Interaction
development
for
personal
daily
support
Design
of
conversational
agent
for
activity
tracking
ByChristerBeke
Abstract
Thisreportistheresultforthefinalprojectofthe CreativeTechnologyBachelor’s programme.TheprimarygoalisassistingRoessinghResearch&Developmentonthe redesignofanactivitytrackingapplicationin alarger“home caringenvironment”called eWALL,whichisintheprocessofbeing movedintoanautonomoustabletsolution.The majorstepstoexecutewerebackgroundresearch, creatingaprototype andvalidationit throughusertesting.Theusertestentaileda4partquestionnaire and23people
1Introduction 4
1.1eWALL 4
1.1.1Design 5
1.1.2Activityapplication 6
1.2Problemdescription 7
1.2.1Challenges 8
2Researchquestions 9
3StateoftheArt 10
3.1Activitytracking 10
3.1.1Goalsetting 11
3.2Conversationalagents 12
3.2.1EmbodiedVirtualAgents 12
3.2.2Scriptedvsrelationalagents 13
3.2.3Whyanagent? 14
4Ideation 15
4.1Domain 15
4.2RobintheRobot 15
4.3Interaction 16
4.3.1Example 16
4.4Scenarios 17
5Method 18
5.1Reporting 18
5.2Evaluation 18
5.3Usertestsetup 18
5.4Usertestprocess 19
6Realisation 20
6.1Criteria 20
6.2Prototypedevelopment 21
6.3Prototypedecisions 22
6.4Prototypescreenshots 23
7Evaluation 24
7.1Hypotheses 24
7.2Results 24
7.2.1Generalquestions 25
7.2.3Scenariousingthevirtualagent 27
7.2.4Evaluation 28
8Conclusionanddiscussion 29
8.1Recommendations 30
9Acknowledgements 31
10Bibliography 32
11Appendices 33
AppendixI:Terminology 33
1
Introduction
Roessingh Research and Development (RRD) is a center for research on rehabilitation, psychology, biomedical science and computer science . They position themselves between1 university and healthcare practice. RRD works on tools to help people manage their own health, focussing on innovative and user friendly applications that offer monitoring of health dataandcoaching.
1.1
eWALL
One of the projects RRD is working on is the eWALL - Home Caring Environment , an 2 European Union funded project designed to contribute to the prolongation of independent living for various types of patients and elderly. The distinction between typical healthcare systems and eWALL is that eWALL tries to do this in an unobtrusive way by using advanced sensing technologies. Current target groups of eWALL are people with age related impairments (ARI), mild cognitive impairments (MCI, or early-stage dementia), and COPD andasthmapatients.
The eWALL system consists of 2 major components: the eWALL Cloud and eWALL Sensing Environment. The Sensing Environment is installed in the patient's home to gather data about activity, or help them with their daily routine. The eWALL Cloud connects the Sensing Environment with other stakeholders like hospitals, government health care systemsandrelatives.
1http://www.rrd.nl/
Figure1:SystemarchitectureofeWALL
1.1.1
Design
Figure2:ThedashboardofeWALL
1.1.2
Activity
application
AnimportantapplicationoneWALLforeveryonewhoneedscoachingwitheveryday physicalactivityistheactivitytracker.Thisappshowsdailyactivitylikestepcount,butalso changeovertime.
1.2
Problem
description
RoessinghResearch&DevelopmentisworkingonacontinuationofeWALL,named
CloudCare2U ,3that hassignificantchanges comparedtothe original project.Oneof theseis thateWALLwillnowberunningonatabletinpeople’sownhome.Thisbringstwo
challengestothetableregardingtheredesignof theactivityapp:screensizeanddomain. Thechangeinscreensizewillaffecttheentiresystem,notonlytheactivity app,butforthe scopeofthisprojectwewill focusonlyonthispart.
Anoutstandingissueisthatthecurrentactivityappdoesnotfallinline withthedesignof eWALL;itisflatasopposedtothereal-world imitatingdesignofthe homescreen.Inorder tomakeitfitwiththerestofeWALLsothatitfeelslikeasingleproduct,theredesignshould usetheseimitationdesignelements.
Furthermore,theredesignedappshouldallowusersto lookbothattheircurrentactivityto seeiftheirdailytargetismet,butalsoshowthemhowthey’redoing overtime.Thistrend analysisisimportanttoletpeoplebeawareoftheirbehaviourandhopefullychangeitto meettheirgoals.
Lastly,theactivityappshouldbeconnectedtothesystem’sconversational agent“Robinthe Robot”.Itshouldbecomeclearintheresearch andtestingwhattypeofinformationpeople wouldwanttousetheconversationalagenttoretrieve, butalsowhichtasks arebettertobe performedbytheuserinterface.
1.2.1
Challenges
Nexttothesedesignspecifications,thereareanumberofsmaller challengesthatbecame apparentfrominterviewswiththeclient.
Firstofall,thesystemisnotinuseinit’scurrentform(alargetouchdisplayonthewall), makingishardtoanswerquestionsrelatedtocomparisonsbetweenthe originalandnew domain(atabletintheuser’sownhome).Thereforetheprojectwillbefocussedsolelyon validatingthetabletsizedredesign.
2
Research
questions
Inordertosolvetheaforementionedproblemfortheclient,the followingresearchquestions havebeendefined:
Primaryresearchquestion
“Whichtasksarebettertobeexecutedvia anembodiedconversationalagentinsteadofthe userinterface?”
Secondaryresearchquestion
3
State
of
the
Art
Thisprojectfocusesontheapplicationofembodied agentsandhowtheycanassistin providingfeedbackandcoachingondailyphysicalactivity,activitygoalsettingand motivationalsupport.Thereforerelevantbackgroundinformationisprovidedonthe subjectsofactivitytracking,goalsetting,conversationalagentsandhuman-computer relationships.
3.1
Activity
tracking
Activitytrackersarewidelyusednowadays.Most ofthesetrackers featureasmartphone applicationcombinedwithasmallwristbandor wristwornsensing device.Thesedevices havesensorsthatkeeptrackofmovementandtheyfeedthisdatabackintothesmartphone application.Theappthentranslatesthatdatainto somethingthatcanbeunderstoodbythe wearer.Thisoftenhappensintheformofgraphs toindicatewhethera certaindailygoalwas met.Therepresentationofthesestatisticsaredifferentperbrand,butmostofthemarevery similar.CurrentlypopularbrandsforpersonalfitnesstrackingareFitBit,Jawboneand AppleWatch,buttherearemanymoresimilardevicesonthemarket.
3.1.1
Goal
setting
Whenactivitygoalsareproperlyset,theycanincreasemotivation,self-regulationand promoteasenseofachievement.Thiscanbe improvedbysettingmeaningfulgoalsusinga methodcalledSMART(Specific,Measurable,Attainable,Relevant,Timely)4.
FitnesssolutionslikeFitBitallowuserstosetgoals intheapp.The ideabehindsettingthese goalsistomotivatepeopletousethedevice moreandbecomehealthier intheprocess.By allowingtosetpersonalizedgoals,theuserfeelsmoreattachedtothegoalsandismore inclinedtofollowtheneededworkoutstoreachthem.Thissatisfies SpecificandRelevant componentsofSMART.Theappallowsyouto trackyourprogressovertime,givingan indicationifyouwillreachyourgoalorif youneedincreaseyour efforts.Thishelpsusers withtheMeasurableandAttainableparts.The timely,orproximal, componentofSMART canalsobecoveredbytheapp,astheuseristotallyfreeinselectingwhentoreachacertain goal(short-orlongtermgoals).
AstudyperformedbyStavrosAsimakopoulos,GrigoriosAsimakopoulosandFrankSpillers
in2017forInformaticsgoesindepthinto themotivationanduserengagementinvolvedwith personalfitnesstrackers.Fortheirstudy,wearables andsmartphoneapps fromFitBitand Jawbonewereused.Theyconcludethatthesuccess ofreachingthesegoalsreliesgreatlyon dataaccuracy,gamificationand thedesignoftheapplicationitself .5Users feltmotivational valuefromseeingstepsandadviceintheuserinterface.Theyalsolikedthelevelof
autonomythesmartphoneapps weregivingthem.
Anotherelementingettinguserstostickto theirworkoutsisgamification.Mostofthe popularappshaveabadgesystem,whereeachtimetheuserreachesacertainmilestone,a smallawardispresentedtoindicatethattheuser isdoingagood job.Oftenthenext milestoneisvisiblewithintheapp,allowinguserstohaveanewtargettoworktowards.
4https://vpal.harvard.edu/publications/setting-goals-who-why-how
Thesecondoptioningamificationisrankings.Users arelistedinhighscoreliststogether withpeopletheyinviteintheapp,usually friendsorfamilymembers.Thisbringsa
competitionelementintothefitnesstrackerandmotivatesusers togetorstay ontopofthe list.AnexampleofthisisthesmartphoneappRunkeeper.Unlikeothersolutionsitdoesnot requirethepurchaseofawristbanddevice,butsimplyusesthe phone’slocationto
determinethedistanceandspeedofeachrun. Thesearethencomparedtosimilarworkouts fromfriends,orevenshareyourprogressonsocialmedia . 6
3.2
Conversational
agents
Aconversationalagentordialogsystemisasoftwaresystemdesigned tointeractwith humansinastructuredway.Thesesystemscan usetext,speech,graphicsorothermethods ofcommunicatingwithahuman.
Well-knownconversationalagentsnowadaysarethe personalassistanttypeprogramsthat smartphonesanddomoticssystemshave.ExamplesoftheseareApple’sSiri, GoogleNow,or AmazonAlexa.
3.2.1
Embodied
Virtual
Agents
Whenintegratingaconversationalagentintoasysteminsuchawaythatit hasadigitalor physicalrepresentationintheenvironmentthatisusestointeractwithit,wecallitan embodiedvirtualagent.ThecurrenteWALLsystemhassuchanagent; RobintheRobot.
Thebiggestdifferencebetweenembodiedvirtualagents andotherdialog systemsisthe amountofinteractionthatispossible.Itcan provideamuchricherwayofcommunication, forexampleusinggesturesorfacialexpressions toconveya message.Whenhumans communicate,muchofourmeaningandintentionsareexpressedviabodylanguage. Bringingthistraittoaconversationalagent greatlyenhancesits capabilitytocommunicate withhumansinanaturalway.
ThecurrentagentineWALLisdesignedlikea1950’srobotthatfitsthelivingroom
environmentofthesystem.Itisimportantforanembodiedagenttofitinit’ssurroundings, sotheeWALLdesignerschosetomake acharacterthat fitstheirtheme.
AnexampleofanembodiedvirtualagentcanbefoundinNon-playercharacters(orNPCs) ingames.InthegameCivilization,NPCsare usedforenemiesandcouncilmembers.You canhaveadialoguewiththemthatchanges dependingonyouranswersortradeoffers. Theseconversationflowsarepre-scripted.However, theseunexpectedresponseskeepthe gameinteresting.Thecharactersarehistoricalfiguresthatleadoneof theenemycomputer players,makingthemrealembodiedagents.
Figure5:Examplesofembodiedagentsas NPCsinthe gameCivilizationV.
3.2.2
Scripted
vs
relational
agents
AconversationalagentsisatypeofHuman-Computer interactionthatisdesignedtohavea long-termrelationshipwiththeuser.Theinteraction betweenthehuman andcomputerover timecanbeseenasarelationship,makinga conversationalagentinthe roleofapersonal assistantarelationalagent.ThisareaofresearchwasfirstdiscussedbyT.BickmoreandR. Picard[Bickmore,Picard,2005].Thisrelationalaspectof anagentsetsitapartfromthe mostlyscriptedagentsthatareusedinthe gamesmentionedabove.Overtime,thecomputer learnsmoreabouttheuserandbuildsupacontexttogivemoremeaningtothe
3.2.3
Why
an
agent?
Whywouldanysystem,eWALLorotherwise,use anembodiedagenttointeractwithusers eventhoughitisperceived asburdensomeonsmartphones?
Thebestapplicationforvirtualassistantsliesintheexecutionofcomplextasks.Onatypical smartphonethesearetaskslikesettingalarms,findingandplayingmusic, oraskingfor weatherforecasts .7Allthesetasksnormallyrequiremultipleactionsintheuserinterface.It isdifficulttodeterminewhichtaskisdifficultand whichtaskiseasy, butgenerallytheline liesatthepointwhereusingtheassistantto doittakesless timeandlesssteps.
Whenitcomestopersonaldailysupport,activity trackingandgoalsetting,theamountof datathatisgatheredovertimemightoverwhelmthe userresultingintheinabilitytomake decisionsbasedonthisdata.Avirtualassistantcanhelporganize thisinformationby intelligentlytranslatingthedatasetintoactionablegoals.
4
Ideation
Duringtheideationphase,severalbrainstormfeedback sessionswereheld withRRDto makesuretheprojectwasgoinginthedirection theclientwanted.An importantstepinthe processwaspivotingfromadesignfocussedprototypetoavirtual agentfocussedprototype. Thevirtualassistantwasmoreinterestingtoworkonasitisusingmoderntechnologiesand willprobablyhavealargerimpactondaily supportsystemslikeeWALLinthefuture.
4.1
Domain
TheoriginaleWALLwasdisplayedonalarge monitorinstalledinalivingroomenvironment inanursinghome.Goingforward,RRDdecidedtousetabletsthatcanbe putinuser’sown homes.Thisdomainchangemustbetakenintoaccountwhendesigning theprototype.With asmallerscreen,itishardertoseefromadistancewhatisdisplayed. Theusereitherneeds tocomeclosertothedevice,oranalternativemethodforinteractionisrequired.Thisbrings ustotheimplementationofRobin theRobotintheprototype.
4.2
Robin
the
Robot
ThevirtualagentintheoriginaleWALLprojectiscalledRobintheRobot.Robinisdisplayed onthehomescreenandwhenclickedonit opensadialogthat showedquestionsthatyou canask.HowevertherewasnorealintelligenceinRobinsoonlythesespecificquestionscan beasked.Alsothequestionsneedtobeclickedasthereisnovoicerecognitionor
text-to-speechavailable.ThismakesinteractingwithRobinnotasseamlessas itshouldbe whenweworkwithembodiedagents.
AfewthingsfromtheoriginalRobinwerefound tobeinterestingto useintheprototype. Thosewerethe50’srobotdesignandtheawarenessofwhichuserwasinteractingwiththe systembydisplaythenamewhenopeningtheRobindialog.
Newideaswerecreatedaswellfortheprototype.Wewantedtoseeifaddingvoice
onthesmallerscreen.ThismeansthattheusercantalktoRobinusing theirvoiceandRobin wouldtalkbackwiththeanswersthatwerefound inthesystem,bringing itclosertoother assistantslikeSiriandGoogleNow.Byaddingthis typeofinteraction,we hopedtoseea reductionofinformationdisplayedintheuserinterface.
4.3
Interaction
Themostimportantpartoftheprototypeis theinteractionwithRobintheRobot.There wereseveraldiscussionsabouthowdeeptheinteractionwithRobinshould go.Forexample whetherRobinshouldbeabletounderstandsmalltalkabouttopics otherthanactivity tracking,orifRobinshouldbeabletoask followupquestionswhen piecesofcontextare missing.Thesetypesofbehaviourswouldresultinamorehuman-like interaction,butthey alsocomplicatetheprototypedevelopment.
Thesecondpointofattentionintheinteractionisthedifficultyofthequestionsbeingasked. Simplequestionslike“HowfardoIneedtowalktoday?”or“Whatis mydailystepgoal?” canbeasked,butthosewouldbejustaseasytofindinthe userinterface.Moredifficult questionslike“WillIreachmygoalthisweek?” wouldbeharderto readfromtheinterface becausemultiplepiecesofinformationareneeded toconstructan answer.Theanswers givenbyRobint0questionsliketheseshouldbeconstructedinsuchaway thattheuserwill beabletointerpret thisinformationeasily.