiologicalmeasuresandtaskperformancemetrics.Eachmethodhasadvantagesanddisadvantages;howeverdisadvantagescanbeovercomebyusingmorethanonemethodofevaluation KiddandBreazeal,2005,Betheletal.,2007b. Theneedformorethanonemethodofmeasurementwasevidentwhenthepsychophysiologicaldatainthisstudyshowedlittleinstatisticallysignicantresults.Hadthatbeentheonlysource ofmeasurementutilized,theoutcomesmayhavebeenverydifferent.Therearetwoothermethods ofassessmentwhichwerenotanalyzedinthisstudy.Theyareastructuredaudiorecordedinterviewandvideoobservations.Thesemethodsofassessmentwillbeanalyzedatalaterdate. ThedesignofaqualityresearchstudyforuseinHRIapplicationsthatproducesresultsthat areveriable,reliable,andreproducibleisamajorchallenge.Theuseofasinglemethodofmeasurementisnotsufcienttointerpretaccuratelytheresponsesofparticipantstoarobotwithwhich theyareinteracting.ItisclearfrompreviousstudiesconductedinHRIthatstandardsneedtobe establishedforconductingreliableandqualitystudiesinwhichmethodsofmeasurementcanbe validatedforusebytheHRIcommunity.Itisessentialtousemultiplemethodsofevaluationto establishstudyvalidity.Additionally,itisimportanttodeterminetheappropriatesamplesizenec136

PAGE 153

essarytoobtainstatisticallysignicantresults.Thiscanbeaccomplishedwithcarefulplanning andutilizingstudydesigntechniques,whicharecustomaryinthepsychologyandsocialscience communities.AsetofrecommendationsforconductinghumanstudiesinHRIispresentedinthe ConclusionsofthisdissertationasacontributionofthisresearchSeeChapter7onpage139. 6.7OpenResearchQuestionsResultingfromtheStudy TheeldofHRIisgrowingandevolving;howeverthereremainsmanyopenquestionsthat requirefurtherinvestigation.Theresultsfromthisstudyindicatethatbodymovements,postures, orientation,andpossiblyilluminatedcolorcanmakeadifferenceinnaturalistichuman-robotinteractionswhenthegoalistoelicitacalmingresponse.Furtherresearchneedstobeconducted todetermineiftheseeffectscanbeextendedtoothersocialhuman-robotinteractions.Thereare threemainquestionsthatwererevealedwhileperformingthisresearchstudy:Canilluminatedcoloredlightingeffectsbeusedeffectivelytoconveyaffectandfornaturalisticsocialhuman-robot interactions?Cantheuseofnon-verbalsounds,tones,and/ormusicbeusedasaneffectivemethod ofaffectexpressionforanaturalistichuman-robotsocialinteraction?andWhatimpactdoesthe inter-agentdistanceorproximityzonehaveonsocialhuman-robotinteraction? Someinitialresearchhasbeenconductedwiththisresearchstudyontheuseofthelightblue lightingeffectontheunder-carriageoftherobotsintheemotivemode.Thewaythestudywas designed,itwasdifculttoisolatetheinuenceofthismechanismforaffectiveexpressioninsocialhuman-robotinteraction.Acrossdisciplines,thereseemstobeverylittleresearchassociated withtheimpactofcolorandhowparticularcolorsareassociatedwithand/orelicitaparticular emotionalresponse.LightbluewasselectedforthisstudybasedonresearchconductedbyArgyleoncoloranditsassociationwithacalmingeffect.Howeveroneparticipantmentioned thathefeltthelightbluelightwascoldanduncaring,andhewouldhavepreferredawarmyellow lightingeffect.Thiswouldlendsupporttotheneedforfurtherresearchintotheeffectsofcolor onhuman-robotinteractions.Additionally,thiswouldneedtobeevaluatedfromacross-cultural perspectivesincedifferentculturesinterpretcolorsdifferently. Thereislittleresearchintheuseofdifferentsounds,tones,and/ormusicasamethodofconveyingaffectforanaturalisticsocialhuman-robotinteraction.Theprimaryexampleoftheuseof soundstoconveyaffectareseeninthemovies.Fromvariousconversationsovertheyears,itap137

PAGE 154

pearsthatwhetherpeoplehaveviewedtheStarWarsmoviesornot,theycaneasilyrecognizethe soundsandtonesmadebytherobotR2D2inthatseriesofmovies.Fromthosesoundsandtones peopleseemtobeawareofwhenR2D2wasalarmed,calm,andangry.Thisisanareaofresearch thatrequiresconsiderableinvestigation.Theremaybesituationsinwhichnon-verbalcommunicationintheformsofsounds,tones,and/ormusicmaybepreferrabletotheuseofverbalcommunication,suchastokeepvictimscalmforthemanyhoursittakestoextractthemafterarobot locatesthemorwithautisticchildrenwhomaybeoverwhelmedbydirectverbalcommunication. Thisareaofresearchdenitelyrequiresfurtherinvestigation. Therehasbeenpreliminaryresearchassociatedwiththeimpactofinter-agentdistancesor proximityonhowarobotbehaveswiththehumanwithwhichitisinteracting.Theresearchassociatedwiththisdissertationhasaddressedthisfromapreliminaryperspective,butfurtherworkis denitelyneededinthisareatoevaluatetheeffectivenessofallnon-facialandnon-verbalmethodsofevaluationbyproximityzone.Thisworkperformedaninitialevaluationoftheuseofbody movements,postures,andorientationbyproximityzoneandadditionallydeterminedthatilluminatedcolorcanbeeffectiveandusedintheintimateproximityzonewhichwasuncertainpriorto thishuman-robotstudy. 6.8Summary Asevidencedbytheresults,participantsreportedfeelingcalmertorobotsoperatedinthe emotivemode;theyviewedtheirinteractionswiththeInuktunmorepositivelycomparedtointeractionswiththePackbot;andtheywerecalmerandviewedtheirinteractionsmorepositivelyto bothrobotswhentheyinteractedwiththePackbotrstfollowedbytheInuktun.Therewerelimitedsignicantresultsassociatedwithheartrateandrespirationratemeandifferences;howeverthe otherphysiologicalresultswerenotstatisticallysignicant.Therewerethreeprimaryfactorsthat impactedtheresults.Participantsmayhavefelttoosafe,participantswereplacedinaproneposition,andparticipantsmayhavebeenfamiliarandcomfortablewiththeWALL-Erobotsincethe moviewasreleasedjustpriortothestartofthestudy.Thereisadiscussionofsuggestedstudydesignandsiteimprovements,andsomeopenresearchquestionsdiscoveredaspartofthisresearch. 138

PAGE 155

Chapter7 ConclusionsandFutureWork Theresearchconductedaspartofthisdissertationwillimpactnotonlythehuman-robotinteractionHRIcommunitybutalsotheroboticscommunityasawhole.Itisinevitablethatrobots willbecomeapartofourdailylivesinmanyforms,anditisimperativethattheHRIcommunity determinehowtheserobotswillinteractinasociallyappropriatemannerwithhumans.Thecontributionsofthisresearchfocusontheuseofnon-facialandnon-verbalmethodsofaffectiveexpressionforsocialhuman-robotinteractionasitrelatestotheurbansearchandrescuedomain. Thischapterpresentssignicantconclusionsandcontributionsresultingfromthisresearchwhich includessignicantresults,asetofprescriptiverecommendationsforusingnon-facialandnonverbalaffectiveexpressionbyproximityzone,andrecommendationsfordesigning,planning,and executinghuman-robotstudies.Thereisabriefdiscussionpresentedofimmediateandlong-term futureresearchgoals. 7.1SignicantConclusionsandContributions Thissectionpresentssignicantresultsandcontributionsresultingfromthisresearch.The resultsdiscussedaredirectlyrelatedtothesignicanceofnon-facialandnon-verbalaffectiveexpressionforuseinsocialhuman-robotinteraction.Signicantcontributionsofthisworkincludes asetofprescriptiverecommendationsfortheappropriateuseofnon-facialandnon-verbalaffectiveexpressionbyproximityzone;validationofthevalenceandarousaldimensionsoftheSelfAssessmentManikinforuseinhuman-robotinteractionstudies;andasetofpracticalrecommendationsfordesigning,planning,andexecutinghuman-robotinteractionstudies.Additionally,there isanarchivaldatasetavailableforfurtherexploration. 139

PAGE 156

7.1.1ImportanceofNon-FacialandNon-VerbalAffectiveExpressionforSocialHumanRobotInteraction Participantswerecalmertorobotsthatwereprogrammedtooperateintheemotivemodecomparedtothoseprogrammedtooperateinastandardmode.Thisresultimpliesthatnon-facialand non-verbalmethodsofaffectiveexpressionthroughtheuseofrobotbodymovements,posture,orientation,andcolordoeshaveanimpactonnaturalisticsocialhuman-robotinteractionsinnon-anthropomorphic,appearance-constrainedrobotswhenthegoalistoelicitacalmingresponse.The researchassociatedwiththisdissertationwastherstcomprehensive,large-scaleuseofnon-facial andnon-verbalaffectiveexpressionmethodsinappearance-constrainednon-anthropomorphicrobots.Althoughmoreresearchneedstobeperformed,theresultsinthisinitialstudydoimplythat thewayarobotmoves,theposturesitdisplays,anditsorientationcanmakeasignicantdifferenceinhowhumansrespondtotherobotwithwhichtheyareinteracting,specicallyintheurban searchandrescuedomainandinotherdomainsrequiringacalmingresponsefromhumansinteractingwithrobots.Itislessclearifthereisanimpactfromtheilluminatedlightblueneonlighting effectoncalmingparticipants.Thecoloredneonlightwasvisibletoparticipants;howeveritisnot clearifitelicitedacalmingresponse.Furtherresearchneedstobeconductedrelatedtotheimpact ofcolorandmorespecicallyilluminatedcoloronsocialhuman-robotinteractions. 7.1.2TheCalibrationEffectforFirstRobotEncounters Fromthisresearch,itisevidentthatparticipantscalibratetheirresponsesbasedontheirrst encounterwitharobot.Itisimportanttoconsiderthiswhenintegratingrobotsintodifferentenvironmentswithhumans.Inthestudy,ifparticipantsviewedorinteractedwiththerobotthatwas perceivedasmorenegative,inthiscasethePackbot,theyviewedbothrobotsmorepositivelyand werecalmertobothrobots;whereasiftheyinteractedwiththecuterobotrstInuktunthey respondedmorenegativelyandexcitedtobothrobotsbutespeciallythePackbotasevidencedby theSAMvalenceandarousalresultsRefertoChapter5onpage79andChapter6onpage121. Fromtheseresultsitappearsthatinthesearchandrescuedomainifhumansarerequiredtointeractwithtwodifferenttypesofrobotsitisbettertohavetheminteractwiththerobotthatmay beperceivedmorenegativelyrst,followedbythesecondinteractionwhichwillbeviewedmore positively. 140

PAGE 157

7.1.3PrescriptiveRecommendationsforNon-FacialandNon-VerbalAffectiveExpression BasedonProximityZone Aspartofthisresearchasetofprescriptiverecommendationsweredevelopedfordeterminingwhatmethodofnon-facialandnon-verbalaffectiveexpressionwasappropriatetousebased ontheinter-agentdistanceorproximitybetweentherobotandthehumanwithwhichitisinteracting.Theseprescriptiverecommendationsformadefactotoolbox,whichprovidesonemechanism forroboticiststoaddaffectretroactivelytoarobotthroughsoftwarechanges,reducingoreliminatingtheneedforphysicalmodicationsordesigninganewrobot.Thesemethodsofaffective expressioncanbeusedinbothnon-anthropomorphic,andanthropomorphicrobotstoaddaffective expressionthroughbodymovements,postures,orientation,illuminatedcolor,and/orsound.Itis recommendedthatfurtherinvestigativestudiesbeconductedtoconrmtherecommendationsfor illuminatedcolorandsound.Thatwillbediscussedmorefullyinthefutureworksectionlaterin thischapter. 7.1.4ValidationoftheSelf-AssessmentManikinSAMasanAssessmentToolforHumanRobotInteractionStudies TheSelf-AssessmentManikinSAMisanassessmenttoolthathasbeenvalidatedandused bythepsychologycommunityformanyyearstomeasurevalence,arousal,anddominanceresponsesfromparticipantsinhumanstudiesBradleyandLang,1994andnowasaresultofthis research,thevalenceandarousaldimensionshavebeenvalidatedasanassessmenttoolforuse intheHRIcommunity.TheSAMassessmentwasrephrasedtocorrespondtothehuman-robot interactiondomain.TwoversionsoftheSAMassessmentweredevelopedwiththeverbagemodiedslightlybetweenthetwoversionssothateachassessmentwouldelicitsimilarresponsesfor validationpurposes.Thetwoversionsoftheassessmentweredesignedtobeconsistentwiththe intentandcontentoftheoriginalSAMassessment.Acorrelationanalysiswasperformedtodetermineiftheresponsescouldbecombinedandaveragedforeaseofanalysesandforvalidation purposes.Theresultsexhibitedpositivecorrelationsbetweenthetwoversionsforeachrobotinteraction;thereforetheresponsestobothassessmentswerecombinedandaveragedforanalysis purposes.Inthehuman-robotstudyassociatedwiththisresearch,theresultsfromparticipants' responsessupportedthevalidityofthisassessmenttoolforuseintheHRIcommunity.Thedomi141

PAGE 158

nancedimensionwasnotvalidatedinthisresearchbecauseapproximately25%oftheparticipants verballyreportedthatthequestionsassociatedwiththisdimensionwereconfusing.Thequestions forthisdimensioneitherneedtoberevisedoranotherassessmenttoolisneededtoevaluatethe dominancedimension. 7.1.5RecommendationsforDesigning,Planning,andExecutingHuman-RobotInteraction Studies Planning,designing,andexecutingahumanstudyforHRIcanbechallenging;howeverwith carefulplanningmanyofthesechallengescanbeovercome.Therearetwomainimprovements thatneedtobemadeinhuman-robotstudiesconductedinHRIandthosearehavinglargesamplesizestorepresentappropriatelythepopulationbeingstudied,andsothatsmalltomediumeffectscanbedeterminedwithstatisticallysignicantresults;andtheuseofmultiplemethodsof evaluationtoestablishreliableandaccurateresultsthatwillhaveconvergentvalidity.Fromtheexperiencegainedincompletingthislarge-scale,complexhuman-robotstudyassociatedwiththisresearch,recommendationsweredevelopedthatfallintothreecategories:AExperimentalDesign Recommendations,BRecommendationsforStudyExecution,andCOtherRecommendations. 7.1.5.1ExperimentalDesignRecommendations Theserecommendationsareprovidedtoassistintheplanninganddesignoflarge-scale,complexhumanstudiesinHRI.Theywillassistresearcherswiththedevelopmentofacomprehensive experimentaldesignthatshouldprovidesuccessfulstudyresults. Determinethemostappropriatetypeofstudyforthehypothesesbeinginvestigatedusingeitherawithin-subjects,between-subjects,ormixed-modeldesign.Thewithin-subjectsdesign increasespowerandreduceserrorvariance;however,apotentialproblemishabituationand practiceeffects.Themixed-modeldesignhassomefactorsthatarewithin-subjectsandsome arebetween-subjects.Animportantissuetoconsideristhatinmixedmodeldesigns;participantsshouldberandomlyassignedtothebetween-subjectsfactorsandthattheorderorsequencethefactorsarepresentedarecounterbalanced. 142

PAGE 159

Performanaprioripoweranalysistoestimatetheappropriatenumberofparticipantsrequired forthestudyinordertohaveabetteropportunitytoobtainstatisticallysignicantresultsthat arevalid,reliable,andaccurate.Thiscanbeaccomplishedthroughpoweranalysistablesor freelyavailablesoftware.Anaprioripoweranalysisisbasedonthenumberofgroupsinthe study,theeffectsize,thealphalevel,andthedesiredstatisticalpower.Itisrecommendedto addafewmoreparticipantstotheestimatednumbertoaccountforproblemswithdata,participantcancellationsthatcannotberescheduled,andparticipantswhodonotcompletethe study. Determinethebestmethodsofevaluationforthehypothesesbeinginvestigated;however,it isrecommendedthatatleasttwoormoremethodsshouldbeutilizedinordertoobtainconvergentvalidityinthestudy.Theresultsfromastudywithmultiplemethodsofevaluation areviewedasmorereliableandaccurate.Additionally,byincorporatingmultiplemethodsof evaluationitwillovercometheinheritproblemsfoundwitheachmethodofevaluation.No onemethodissufcientformeasuringaccuratelyparticipants'responses. Designanenvironmentorstudyspacethatcloselyreectsthereal-worldthatisbeingtested formorenaturalparticipantresponses.Ifareal-worldenvironmentisnotpossiblefortesting, thenensurethetestenvironmentisashighdelityaspossiblewiththeuseofprops,lighting, and/orsoundeffects.ThiswillneedtobewithinIRBallowablestandardsofacceptableparticipanttreatment.Whenconductingpsychophysiologystudiesusingskinconductanceresponse, itisessentialtohaveatemperature-controlledenvironment. Wheneverpossibleperformthestudywithmorethanonetypeofrobot.Thiswillhelpwith generalizingresultsacrossdifferentrobottypesversusresultsthatarespecictoaparticular robot.TheresultswillbemoreusefultotheHRIandroboticscommunityifmorethanone robottypeisutilized. TheserecommendationsshouldassistHRIresearchersindeterminingthethebestapproachto designingacomprehensivestudyandincreasetheprobabilityofobtainingstatisticallysignicant studyresults.Forstudysuccessitisessentialtohaveanappropriatesamplesizeandtouseatleast twoormoremethodsofevaluation. 143

PAGE 160

7.1.5.2RecommendationsforStudyExecution ThefollowingrecommendationsareprovidedtoHRIresearcherstofacilitatetheexecution oftheexperimentaldesignforthestudy.Theserecommendationswillassistinrevealingpotentialawsintheexperimentaldesignsothatcorrectionscanbeimplementedresultinginasmooth running,efcientstudy.However,evenwiththebestdesignsyoucanexpectequipmentfailures, participantsandassistantsarrivinglateornotatall,andotherpitfalls.Thekeyistohavecontingencyplansinplaceandanticipateworstcasescenariosbecausetheydooccur. Developawrittenstudyprotocolofallinstructions,assessmentswithordering,participant tasksinorderofexecution,timingofevents,coordinationofdatacollection,andanyassociatedactivities.ThisdocumentwillbeusedwhenpreparingIRBpaperwork,creatinginstructionsforparticipants,andpreparinginformedconsentdocuments. Performmultipletestrunsoftheplannedstudyprotocoluntilallglitchesandproblemshave beendiscoveredandresolvedandthereisasmoothrunningsysteminplace.Developinga checklistsystemfortheresearcherandresearchassistantsisrecommendedoncethestudy executionisnalized. Makesurethatthereisredundancyinallequipmentthatisrequiredforthestudybecausefailuresarecommon.Testandverifythatbackupequipmentisworkingproperlysothatfaulty equipmentcanbereplacedimmediately. Alwaysprepareforandexpectequipmentfailures,participantsand/orresearchassistantsnot arrivingattheirdesignatedtimes,orjustaboutanythingelsethatmightunexpectedlyoccur. Havingcontingencyplansinplacesuchasredundantequipment,doubleschedulingvolunteerassistantsorestablishanon-calllist,anddevelopanon-calllistforparticipantswhoare availableonshortnotice. Alwaysallowtimeforstudydelays,participantsarrivinglate,and/orequipmentfailuresthat maycausethecancellationofparticipantsanddelayoftheoverallstudy. 144

PAGE 161

7.1.5.3OtherRecommendations Thefollowingarerecommendationsfortherecruitmentofparticipantsandvolunteerresearch assistantsandarebasedonourexperiencesandmaynotapplytoallresearchersanduniversities. TheywereexcellentresourcesforourparticularstudyandweareawareofsimilarprogramsavailableatmanyUnitedStatesandEuropeanuniversitiesandinstitutions. AgoodsourceofrecruitingqualityvolunteerresearchassistantsisfromanHonorsCollegeor Programifavailableattheuniversityorinstitution.Additionally,pre-medicalandpsychology studentsoftenhavevolunteerhoursrequirementsandwillvolunteertheirtimetofulllthese requirements. Recruitingparticipantstoreachtheestimatednumberrequiredbythepoweranalysisrequires multiplemethodsofcontactsuchasyerspostedacrosscampus;wordofmouthtofriends, family,andassociates;offeringincentivessuchasdoorprizes,payforparticipation,andextra creditincoursesforparticipation;signingupforresearchstudyparticipantpoolsthroughthe psychologydepartmentoncampusifoffered. Conductinghumanstudiescanbechallengingandalsoveryrewarding.Carefulplanningand designcanmaketheexperiencemorepositiveandsuccessful.Followingtheaboverecommendationsshouldimprovethechancesofhavingasuccessfulstudywithaccurateandreliablestatisticallysignicantresults.Throughtheuseofappropriatesamplesizesandmultiplemethodsof evaluation,convergentvalidityshouldbeobtained.ReadersaredirectedtoElmes,Kantowitz,and RoedigerIII,2006,JohnsonandChristensen,2004,Stevens,1999forfurtherreference. 7.1.6ArchivalDataSet Anothersignicantcontributionofthisresearchisanextensivearchivaldataset.Thedataconsistsofmultipleself-assessmentsthatwerenotanalyzedaspartofthisdissertation,psychophysiologydata,videodataofeachparticipantfromfourdifferentcameraangles,andaudiorecorded structuredinterviewles.Thisdatacanbeanalyzedfromdifferentperspectivestoevaluatevarious researchquestions. 145

PAGE 162

7.2FutureWork Thereareseveralimmediateandlong-termgoalsassociatedwithfutureworkforthisresearch. Theimmediategoalsfocusonfurtheranalysesofthedatacollectedfromthisresearchstudy. Long-termgoalsaddressitemsthatrequirefurtherinvestigationoverthenextveyearsorlonger. 7.2.1ImmediateFutureResearchGoals Intheimmediatefuture,furtheranalysesneedstobeconductedwiththecurrentdatacollected.Theseanalysesincludethecodingofvideo-recordedbehavioraldata,transcriptionand analysisoftheaudio-recordedstructuredinterview,andanalysisofthepsychophysiologydatafor heartratevariability,respiratorysinusarrhythmiaRSA,interbeatintervalIBI,andalsoskin conductanceresponseSCR. Behavioralcodingwillbeconductedbyindependentratersatalatertime.CodingwillbeperformedusingtheNolduscodingsystemforpositive,neutral,andnegativeresponsesoftheparticipantstotherobots.Positivecodingwouldincludeanorientationtowardtherobot,smiling expression,relaxedbodypostureormotions.Negativecodingwillincludefrowning,inching movements,movementsthatretreatawayfromtherobot,andtighteningofmuscles.Neutralcodingincludesnofacialorbodyexpressionspresent.Theaudiorecordedstructuredinterviewdata willneedtobeaccuratelytranscribedandanalyzedthroughqualitativeevaluationmethods. Heartratereactivityneedstobecalculatedusingthedifferencebetweenthebaselineand evokedECGreadings.TheinterbeatintervalIBIserieswillbederivedfromtheECGandwill behandcorrectedforartifactsandectopicbeatsusingtheQRSToolandCMETsoftwareavailable athttp://www.psychozz.orgAllen,2002,Allen,Chambers,andTowers,2007. 7.2.2Long-TermFutureResearchGoals Long-termfutureresearchneedstoexpanduponthisseminalworkonnon-facialandnonverbalaffectiveexpressionwithappearance-constrainednon-anthropomorphicrobotsinthree primarydirections.First,itwouldbebenecialtodevelopnon-verbalexpressions,tones,and/or musicthatwillbeuniquetoeachparticulartypeofrobote.g.,snakestylerobots,leggedrobots, andevenunmannedaerialvehiclesthatmightbeusedforcrowdcontrol.Thiswillalsoprovide 146

PAGE 163

anothermethodofsocialhuman-robotinteraction.Additionally,human-robotstudiesneedtobe performedtoconrmtheprescriptiverecommendationsforsoundbyproximity. Second,thereneedstobecontinuedexplorationofmovements,postures,andorientationfor affectiveexpressionusingdifferenttypesofrobotsandforuseindifferentdomainsandapplications.Additionally,itwouldbeusefultodevelopataxonomyofmovementsandposturesthatcan beappliedtoparticularaffectiveexpressions,similartothetaxonomyoffacialexpressionsdevelopedbyBreazealforherrobotKismet. Third,thereneedstobeanin-depthstudyoftheuseofcolortoexpressaffectinrobotsand agentsandtoelicitresponsesfromthehumanswithwhichtheyareinteracting.Additionalhuman-robotstudiesneedtobeconductedtoconrmtheprescriptiverecommendationsfortheuse ofcolorbyproximity.Asappearance-constrainednon-anthropomorphicrobotsbecomeanintegralpartofoursocietyitisimperativetoexplorehowtheywillimpactandinteractwithhumans. Eventhoughthisresearchwasfocusedonappearance-constrainednon-anthropomorphicrobots, theconceptswillbeapplicabletoanthropomorphicrobotsandotheragentsaswell. Alloftheaforementionedresearchdevelopmentswillneedtobetestedeitherinahighdelity labsettingorineldenvironmentstosimulatereal-worldinteractionsandresponses.Thiswill requireresearcherstodesignandconductcomplex,comprehensive,andlarge-scalehuman-robot andhuman-computerinteractionstudiestotestnewdevelopmentsandmeasuretheresponsesof thehumansthatwillbeinteractingwiththerobotsandanynewtechnology.Thesetypesofstudies areessentialtobringcredibilitytotheresearchareasofHRI,HCI,androbotics. 7.3Summary Themostsignicantresultofthisresearchisthatparticipantsreportedfeelingcalmerwhen interactingwithrobotsprogrammedtooperateinanemotivemodethroughtheuseofappropriate bodymovements,postures,orientation,andilluminatedlightblueneonlighting.Anothernding ofimportanceisthatparticipantsappeartocalibratetheirresponsestorobotsbasedontheirrst robotencounter.Aspartofthisresearchasetofprescriptiverecommendationsweredeveloped fordeterminingtheappropriateuseofnon-facialandnon-verbalaffectiveexpressionbyproximityzone.Theserecommendationscanbeappliedretroactivelythroughsoftwareimplementation reducingoreliminatingtheneedforphysicalmodicationofarobotordesigninganewrobot. 147

PAGE 164

AdditionallythisworkvalidatedthevalenceandarousaldimensionsoftheSelf-AssessmentManikinforuseasameasurementtoolforfuturehuman-robotinteractionHRIstudies.Apracticalset ofrecommendationsispresentedfordesigning,planning,andexecutinghuman-robotstudiesin HRI.Thereisadiscussionofimmediateandlong-termresearchgoalsaspartoffuturework. 148

PAGE 165

References Argyle,M.1975. BodilyCommunication .London:Methuen. Balkwill,L.;Thompson,W.F.;andMatsunaga,R.I.E.2004.Recognitionofemotionin japanese,western,andhindustanimusicbyjapaneselisteners. JapanesePsychologicalResearch 46:337. Bartneck,C.2000. AffectiveExpressionsofMachines .Mastersthesis,EindhovenUniversityof Technology. Bethel,C.L.,andMurphy,R.R.2006a.Affectiveexpressioninappearance-constrainedrobots.In 1stACMSIGCHI/SIGARTConferenceonHuman-RobotInteractionHRI2006 ,327.Salt LakeCity,UT:ACMPress. Bethel,C.L.,andMurphy,R.R.2006b.Auditoryandothernon-verbalexpressionsofaffect forrobots.In 2006AAAIFallSymposiumSeries,AurallyInformedPerformance:Integrating MachineListeningandAuditoryPresentationinRoboticSystems .Washington,DC:AAAI. Bethel,C.L.,andMurphy,R.R.2007.Non-facial/non-verbalmethodsofaffectiveexpression asappliedtorobot-assistedvictimassessment.In 2ndACM/IEEEInternationalConferenceon Human-RobotInteractionHRI2007 ,287.Arlington,VA:ACMPress. Bethel,C.L.,andMurphy,R.R.2008.Surveyofnon-facial/non-verbalaffectiveexpressionsfor appearance-constrainedrobots. IEEETransactionsonSystems,Man,Cybernetics,PartC 38. Bethel,C.L.;Salomon,K.;Burke,J.L.;andMurphy,R.R.2007a.Psychophysiologicalexperimentaldesignforuseinhuman-robotinteractionstudies.In The2007InternationalSymposium onCollaborativeTechnologiesandSystemsCTS2007 .Orlando,FL:IEEE. 149

PAGE 166

Bethel,C.L.;Salomon,K.;Murphy,R.R.;andBurke,J.L.2007b.Surveyofpsychophysiology measurementsappliedtohuman-robotinteraction.In 16thIEEEInternationalSymposiumon RobotandHumanInteractiveCommunication Bickmore,T.W.,andPicard,R.W.2004.Towardscaringmachines.InDykstra-Erickson,E., andTscheligi,M.,eds., ConferenceonHumanFactorsinComputingSystemsCHI'04Extended AbstractsonHumanfactorsinComputingSystems ,1489.Vienna,Austria:ACMPress, NewYork,NY,USA. Blake,R.L.J.,andMcKay,D.A.1986.Asingle-itemmeasureofsocialsupportsasapredictorof morbidity. JournalofFamilyPractice 22:82. Bradley,M.M.,andLang,P.J.1994.Measuringemotion:Theself-assessmentmanikinandthe semanticdifferential. JournalofBehavioralTherapyandExperimentalPsychiatry 25:49. Breazeal,C.,andFitzpatrick,P.2000.Thatcertainlook:Socialamplicationofanimatevision.In Dautenhahn,K.,ed., AAAI2000FallSymposium ,volumeTechnicalReportFS-00-04,1822. AAAI,MenloPark,CA. Breazeal,C.L.2002. Designingsociablerobots .Intelligentrobotsandautonomousagents. Cambridge,Mass.:MITPress. Breazeal,C.2004.Socialinteractionsinhri:therobotview. IEEETransactionsonSystems,Man andCybernetics,PartC 34:181. Brownley,K.A.;Hurwitz,B.E.;andSchneiderman,N.2000.Cardiovascularpsychophysiology. InCacioppo,J.T.;Tassinary,L.G.;andBerntson,G.G.,eds., HandbookofPsychophysiology USA:CambridgeUniversityPress,2ndedition.224. Bruce,A.;Nourbakhsh,I.;andSimmons,R.2002.Theroleofexpressivenessandattentionin human-robotinteraction.In IEEEInternationalConferenceonRoboticsandAutomation,ICRA '02 ,volume4,4138. Buck,R.1984. Thecommunicationofemotion .NewYork:GuilfordPress. Bull,P.1987. PostureandGesture .OxfordOxfordshire;NewYork:PergamonPress,1stedition. 150

PAGE 167

Burke,J.L.;Murphy,R.R.;Riddle,D.R.;andFincannon,T.2004.Taskperformancemetrics inhuman-robotinteraction:Takingasystemsapproach.In PerformanceMetricsforIntelligent Systems Buysse,D.J.;Reynolds,C.F.;Monk,T.H.;Berman,S.R.;andKupfer,D.J.1989.Pittsburgh sleepqualityindex:Anewinstrumentforpsychiatricpracticeandresearch. PsychiatryResearch 28:193. Ca namero,L.D.,andFredslund,J.2000.Howdoesitfeel?emotionalinteractionwithahumanoidlegorobot.InDautenhahn,K.,ed., AAAI2000FallSymposium-SociallyIntelligent Agents:TheHumanintheLoop ,volumeTechnicalReportFS-00-04,23.AAAI,Menlo Park,CA. Cohen,S.;Kamarck,T.;andMermelstein,R.1983.Aglobalmeasureofperceivedstress. Journal ofHealthandSocialBehavior 24:385. Cohen,J.1988. Statisticalpoweranalysisforthebehavioralsciences .Hillsdale,NJ:Lawrence EarlbaumAssociates,2ndedition. Dautenhahn,K.;Walters,M.;Woods,S.;Koay,K.L.;Nehaniv,C.L.;Sisbot,A.;Alami,R.;and Sim eon,T.2006.Howmayiserveyou?:Arobotcompanionapproachingaseatedperson inahelpingcontext.In 1stACMSIGCHI/SIGARTConferenceonHuman-RobotInteraction HRI2006 ,172.SaltLakeCity,UT:ACMPress,NewYork,NY,USA. Dautenhahn,K.;Ogden,B.;andQuick,T.2002.Fromembodiedtosociallyembeddedagentsimplicationsforinteraction-awarerobots. CognitiveSystemsResearch 3:397. Dawson,M.E.;Schell,A.M.;andFilion,D.L.2000.Theelectrodermalsystem.InCacioppo, J.T.;Tassinary,L.G.;andBerntson,G.G.,eds., HandbookofPsychophysiology .USA:CambridgeUniversityPress,2ndedition.200. Ekman,P.,andFriesen,W.V.1967.Headandbodycuesinthejudgementofemotion:Areformulation. PerceptualandMotorSkills 24:711. 151

PAGE 168

El-Nasr,M.S.,andSkubic,M.1998.Afuzzyemotionalagentfordecision-makinginamobile robot.In The1998IEEEInternationalConferenceonFuzzySystemsProceedings,IEEEWorld CongressonComputationalIntelligence ,volume1,135. Elmes,D.G.;Kantowitz,B.H.;andRoedigerIII,H.L.2006. ResearchMethodsinPsychology Belmont,CA:Thomson-Wadsworth,8thedition. Fast,J.1970. Bodylanguage .NewYork:M.Evans;distributedinassociationwithLippincott. Fincannon,T.;Barnes,L.E.;Murphy,R.R.;andRiddle,D.L.2004.Evidenceoftheneedfor socialintelligenceinrescuerobots.In IEEE/RSJInternationalConferenceonInternational ConferenceonIntelligentRobotsandSystemsIROS2004 ,volume2,1089. Fong,T.;Nourbakhsh,I.;andDautenhahn,K.2003.Asurveyofsociallyinteractiverobots. RoboticsandAutonomousSystems 42-4:143. Frijda,N.H.2000.Thepsychologists'pointofview.InLewis,M.,andHaviland-Jones,J.M., eds., Handbookofemotions .NewYork:GuilfordPress,2ndedition.xvi. Garreau,J.2007.Botsontheground:Intheeldofbattleorevenaboveit,robotsareasoldier's bestfriend. Hall,E.T.1966. TheHiddenDimension .NewYork:Doubleday. Itoh,K.;Miwa,H.;Nukariya,Y.;Zecca,M.;Takanobu,H.;Roccella,S.;Carrozza,M.C.;Dario, P.;andAtsuo,T.2006.Developmentofabioinstrumentationsystemintheinteractionbetween ahumanandarobot.In InternationalConferenceofIntelligentRobotsandSystems ,2620 2625. Johnson,B.,andChristensen,L.2004. EducationalResearchQuantitative,Qualitative,andMixed Approaches .Boston,MA:PearsonEducation,Inc.,2ndedition. Juslin,P.N.1997.Emotionalcommunicationinmusicperformance:Afunctionalistperspective andsomedata. MusicPerception 14[Summer1997]:383. Kidd,C.D.,andBreazeal,C.2005.Human-robotinteractionexperiments:Lessonslearned.In ProceedingofAISB'05SymposiumRobotCompanions:HardProblemsandOpenChallenges inRobot-HumanInteraction ,141. 152

PAGE 169

Kuli c,D.,andCroft,E.A.2003.Estimatingintentforhuman-robotinteraction.In The11th InternationalConferenceonAdvancedRoboticsICAR2003 ,810. Kuli c,D.,andCroft,E.2005.Anxietydetectionduringhuman-robotinteraction.In IEEE/RSJ InternationalConferenceonIntelligentRobotsandSystemsIROS2005 ,616. Kuli c,D.,andCroft,E.2006.Physiologicalandsubjectiveresponsestoarticulatedrobotmotion. Robotica Forthcoming:15. Lang,P.J.;Bradley,M.M.;andCuthbert,B.N.1993.Emotion,arousal,valence,andthestartle reex.InBirbaumer,N.,and Ohman,A.,eds., Thestructureofemotion:psychophysiological, cognitive,andclinicalaspects .Seattle:HogrefeandHuber.vi. Lisetti,C.L.;Brown,S.M.;Alvarez,K.;andMarpaung,A.H.2004.Asocialinformaticsapproachtohuman-robotinteractionwithaservicesocialrobot. IEEETransactionsonSystems, ManandCybernetics,PartC 34:195. Lisetti,C.L.,F.N.2002.Maui:amultimodalaffectiveuserinterface.In InternationalMultimedia Conference,Proceedingsofthe10thACMinternationalconferenceonMultimedia ,161. Juan-les-Pins,France:ACMPress,NewYork,NY,USA. Liu,C.;Rani,P.;andSarkar,N.2006.Affectivestaterecognitionandadaptationinhuman-robot interaction:Adesignapproach.In InternationalConferenceonIntelligentRobotsandSystems IROS2006 ,3099. Maeda,Y.1999.Fuzzyruleexpressionforemotionalgenerationmodelbasedonsubsumption architecture.In 18thInternationalConferenceoftheNorthAmericanFuzzyInformationProcessingSociety,NAFIPS ,781. Makatchev,M.;Lee,M.K.;andSimmons,R.2009.Relatinginitialturnsofhuman-robotdialoguestodiscourse.In InternationalConferenceonHuman-RobotInteraction Michaud,F.;Audet,J.;Letourneau,D.;Lussier,L.;Theberge-Turmel,C.;andCaron,S.2001. Experienceswithanautonomousrobotattendingaaai. IntelligentSystems,IEEE[seealsoIEEE IntelligentSystemsandTheirApplications] 16:23. 153

PAGE 170

Mizoguchi,H.;Sato,T.;Takagi,K.;Nakao,M.;andHatamura,Y.1997.Realizationofexpressive mobilerobot.In 1997IEEEInternationalConferenceonRoboticsandAutomation ,volume1, 581. Moshkina,L.,andArkin,R.C.2005.Humanperspectiveonaffectiveroboticbehavior:Alongitudinalstudy.In IEEE/RSJInternationalConferenceonIntelligentRobotsandSystemsIROS 2005 ,2443. Murphy,R.R.;Riddle,D.;andRasmussen,E.2004.Robot-assistedmedicalreachback:asurvey ofhowmedicalpersonnelexpecttointeractwithrescuerobots.In 13thIEEEInternational WorkshoponRobotandHumanInteractiveCommunicationRO-MAN2004 ,301. Mutlu,B.;Osman,S.;Forlizzi,J.;Hodgins,J.K.;andKiesler,S.2006.Taskstructureanduser attributesaselementsofhuman-robotinteractiondesign.In 15thIEEEInternationalWorkshop onRobotandHumanInteractiveCommunicationRO-MAN2006 .UniversityofHertfordshire, Hateld,UK:IEEE. Mutlu,B.;Hodgins,J.K.;andForlizzi,J.2006.Astorytellingrobot:Modelingandevaluationof human-likegazebehavior.In 2006IEEE-RASInternationalConferenceonHumanoidRobots HUMANOIDS'06 .Genova,Italy:IEEE. Nass,C.I.,andBrave,S.2005. WiredforSpeech:howvoiceactivatesandadvancesthehumancomputerrelationship .Cambridge,Mass.:MITPress. Nass,C.,andMoon,Y.2000.Machinesandmindlessness:Socialresponsestocomputers. Journal ofSocialIssues 56:81. Norman,D.A.2004. Emotionaldesign:whyweloveorhateeverydaythings .NewYork:Basic Books. Nourbakhsh,I.R.;Kunz,C.;andWilleke,T.2003.Themobotmuseumrobotinstallations:a veyearexperiment.In IEEE/RSJInternationalConferenceonIntelligentRobotsandSystems IROS2003 ,volume4,3636. Ortony,A.;Clore,G.L.;andCollins,A.1988. Thecognitivestructureofemotions .Cambridge England;NewYork:CambridgeUniversityPress. 154

PAGE 171

Paffenbarger,R.S.J.;Wing,A.;andHyde,R.1978.Physicalactivityasanindexofheartattack riskincollegealumni. AmericanJournalofEpidemiology 108:161. Picard,R.W.;Vyzas,E.;andHealey,J.2001.Towardmachineemotionalintelligence:analysisof affectivephysiologicalstate. IEEETransactionsonPatternAnalysisandMachineIntelligence 23:1175. Picard,R.W.1997. Affectivecomputing .MITPress. Rani,P.;Sims,J.;Brackin,R.;andSarkar,N.2002.Onlinestressdetectionusingpsychophysiologicalsignalsforimplicithuman-robotcooperation. Robotica 20:672. Rani,P.;Sarkar,N.;Smith,C.A.;andAdams,J.A.2003.Affectivecommunicationforimplicit human-machineinteraction.In IEEEInternationalConferenceonSystems,Man,andCybernetics ,volume5,4896. Rani,P.;Sarkar,N.;Smith,C.A.;andKirby,L.D.2004.Anxietydetectingroboticsystemtowardsimplicithuman-robotcollaboration. Robotica 22:85. Reeves,B.,andNass,C.1998. TheMediaEquation:HowPeopleTreatComputers,Television, andNewMediaLikeRealPeopleandPlaces .NewYork,NY:CambridgeUniversityPress. Riddle,D.R.;Murphy,R.R.;andBurke,J.L.2005.Robot-assistedmedicalreachback:using sharedvisualinformation.In IEEEInternationalWorkshoponRobotandHumanInteractive CommunicationROMAN2005 ,635.Nashville,TN:IEEE. Russell,J.A.,andLemay,G.2000.Emotionconcepts.InLewis,M.,andHaviland-Jones,J.M., eds., Handbookofemotions .NewYork:GuilfordPress,2ndedition.xvi. Russell,J.A.;Weiss,A.;andMendelsohn,G.A.1989.Affectgrid:Asingle-itemscaleofpleasureandarousal. PersonalityandSocialPsychology 57:493. Sarkar,N.2002.Psychophysiologicalcontrolarchitectureforhuman-robotcoordinationconceptsandinitialexperiments.In IEEEInternationalConferenceonRoboticsandAutomationICRA'02 ,volume4,3719. 155

PAGE 172

Scheeff,M.;Pinto,J.;Rahardja,K.;Snibbe,S.;andTow,R.2000.Experienceswithsparky,a socialrobot.In ProceedingsoftheWorkshoponInteractiveRobotEntertainment Scherer,K.R.,andOshinsky,J.S.1977.Cueutilizationinemotionattributionfromauditory stimuli. MotivationandEmotion 1:331. Scherer,K.R.1979.Nonlinguisticvocalindicatorsofemotionandpsychopathology.InIzard, C.E.,ed., EmotionsinPersonalityandPsychopathology .NewYork:PlenumPress.495. Schubert,E.2004.Modelingperceivedemotionwithcontinuousmusicalfeatures. MusicPerception 21:561. Shimokawa,T.,andSawaragi,T.2001.Acquiringcommunicativemotoractsofsocialrobotusing interactiveevolutionarycomputation.In IEEEInternationalConferenceonSystems,Man,and Cybernetics ,volume3,1396. Spiegel,J.P.,andMachotka,P.1974. MessagesoftheBody .NewYork,NY:FreePress. Spielberger,C.;Gorsuch,R.;andLushene,R.1970. ManualfortheState-TraitAnxietyInventory PaloAlto,CA:ConsultingPsychologistPress. Steinfeld,A.;Fong,T.;Kaber,D.;Lewis,M.;Scholtz,J.;Schultz,A.;andGoodrich,M.2006. Commonmetricsforhuman-robotinteraction.In 1stACMSIGCHI/SIGARTConferenceon Human-RobotInteraction .SaltLakeCity,Utah,USA:ACMPress. Stern,R.M.;Ray,W.J.;andQuigley,K.S.2001. PsychophysiologicalRecording .NewYork,NY, USA:OxfordUniversityPress,2ndedition. Stevens,J.P.1999. IntermediateStatisticsAModernApproach .Mahwah,NJ:LawrenceErlbaum Associates,Publishers,2ndedition. Sugano,S.,andOgata,T.1996.Emergenceofmindinrobotsforhumaninterface-research methodologyandrobotmodel.In 1996IEEEInternationalConferenceonRoboticsandAutomation ,volume2,1191. Tassinary,L.G.,andCacioppo,J.T.2000.Theskeletomotorsystemsurfaceelectromyography. InCacioppo,J.T.;Tassinary,L.G.;andBerntson,G.G.,eds., HandbookofPsychophysiology USA:CambridgeUniversityPress,2ndedition. 156

PAGE 173

Thrun,S.1998.Whenrobotsmeetpeople. IntelligentSystemsandTheirApplications,IEEE[see alsoIEEEIntelligentSystems] 13:27. Vel asquez,J.D.1997.Modelingemotionsandothermotivationsinsyntheticagents.In 14th NationalConferenceonArticialIntelligence ,10.Providence,RhodeIsland:TheAAAIPress, MenloPark,California. Walters,M.L.;Dautenhahn,K.;teBoekhorst,R.;KhengLee,K.;Kaouri,C.;Woods,S.;Nehaniv, C.;Lee,D.;andWerry,I.2005.Theinuenceofsubjects'personalitytraitsonpersonalspatial zonesinahuman-robotinteractionexperiment.In IEEEInternationalWorkshoponRobotand HumanInteractiveCommunicationRO-MAN2005 ,347. Watson,D.;Clark,L.A.;andTellegen,A.1988.Developmentandvalidationofbriefmeasures ofpositiveandnegativeaffect:thepanasscales. JournalofPersonalityandSocialPsychology 54:1063. Zimmerman,J.;Forlizzi,J.;andEvenson,S.2004.Taxonomyforextractingdesignknowledge fromresearchconductedduringdesigncases.In Futureground04ConferenceoftheDesign ResearchSociety 157

PAGE 174

Appendices 158

PAGE 175

AppendixACopyrightPermissions ThefollowingarepermissionsgrantedbyThoughtTechnologyLtd.touseimagesforsensor placementfromtheironlineProComp5InitiManual(pages25,30,and33)locatedat http://www.thoughttechnology.com/manual.htm.RefertoFigures38and39. Figure38.LetterofpermissionforuseofcopyrightedmaterialfromHalMyers,Ph.D.,President ofThoughtTechnologyLtd. 159

PAGE 176

AppendixAContinued Figure39.Electronice-mailpermissionforuseofcopyrightedmaterialfromLawrenceKlein, VicePresidentofThoughtTechnologyLtd. 160

PAGE 177

AppendixBIRBLettersforHumanSubjectsClearance ThefollowingaretheoriginalandcontinuingreviewapprovallettersfromtheUniversityof SouthFlorida'sDivisionofResearchIntegrityandComplianceInstitutionalReviewBoard.The originalapprovalperiodwasfrom11/29/2007through11/27/2008andthecontinuingreviewapprovalperiodisfrom11/18/2008through11/17/2009.TheIRBnumberforthisresearchstudyis: 106255. 161

PAGE 178

AppendixBContinued Figure40.InstitutionalReviewBoardapprovalletterforinitialreviewforIRB106255page1of 2. 162

PAGE 179

AppendixBContinued Figure41.InstitutionalReviewBoardapprovalletterforinitialreviewforIRB106255page2of 2. 163

PAGE 180

AppendixBContinued Figure42.InstitutionalReviewBoardapprovalletterforcontinuingreviewforIRB106255page 1of2. 164

PAGE 181

AppendixBContinued Figure43.InstitutionalReviewBoardapprovalletterforcontinuingreviewforIRB106255page 2of2. 165

PAGE 182

AbouttheAuthor CindyL.BethelisadoctoralcandidateinComputerScienceandEngineeringattheUniversityofSouthFlorida.SheisanIEEERoboticsandAutomationSocietyRASGraduateFellow andanNSFGraduateResearchFellow.Hermajoremphasisinresearchfocusesintheareasof human-robotinteraction,affectivecomputing,robotics,andarticialintelligence,withaminorin theareasofpsychologyandappliedstatistics.Ms.BethelreceivedaB.S.inComputerScience SummaCumLaudefromtheUniversityofSouthFloridainMay2004.