Comparing spatially static and dynamic vibrotactile take-over requests in the driver seat

(1)

Delft University of Technology

Comparing spatially static and dynamic vibrotactile take-over requests in the driver seat

Petermeijer, S. M.; Cieler, S.; de Winter, J. C F

DOI

10.1016/j.aap.2016.12.001

Publication date

2017

Document Version

Final published version

Published in

Accident Analysis & Prevention

Citation (APA)

Petermeijer, S. M., Cieler, S., & de Winter, J. C. F. (2017). Comparing spatially static and dynamic

vibrotactile take-over requests in the driver seat. Accident Analysis & Prevention, 99(Part A), 218-227.

https://doi.org/10.1016/j.aap.2016.12.001

Important note

To cite this publication, please use the final published version (if applicable).

Please check the document version above.

Copyright

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy

Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.

This work is downloaded from Delft University of Technology.

(2)

Green Open Access added to TU Delft Institutional Repository

‘You share, we take care!’ – Taverne project

https://www.openaccess.nl/en/you-share-we-take-care

Otherwise as indicated in the copyright section: the publisher

is the copyright holder of this work and the author uses the

Dutch legislation to make this work public.

(3)

ContentslistsavailableatScienceDirect

Accident

Analysis

and

Prevention

j o ur na l h o me pa g e :w w w . e l s e v i e r . c o m / l o c a t e / a a p

Comparing

spatially

static

and

dynamic

vibrotactile

take-over

requests

in

the

driver

seat

S.M.

Petermeijer

a,∗

_,

_S.

_Cieler

b

_,

_J.C.F.

_de

_Winter

c

a_Lehrstuhl_für_Ergonomie,_Fakultät_für_{Maschinenwesen,}_Technische_Universität_München,_{Boltzmannstraße}_15,_85747,_Garching,_Germany b_Division_Interior,_Interior_Electronics_Solutions,_Continental_Automotive,_Babenhausen,_Germany

c_Department_of_{Biomechanical}_Engineering,_Faculty_of_Mechanical,_Maritime_and_Materials_Engineering,_Delft_University_of_Technology,_Delft,_The Netherlands

a

r

t

i

c

l

e

i

n

f

o

Articlehistory: Received26July2016 Receivedinrevisedform 29November2016 Accepted3December2016 Availableonline12December2016 Keywords:

Tactile/hapticdisplays Highlyautomateddriving Take-overrequest Simulator

a

b

s

t

r

a

c

t

Vibrotactilestimulicanbeeffectiveaswarningsignals,buttheireffectivenessasdirectionaltake-over requestsinautomateddrivingisyetunknown.Thisstudyaimedtoinvestigatethecorrectresponserate, reactiontimes,andeyeandheadorientationforstaticversusdynamicdirectionaltake-overrequests presentedviavibratingmotorsinthedriverseat.Inadrivingsimulator,eighteenparticipantsperformed threesessions:1)asessioninvolvingnodriving(Baseline),2)drivingahighlyautomatedcarwithout additionaltask(HAD),and3)drivingahighlyautomatedcarwhileperformingamentallydemanding task(N-Back).Persession,participantsreceivedfourdirectionalstatic(intheleftorrightpartofthe seat)andfourdynamic(movingfromonesidetowardstheoppositeleftorrightoftheseat)take-over requestsviatwo6×4motormatricesembeddedintheseatbackandbottom.IntheBaselinecondition, participantsreportedwhetherthecuewasleftorright,andintheHADandN-Backconditionsparticipants hadtochangelanestotheleftortotherightaccordingtothedirectionalcue.Thecorrectresponserate wasoperationalizedastheaccuracyoftheself-reporteddirection(Baselinesession)andtheaccuracyof thelanechangedirection(HAD&N-Backsessions).Theresultsshowedthatthecorrectresponserate rangedbetween94%forstaticpatternsintheBaselinesessionand74%fordynamicpatternsinthe N-Backsession,althoughtheseeffectswerenotstatisticallysigniﬁcant.Steeringwheeltouchandsteering inputreactiontimeswereapproximately200msfasterforstaticpatternsthanfordynamicones.Eye trackingresultsrevealedacorrespondencebetweenhead/eye-gazedirectionandlanechangedirection, andshowedthatheadandeye-gazemovementswhereinitiatedfasterforstaticvibrationsthanfor dynamicones.Inconclusion,vibrotactilestimulipresentedviathedriverseatareeffectiveaswarnings, buttheireffectivenessasdirectionaltake-overrequestsmaybelimited.Thepresentstudymayencourage furtherinvestigationintohowtogetdriverssafelybackintotheloop.

1. Introduction

1.1. Highlyautomateddrivingandtake-overmaneuvers

Highly automated cars may be introduced on public roads withinthenext5–10years(ERTRACTaskForce,2015).Inhighly automateddriving,thecardrivesitselfformostofthetime,butthe drivermustintervenewhentheautomationprovidesatake-over request(Gasseretal.,2012).Howtogetahumanoperatorbackinto thecontrolloopafteraperiodofpassivemonitoringisaclassical issueinhumanfactorsscience(Bainbridge,1983)thathasbecome

∗ Correspondingauthor.

E-mailaddress:petermeijer@lfe.mw.tum.de(S.M.Petermeijer).

pertinentintheautomateddrivingdomain(e.g.,GoldandBengler, 2014;Kerschbaumetal.,2015;Zeebetal.,2015).

Whentheautomationprovidesatake-overrequest,thedriver hastogetbackintothecontrolloopby1)shiftinghis/herattention totheroad,2)cognitivelyprocessingthetrafﬁcsituationand select-inganappropriateaction,3)repositioninghim/herselfinorderto takebackcontrolofthevehicle,and4)implementingtheaction viathesteeringwheeland/orpedals(Goldetal.,2013;Petermeijer etal.,2015;Zeebetal.,2015).Mosttake-overrequestsin previ-ousstudieshavebeenalarmsthatinformthedriverthatheorshe hastotakebackcontrol(e.g.,BanksandStanton,2015;Goldetal., 2013;Melcheretal.,2015).Atake-overrequestmaybedesigned insuchawaythatitdoesnotonlywarnthedriverthatatake-over isrequired,butalsoassistshim/herintheaforementioned‘shifting ofattention’and‘cognitiveprocessingandactionselection’phases.

Lorenzetal.(2014),forexample,proposedahead-updisplaythat

http://dx.doi.org/10.1016/j.aap.2016.12.001

(4)

indicatedsafeorunsafe‘corridors’ontheroadafteratake-over request.

1.2. Thepotentialofvibrotactiletake-overrequests

Inthefuture,thedriverofahighlyautomatedcarmaybe per-mittedtoengageinnon-drivingtaskssuchaseating,resting,or talkingonthephone,andempiricalstudiesindicatethatdrivers arelikelytodoso(Llanerasetal.,2013;Meratetal.,2012).Visual andauditorywarningsmaynotbesuitableastake-overrequests inhighlyautomateddriving:Whendriversarenolongerrequired tolookattheroad,theyarelikelytomissvisualindicationsonthe dashboardoronahead-updisplay.Similarly,auditorywarnings mightgounnoticedwhenengaginginnon-drivingtaskssuchas talkingtopassengersorlisteningtomusic.

Becausevibrotactilestimulidonothavetobeinthedriver’s visualﬁeld,theyareaviablecomplementtoauditoryandvisual displaysforassistingadriverinatake-overscenario.Bypresenting take-overrequestsinamultimodalmanner,theredundancyofthe warningisincreasedandconsequentlytheprobabilityofmissesis reduced(Wickensetal.,2012;Hancocketal.,2013).Prewettetal. (2012)foundinameta-analysisthatvibrotactilewarningsyield performanceadvantages(e.g.,fasterreactiontimes)whenaddedto abaselinetaskortovisualcues.Furthermore,inaprevious simula-torstudy,ithasbeenshownthatvisualandvibrotactilewarnings combinedyieldedfasterreactiontimesthanvisual-onlytake-over requests(Petermeijeretal.,2016).

Inordertoassistadriverinamaneuver,aninterfaceshould beabletoconveymoreinformationthanabinarywarning.Visual andauditorydisplaysaretraditionallyconsideredmoresuitable forcommunicatingsemanticsthantactiledisplays(Baldwinetal., 2012).Nonetheless,tactiledisplaysaresuitableforproviding direc-tionalinformation(VanErpetal.,2005)orsimplemessagesusing so-called tactons(i.e., byencoding theinformation in terms of thefrequency,timing,intensity,and/orlocationofthevibrotactile stimulus;BrewsterandBrown,2004).Similarly,vibrotactile dis-playsmaybeusefultoconveydirectionalinformation(e.g.,steer rightorleft)tothedriverduringatake-overrequest.

1.3. Staticanddynamicdirectionalcuesinvibrotactilewarnings Oneapproachtoassistthedriverinthetake-overprocedure couldbetoprovideadirectionalcue,thatis,toembed“directional informationintothetactilewarningsignalsinordertoorientthe driver’sspatialattentionindifferentlocations”(Mengetal.,2015, p. 336; seealso,Gray et al.,2014; Hoet al., 2005;Petermeijer etal.,2015).Vibrotactilestimuliwithdirectionalcueshavebeen studiedinavarietyofdrivingapplications,includinglanekeeping assistance(Beruschaetal.,2010),blindspotwarning(Morrelland Wasilewski,2010),andrearcollisionwarningsystems(Hoetal., 2005).HwangandRyu(2010)provideddirectionalcuesviathe steeringwheel,andparticipantswereaskedtoreactbyturning thesteeringwheelleftorrighttowardsthesideofthevibration. Theyfoundanaveragecorrectresponserateofabout90%,withan averageresponsetimeof2s.

Directionalcuescanbepresentedbymeansofstaticpatterns (i.e.,stimuliatonelocationonthehumanbody)ordynamicones (i.e.,asequenceofstimuliatdifferentlocationsonthebody,to sim-ulateamovement),seealsoPetermeijeretal.(2015)fora catego-rizationofvibrationpatterns.InastudybyPetermeijeretal.(2016), participantsdidnotseemtonoticethatthestaticvibrationsinthe driverseatwereprovidedontheleftorrightside.Sayeretal.(2005)

foundsimilarresultstoPetermeijeretal.;intheiron-roadstudy, severalparticipantshaddifﬁcultydiscriminatingstaticdirectional cues(i.e.,leftandright)ofvibrotactilestimuliintheseatbottom.

MengandSpence(2015)alsonotedthatstaticvibrotactile

stim-ulimightbelimitedintheirabilitytoconveydirectionalcuesand arguedthat“dynamictactilecuesmightbeusedtoimprovedrivers’ localizationanddifferentiationtothetactilewarnings”(p.339).

Two recent driving simulator studies showed that dynamic patternsthat movetowardsthedriver’s torso(usingfour stim-ulus locations, namely the two wrists, and two on the waist) evokedfasterreactionthanaway-from-torsocuesorstatic vibra-tions (Meng et al., 2014; Meng et al., 2015). In Schwalk et al. (2015),participantsreportedthatdynamic vibrationsthat trav-elledfromthetopofthebackresttowardsthefront oftheseat bottomwereappropriateforsignalingdriver-to-automation con-troltransitions.Conversely,dynamicstimuli thattravelledfrom theseatbottomtothebackrestwereregardedasappropriatefor automation-to-drivertransitions.Thus,basedontheavailable lit-erature,itappearsthatdynamicvibrationpatternsholdpromiseas take-overrequests.

1.4. Aim

Theaimofthisstudywastoevaluatehowaccuratelyhumans areabletorespondtovibrotactilestimuliinthedriverseat.The vibrations(i.e.,take-overrequests)containedadirectionalcueto informtheparticipantthathe/shehadtochangelanestoeitherthe leftortheright.Weevaluatedthecorrectresponserateofstatic vibrations(i.e.,non-movingvibrationsthatwerepresentedonthe leftorontheright)anddynamicvibrations(i.e.,vibrationsthat movedtotheleftortotheright)inadrivingsimulatorexperiment. Thecorrectresponseratewasmeasuredinthreeconditions:1)a baselineconditioninwhichtheparticipantsweresittinginthe sim-ulatorbutnotdriving(lowmentaldemand),2)whileparticipants weredrivingahighlyautomatedvehicle(mediummentaldemand), and3)whileparticipantsweredrivingahighlyautomatedvehicle andengagedinamemorytask(highmentaldemand).Inthe Base-linecondition,participantshadtoreport‘left’or‘right’,whereasin thetwodrivingconditions,participantshadtomakealeftorright lanechangeafterhavingreceivedthedirectionaltake-overrequest. Thesethreeconditionswereincludedtoinferwhetherthe direc-tionalcueswerestillperceivableiftheparticipantswereengaged inamentallydemandingtask,forthis willlikely bethecasein real-worldtake-overscenarios.Driverbehaviorwasfurther oper-ationalizedintermsoftake-overtime,steeringwheelinput,and eyeandheadmovements.Wehypothesizedthatdynamicpatterns wouldyieldhighercorrectresponseratesthanstaticones,andthat thecorrectresponserateofvibrotactilepatternswoulddecrease whenthedriverisunderincreasedmentaldemand.

2. Method

2.1. Participants

Eighteenparticipants(ﬁvefemale)holdingadriverlicense par-ticipated in the study. The participants were between 22 and 78 years old (M=43.0years; SD=15.2years). Two participants indicatedtheydrovelessthan2000kmperyear,twelve partici-pantsreportedayearlymileageof5000–20,000kmperyear,and theremaining fourparticipants reporteda yearly mileage over 20,000km.Sevenparticipantsindicatedthattheywearglassesor contactlenseswhendriving,andnonereportedtobecolorblind. Twoparticipantswereleft-handed. Allparticipantshad partici-patedinadrivingsimulatorstudyatleastoncebefore.

2.2. Simulator

The experiment was conducted in a ﬁxed-base simulator, locatedatContinental,Babenhausen,Germany.Thesimulator con-sistedofaBMW5-serieschassisinfrontofthreeprojectorsthat

(5)

Fig.1.Top:Drivingsimulatorusedinthisstudy.Bottom:Thevisualdisplayshown ontheinstrumentclusterofthesimulator.Thecentershowedthespeedometer,and ontherightsideaniconindicatedtheautomationstatus(blueicon=active;grey icon=inactive).(Forinterpretationofthereferencestocolourinthisﬁgurelegend, thereaderisreferredtothewebversionofthisarticle.)

providedthefrontviewofapproximately180◦(Fig.1).Small TFT-screensactedasrear-viewmirrors.AmountedTFTscreenanda touchscreenrepresentedtheinstrumentclusterandcenter con-sole.Afour-cameraeye-trackingsystemrunningat60Hz(Smart EyePro,version6.1.13;SmartEye,2016)wasusedtotrackthe par-ticipants’headandgazemotion.Theparticipantscouldtogglethe automationonandoffbymovingtheACClevereitherupordown. Aniconontheinstrumentcluster’srightsideindicatedthe automa-tionmode(Fig.1).ThesimulationranonSILABsoftware(WIVW, 2016).

2.3. Vibrotactileseat

Vibrationswerepresentedtotheparticipantsviaavibrotactile seatconsistingofa Velcromatthatcovered48 eccentric rotat-ingmassmotors(PicoVibemodelnumber:307-103,dimensions: 9×25mm).Themotorswereconﬁguredinto6×4matricesinthe seatbottomandseatback,respectively(Fig.2).Theinter-motor distancewasapproximately45mmbetweenthesixrowsand30, 50,and30mmbetweenthefourrespectivecolumns.Thevoltage totheindividualmotorswascontrolledusingthreePulseWidth Modulator(PWM)controllers,whichinturnwerecontrolledbyan ArduinoMegaconnectedtotheserverofthedrivingsimulator. 2.4. Staticanddynamicvibrationpatterns

Participantswereprovidedwithstaticordynamicvibrations, whichcontainedaleftorrightdirectionalcue(Fig.2).Furthermore, vibrationpatternswerepresentedateithertheseatbottomorthe seatback.Thus,therewerefourstaticpatterns(i.e.,1:backleft,2: backright,3:bottomleft,and4:bottomright)andfourdynamic patterns(i.e.,5:backmovingleft,6:backmoving right,7: bot-tommovingleft,and8:bottommovingright).Themotors,when activated,vibratedatapproximately60Hz.

Astaticpatternwasprovidedbythreevibrationpulses(500ms on/offintervals)intwocolumns(i.e.,12motors)(Fig.3).Adynamic

patternwasprovidedbyactivatingthecolumnsinsuccessionfrom thelefttotheright,orviceversa.Acolumnwasactivefor200ms andevery100msanadjacentcolumnactivated,creatingapattern thatmovedfromonesidetotheotherwithamaximumoftwo columnsvibratingatthesametime.

2.5. Experimentaldesign

Awithin-subjectdesignwasusedtoevaluatetheeffectofthe vibrationtype(i.e.,staticvs. dynamic)and mentaldemand(i.e., low,medium,andhigh).Eachparticipantexecutedthreesessions: 1) Baseline, 2) HAD, and 3) N-Back. The Baseline session was completedﬁrst,andtheHADandN-Backsessionswere counterbal-ancedacrossparticipants.Persession,theparticipantexperienced theeightvibrationpatterns(fourstaticonesandfourdynamicones, seeSection2.4)incounterbalancedorder.

1)TheBaselinesession(lowmentaldemand):Theeightvibration patternswerepresented totheparticipantin thedriverseat whilenovirtualsimulationwasrunning.Aftereachpattern,the participantwasaskedtoﬁlloutthreemultiple-choicequestions regardingthelocationanddirectionofthepattern:‘Ifeltthe vibration(1a.intheseatbottom,1b.intheseatback,1c.inthe wholeseat;2a.ontheleftside,2b.ontherightside,2c.onboth sides;3a.travellingtotheleft,3b.travellingtotheright,3c.not travelling).

2)TheHAD session (mediummental demand). The participant drovea highlyautomatedvehicleonthehighwayand expe-riencedeighttake-overrequestsviathevibrotactile seat.The take-over request warned the driverof a stationary vehicle ahead. The participant was instructedto change to the left orrightlaneaccordingtothedirectionalcueofthevibration pattern.Ifthevibrationpatternwasstatic,theparticipanthad tochangelanestowardsthesame sideas thevibration was presented.Incaseofadynamicpattern,theparticipanthadto changelanestothesidethevibrationsweremovingtowards. Note that thetake-over request was only presented viathe vibrotactileseat;noadditionalauditoryorvisualwarningwas presented.

3)TheN-Backsession(highmentaldemand).Thissessionwasthe sameastheHADsession,buttheparticipantperformedan addi-tionalN-Backtaskwhenthevehiclewasdrivingautomatically. TheN-Backtaskisawidelyusedtechniqueforimposingmental demandsinautomateddrivingresearch(e.g.,Goldetal.,2016; Radlmayretal.,2014;Louwetal.,2016).Inourstudy,the partici-pantperformeda2-BacktaskasspeciﬁedbyMehleretal.(2011). Apre-recordedfemalevoiceutteredrandomdigitsbetween0 and9withanintervalof2.25sbetweenthepresentationofeach digit.Theparticipanthadtorepeatthenumberthatwasuttered two digits before the current digit. The task automatically startedapproximately700mafterthestartofthesession,and 900maftereverytake-overrequest.TheN-Backtaskstopped automaticallywhenatake-overrequestwaspresented.The N-Backwasusedtoinvestigatetheeffectofamentallydemanding non-drivingtaskonthecorrectresponserateandreactiontimes. 2.6. Drivingscenario

DuringtheHADandN-Backsessions,theparticipantsdrovein highlyautomatedmodeonathreelanehighway,withlanewidths of3.75m. Atthestartofthesession,thesimulatedvehiclewas parkedatthesideoftheroad.Theparticipantwasaskedvia inter-comtomergeontothehighwayandactivatetheautomationwhen drivinginthemiddlelane.Theautomationkeptaspeedof120km/h andstayedinthecenterofthelane.Duringeachsession,the partic-ipantreceivedeighttake-overrequests,whichwereapproximately

(6)

Fig.2.Vibrationlocationsintheseatbackandseatbottom.Top:Thevibrationlocationsforthefourstaticpatterns.Bottom:Thevibrationlocationsforthefourdynamic patterns.

Fig.3.Schematicofthetemporalandspatialconﬁgurationofstaticanddynamicpatternswitha‘left’directionalcue.Onedotrepresentsasinglevibrationmotor.Theﬁlled circlesrepresentactivemotorsandtheemptycirclesrepresentinactivemotors.

1.5minapart.Atthemomentthatthetake-overrequestwas pre-sented,astationarycarappeared223minfrontoftheparticipant’s car(i.e.,inthemiddlelane),andtheautomationwasautomatically deactivated.Ataspeedof120km/hthiscorrespondedtoalead timeofabout7s(forstudiesusingthesametake-overparameters seeGoldetal.,2013;Goldetal.,2016;Petermeijeretal.,2016).All take-overrequestswereprovidedonstraightroadsegments,which meantthatnoimmediateactionoftheparticipantwasneededto stabilizethecarinitslane.Theparticipanthadtobrakeand/orsteer toavoidcollidingwiththestationarycarahead.

Betweenthetake-overrequests,theparticipant’scartravelled pasttwoorthreeslowermovingvehiclesontherightlane,whereas afastermovingvehicleontheleftlaneovertooktheparticipant’s car.Notrafﬁcwasaroundtheparticipant’scarwhenatake-over requestwaspresented.

2.7. Procedureandinstructions

Atarrival,theparticipantwaspresentedwithwritten instruc-tions and a consent form. After signing the consent form the participantcompletedanintroductory questionnairewith ques-tions about the participant’s gender, age, driving experience, drivingstyle,pastexperienceindrivingsimulators,andpresumed preferenceandperceivedurgencyofauditory,visual,and vibrotac-tiletake-overrequests.

Next,theparticipantperformedtheBaselinesessioninwhich he/she experienced all eight vibration patterns. Then, the eye-trackingsystemwascalibrated,andtheparticipantdroveatraining

of 2minto familiarize withthesimulator, a take-overrequest, andhowtoreactivatetheautomation.Afterapproximately1min ofdrivinginthetrainingsession,theN-Backtaskautomatically started.Atotalof20digitswerepresented,afterwhichthe partici-pantreceivedadynamicvibrotactiletake-overrequestintheback oftheseat,requestingtheparticipanttochangelanestotheleft.

Next,theparticipantperformedtheHADandN-Backsessions (incounterbalancedorder)inthedrivingsimulatorwitha5min breakinbetween.ToverifywhethertheN-Backtaskprovided addi-tionalworkload,theparticipantcompletedaNASATaskLoadIndex (NASA-TLX)aftereachofthetwodrivingsessions.Afterallthree sessionswerecompleted,theparticipantcompletedfour question-naires:(1)aquestionnaireonacceptance(VanderLaanetal.,1997) concerningthestaticpatterns,(2)thesameacceptance question-naire but now for thedynamic patterns,(3) a User Experience Questionnaire(UEQ)concerningthetactileseat,and(4)a question-nairethatpresentedthesamequestionsabouttake-overrequest modality astheintroductory questionnaire.Thisﬁnal question-nairewasusedtoevaluatewhethertheparticipants’preference had changed after actually experiencing vibrotactile take-over requests.

Theparticipantwasinstructedinwritingandverballyatthe beginningoftheﬁrstdrivingsessiontokeepthehandsandfeet off the steering wheel and pedals and not to intervene unless theautomationprovidedatake-overrequest.Whenatake-over requestwaspresented,theparticipanthadtoplacethehandsback onthesteeringwheel,performthenecessarysafetychecks,and avoidcollidingwiththecaraheadbybrakingand/orsteering.The

(7)

participantwasalsoaskedtofocusontheN-Backtaskwhenthe automationwasactiveduringtheN-Backsessionandtostopdoing theN-Backtaskin caseofatake-over.Theparticipantwasalso informedthat duringa take-overprocedurethere wouldbeno trafﬁcaround.Lastly,theparticipantwasaskedtochangelanes accordingtothedirectionalcue inthevibrationpattern andto reactivatetheautomationafterhavingpassedthestationarycar andbeingbackinthemiddlelane.

2.8. Dependentvariables 2.8.1. Objectivemeasures

Thecorrectresponserateofthepatternswasdeﬁnedasthe percentageofvibrotactilewarningsinwhichtheparticipants cor-rectlyidentiﬁedthedirectionalcueofthevibrationpattern.Inthe Baselinesession,ananswerwasmarkedascorrectwhenthe par-ticipantindicatedthecorrectside(forstaticpatterns)ordirection (fordynamicpatterns)inthemultiple-choicequestion.Duringthe drivingsessions,aresponsewasmarkedascorrectwhenthe partic-ipantmadealanechangetothesamesideasthevibrationpattern’s sideordirection.

Thefollowingmeasuresofreactiontimewereusedtoassess howquicklytheparticipantstookbackcontrolofthevehicleafter atake-overrequest:(1)Steer-touch:absolutesteeringwheelangle greaterthan0.25deg.This0.25degthresholdwasusedinanearlier studybyPetermeijeretal.(2016)asameasureofhowfast partic-ipantstouchedthesteeringwheelafterthetake-overrequest;(2) Steer-turn:absolutesteeringwheelanglegreaterthan2deg.The 2degthresholdwasusedtorepresenttheinitiationofasteering action(2.0degwasalsousedbyGoldetal.,2013);(3)Lanechange: absolutedeviationfromthelanecentergreaterthan1.875m(i.e., halfalanewidth).

Thegazeheadingrepresentstheleft/rightanglebetweenthe eye-gazevectorandavectorpointingforwardstothesimulator screen.Similarly,theheadheadingistheleft/rightangleofthe ori-entationoftheheadwithrespecttoavectorpointingforwardsto thesimulatorscreen.Thegazeandheadheadingwereanalyzedto investigatewhetherstaticordynamicvibrotactilewarningsevoke fastergazereactionstowardsacertaindirection.Theeye-tracking datawereﬁlteredwithafourth-orderlow-passButterworthﬁlter havingacut-offfrequencyof5Hz.

2.8.2. Self-reportmeasures

AllquestionnaireswereofferedinapaperformatinGerman language.TherawNASA-TLXincludedsixitems,namely,mental demand,physicaldemand,temporaldemand,performance,effort, andfrustration.Allitemsconsistedofa21-tickhorizontalbarwith anchorsontheleft(i.e.,low/good)andright(i.e.,high/poor)sides (Vertanen,2016).

Theacceptancequestionnairewasofferedtodeterminethe per-ceivedusefulnessandsatisfactionofstaticanddynamicvibration patterns.Theusefulnessscorewasdeterminedacrossthefollowing fiveitems:1.useful−useless,3.bad−good,5.effective− super-fluous,7.assisting− worthless,and9.raisingalertness−sleep inducing.Thesatisfactionscorewasdeterminedbythefollowing fouritems:2.pleasant−unpleasant,4.nice−annoying,6.irritating −likeable,8.undesirable−desirable.Allitemswereonafive-point scale.Signreversalswereperformedforitems1,2,4,5,7,and9,so thathigherscoresindicateahigherusefulness/satisfactionscore. 2.9. Statisticalanalyses

Ofthedependentmeasures,weobtainedforeverysessionan 18_×8matrix(i.e.,participantsxvibrationpatterns).Forthecorrect responserateandusefulnessandsatisfactionscores,thenumbers inthematrixwereranktransformedtoaccountfornon-normal

Fig.4.Meancorrectresponserateforthestaticanddynamicpatternsper par-ticipant.Theblackmarkersrepresentthecorrectresponseratesfortheindividual participants.Themean(SD)percentagesperconditionareasfollows:Baselinestatic: 93.8(11.2),Baselinedynamic:89.1(20.3),HADstatic:86.6(29.0),HADdynamic: 79.2(28.8),N-Backstatic:80.6(32.7),N-Backdynamic:73.6(30.3).

distributions.Forthecorrectresponserate,thedifferencesbetween thethreesessions(i.e.,Baseline,HAD,andN-Back)wereassessedby meanofarepeated-measuresANOVA.Fortheremainingmeasures, thedifferencesbetweenthetwotypesofvibrationpatterns(i.e., staticanddynamic)andbetweenthetwodrivingsessions(i.e.,HAD andN-Back)wereassessedbymeansofpairedttests.Additionally, Cohen’sd_zwasusedtodescribethesizeofthewithin-subjecteffect (Fauletal.,2007).Effectsweredeclaredstatisticallysigniﬁcantif p<0.05.Ifmultipleeffectswerecomparedasafunctionoftravelled distance,amorestringentsigniﬁcancelevelof0.01wasadopted.

3. Results

DuringtheBaselinesession,thecorrectresponseratesfortwo participants werenot recorded correctly; therefore, these data wereimputedwiththemeanvalueoftheremaining16 partici-pants.Oneparticipantdidnotevadethestationarycarwhenthe ﬁrsttake-overrequest(HADsession)waspresented.Dataofthis particulartake-overmaneuverwereexcluded.Eye-trackingdataof threeparticipantswereunavailablebecauseoftechnicaldifﬁculties (e.g.,becausetheparticipantwaswearingglasses).

3.1. Correctresponserate

Duringapreliminaryanalysis,itwasfoundthattherewereno statisticallysigniﬁcantdifferencesbetweenthecorrectresponse ratesofpatternsintheseatbottomandseatback.Therefore,these resultshave been aggregated, and are not reported separately.

Fig.4showsthecorrectresponseratesforthethreesessionsand thestaticversusdynamic patterns.Itcanbeseen thatthe cor-rectresponseratedecreaseswithincreasedmentaldemand(i.e., decreasingfromBaselinetoHADtoN-Back),buttheseeffectswere notsigniﬁcantlydifferent,F(2,34)=2.01,p=0.149.Therewasalso nostatisticaldifferencebetweenthestaticanddynamicvibration patterns,F(1,17)=3.36,p=0.084.

3.2. Reactiontimes

Table3showsthereactiontimesfromthemomentofthe take-overrequesttothefirststeeringwheeltouchandthelanechange. PairedttestsyieldednosignificanteffectsbetweentheHADand N-Backsessions(t(17)=1.12,p=0.278;t(17)=1.21,p=0.241;and t(17)=1.47,p=0.160,forsteer-touch,steer-turn,andlanechange, respectively).However,thereactiontimeswereslightlyfasterfor staticpatternsthanfordynamicpatterns(t(17)=−2.69,p=0.016; t(17)=_−2.54, p=0.021; t(17)=_−2.18, p=0.043, for steer-touch, steer-turn,andlanechange,respectively).Duringtheentirestudy, thebrakepedalwasappliedonlyfivetimesduringa take-over

(8)

maneuver(twiceintheHADsessionandthreetimesintheN-Back session).

3.3. Drivingvariables

Fig.5showsthemeanlateralpositionandthemeansteering wheelangleacrossparticipantsasafunctionoftravelleddistance forthestaticanddynamicpatterns(left)andfortheHADand N-Backsessions(right).Itcanbeseenthattheparticipantssteered aroundthestationarycar,withthesteeringwheelanglefollowing apatternthatischaracteristicofadoublelanechange.Theresults inFig.5areconsistentwiththereactiontimes(Table1)inthat dynamicvibrationsyieldedslightlyslowerlanechangesthanstatic vibrations.Speciﬁcally,themaximumsteeringangleishigherand occursearlierforstaticvibrationsthanfordynamicones.The bot-tomtwoplotsinFig.5showthepvaluesofpairedttestsbetween thestaticanddynamicpatterns(left)andbetweentheHADand N-Backcondition(right).ThesegraphsareinspiredbyManhattan plots(e.g.,Tanikawaetal.,2012).Highvaluesontheinverted log-arithmicscalerepresentlowpvalues.Thepvaluesregardingthe dynamicversusstaticpatterns(bottomleft)showtwopeaksthat exceedthe0.01threshold,whereasthepvaluesbetweentheHAD andN-Backsessions(bottomright)exceedthisthresholdonce. 3.4. Eye-trackingdata

Fig.6showsthemeanheadingangleofeye-gazeandthehead, includingthestandarddeviationsacrossthemeanofparticipants. Shortlyafterthetake-overrequest,thestandarddeviationofthe gazeheadingdecreases,suggestingthattheparticipantsfocused ontheroadahead.Afterthis(i.e.,fromabout50mafterthe take-overrequestwaspresented),participantsshiftedtheirattentionto theleftorrightdependingonthedirectionofthelanechange.The secondpeakinheadingoccurswhentheparticipantsreturnedto themiddlelane.Basedonthepvalues,itseemsthatboththegaze andheadheadingforthestaticpatternsdivertedearliertowards theleft/rightthandynamicpatterns.

3.5. Self-reportquestionnaires

Themean (SD) workload acrossthesix scales of the NASA-TLX was 21.7% (16.3%) and 35.7% (15.2%) for the HAD and N-Back session, respectively. A paired t test showed a signiﬁ-cantdifferencebetweenthesetwosessions,t(17)=−4.31,p<0.001, dz=−1.02. Fig. 7 shows that the participants rated the vibro-tactilefeedback positiveintermsof usefulnessandsatisfaction, but no differences between static and dynamic patterns were found(usefulness:t(17)=−0.32,p=0.749,dz=−0.08;satisfaction: t(17)=−0.12,p=0.906,d_z=−0.03).In theintroductoryand ﬁnal questionnaire,61%and72%ofparticipants,respectively,reported that take-overrequestsshouldbe provided bymeansof vibra-tionsincombinationwithauditoryand/orvisualwarnings.Inboth questionnaires17%ofparticipantsreportedthattakeoverrequests shouldbeprovidedbymeansofvibrationsonly.

4. Discussion

Theaimofthisstudywastoinvestigatethecorrectresponse rates,reactiontimes,andeye/headorientationinresponsetostatic anddynamicvibrationpatternsconveyedviaavibrotactileseat. Weconductedadrivingsimulatorexperimentwiththreesessions: Baseline,HAD,andN-Back.TheBaselinesessionwasusedto mea-sureparticipants’responserateswithlowmentaldemand,whereas theN-Backtaskimposedextramentaldemandontheparticipant (asconﬁrmedbytheresultsofNASA-TLX).

4.1. Theeffectofmentaldemand

The Baseline session yielded the highest average correct responserates(91%),followedbytheHAD(83%)andN-Back ses-sions(77%).Thus,whenparticipantswerenotengagedinadriving task,theywerereasonablywellabletodistinguishleftversusright vibrations,butwhenmentaldemandincreased,theabilityto dis-tinguishthedirectionalityofthevibrotactilestimulidiminished.It shouldbenotedthatthesedifferenceswerenotstatistically signif-icant.

Thereactiontimesshowednosigniﬁcantdifferencesbetween theHADandN-Backsessions(Table1).Thesecondarytaskthatthe participantsperformedrequiredcognitiveprocessingand a ver-balresponse;participantsdidnothavetousetheirhandsanddid nothavetolookawayfromtheroad.Ourﬁndingsareinlinewith arecentstudyperformedbyGoldetal.(2016),whichalsofound thattake-overtimeswerehardlyaffectedbyamentally demand-ingnon-drivingtask.Insomepreviousstudies,participantslosta largeamountoftimewithdisposingobjects(e.g.,aphone,tablet, orbook)orwithre-attendingtotheroad.Forexample,Melcher etal.(2015)foundanaveragereactiontimeof3.5stoatake-over requestwhenparticipantsheldamobilephone.InGoldetal.(2013)

andPetermeijeretal.(2016),theparticipantswereperforminga SurrogateReferenceTask(SuRT)havingtheireyesdivertedfrom theroad,buttheirhandsfree,whenthetake-overrequestwas pre-sented.Thesetwostudiesreportedsimilartouch-reactiontimes butsubstantiallyhighersteer-turnreactiontimesthanthepresent study.Insummary,itappearsthatbiomechanicaldistractioncauses largeimpairmentsinreactiontimes.Visualdistraction,ontheother hand,doesnotappeartohaveaninﬂuenceonthetimetoachieve motorreadiness(seealsoZeebetal.,2016),butincreasesthetime togetcognitivelybackintotheloop(asoperationalizedbya ‘con-scious’steering action).Finally,cognitivedistraction(asapplied inourN-Backcondition)seemstohaveonlyminoreffectsonthe reactiontimesinatake-overscenario.

4.2. Staticanddynamicpatterns

Staticvibrationpatternsyieldedhighercorrectresponserates thandynamicpatterns,buttheeffectwassmallandnot statisti-callysigniﬁcant.Fig.4illustratedthehighvariabilityofthecorrect responserateamongparticipants,bothwiththestaticanddynamic patterns.

ContrarytotheresultsofMengetal.(2014,2015),staticpatterns showedsignificantlyfasterreactiontimesthandynamicpatterns. ThisdiscrepancybetweenourresultsandthoseofMengetal.can beexplainedasfollows.First,inthepresentexperimentthe partici-pantshadtorecognizethedirectionalcueofthevibrationpatterns inordertomakealanechangeinthecorrectdirection,whereas inthestudiesofMengetal.,theparticipantswereinstructedto react(i.e.,tobrake)asquicklyaspossiblewhentheyperceivedthe vibrotactilewarning.Itprobablytakestimeforadrivertorecognize thedirectionofthedynamicpattern.Toillustrate,thefirst200ms ofadynamicstimulustotheleftishardlydistinguishablefroma staticstimulustotheright;onlyafter200msitbecomesclearthat thedynamicstimulusmovestotheoppositeside(seeFig.3).This potentialconfusionbetweenvibrationonsetanddirectionoftravel maybeinherenttomanytypesofdynamicvibrations,andcould representasignificantdrawbackascomparedtostaticvibrations. Asolutiontothisconfusioninourcasewouldhavebeentopresent thedynamicvibrationsononesideoftheseatonly(i.e.,the vibra-tionscouldtravelfromleftofthecenterfurtheroutwardtothe left,orviceversafromtherightofthecenterfurtheroutwardto theright).However,thiswouldhavelimitedtherangeoftravel (andthereforetheperceptibility)ofthedynamicvibrations,and stilldoesnotdoawaywiththefactthatbydefinitionittakestime

(9)

Fig.5.Top:Meanlateralpositionofthevehicleacrossparticipantsasafunctionoftravelleddistancesincethetake-overrequest(thetake-overrequestwaspresentedata distanceof0m).Theverticallineat223mrepresentsthestationarycarinthemiddleofthelane.Thehorizontaldashedlinesat1.875mand−1.875mrepresentthelane markingsontheroad.Middle:Themeansteeringwheelpositionindegrees.Ifthelanechangewasmadetotheright,thevalueswereinverted.Bottom:pvaluesofpairedt testsforthesteeringwheelangle.Thehorizontaldashedlineinthebottomplotsindicatesapvalueof0.01.

Table1

Meansandstandarddeviations(inseconds)ofthereactiontimesandeffectsizes(Cohen’sdz)betweenconditions.

HAD N-Back Cohen’sdz

StaticM(SD) DynamicM(SD) StaticM(SD) DynamicM(SD) Staticvs.Dynamic HADvs.N-Back

Steer-touch(s) 1.95(0.41) 2.10(0.58) 1.80(0.40) 2.07(0.62) −0.633 0.264

Steer-turn(s) 2.15(0.45) 2.29(0.60) 1.97(0.46) 2.25(0.65) −0.599 0.286

Lanechange(s) 4.31(0.49) 4.48(0.65) 4.13(0.54) 4.40(0.67) −0.515 0.347

Fig.6.Topplots:meangazeandheadheadingasafunctionoftravelleddistancesincethetake-overrequest(thetake-overrequestwaspresentedatadistanceof0m).The shaderepresentsthemean±1standarddeviationacrossthemeanofparticipants.Aheadingofzeroimpliesthattheparticipantwaslookingstraightahead,andapositive valueindicatesthattheparticipantwaslookingtotheleft.Ifthelanechangewasmadetotheright,thevalueswereinverted.Bottomplots:pvaluesofthettestbetween dynamicandstaticpatterns.Inallplots,theverticallineatadistanceof0mindicatesthemomentofthetake-overrequest.Thehorizontaldashedlineinthebottomplots indicatesapvalueof0.01.

(10)

Fig.7. Meanscoresacrossparticipantsontheusefulnessandsatisfactionitems,rangingfrom−2to+2.Theerrorbarsshowthestandarddeviationacrossparticipants.

tobeabletodistinguishthedirectionoftravelofadynamic

vibra-tion.Second,Mengetal.presentedthevibrationsonthewristand

waistinordertoachieveamovementtowardsorawayfromthe

body.Theyfoundthat‘towardsthebody’vibrationselicited

signif-icantlyfasterreactiontimesthanstaticpatterns,but‘awayfrom

thebody’cuesdidnot.Thedirectionalcuesinthepresentstudy

ﬁrstmovedtowardsandthenmovedawayfromthetorso’s

mid-point,possiblyrenderingthe‘towardsthebodycue’ineffective.A

thirdexplanationforthelongreactiontimestodynamicvibrations

couldbethatthenumberofactivatedmotorsintheﬁrst100msof

thedynamicpatternwashalfoftothestaticpattern.Humansare

moresensitivetovibrationsthatstimulatealargerarea,aneffect

alsoknownasspatialsummation(Geschneideretal.,2002).

Col-lectively,thesethreepointsareworthconsideringbydesignersof dynamicvibrotactilewarnings.

Staticanddynamicvibrotactilepatternsalsoyieldeddifferences regardingtheparticipants’orientingbehavior (Fig.6):thestatic patternsevokedafastergazeandheadreactiontowardsthe direc-tionofthelanechange.AcomparisonofFigs.5and6showsthat thesteeringmovementappearstolagabout30m(correspondingto about1sat120km/h)behindtheeyeandheadmovement,which isinlinewithGoldetal.(2013)whofoundthatthegazereaction timewasabout1sfasterthanthehands-ontime.

Generally, before the take-over request, the gaze heading showedalargedeviationaroundzero(Fig.6,right).Shortlyafter thetake-overrequest,theaverageheadinganglearoundzeroand thedropinheadinganglestandarddeviationindicatethatthe par-ticipantsfocusedontheroadahead.Theseresultsareinlinewith astudybyMorandoetal.(2016)whichanalyzedgazedataduring naturalisticdriving.Theseauthorsfoundasimilarshiftofattention towardstheroadaheadafteraForwardCollisionWarning(FCW) wasproduced.Afterattendingtotheroad,participantsgazedinto thesamedirectionasthelanechange(Fig.6).Thismaybea man-ifestationofthefactthat driverstendtolookwheretheysteer (WannandSwapp,2000)oritmaybeaconsequenceofthefact thatdriversscannedthemirrorsoradjacentlanetoseewhetherit isfree.Insummary,participants’eyemovementshowedapattern ofre-attendingtotheroadfollowedbylookingintothedirection ofsteering,withstatictake-overrequestsyieldingfasterreactions thandynamicones.

4.3. Vibrotactilestimulitocomplementvisualandauditory displays

Thereactiontimesoftheparticipantsseemedtobeunaffected byadditionalmentaldemand,whichindicatesthatthevibrotactile stimuliwereeffectiveaswarnings.However,thecorrectresponse rates of thepresent experiment ranged between74% and 94%, withlarge individual differences.These response rates maybe insufﬁcientforsafedrivingifoneofthetwolanesisblocked.As indicatedbySchwalketal.(2015)“underrealconditionsin

pub-lictrafﬁctherecognitionratesshouldbecloseto100%,especially whenitcomestocrucialwarnings”(p.1434).Thus,itseemsthat unimodalvibrotactiletake-overrequests,asimplementedinthe presentexperiment,arenotsuitableforconveyingsemantic infor-mation,like‘changelanestotheleft’.Presumably,symbols(e.g., arrows) orvoicecommandsmaybemoreeffectiveassemantic messages.Salzeretal.(2011),forexample,foundcorrectresponse ratesofapproximately74%fora unimodalvibrotactile stimulus presentedviaeightmotorslocatedontheparticipant’sthigh.The correctresponserateincreasedto95%whenamultimodal stimu-lus(i.e.,vibrotactile,auditory,andvisual)waspresented.Moreover, correctresponseratesof99%werefoundforunimodalvisual stim-uli.However,inSalzeretal.,theparticipants’taskwastoreactto thestimulusasquicklyaspossiblebypressingabutton,afterwhich theyindicatedthedirectionofthevibrationonatouch-screen. Pre-sentingadditionalvisualinformationduringatake-overscenario, whichrequiresconsiderablevisualattentiontotheroad,mightbe lesseffective.

Previousstudies(e.g.,Lifetal.,2014;VanErpetal.,2002Van Erpetal.,2005)havefoundvibrotactilefeedbackwithdirectional cuestobeeffectiveinassistinganoperatorinperformingatask. However,inthesestudiesthevibrotactilefeedbackassisted the operatorinaprimarytask(e.g.,hoveringahelicopter),whereasin thepresentstudythevibrationpatternswerepresentedduringa transitionofcontrol.Possibly,thedriversinourexperimentwere busywithcognitivelyprocessingthetrafﬁcsituation,whichmay havediminishedtheirabilitytorecognizethepatterns.

It maybepossibletoimprovethevibrationpatternssothat theyyieldhighercorrectresponserates.Consistentwith recom-mendationsbyJonesandSarter(2008),ourmotorswereplacedat inter-motordistances(30or50mmlaterallyand45mm longitudi-nally)thatwerelargerthanthetwo-pointdiscriminationthreshold (10mmforsuccessivevibrationsonone’sback;Eskildsenetal., 1969).Inourstudy,thedynamicpattern travelleda distanceof about210mm. Patternswithanevenlargertravellingdistance, like front toback, might be easierto recognize. Schwalket al. (2015)showedthatthecorrectresponseratefora patternthat movedfromthefrontoftheseatbottomtowardsthetopoftheseat backwasbetterrecognizedthanpatternstravellingintheopposite direction.Elsewhere,Schwalketal.(2016)showedanincreased correctresponserateforstaticvibrationswhenasmallernumber ofmotorswasactivated(i.e.,onecolumnofactivatedmotorsonthe leftinsteadoftwo).Furtheradjustmentsofthefrequency, ampli-tude,location,andtimingofthepatternsmightimprovethecorrect responseratesofstaticanddynamicwarnings.

5. Conclusionsandrecommendations

The vibration patterns used in this study were effective as a warning to prompt drivers to quickly reclaim manual con-trol, but participantsdidnotreliably detectthedirectionalcue

(11)

thatwasembeddedinthestimulus.Furthermore,staticpatterns yieldedfasterreactiontimesthandynamicpatterns.Basedonthese ﬁndings,vibrotactilefeedbackmaybeavaluablesupplementto auditoryandvisualdisplays,butwerecommendthatdirectional instructions in a take-over scenarioshould notbe provided by meansofavibrotactileseatalone.

Futurestudiesshouldinvestigatehowthefourdimensionsof vibrotactilestimuli(i.e.,frequency,amplitude,location,and tim-ing)shouldbetunedtomakedirectionalcueseasiertorecognize. In ourstudy, theparticipantsreceived a highnumber of take-overrequests,eachwithanidenticalroadobstruction.Thismay haveallowedthemtopreparetothevibrotactilestimuli.In real-ity,thetake-overrequestswillbemorevariableandoccurwith lowerfrequency.Toinvestigatewhetherthepresentresults gen-eralizetorealdrivingconditions,longdrivingsessionswithrare safety-criticaleventsarerequired.

Acknowledgements

The authors are involved in the European Marie Curie ITN project HFAuto − Human Factors of Automated driving (PITN–GA–2013–605817).

AppendixA. Supplementarydata

Supplementarydataassociatedwiththisarticlecanbefound,in theonlineversion,athttp://dx.doi.org/10.1016/j.aap.2016.12.001.

References

Bainbridge,L.,1983.Ironiesofautomation.Automatica19,775–779,http://dx.doi. org/10.1016/0005-1098(83)90046-8.

Baldwin,C.J.,Spence,C.,Bliss,J.P.,Brill,J.C.,Wogalter,M.S.,Mayhorn,C.B.,Ferris, T.K.,2012.Multimodalcueing:therelativebeneﬁtsoftheauditory,visual,and tactilechannelsincomplexenvironments.ProceedingsoftheHumanFactors andErgonomicsSociety56thAnnualMeeting,1431–1435,http://dx.doi.org/ 10.1177/1071181312561404.

Banks,V.A.,Stanton,N.A.,2015.Keepthedriverincontrol:automating

automobilesofthefuture.Appl.Ergon.53(PartB),389–395,http://dx.doi.org/ 10.1016/j.apergo.2015.06.020.

Beruscha,F.,Wang,L.,Augsburg,K.,Wandke,H.,2010.Dodriverssteertowardsor awayfromlateraldirectionalvibrationsatthesteeringwheel?Proceedingsof the2ndEuropeanConferenceonHumanCentredDesignforIntelligent TransportSystems,227–236.

Brewster,S.,Brown,L.M.,2004.Tactons:stucturedtactilemessagesfornon-visual informationdisplay.AustralasianInterfaceConference2004,18–22.

ERTRACTaskForce,2015.AutomatedDrivingRoadmap(Retrievedfrom)http:// www.ertrac.org/uploads/documentsearch/id38/ERTRAC Automated-Driving-2015.pdf.

Eskildsen,P.,Morris,A.,Collins,C.C.,Bach-Y-Rita,P.,1969.Simultaneousand succesivecutaneoustwo-pointthresholdsforvibration.Psychonom.Sci.14, 146–147,http://dx.doi.org/10.3758/BF03332755.

Faul,F.,Erdfelder,E.,Lang,A.G.,Buchner,A.,2007.G*Power3:Aﬂexiblestatistical poweranalysisprogramforthesocial,behavioral,andbiomedicalsciences. Behav.Res.Methods39,175–191,http://dx.doi.org/10.3758/BF03193146. Gasser,T.,Arzt,C.,Ayoubi,M.,Bartels,A.,Bürkle,L.,Eier,J.,Flemisch,F.,Häcker,D.,

Hesse,T.,Huber,W.,Lotz,C.,Maurer,M.,Ruth-Schumacher,S.,Schwarz,J., Vogt,W.,2012.RechtsfolgenzunehmenderFahrzeugautomatisierung.In: BerichtederBundesanstaltfürStraßenwesen,Heft,p.F83.

Geschneider,G.A.,Bolanowski,S.J.,Pope,J.V.,Verrillo,R.T.,2002.Afour-channel analysisofthetactilesensitivityoftheﬁngertip:frequencyselectivity,spatial summation,andtemporalsummation.Somatosens.MotorRes.19,114–124,

http://dx.doi.org/10.1080/08990220220131505.

Gold,C.,Bengler,K.,2014.Takingovercontrolfromhighlyautomatedvehicles. Proceedingsofthe5thInternationalConferenceonAppliedHumanFactors andErgonomics,3662–3667,http://dx.doi.org/10.1177/0018720816634226. Gold,C.,Damböck,D.,Lorenz,L.,Bengler,K.J.,2013.Takeover!Howlongdoesit taketogetthedriverbackintotheloop?ProceedingsoftheHumanFactors andErgonomicsSocietyAnnualMeeting57(1),1938–1942,http://dx.doi.org/ 10.1177/1541931213571433.

Gold,C.,Körber,M.,Lechner,D.,Bengler,K.,2016.Takingovercontrolfromhighly automatedvehiclesincomplextrafﬁcsituations:theroleoftrafﬁcdensity. Hum.Factors:J.Hum.FactorsErgonom.Soc.58,642–652,http://dx.doi.org/10. 1177/0018720816634226.

Gray,R.,Ho,C.,Spence,C.,2014.Acomparisonofdifferentinformativevibrotactile forwardcollisionwarnings:doesthewarningneedtobelinkedtothecollision event?PLoSOne9,e87070,http://dx.doi.org/10.1371/journal.pone.0087070.

Hancock,P.A.,Mercado,J.E.,Merlo,J.,VanErp,J.B.,2013.Improvingtarget detectioninvisualsearchthroughtheaugmentingmulti-sensorycues. Ergonomics56,729–738,http://dx.doi.org/10.1080/00140139.2013.771219. Ho,C.,Tan,H.Z.,Spence,C.,2005.Usingspatialvibrotactilecuestodirectvisual

attentionindrivingscenes.Transp.Res.PartF:TrafﬁcPsychol.Behav.8, 397–412,http://dx.doi.org/10.1016/j.trf.2005.05.002.

Hwang,S.,Ryu,J.-H.,2010.Thehapticsteeringwheel:vibro-tactilebased navigationforthedrivingenvironment.20108thIEEEInternational ConferenceonPervasiveComputingandCommunicationsWorkshops (PERCOMWorkshops),660–665,http://dx.doi.org/10.1109/PERCOMW.2010. 5470517.

Jones,L.A.,Sarter,N.B.,2008.Tactiledisplays:guidancefortheirdesignand application.Hum.Factors:J.Hum.FactorsErgonom.Soc.50(1),90–111,http:// dx.doi.org/10.1518/001872008x250638.

Kerschbaum,P.,Lorenz,L.,Bengler,K.J.,2015.Atransformingsteeringwheelfor highlyautomatedcars.ProceedingsofIEEEIntelligentVehicleSymposium(IV), 1287–1292,http://dx.doi.org/10.1007/978-3-540-89350-96.

Lif,P.,Oskarsson,P.-A.,Hedström,J.,Andersson,P.,Lindahl,B.,Palm,C.,2014. Evaluationoftactiledriftdisplaysinhelicopter.Hum.Comput.Interact.Adv. Interact.Modal.Techn.,578–588, http://dx.doi.org/10.1007/978-3-319-07230-2.

Llaneras,R.E.,Salinger,J.,Green,C.A.,2013.HumanFactorsissuesassociatedwith limitedabilityautonomousdrivingsystem:drivers’allicationofvisual attentiontotheforwardsroadway.ProceedingsoftheSeventhInternational DrivingSymposiumonHumanFactorsinDriverAssessment,Training,and VehicleDesign,92–98http://drivingassessment.uiowa.edu/sites/default/ﬁles/ DA2013/Papers/014Mueller0.pdf.

Lorenz,L.,Kerschbaum,P.,Schumann,J.,2014.Desigingtakeoverscenariosfor automateddriving:howdoesaugmentedrealitysupportthedrivertogetback intotheloop?ProceedingsoftheHumanFactorsandErgonomicsSociety58th AnnualMeeting,1681–1685,http://dx.doi.org/10.1177/1541931214581351. Louw,T.,Madigan,R.,Carsten,O.,Merat,N.,2016.Weretheyintheloopduring

automateddriving?Linksbetweenvisualattentionandcrashpotential.Inj. Prev.,http://dx.doi.org/10.1136/injuryprev-2016-042155.

Mehler,B.,Reimer,B.,Dusek,J.A.,2011.MITAgeLabDelayedDigitRecallTask (n-back).Workingpaper2011-3B,Retrievedfromhttp://agelab.mit.edu/ system/ﬁles/Mehleretaln-back-white-paper2011B.pdf.

Melcher,V.,Rauh,S.,Diederichs,F.,Bauer,W.,2015.Take-overrequestsfor automateddriving.6thInternationalConferenceonAppliedHumanFactors andErgonomics(AHFE2015),4219–4225,http://dx.doi.org/10.1016/j.promfg. 2015.07.788.

Meng,F.,Spence,C.1,2015.Tactilewarningsignalsforin-vehiclesystems.Accid. Anal.Prev.75,333–346,http://dx.doi.org/10.1016/j.aap.2014.12.013. Meng,F.,Gray,R.,Ho,C.,Ahtamad,M.,Spence,C.,2014.Dynamicvibrotactile

signalsforforwardcollisionavoidancewarningsystems.Hum.Factors:J.Hum. FactorsErgonom.Soc.57,329–346,http://dx.doi.org/10.1177/

0018720814542651.

Meng,F.,Ho,C.,Gray,R.,Spence,C.,2015.Dynamicvibrotactilewarningsignalsfor frontalcollisionavoidance:towardsthetorsoversustowardsthehead. Ergonomics58,411–425,http://dx.doi.org/10.1080/00140139.2014.976278. Merat,N.,Jamson,A.H.,Lai,F.C.H.,Carsten,O.,2012.Highlyautomateddriving,

secondarytaskperformance,anddriverstate.Hum.Factors:J.HumanFactors Ergonom.Soc.54,762–771,http://dx.doi.org/10.1177/0018720812442087. Morando,A.,Victor,T.,Dozza,M.,2016.Driversanticipatelead-vehicleconﬂicts

duringautomatedlongitudinalcontrol:sensorycuescapturedriverattention andpromoteappropriateandtimeresponses.Accid.Anal.Prev.97,206–219,

http://dx.doi.org/10.1016/j.aap.2016.08.025.

Morrell,J.,Wasilewski,K.,2010.Designandevaluationofavibrotactileseatto improvespatialawarenesswhiledriving.2010IEEEHapticsSymposium, 281–288,http://dx.doi.org/10.1109/HAPTIC.2010.5444642.

Petermeijer,S.M.,DeWinter,J.C.F.,Bengler,K.J.,2015.Vibrotactiledisplays:a surveywithaviewonhighlyautomateddriving.IEEETrans.Intell.Transp. Syst.99,1–11,http://dx.doi.org/10.1109/TITS.2015.2494873.

Petermeijer,S.M.,Bazilinskyy,P.,Bengler,K.J.,DeWinter,J.C.F.,2016.Take-over again:InvestigatingmultimodalanddirectionalTORstogetthedriverback intotheloop.Manuscriptsubmittedforpublication.

Prewett,M.S.,Elliott,L.R.,Walvoord,A.G.,Coovert,M.D.,2012.Ameta-analysisof vibrotactileandvisualinformationdisplaysforimprovingtaskperformance. IEEETrans.Syst.ManCybern.PartC(Appl.Rev.)42,123–132,http://dx.doi. org/10.1109/TSMCC.2010.2103057.

Radlmayr,J.,Gold,C.,Lorenz,L.,Farid,M.,Bengler,K.,2014.Howtrafﬁcsituations andnon-drivingrelatedtasksaffectthetake-overqualityinhighlyautomated driving.ProceedingsoftheHumanFactorsandErgonomicsSocietyAnnual Meeting58,2063–2067,http://dx.doi.org/10.1177/1541931214581434. Salzer,Y.,Oron-Gilad,T.,Ronen,A.,Parmet,Y.,2011.Vibrotactileon-thighalerting

systeminthecockpit.Hum.Factors:J.HumanFactorsErgonom.Soc.53(2), 118–131,http://dx.doi.org/10.1177/0018720811403139.

Sayer,T.B.,Sayer,J.R.,Devonshire,J.M.,2005.Assessmentofadriverinterfacefor lateraldriftandcurvespeedwarningsystems:mixedresultsforauditoryand hapticwarnings.ProceedingsoftheThirdInternationalDrivingSymposiumon HumanFactorsinDriverAssessment,Training,andVehicleDesign,218–224

http://drivingassessment.uiowa.edu/DA2005/PDF/32TinaSayerformat.pdf. Schwalk,M.,Kalogerakis,N.,Maier,T.,2015.Driversupportbyavibrotactileseat

matrix−Recognition,adequacyandworkloadoftactilepatternsintake-over scenariosduringautomateddriving.6thInternationalConferenceonApplied

(12)

HumanFactorsandErgonomics(AHFE2015),1427–1434,http://dx.doi.org/10. 1016/j.promfg.2015.07.507.

SchwalkM.,CuiH.,MaierT.,2016.InformationskodierungmittelsTaktiler Sitz-Matrix(TSM)−WieguterkennenwirvibrotaktileMuster?In:Arbeitin komplexenSystemen−Digital,vernetzt,human?!,62.Frühjahrskongressder GesellschaftfürArbeitswissenschaft(GfA),1–6.https://www.researchgate.net/ publication/297155281InformationskodierungmittelsTaktilerSitz-Matrix TSM - WieguterkennenwirvibrotaktileMuster.

SmartEye,2016.SmartEyeProProductSheet,Retrievedfromhttp://smarteye.se/ wp-content/uploads/2016/04/Smart-Eye-Pro.pdf.

Tanikawa,C.,Urabe,Y.,Matsuo,K.,Kubo,M.,Takahashi,A.,Ito,H.,Tajima,K., Kamatani,N.,Nakamura,Y.,Matsuda,K.,2012.Agenome-wideassociation studyidentiﬁestwosusceptibilitylociforduodenalulcerintheJapanese population.Nat.Genet.44,430–434,http://dx.doi.org/10.1038/ng.1109. VanErp,J.B.F.,Veltman,J.A.,VanVeen,H.A.H.C.,Oving,A.B.,2002.Tactiletorso

displayascountermeasuretoreducenightvisiongogglesinduceddrift.In: SymposiumonSpatialDisorientationinMilitarryVehicles:Cuases, ConsequencesandCures,LaCoruna,Spain,pp.15–17http://www.dtic.mil/ dtic/tr/fulltext/u2/p013888.pdf.

VanErp,J.B.F.,VanVeen,H.A.H.C.,Jansen,C.,Dobbins,T.,2005.Waypoint navigationwithavibrotactilewaistbelt.ACMTrans.Appl.Percept.2,106–117,

http://dx.doi.org/10.1145/1060581.1060585.

VanderLaan,J.D.,Heino,A.,DeWaard,D.,1997.Asimpleprocedureforthe assessmentofacceptanceofadvancedtransporttelematics.Transp.Res.PartC: Emerg.Technol.5,1–10,http://dx.doi.org/10.1016/S0968-090X(96)00025-3. Vertanen,Keith,2016.ImplementationoftheNASA-TLXinHTMLand

JavaScript—NASA-TLXGermanversion,link: www.keithv.com/software/nasatlx.

WIVW,2016.WürzburgerInstitutefürVerkehrswissenschaften—SILAB5.0,link:

https://wivw.de/de/silab.

Wann,J.P.,Swapp,D.K.,2000.Whyyoushouldlookwhereyouaregoing.Nat. Neurosci.3,647–648,http://dx.doi.org/10.1038/76602.

Wickens,C.D.,Hollands,J.G.,Banbury,S.,Parasuraman,R.,2012.Engineering PsychologyandHumanPerformance,fourthedition.PsychologyPress, ISBN-13:978-0205021987.

Zeeb,K.,Buchner,A.,Schrauf,M.,2015.Whatdeterminesthetake-overtime?An integratedmodelappraoachofdrivertake-overafterautomateddriving.Accid. Anal.Prev.78,212–223,http://dx.doi.org/10.1016/j.aap.2015.02.023. Zeeb,K.,Buchner,A.,Schrauf,M.,2016.Istake-overtimeallthatmatters?The

impactofvisual-cognitiveloadondrivertake-overqualityafterconiditionally automateddriving.Accid.Anal.Prev.92,230–239,http://dx.doi.org/10.1016/j. aap.2016.04.002.

SebastiaanM.PetermeijerreceivedtheMScdegree(cum laude)inmechanicalengineeringfromDelftUniversityof Technology,Delft,theNetherlands,inApril2014.Heis currentlyaPhDcandidateintheErgonomicsDepartment attheTechnischeUniversitätMünchen,workinginthe HFAutoproject.Withinthisproject,hefocusesonhaptic human-machineinteractioninahighlyautomated vehi-cle.S.M.Petermeijerwonthe2014HumanFactorsPrize withthepapertitled‘Shoulddriversbeoperatingwithin anautomation-freebandwidth?Evaluatinghaptic steer-ingsupportsystemswithdifferentlevelsofauthority’.

StephanCielergraduatedinpsychologyatthe Univer-sityofMünsterin1991.AfterworkingfortheInstitute forTrafﬁcSafetyofTÜVRheinland(Köln)hejoinedthe DivisionCockpitModulesofSiemensVDOinBabenhausen (Germany).TodayheismanagerforHMIandDesignin theInteriorElectronicsSolutionsBusinessUnitof Conti-nentalinBabenhausen(Germany).Hisﬁeldsofactivity includeuserneedsanalyses,conceptresearchforHMI, driverassistance,validationandsimulatorstudies.He ismemberoftheISO/DINworkinggroup‘ManMachine Interface’oftheGermanStandardsCommitteefor Auto-motiveEngineering.

JoostC.F.deWinterreceivedtheMScdegreeinaerospace engineeringandthePhDdegree(cumlaude)fromthe DelftUniversityofTechnology,Delft,theNetherlands, in2004and2009,respectively.Hisresearchinterestis humanfactorsandstatisticalmodelling,includingthe studyofindividualdifferences,driverbehaviormodelling, multivariatestatistics,andresearchmethodology.Heis currentlyassociateprofessoratthefacultyofMechanical, MaritimeandMaterialsEngineeringattheDelft Univer-sityofTechnology.