Delft University of Technology
Does haptic steering guidance instigate speeding? A driving simulator study into causes
and remedies
Melman, Timo; de Winter, Joost; Abbink, David
DOI
10.1016/j.aap.2016.10.016
Publication date
2017
Document Version
Final published version
Published in
Accident Analysis & Prevention
Citation (APA)
Melman, T., de Winter, J., & Abbink, D. (2017). Does haptic steering guidance instigate speeding? A driving
simulator study into causes and remedies. Accident Analysis & Prevention, 98, 372–387.
https://doi.org/10.1016/j.aap.2016.10.016
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.
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.
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
Does
haptic
steering
guidance
instigate
speeding?
A
driving
simulator
study
into
causes
and
remedies
T.
Melman,
J.C.F.
de
Winter
∗,
D.A.
Abbink
DepartmentofBioMechanicalEngineering,DelftUniversityofTechnology,Mekelweg2,2628CDDelft,theNetherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received22May2016 Receivedinrevisedform 17September2016 Accepted13October2016 Keywords:
Behavioraladaptation Hapticsteeringguidance Human-automationinteraction Drivingsimulator
a
b
s
t
r
a
c
t
Animportantissueinroadtrafficsafetyisthatdriversshowadversebehavioraladaptation(BA)todriver assistancesystems.Hapticsteeringguidanceisanupcomingassistancesystemwhichfacilitates lane-keepingperformancewhilekeepingdriversintheloop,andwhichmaybeparticularlypronetoBA. Thusfar,experimentsonhapticsteeringguidancehavemeasureddriverperformancewhilethevehicle speedwaskeptconstant.Theaimofthepresentdrivingsimulatorstudywastoexaminewhetherhaptic steeringguidancecausesBAintheformofspeeding,andtoevaluatetwotypesofhapticsteeringguidance designednottosufferfromBA.Twenty-fourparticipantsdrovea1.8mwidecarfor13.9kmonacurved road,withconesdemarcatingasingle2.2mnarrowlane.Participantscompletedfourconditionsina counterbalanceddesign:noguidance(Manual),continuoushapticguidance(Cont),continuousguidance thatlinearlyreducedfeedbackgainsfromfullguidanceat125km/htowardsmanualcontrolat130km/h andabove(ContRF),andhapticguidanceprovidedonlywhenthepredictedlateralpositionwasoutside alateralbandwidth(Band).Participantswerefamiliarizedwitheachconditionpriortotheexperimental runsandwereinstructedtodriveastheynormallywouldwhileminimizingthenumberofconehits. ComparedtoManual,theContconditionyieldedasignificantlyhigherdrivingspeed(onaverageby 7km/h),whereasContRFandBanddidnot.Allthreeguidanceconditionsyieldedbetterlane-keeping performancethanManual,whereasContandContRFyieldedlowerself-reportedworkloadthanManual. Inconclusion,continuoussteeringguidanceenticesdriverstoincreasetheirspeed,therebydiminishing itspotentialsafetybenefits.ItispossibletopreventBAwhileretainingsafetybenefitsbymakingadesign adjustmenteitherinlateral(Band)orinlongitudinal(ContRF)direction.
©2016PublishedbyElsevierLtd.
1. Introduction
AdvancedDriverAssistanceSystems(ADAS)supportdriversin taskssuchaslane keeping,car following,braking,and obstacle avoidance(e.g.,EichelbergerandMcCartt,2016;Fergusonetal., 2008).Generally,ADASaredevelopedwiththegoaltoincrease comfortandsafety,andnumeroussimulator-basedandtest-track studieshaveindeedshownsuchbenefits(Bengleretal.,2014;Piao andMcDonald,2008).Inreality,however,theanticipatedsafety benefitsare often diminished because drivers showbehavioral adaptation(BA),suchasdrivingwithahigherspeed,drivingcloser toaleadvehicle,performingdistractivenon-drivingtasks,or driv-inglongertripsascomparedtodrivingwithoutADAS(Elvik,2013; Hiraokaetal.,2010;MartensandJenssen,2012;Mehleretal.,2014; OECD,1990;Saad,2006).
∗ Correspondingauthor.
E-mailaddress:j.c.f.dewinter@tudelft.nl(J.C.F.deWinter).
Theabilitytoadaptisintrinsictohumans,andalthough adap-tationcanhavepositiveeffectsincertaincircumstances(e.g.,close followingmaybebeneficialintermsofhighwaycapacity),most transportationresearchers are concernedwithadaptations that degradethesafetybenefitsthatcanbeachievedwithADAS.For example,Sagbergetal.(1996)observedareducedtimeheadway among taxisequipped withanAnti-lockBraking System(ABS), comparedtotaxiswithoutABS.Theirresultssuggestthatthetaxi driversexploitedthefactthatABSreducesthebrakingdistanceby drivingclosertothevehicleinfront.SuchBAwithnegative con-sequenceshasbeenimplicatedinmanytypesofADASincluding notonlyABS,butalsoadaptivecruisecontrol(Panouetal.,2007), lanedeparturewarningsystems(Rudin-BrownandNoy,2002),and collisionavoidancesystems(JanssenandNilsson,1993).
ThepsychologicalmechanismsbehindBAareyettobe eluci-dated,butithasbeenpostulatedthatdriversexhibitatrade-off betweentwoconflictingmotivations,namelyarrivingata desti-nationintime (efficiency)versusavoiding dangeroussituations (safety),andwherebythedriver’slevelofsubjectiverisk(Näätänen
http://dx.doi.org/10.1016/j.aap.2016.10.016 0001-4575/©2016PublishedbyElsevierLtd.
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 373 and Summala,1974; Wilde, 2013,1998), taskdifficulty (Fuller,
2005), or time/safety margins (Gibson and Crooks, 1938; Van Winsumetal.,1999)areimportanthomeostaticvariables. Accord-ingly,driversadoptahigherspeedorashorterheadwaywhenthe drivingtaskbecomeseasier,lessrisky,orlesstemporally demand-ingdue to a changein theroad-vehicle-driversystem,suchas improvedenvironmentalconditions(e.g.,whenaddingroad light-ing;Assumetal.,1999)orincreasedassistanceinthecardriving task(e.g.,whenusingadaptivecruisecontrol;Dragutinovicetal., 2005).
ThemagnitudeoftheBAeffectisthoughttodependonthetime drivenwiththeADAS,thedriver’sattitudetowardstheADAS(e.g., whetherthedriverusesthesystemtodrivetothelimit),driver experience,andthedesignofADAS(Carstenetal.,2012;Saadetal., 2004;Sullivanetal.,2016).Onesupposedlyimportantpredictorof BAisthe‘noticeability’oftheADAS:IthasbeensaidthatADAS which causedirectlynoticeabledifferences inthe road-vehicle-driversystemsufferfromBAtoagreaterextentthanADASthat donot(Elviketal.,2004a,b).Thatis,ifdriversaremoreawarethat ADASinterfereswiththeirdrivingtask,itismorelikelythatthey willadapttheirbehavior.Forexample,largerBAeffectshavebeen demonstratedfordrivingwithanightvisionenhancementsystem thanforanon-visiblefeaturesuchaselectronicstabilitycontrol (e.g.,Hiraokaetal.,2010;Jiménezetal.,2008).Basedonthese find-ingsitisexpectedthatADASthatcontinuouslyinteractwiththe driveraremorelikelytosufferfromBAthanforinstanceemergency systems.
OnetypeofADASthatisgrowinginpopularityandwhichmay beparticularlypronetoBAishapticsteeringguidance.The phi-losophyofhapticsteeringguidanceistousethecontrolinterface asamediumofcooperationbetweenthedriverandanintelligent vehicle,withtheaimtokeep thedriverinformedand involved inthedrivingtask,andtopreventtheout-of-the-loopproblems thatoccurinhands-freeautomateddriving(Abbinketal.,2012; Flemischetal.,2008;GriffithsandGillespie,2005;Johnsetal.,2016; Marsetal.,2014a;O’Malleyetal.,2006;Soualmietal.,2014,see
Petermeijeretal.,2015bforareview).Concretely,haptic steer-ingguidancecontinuouslyassistsdriversinthesteeringtaskby providingtorquesonthesteeringwheelbasedonthetarget steer-ingbehaviorofanautomatedcontroller.Thedrivermay‘relax’his musclesandconformtotheappliedtorque,ormaysteeragainst it.Thus,thehumanandthemachinearejointlysteeringthecar, andthedegreeofsupportcanvaryalongacontinuousscalefrom driver-in-control(i.e.,thedriverhasafirmgriponthesteering wheelandoverridestheappliedtorques)tomachine-in-control (i.e.,thedriverhasaverylightgriponthesteeringwheel). Pre-viousresearchhasshownbeneficialeffectsintermsofimproved lane-keepingperformance,increasedsafetymargins,andreduced self-reportedworkloadfordrivingwithsteeringguidanceas com-paredtounsupporteddriving(Marsetal.,2014b; Mulderetal., 2012;O’Malleyetal.,2006).Insummary,duetothecontinuous interaction,increasedcontrollability,andreducedworkload,haptic steeringguidancemaybehighlysusceptibletoBA.
Recently,researchershavestartedtoinvestigatethehypothesis that the beneficial effects of hapticguidance might be accom-paniedbyunintendedsideeffects.Adrivingsimulatorstudyby
Petermeijeretal.(2015a) foundthatdrivers showeddangerous steering oscillations, alsocalled ‘aftereffects’, after the steering guidancefailedpriortoenteringacurve.Aswithmostresearch onhaptic steering guidance (e.g., Griffiths and Gillespie, 2005; Mohellebietal.,2009;Mulderetal.,2012),thevehiclespeedin thisstudywasheldconstant.Itisyetunknownwhether partici-pantsdrivingwithhapticsteeringguidancewillshowBAinterms ofincreaseddrivingspeedwhentheguidancesystemisactiveand functioningnormally.TheonlystudyonthistopicfoundnoBAwith continuoushapticsteeringguidancecomparedtomanualdriving
(Marsetal.,2014b).Theauthorscomparedtwogroupsof partici-pantsinadrivingsimulator;onegroupdrovewithhapticsteering guidanceandtheotherdrovewithout.Nostatisticallysignificant speed differencewasfoundbetweenthetwogroups;however, duetothebetween-subjectdesign,thisparticularstudymayhave lackedthestatisticalpowertodetectadifferenceinmeandriving speed.
Theaimofthepresentresearchwastwofold.Asindicatedabove, hapticsteering guidanceis a noticeabletype of ADASand may thereforebehighlysusceptibletoBA.Ourfirstaimwastotestthe hypothesisthathapticsteeringguidancecausesBA operational-izedasdrivingspeed.DrivingspeedisaprimemeasureofBAwith strongimplicationsforroadsafety(Elvik,2013):Anincreaseof speedreducesadriver’stimetorespondinanemergencyscenario, increasestheprobabilityofbeinginvolvedinacrash,increasesthe driver’sseverityofinjuryifacrashoccurs,andincreasesthe sever-ityofinjuryof(vulnerable)roadusersthatarehitbythedriver (AartsandVanSchagen,2006;Elviketal.,2004a,b;Hedlund,2000). Oursecondaim,anticipatingonthehypothesizedBAcaused byhapticsteeringguidance,wastoinvestigatetheeffectiveness oftwotypesofhapticsteeringguidancethatweredevelopedto mitigatespeedingwithoutcompromisingthebeneficialeffectsof guidanceonsafetyandcomfort.Thefirstdesign(Band) incorpo-ratesalateralbandwidthwherebytheguidanceengagesonlywhen thevehicledeviatessubstantiallyfromthelanecenter.Thisdesign waspreviouslytestedataconstantdrivingspeedandwasfound tomitigateeffectsof over-relianceincase thesystemsuddenly failed(Petermeijeretal.,2015a).Theseconddesignisalongitudinal boundarysystem(ContRF)thatremovesthecontinuousguidance whendrivingfasterthanapre-definedspeedthreshold.These fun-damentallydifferentsystemswerebothhypothesizedtoreduce speeding:theBand conditionisequivalenttodriving manually unlessmakingalargelateralerror(therebyprovidingguidanceonly whenneeded),andtheContRFconditionprovidesguidancein nor-malconditions,butceasestofunctionwhenthedriveradoptsahigh speed(therebyremovingthebenefitsof guidancewhendriving fast).
Thisstudyevaluateddrivingbehaviorwhendrivingwith hap-ticsteeringguidancesystemsonanarrowroadwithconesalong theentireroad,comparedtounsupporteddriving.Priortoeach guidancecondition,drivers werefamiliarizedwiththeworking mechanismsofthesteeringguidance.ThiswasdonebecauseaBA effectmayappearonlyafteralearningperiodthatallowsdrivers todevelopamentalmodelofthesystem(Beggiato etal.,2015; Bianchi Piccininiet al.,2014;Martens and Jenssen,2012; Saad, 2006;Sullivanetal.,2016).Toenhancethefamiliarizationprocess, eachguidanceconditionwasexplainedtotheparticipantsindetail. Duringtheactualexperiment,driverswereinstructedtodriveas theynormallywouldwhileminimizingthenumberofconehits. Driversreceivedreal-timefeedbackontheirlane-keeping perfor-mance:aconehitwasindicatedbymeansofareddotappearing onthescreen.Theaugmentedfeedback(i.e.,reddots)andnarrow roadwereassumedtoenhancethesubjectiveriskandnoticeability ofthelane-keepingbenefitsofthehapticguidance,andto discour-ageparticipantsfromdrivingatfullspeed(seeZhaietal.,2004for aspeed-accuracytrade-offinlanekeeping).Duetothesefactors,it wasexpectedthatifhapticsteeringguidancesuffersfromBA,this effectwouldbedetectedsooner.Toinvestigatethepotentialrisks ofspeeding,asharpcurvewasintroducedattheendofthetrial trajectory.
The aimof this study wasto investigatethe effectof three differentdesignsofhapticsteeringguidanceonspeeding.Itwas hypothesizedthat whendrivingwithcontinuoussteering guid-anceparticipantswouldadoptahigherspeedthanwhendriving manuallywithoutsupport.Furthermore,alateraland longitudi-nalalternativesteeringguidanceweretested.Bothdesignswere
Fig.1.Simulatorenvironmentincludingthecarfrontandconehitwarning(i.e.,red dot).(Forinterpretationofthereferencestocolourinthisfigurelegend,thereader isreferredtothewebversionofthisarticle.)
hypothesizedtonotsufferfromspeedadaptationswhile retain-inga highlane-keepingperformance comparedtounsupported driving.Inordertoofferacomprehensiveevaluationand compar-isonbetweenconditions,eachdesignwasassessedwithrespectto fivecategoriesofmeasures:speed,lane-keepingaccuracy,safety margin,workload,andsystemacceptance.
2. Method 2.1. Participants
Twenty-fourparticipants(7female)between23and52years old(M=28.0,SD=9.6)withnormalorcorrected-to-normalvision volunteeredfora drivingsimulatorexperiment. Allparticipants hadtheirdriver’slicenseforatleastfiveyears.Inresponsetothe questionofhowoftentheydroveinthepast12months,6 partici-pantsreportedtodriveeveryday,4drove4–6daysaweek,5drove 1–3daysperweek,5droveonceamonth,3drovelessthanoncea month,and1never.Regardingmileageinthepast12months,the mostfrequentlyselectedresponsecategorywas1.001–5.000km (8respondents),followedby10.001–15.000km(6respondents), and25.001–35.000km(3respondents).Inanattempttomeasure participants’familiaritywithautomateddrivingsystems,weasked themwhethertheyhadeverheardoftheGoogleDriverlessCar. Themajorityofparticipants(21of24,or88%)indicated‘yes’tothis question,whichishigherthanapreviouslymeasuredglobal aver-ageof52%obtainedviaaninternationalInternetsurvey(Kyriakidis etal.,2015).
2.2. Apparatus
Theexperimentwasconductedinafixed-basesimulatoratthe ControlandSimulation Department atthefaculty ofAerospace Engineering,Delft UniversityofTechnology.Thesteering wheel waselectronicallyactuatedbyaMOOGFCSECol8000SActuator runningat2500Hz.Vehicledynamicsweresimulatedwitha single-trackmodel (heavy sedan of 1.8mwide), havingan automatic gearbox,and a maximum speed of 160km/h. Thescenery was visualizedusingthreeLCDprojectorswithahorizontaland ver-ticalfield-of-viewofrespectively180◦and40◦.Thevisualswere refreshedat50Hz,whereasthesimulationanddataloggingwere updatedat100Hz.Acarfrontwasvisualizedtofacilitateperception ofthecar’spositionrelativetotheroadboundaries.Carvibrations (‘roadrumble’)weresimulatedwithaseatshakerimplementedin thedriver’sseat.
2.3. Designsofhapticsteeringguidance
InadditiontotheManualcondition,whichsimulatednatural self-alignmenttorques,threedifferentmethodswereusedto pro-videsuperimposedhapticguidancetorquesonthesteeringwheel. Eachof thesethreemethods useda two-levelalgorithm which wasidenticaltopreviouslypublishedresearch(AbbinkandMulder, 2009;Mulderetal.,2008;Petermeijeretal.,2015a).Thefirstlevel calculatedthedesiredsteeringanglebasedonatwo-parameter modelthatpredictsthefuturelateralerrorbetweenthelane cen-terandthemiddleofthecar(efuture lat)andthefutureheadingerror
ofthecar(efuture heading)atalook-aheadtimeof0.7s.Thefirstlevel
wasidenticalforeachofthethreetestedguidanceconditions.At thesecondlevel,thetwovariablescalculatedinthefirststepwere convertedtofeedbacktorquesaccordingtoanalgorithmthatwas differentforeachofthethreeguidanceconditions.
2.3.1. Continuoussteeringguidance(Cont)
ThesystemContformsthebaselineforthehapticsteering guid-ance.Itprovidescontinuousfeedbacktorquesonthesteeringwheel usingthetwo-levelarchitecturedescribed above,forwhichthe secondlevelisshowninEq.1.
Tfeedback=
efuture lat·P+efuture heading·D
·Kf (1)
ThefeedbackgainswereidenticaltoMulderetal.(2008),with theforcefeedbackgain(Kf)=2.0,theproportionalgain(P)=0.08,
andthederivativegain (D) =0.9.
2.3.2. Continuoussteeringguidancewithareducingfeedback gain(ContRF)
The ContRFis a speed-dependent version of thecontinuous guidanceCont.Atspeeds below125km/h theContRFcondition functionsidenticallytoCont.Ifthespeedisgreaterthan125km/h, thefeedbacktorque(Tfeedback)linearlyreducestozero,andbeyond
speedsof130km/hitisidenticaltotheManualcondition(seeEq.
2).TheworkingprincipleofContRFistoremovetheguidancewhen drivingatexcessivespeeds,therebytheoreticallymitigatingspeed adaptation.Theboundaryof125–130km/hwaschosenbasedon resultsofpilotstudies.
Tfeedback=
⎧
⎪
⎪
⎪
⎨
⎪
⎪
⎪
⎩
efuturelat·P+efutureheading·D
·Kf forv < 125
efuturelat·P+efutureheading·D
·Kf· 130−v 5 for125 ≤v ≤130 0 forv > 130 (2)2.3.3. Bandwidthguidance(Band)
Thebandwidthguidancewassimilartothe‘doublebandwidth’ systempreviouslyintroducedbyPetermeijeretal.(2015a).This designwasshowntomitigateover-relianceonhapticguidance, andmaybeaviablesolutiontospeedadaptationaswell.TheBand conditionhastwostatesofoperation.InState1theBandsystem doesnotexertanytorquewhenthevirtualcarisinthelane(i.e., absoluteefuturelatissmallerthan0.2m).Oncetheefuturelatexceeds
thisthreshold,thesystemswitchestoState2.InState2thesystem exertstorqueuntiltheabsoluteefuturelatisbelow0.1mofthelane
center,asshowninEqs.3and4.
Tstate1feedback=
0 for|efuturelat| < 0.2
efuturelat·P·Kf for|efuturelat| ≥ 0.2
(3)
Tstate2feedback=
0 for|efuturelat| < 0.1
efuturelat·P·Kf for|efuturelat| ≥ 0.1
T. Melman et al. / Accident Analysis and Prevention 98 (2017) 372–387 375 Table1
Means(M),standarddeviations(SD),effectsizes(dz),andresultsoftherepeatedmeasuresANOVA(F,p)perdependentmeasure.
Manual(1) ContRF(2) Band(3) Cont(4) Pairwisecomparisons
1–2 1–3 1–4 2–3 2–4 3–4
M(SD) M(SD) M(SD) M(SD) pvalueF(3,69) p(dz) p(dz) p(dz) p(dz) p(dz) p(dz)
Speed
Meanspeed(km/h) 105.7(12.7) 106.4(9.0) 108.3(11.3) 113.3(13.1) p=0.001F=5.96 (0.22) (0.31) xx(0.74) (0.19) x(0.71) (0.43) Percentageoftimeabove125km/h(%) 9.8(22.2) 5.3(10.6) 13.4(24.2) 23.8(31.6) p=6.57×10−4F=6.44 (0.20) (0.28) x(0.72) (0.43) xx(0.76) (0.38)
Lane-keepingperformance
Percentagetimeoff-road(%) 7.21(4.36) 3.32(2.24) 3.92(2.49) 3.40(2.67) p=2.51×10−10F=22.68 xxx(1.58) xxx(1.07) xxx(1.36) (0.37) (0.00) (0.32) Meanabsolutelateralerror(m) 0.087(0.014) 0.074(0.010) 0.086(0.011) 0.074(0.012) p=1.08×10−11F=27.10 xxx(1.47) (0.01) xxx(1.13) xxx(1.33) (0.06) xxx(1.31) Maximumabsolutelateralerror(m) 0.47(0.15) 0.33(0.06) 0.37(0.14) 0.38(0.24) p=2.17×10−7F=14.42 xxx(1.32) xx(0.75) xxx(1.22) (0.35) (0.04) (0.24) Meanlanereturntime(s) 3.19(1.64) 2.03(1.16) 2.00(1.07) 1.72(0.73) p=4.41×10−6F=11.22 xx(0.85) x(0.70) xxx(1.35) (0.02) (0.28) (0.22)
Workload
Meanabsolutefeedbacktorque(Nm) 0.19(0.03) 0.06(0.03) 0.21(0.04) p=1.95×10−22F=179.15 xxx(3.38) x(0.60) xxx(3.90) Steeringreversalrate(Hz) 0.73(0.23) 0.49(0.17) 0.64(0.17) 0.51(0.17) p=6.04×10−14F=35.31 xxx(2.09) (0.57) xxx(1.58) xxx(1.45) (0.17) xx(0.83) NASA-TLX(%) 47.78(12.12) 33.26(11.70) 42.19(16.07) 32.81(13.19) p=2.64×10−6F=11.74 xxx(1.22) (0.42) xx(0.86) (0.58) (0.00) (0.50) Meanabsolutedrivertorque(Nm) 0.75(0.07) 0.72(0.06) 0.75(0.06) 0.76(0.08) p=0.047F=2.78 (0.37) (0.06) (0.16) (0.37) x(0.61) (0.20)
Systemacceptance
Satisfactionscale(−2,2) 0.88(0.52) 0.52(0.69) 0.79(0.73) p=0.076F=2.72 (0.45) (0.10) (0.31) Usefulnessscale(−2,2) 0.98(0.35) 0.83(0.44) 0.92(0.50) p=0.410F=0.92 (0.23) (0.12) (0.17) x:p<0.05,xx:p<0.01,xxx:p<0.001.
Fig.2.Fromtoptobottom:First:curvature(1/curveradius)ofthetrajectory.Second:meanspeedacrossallparticipantspercondition.Thehorizontaldashedlinesindicate thespeedthresholdsoftheContRFcondition.Third:cumulativenumberofconehitsforallparticipantscombinedpercondition.Coneswere8mapart.Theconehitresults fortheContandContRFconditionsareoverlapping.Fourth:thepercentageofparticipantswhodrovefasterthan125km/hpercondition.
2.4. Roadenvironment
Allparticipantsdroveeachtrialonthesamenarrowsingle-lane road(2.2mwideand13.9kmlong),drivinginoneofthefour con-ditions(Manual,ContRF,Band,orCont).Theroadwidthof2.2m andcarwidthof1.8mallowed0.2monbothsidesofthecarbefore aconewouldbehit.Thefirst12km ofthetrajectorycontained threetypesofcurveswithinnerradiiof1500m,750m,and500m, respectively.Thisroaddesignassuredthatnobrakingwasrequired beforecurves(i.e.,curvescouldbetakenfullthrottle),andthatthe lateralaccelerationsstayedatalltimesinthelinearregionwhere thesimulatedcardynamicsarevalid(Dixon,1988).Toinvestigate thedownsidesofpotentialspeeding,asharpcurvetotherightwas introducedattheendofthetrajectory(innerradiusof300mand 300mlong)forwhichthephysicallymaximumspeedwas approxi-mately125km/h.Thatis,drivingfasterthan125km/hwouldresult inthecarveeringofftheroadontheoutsideofthecurve.Before eachexperimentaltrial,participantswerefamiliarizedwiththe guidancebymeansofatrainingrun.Theroadsofthetrainingruns wereidenticaltothefirstthreequarters(10.5km)oftheroadinthe subsequentexperimentaltrials.Speedperceptionwasenhanced bymeansoftreesalongsidetheroad.Coneswereplacedalongthe entireroadwithadistanceof8mbetweencones.Aconehitwas visualizedwithareddotonthesidewherethecarhitacone(Fig.1). Noon-roadobstaclesandnotrafficweresimulated.
2.5. Experimentaldesign
The four conditions were presented in a counterbalanced within-subjectsdesign.Priortotheexperimentparticipantsread and signed an informed consent form, explaining thepurpose, instructions, and procedures of the study. Participants were informedabouttheavailabilityofeachsteeringguidanceandwere toldtokeep bothhands onthesteering wheel ina ten-to-two positionatalltimes.Participantswereinstructedtodriveasthey normallywouldandtominimizethenumberofconehits.Nospeed advicewas given and any questions regarding speed were not answered.
Beforeenteringthedrivingsimulator,participantscompleteda questionnaireregardingtheirdrivingexperienceaswellasaDriver BehaviourQuestionnaire(DBQ)containingsevenviolationitems (DeWinterandDodou,2016).Apreviousmeta-analysisindicated thattheDBQviolationsscalehasamoderatelystrongrelationship (r=0.24)withrecordedmeasuresofspeedandspeeding(DeWinter etal.,2015).
Priortoeachtrial,atrainingrunofapproximatelysixminutes wasperformed(i.e.,fixeddistanceof10.5km).Sixminuteswas consideredsufficienttobecomefamiliarwithaguidancesystem (McGeheeetal.,2004).Toenhancethefamiliarizationprocess,two actionsweretaken.First,theexperimenterexplainedthe work-ingmechanismofeachguidancesystem,butnottheunderlying hypothesis.Second,participantswerestimulatedtoexperiencethe guidance’sworkingmechanismbyallowingthemtodrivewithout negativeconsequences(i.e.,conehitswerenotcountedbutstill visualized).Toemphasizetheimportanceofunderstandingeach guidancecondition,theexperimenterorallymotivatedthedriver toexperiencethemechanismofeachguidanceconditionatleast once.ForContRF,thismeantthatthedrivingspeedwasatleast onceabove130km/h,sothattheparticipantscouldfeelthe steer-ingguidancebeingabsentwhendrivingfast.ForContthiscame downtodrivingwithlargelateralerrorstofeelthefeedbackforce increasing,andforBanddriverswereaskedtoletgoofthesteering wheeltoobservethattheguidanceturnsonjustbeforehittingthe cones.
Aftereachtrial,participantswereinformedaboutthenumber ofconehitsandwererequestedtostepoutofthesimulatorfora 5minbreakandtofilloutthreequestionnaires:aNASATaskLoad Index(NASA-TLX)(HartandStaveland,1988)toassessworkload, anacceptancequestionnaire(VanderLaanetal.,1997)toassess satisfactionandusefulnessoftheguidance,andasimulator sick-nessitem.Inthelatter,participantsneededtoindicatewhether theywerefeelingsimulatorsicknessonascalefrom1to6(1=not experiencinganynausea,nosignofsymptoms,2=arising symp-toms(likeafeelingintheabdomen),butnonausea,3=slightly nauseous,4=nauseous,5=verynauseous,retching,6=vomiting). Aresponseof4orhigherwouldstoptheexperiment.Thetotal
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 377
Fig.3. Theparticipants’meandrivingspeeds(km/h)fortheContRFconditioncomparedtoContcondition.Thebluelineindicatesequalspeedforbothconditions.The differentmarkersdesignatethenumberoftimesaparticipantexceededthe125km/hspeedthresholdwhendrivingwithContRF.18outofthe24participantsdrovefaster withContthanContRF.10participantsexperiencedthereductionofContRFguidance(circlesandtrianglescombined).
experiment,includingfillingoutallquestionnaires,took approxi-mately1.5hperparticipant.
2.6. Dependentmeasures
Thedatameasuredonthefirstandlast400mofthetrajectory werediscarded,becauseoftheinitialaccelerationsandfinal decel-erationsofthesimulatedvehicle.Thisresultedina13.1kmlong trajectorythatwasusedintheanalysis.Thedependentmeasures thatwerecalculated werecategorizedintospeed, performance, workload,safety,andsystemacceptance.
2.6.1. Vehiclespeed
MeanSpeed(km/h).ThiswastheprimaryBAeffectofinterest. PercentageofTimeAbove125km/h(%).The125km/hthreshold correspondstothelowerspeedthresholdoftheContRFguidance.
2.6.2. Performance
PercentageTimeOff-Road(%),whichistheamountoftimethat thecardrivesoutsidetheconeboundaries(i.e.,themiddleofthe cardeviatesmorethan0.2mfromthelanecenter),expressedasa percentageofthetotaldrivingtime.
MeanandMaximumAbsoluteLateralError(m).Theabsolute lat-eralerrorwasdefinedasthedistancefromthemiddleofthecar towardsthecenterofthelane.Theabsolutelateralerrorandtime off-roadaremeasuresoflane-keepingaccuracy.
MeanLaneReturnTime(s).Thismeasurerepresentsthetime fromthemomentthecarcrossestheconeboundary(i.e.,centerof thecarwithin0.2mofthelanecenter)tothemomentofreturning thecarbackwithintheconeboundariesforatleast5s.Thisserves asmeasureofcontrollability.
2.6.3. Workload
NASA-TLX Subjective Workload (%). After each trial, partici-pantswereaskedtoindicatetheirworkloadonsixitems:Mental Demand,PhysicalDemand,TemporalDemand,Performance,Effort, andFrustration.Itemswerescoredona21-pointscalefromvery lowtoveryhigh,exceptforPerformance,whichrangedfrom per-fecttofailure.Theoverallworkloadwascalculatedasthearithmetic meanofthesixitems(Cain,2007;HartandStaveland,1988).
Steering ReversalRate (reversals/s).Thesteering reversalrate wasdefinedasthenumberoftimesthatthesteeringwheelwas reversedbyamagnitudegreaterthan2deg(McLeanandHoffmann, 1975).Itwascalculatedbydeterminingthelocalminimaand max-imaofthesteeringwheelangle,andifthedifferencebetweentwo adjacentpeakswasgreaterthan2deg,itwascountedasareversal. Thismeasurecanbeconsideredanobjectiveindicatorofworkload (Johanssonetal.,2004).
MeanAbsoluteFeedbackTorque(Nm).Thefeedbacktorqueisthe torquesuperimposedonthedriverbythehapticsteering guid-ance.Ahighmeanfeedbacktorquemeansthatmoreguidancewas applied.
Mean Absolute Driver Torque(Nm). The driver torque is the torqueappliedbythedriveronthesteeringwheel,andwas con-sideredasameasureofthedriver’sphysicaleffort.
2.6.4. Safetymargins
Median Time toLine Crossing (TLC)(s). Themedian TLC was approximatedusingthelateralspeedandlateralacceleration(Van Winsumetal.,2000).TheTLCwassetat0swhendrivingoutside thelaneboundaries.BecauseTLCvariesmuchduringlane keep-ing,anadditionalfine-grainedanalysiswasperformedbybinning theTLCvaluesintofourgroupsofsafetymargins:outofbound (TLC=0s),lowsafetymargin(0s<TLC≤2s),moderatesafety mar-gin(2s<TLC≤4s),andhighsafetymargin(TLC>4s).TheseTLC binswereroughlybasedonpreviousstudieswhichhaveassessed
Fig.4.Speeddistribution(km/h)ofallparticipantscombined.Thebinwidthis5km/h.Thefractionisplottedinthemiddleofeachbin.Theredandgreenlinesindicatethe speedthresholdsofContRF.Thesumofallfractionsequals1foreachcondition.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderisreferredto thewebversionofthisarticle.)
TLCortime-to-collision(TTC)intermsofsubjectiverisk(Kondoh etal.,2006),criteriafordriverimpairment(Brookhuisetal.,2003), andself-chosenocclusiontimes(Godthelpetal.,1984).
2.6.5. Systemacceptance
Anacceptance questionnaire(VanderLaanetal.,1997)was usedtoassesssystemacceptanceontwodimensions,ausefulness scaleand an affectivesatisfaction scale.Thisquestionnaire con-sistedofnineitems,scoredbetween+2and−2.Theusefulnessscale wasobtainedbytakingtheaveragescorefortheitems: Useful-Useless, Bad-Good*, Effective-Superfluous, Assisting-Worthless, andRaisingAlertness-Sleep-inducing.Thesatisfactionscalewasthe averagescorefortheitems:Pleasant-Unpleasant,Nice-Annoying, Irritating-Likable*, and Undesirable-Desirable*.Appropriatesign reversalswereconductedfortheitemsindicatedwithanasterisk.
2.7. Statisticalanalyses
Foreachdependentmeasure,amatrixof24×4numberswas obtained(24participantsand4conditions).Thismatrixwas rank-transformedaccordingtoConoverandIman(1981)toaccountfor possibleviolationsoftheassumptionofnormalityassociatedwith parametrictests.Therank-transformedmatrix,consistingof num-bersfrom1to96,wassubmittedtoarepeatedmeasuresANOVA withthefourconditionsaswithin-subjectsfactor.Bonferroni cor-rectionswereappliedtothesixpairwisecomparisonsbetweenthe conditions.Thedzeffectsizesforpairwisecomparisonswere
cal-culatedusing Eq.5 (Faul etal.,2007), where x−y is themean ofthe differenceand x−y is thestandard deviation ofthe
dif-ference.Redundancies/associationsbetweendependentmeasures wereassessedbymeansofSpearmanrank-ordercorrelation coef-ficients.
dz= |x−y|
x−y (5)
3. Results
During the training run, all participants had experienced the mechanism of each guidance condition at least once. This exploratorybehaviorduringtrainingwasnotanalyzed.
3.1. Vehiclespeed
Table1showsthemeansandstandarddeviationsforall depen-dentmeasures,theresultsofrepeatedmeasuresANOVA,andthe pairwisecomparisons.Participants’meanspeedsweresignificantly different between the four conditions, F(3,69)=5.96, p=0.001. WhensupportedbyCont,participantsdrovesignificantlyfaster(on averageby7km/h,withmediumeffectsizes)comparedto Con-tRFandManual.Nostatisticallysignificantspeeddifferenceswere observedbetweenManualandthetwoguidanceconditionsthat werehypothesizednottosufferfromBA(i.e.,ContRFandBand).
Fig.2showsthemeanspeed,cumulativenumberofconehits,and percentageofparticipantsexceedingthe125km/hthresholdasa functionoftravelleddistance.Itcanbeseenthatparticipantsinall fourconditionsdrovefasteronthelongstraightthanontheother partsoftheroute.ThedifferenceinmeanspeedsbetweenContand thethreeotherconditionsoccurredbothonstraightsandincurves (Fig.2).
Theaverage(SD)completiontimes ofthe13.1km trajectory were440s(52),435s(39),428s(45),and411s(48),forManual, ContRF,Band,andCont,respectively.Notethatthemeanspeed andcompletiontimesshowaperfectnegativeSpearman correla-tion(=−1),andsothecompletiontimeswerenotsubjectedto statisticaltests.
Participants in the ContRFcondition drove slower than the 125km/hthresholdforonaverage94.7%ofthetime,effectively resultinginanidenticalguidanceasthecontinuousguidance.A majorityof14outof24participantsintheContRFconditionnever exceededthe lower speed thresholdof 125km/h (Fig.3).Even thoughContRFandConteffectivelywereidenticalconditionsfor mostofthe drivingtime,the speeddistributionbetweenthese twoconditionswasnotablydifferent(Fig.4).Fig.4showsadrop fortheContRFjustbeforethelowerspeedthreshold,whereasfor
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 379
Fig.5. MeanscoresontheNASA-TLX.x:p<0.05,xx:p<0.01,xxx:p<0.001.
theContcondition,nosuchdropcanbeseen.Fig.4furthershows thatalargerfractionofthespeeddistributionoftheCont condi-tionislocatedabovethe125km/hthresholdascomparedtothe threeotherconditions,aneffectthatisstatisticallysignificantwith respecttotheManualandContRFconditions(Table1).
3.2. Performance
All three steering guidance conditions yielded strongly improvedlane-keepingperformancecomparedtotheManual con-dition,intermsofalowertimeoff-road,lowermaximumabsolute lateralerror,andlowerlanereturntime(Table1;Fig.2).Bandand Manualyieldedsignificantlyhighermeanabsolutelateralerrors thanContRFandCont(Table1),whichmaybecausedbythefact thatBandprovided noguidanceforonaverage84%ofthe driv-ingtimeandthereforemostlyfunctionedidenticallytotheManual condition.Nostatisticallysignificantdifferencesinlane-keeping performancewerefoundbetweenContRFandCont.
3.3. Workload
Table1showsthattheself-reportedworkload(NASA-TLXscore) andobjectiveworkload(steeringreversalrate)weresignificantly higherforManualthanforContandContRF.Foreachofthesix NASA-TLXitems,theManualconditionyieldedhigherworkload scoresthantheContandContRFconditions,withstatistically sig-nificanteffects fortheMental Demand,Performance,andEffort items(Fig.5).Furthermore,theBandconditionyieldedsignificantly higherMentalDemandthantheContcondition.
Themeanfeedbacktorqueprovidedbytheguidancewas signif-icantlydifferentbetweenallconditions.Bandyieldedsignificantly lessfeedbacktorque(M=0.06Nm)thanthecontinuousguidance conditions(ContRFM=0.19NmandContM=0.21Nm).No feed-backtorquewasappliedduringtheManualcondition.Moreover, theresultsshowedthatlowerphysicaleffort(drivertorque)was obtainedfordrivingwithContRFthanfordrivingwithCont guid-ance.
3.4. Safetymargins
SubstantiallyhighersafetymarginsintermsofmedianTLCwere foundforContRFandContcomparedtoManualandBand(Table2). Additionally,slightlyhighersafetymarginswereadoptedfor Con-tRFthanCont,althoughnotstatisticallysignificant(p=0.071).In theContRFandContconditions,participantsdrovelessoftenwith alowsafetymarginbutmoreoftenwithahighsafetymargin,as comparedtotheManualandBandconditions.Overalldrivingwith
ContRFandContresultedinthehighestsafetymarginsintermsof medianTLCandTLC>4s.
3.5. Systemacceptance
Table1showsalowersatisfactionscoreforBandthanfor Con-tRFandCont,althoughnotstatisticallysignificant(F(2,46)=2.72, p=0.076).Nodifferencewasfoundfortheusefulnessscaleeither (F(2,46)=0.92,p=0.410).
3.6. Sharpcurve
Theperformancemeasureswerecalculatedseparatelyforthe 300mlongsharpcurvesegmentattheendofthetrajectory.The resultsdidnotshowsignificantdifferencesinlane-keeping perfor-mancebetweenthefourconditions.Nevertheless,thetwolargest absolutelateralerrors(1.4mand0.6m,respectively)werefound forthetwoparticipantsdrivingintheContcondition.Thesetwo participantsadoptedrelativelyhighspeedsattheentranceofthe sharpcurveof135km/h(rank1/96)and117km/h(rank12/96), respectively.Fig.6shows(1)thecurvature,(2) themeanspeed (km/h), (3)themeanabsolutelateralerror(m),(4)themedian TLC(s),and(5)thestandarddeviationofthesteeringwheelangle amongtheparticipantsasafunctionoftravelleddistanceinthe sharpcurve.Thesharpcurveresultedintwodistinctpeaksinthe meanabsolutelateralerror.Thefirstpeak(about 10mintothe curve)iscausedbymostparticipantsslightlycuttingthecurveon theinside,whereasthesecondpeak(70mintothecurve)ismainly causedbymostparticipantsveeringtotheoutsideofthecurve. Criticalsafetymargins(medianTLC<1s)wereobservedforall con-ditionswhenenteringthesharpcurve.Forcontinuousguidance, therewerelargesteeringangledifferencesbetweenparticipants. Largemeanmaximumabsolutelateralerrors wereobtainedfor theCont(0.20m)andManual(0.19m)conditionscomparedtothe ContRF(0.16m)andBand(0.16m)conditions.
3.7. Supplementaryanalyses
TheSpearmancorrelationcoefficientsbetweenthemeanspeed andthemeanabsolutelateralerrorwere0.40,0.31,0.14,and0.11 fortheManual,ContRF,Band,andContconditions,respectively (see Appendix A). This suggeststhat driving speed moderately yetconsistentlyinfluencedtaskperformance,presumablydueto a speed-accuracytrade-off(cf.Zhaiet al.,2004).In the simula-torsicknessitem,noneoftheparticipantsresponded4(nauseous) orhigher,andthuseveryonefinishedtheexperiment.Specifically, thenumberofresponsesbeing1(notexperiencinganynausea),2
Fig.6. Meanspeed(km/h),meanabsolutelateralerror(m),medianTLC(s),andstandarddeviationofthesteeringwheelangle(deg)amongtheparticipantsasafunction oftraveleddistanceinthesharpcurve.Thetopfigureshowsthecurvature(i.e.,1/radiusinmeters).
(arisingsymptoms),and3(slightlynauseous) were61,27, and8, respectively.TheSpearmancorrelationcoefficients betweenthe meanspeedontheonehand,andthemeanDBQviolationsscore, thedrivingfrequency,andthemileageinthelast12months,onthe other,rangedbetween−0.06and0.25(AppendixA).When averag-ingthespeedacrossthefourconditions,theDBQ-speedcorrelation was0.24,whichisinlinewithapreviouslypublishedmeta-analysis (DeWinteretal.,2015)butnotsignificantlydifferentfromzero (p=0.262).Thesefindings suggestthatthedegreeofbehavioral adaptationisnotassociatedwiththesepersonalcharacteristicsin apracticallysignificantmanner.
4. Discussion 4.1. Mainresults
Theaimofthisstudywastoinvestigatetheeffectsofhaptic steeringguidanceonspeeding,andtoevaluatetwovariationsof hapticdriversupportthatweredesignedtomitigatesuch speed-ing.Tounderstanddrivingbehaviorbetter,wealsoassessedthe effects onlane-keepingperformance, safety margins,workload, anddriveracceptance.ThemeanspeedsintheManual,ContRF, Band,andContconditionswere105.7,106.4,108.3,and113.3km/h, respectively,withstatisticallysignificantdifferencesbetweenCont andManualandbetweenContandContRF.Theseresultsconfirm thehypothesisthat continuoushaptic steeringguidancecauses driverstodrivefaster,whichdiminishesthesafetybenefitsofthis assistivetechnologyascomparedtofixed-speedsimulatorstudies. TheresultsareinaccordancewithresultsfromearlierBAstudies regardingothertypesofADAS,suchasobstacleavoidancesystems andnightvisionenhancementsystem(Hiraokaetal.,2010;Janssen andNilsson,1993).
Thelaterallyadjusted(Band)andlongitudinallyadjusted (Con-tRF)guidanceconditionsthatweredesignedtomitigatespeeding werebothsuccessfulinachievingthisobjective,whileretaininga highlane-keepingperformance.ComparedtotheManualandBand
conditions,participantsdrivingwithContandContRFwere bet-terabletocenterthecarinthemiddleoftheroadatareduced workload.Theresultsfurthershowedthatcomparedtothe Man-ualcondition,Contprovidedincreasedsafetymarginsintermsof TLCdespiteahighermeanspeed(whichcorrelatesnegativelywith TLC,seeSupplementarymaterials),signifyingthatdrivers could affordtosafelyincreasetheirspeedduetothebenefitsofferedby thehapticsteeringguidance.ContRFhadevenslightlyhighersafety marginsthanCont,possiblyduetothelowerdrivingspeedinthis condition.
4.2. EffectivenessofContRFinpreventingspeeding
EventhoughContandContRFwereeffectivelyidenticalsystems for95%ofthedrivingtime,ContRFsuccessfullypreventedspeed adaptation.TheContRFthresholdof125km/hwaswellabovethe averagedrivingspeedof108.4km/h,and14ofthe24participants neverexperiencedthereductionofguidanceduringtheirtrials(i.e., theirspeedwasalwaysbelow125km/h).Theeffectivenessofthe ContRFguidancedespitethefactthatdriversonlyrarely experi-enceditsuggeststhatBAdoesnotnecessarilymanifestitselfas afunctionofcurrentADASinterventionandvisibilitybutrather thatexpected(loss)offunctionalityoffersaremedyagainstBA.The effectsmaybeexplainedwiththehelpofthetheoriesofsafety marginsandriskcompensationintroducedabove:arguably,drivers intheContconditionspeededupcomparedtotheManual condi-tionbecausetheguidanceloweredtheirsubjectiveriskleveland increasedsafetymargins.WithContRF,participantsdidnot expe-riencesuchareductioninsubjectiveriskbecausetheyhadbeen instructedandtrained thatthebenefitsofthis assistive system woulddisappearwhendrivingfast.
TheworkingmechanismoftheContRFsystemmaybefurther explainedbymeansofananalogypreviouslyintroducedbyWilde (1998):ontheonehandengineersmakedrivingsaferbyoffering forgivenessincaseofanaccident(e.g.,seatbelts,airbags, crash-worthycar design, etc.),yet onthe otherhand theymake the
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 381 Table 2 Means (M ), standard deviations (SD ), effect size s (dz ), and results of the repeated measures ANOVA (F , p ) for time to line crossing (TLC). Manual (1) ContRF (2) Band (3) Cont (4) Pairwise comparisons 1–2 1–3 1–4 2–3 2–4 3–4 M (SD ) M (SD ) M (SD ) M (SD ) p value F (3,69) p (dz ) p (dz ) p (dz ) p (dz ) p (dz ) p (dz ) Median TLC (s) 1.32 (0.48) 2.09 (0.67) 1.36 (0.46) 1.89 (0.65) p = 4.73 × 10 − 17 F = 48.86 xxx (1.97) (0.07) xxx (1.56) xxx (1.90) (0.56) xxx (1.33) Percentage low safety margin (0 s < TLC ≤ 2 s) (%) 54.20 (6.59) 46.87 (7.01) 56.32 (6.78) 49.27 (7.57) p = 1.61 × 10 − 13 F = 33.67 xxx (1.79) (0.45) xx (0.75) xxx (2.06) x (0.63) xxx (1.10) Percentage moderate safety margin (2 s < TLC ≤ 4 s) (%) 14.08 (2.68) 15.97 (1.47) 13.90 (2.26) 15.51 (1.75) p = 2.48 × 10 − 7F = 14.24 xx (0.83) (0.23) x (0.62) xxx (1.20) (0.31) xxx (1.10) Percentage high safety margin (TLC > 4 s) (%) 24.51 (6.46) 33.84 (6.84) 25.86 (5.68) 31.82 (7.22) p = 2.07 × 10 − 16 F = 45.83 xxx (2.46) (0.23) xxx (1.66) xxx (1.80) (0.52) xxx (1.03) x: p < 0.05, xx: p < 0.01, xxx: p < 0.001. Note : The results for TLC = 0 are identical to the time off-road results in Table 1 .
consequencesofdangerousbehaviormoresevere(e.g.,by imple-mentingspeedbumps).Asimilartypeofconflictingsafetypolicy appliestotheContRFsystem,whereontheonehanddrivingsafety isenhancedbyofferingguidanceonthesteeringwheel,yet speed-ing is discouraged bytaking away the sameguidance. Perhaps peopleneedsuchopposingmotivationtouseADASinaresponsible manner.
Atpresent,itisdifficulttoestablishwhethertheeffectivenessof ContRFiscausedbyconscious(‘explicit’)mechanismsorby uncon-scious(‘implicit’)mechanisms(cf.Evans,2008).Regardingimplicit mechanisms,itmaybearguedthattheContRFsystemgives physi-calfeedbackabouttheobjectivelevelofrisk(speed)atasubcortical neuromuscular level(cf. Abbinket al.,2011).Loosely speaking, becauseparticipantsreceivedthehapticfeedbackdirectlyontheir hands,theymayhavereflexivelyandhabituallyrespondedtothis feedback,withoutbeingconsciouslyawareofthis.Alternatively, participantsmayhavebeenexplicitlyawareofthefactthatthe hapticfeedbackdisappearswhendrivingfast,eitherthroughthe trainingtheyhadreceivedorthroughthefactthatthe disappear-anceoffeedbackisemotionallyarousingandleavesaconsciously accessible trace in memory. Futureresearch should investigate which implicitorexplicitmechanismsareunderlyingfactorsin BApreventiontechnologies.Forexample,togainmoreinsightinto theexplicitfactorsbehindabehavioralchange,averbalprotocol methodcouldbeused(Banksetal.,2014).
4.3. SpeedparametersoftheContRFsystem
TheContRFsystemdoesnotnecessarilyrepresenttheoptimal solutiontopreventBAandmayberefinedinvariousways.Inthis study,thelowerspeed thresholdwassetfairlyhigh(125km/h) withthereductionoccurringoverarelativelysmallspeedrange (125–130km/h),inordertokeepthebenefitsofcontinuous guid-anceandtoensureanoticeablefeedbackreduction.Thequestion remainswhatwouldhappenifonechangesthesedesign param-eters. For example,it ispossibletolower thespeed thresholds towardstheaveragespeedinmanualdriving,sothatalmostall par-ticipantshavetomakeadecisionbetweenusingguidanceversus adoptingahighspeed.Similarly,itispossibletoconceiveasystem thatreducestheguidanceoverabroadspeedrangesothateach driverhastoachieveatrade-offalongacontinuumbetweensafety andefficiency.Thesetopicscanbeaddressedinfutureresearchto improvetheunderstandingofBApreventingtechnologies.
4.4. Effectivenessofthebandsystem
In accordance withpreviousresearch,drivingwithboth the continuousguidanceandthebandwidthguidancewasfoundto improvedrivers’lane-keepingperformancecomparedtomanual driving(Flemischetal.,2008;Kienleetal.,2012;Marchal-Crespo et al.,2010;Petermeijeret al.,2015a).More specifically,in our study,theBandconditionyieldedsubstantiallyimprovedtime off-roadcomparedtotheManualcondition,butitdidnotimprovethe meanabsolutelateralerror.Furthermore,nostatisticallysignificant differenceinoverallself-reportedworkloadwasfoundbetween ManualandBand.Theseeffectscanbeexplainedbythefactthat BandandManualwerelargelyidenticalwhendrivinginsidethe lane,withnohapticguidanceofferedin theBandconditionon averagefor84%of thetime.Theworkingprincipleof theBand systemcorrespondstomarketedlane-keepingassistancesystems (e.g.,Daimler,2013;VolvoCarCorporation,2015),whichalsoguide thedriverawayfromlaneboundariesratherthantowardsthelane centerastheContsystem.
ThefactthattheBandsystemcloselyresemblesmanual driv-ingforthemajorityofthedrivingtimemaybethereasonthat itpreventsBAintheformofspeedingandaftereffectswhenthe
systemfailsordisengages.Thelatterstatementisinlinewitha previousfieldstudyconductedbyBreyeretal.(2010),whichdid notfindevidenceofadverseaftereffectsforacorrectivesteering systemthatwasdeactivatedafteraprolongedexposuretothe sys-tem.OurbandwidthconditionfunctionedsimilarlytoBreyeretal.’s correctivesteeringsystembysupportingthedriverwhen cross-inga laneboundary,yetallowingthedrivertosway withinthe laneboundaries.Insummary,thesatisficingapproach(i.e., keep-ingthedriverfullyinchargewhendrivinginthelane;seealso
Goodrichetal.,2000;Summala,2007),asopposedtothe optimiz-ingapproach(i.e.,continuousguidance),hastheadvantagethatno BAoccursandnoadversesteeringaftereffectsareevokedduringa suddentransitionofcontroltomanualdriving(Breyeretal.,2010; Petermeijeretal.,2015a).However,thisoccursatthecostofan ele-vatedworkloadandworselane-centeringperformancecompared tocontinuousguidance.
4.5. Experimentalconditionsthatgiverisetobehavioral adaptation
Thespeedadaptationof7km/hforcontinuousguidanceis dif-ferentfrom findingsby Marsetal. (2014b)who didnotfind a statisticallysignificantspeeddifferencebetweenagroupof12 par-ticipantsdrivingwithhapticsteeringguidanceandanothergroup of12participantsdrivingmanually.First, thefocusof theMars etal.studyisdifferentfromours:whereasMarsetal.measured howdrivingspeedevolveswithpracticeacrosstwelve4.3kmlaps over twodays, thepresent studyinvestigated more immediate speed adaptations in 24.4km of driving percondition(10.5km trainingplusanexperimentaldriveof13.9km).Second,although abetween-subjectsdesignismethodologicallystrongbecauseit preventscarry-overeffectsbetweenconditions,largeindividual differencesinspeedmakeitlesslikelytoobservestatistically sig-nificanteffects in a between-subjects design as compared to a within-subjectdesign.Third,inourexperiment,theroadwas nar-rowandsalientconcurrentperformancefeedbackwasprovided bymeansofareddotwhenhittingacone,whereasinthestudy byMarsetal.thiswasnotthecase.Duetothenarrowroadand knowledge-of-resultsfeedback,drivers canbeassumed tohave beenwellawareoftheimprovedlane-keepingperformance facili-tatedbyhapticguidance.Thenoticeabilityoftheguidancebenefits combinedwiththeextendedfamiliarizationperiodcouldexplain whystrongspeeddifferenceswereobservedinthepresent1.5h longexperiment.
4.6. Risksofbehavioraladaptation
Despiteitshigher speed,Contyielded highersafetymargins andbetterlane-keepingperformance(intermsoflowertime off-road,maximumlateralerror,andabsolutelateralerror)thanthe Manualcondition.Thisraisesthefundamentalquestion:whyisBA regardedasundesirableifdrivingperformanceandsafetymargins areactuallyimproved?Whenhapticsteeringguidanceiswithin its operational limits it can indeed be considered favorable to unsupporteddriving.However,whenhapticsteeringguidanceis outsideitsoperationallimitsahigherspeedimplieshighercrash risk,higherinjuryseverity,andlowertimetorespond.Infact,for agivenspeedthecrashriskmaybeevenworsethaninmanual drivingduetotheaforementionedaftereffects(Petermeijeretal., 2015a).Theadverseeffectsofspeedingwereillustratedbyalarge laneexitwhenenteringasharpcurveathighspeed(Fig.6). Hav-ingtheunrealistictrustthathapticsteeringguidance(oranyother intelligentvehicleorautomateddrivingsystemforthatmatter)can anticipatesharpcurvesatalltimes,mayleadtodangerous situa-tions.Thesharpcurveismerelyoneexample;therearenumerous otherexampleslikesensorfailure,anobstacleontheroad,or
com-puterfailure,thatcanunexpectedlypushhapticsteeringguidance outside its operationalenvelope. Thephilosophy behindhaptic steeringguidanceistoincorporatethebestofboth—ahuman’s cre-ativesolutionscombinedwithamachine’saccurateandconsistent performance.However,ifdriversover-trustthemachineandadopt anexcessivespeedthecalibrationinthisteamperformanceisoff, whichmayinturnresultinalossofcontrol.
4.7. Self-reportedsatisfactionandusefulnessofthehaptic steeringguidance
Asupportsystemshouldbeperceivedusefulandsatisficingto beacceptedbydriversinacommercialvehicle.Theresultsofthe acceptancequestionnaireshowedthatallthreeguidance condi-tionswerewellliked,withaveragescoresabovezeroforboththe usefulnessandsatisfactiondimensions.TheBandsystemshowed slightlyandnon-significantlyloweracceptancethanContand Con-tRF;itmightbethatparticipantsexperiencedthesuddenincrease infeedbackforcewhencrossingalaneasannoying.Thisisin accor-dancewithresultsbyNavarroetal.(2010)andDeWinteretal. (2008),whofoundthatparticipantsgaverelativelylowratingsof acceptancetoadiscontinuoushapticinterventiononthesteering wheelorgaspedal, respectively.Discontinuoushapticfeedback maycauseunwantedreflexes/overshoots,excessivewearof hard-ware,ortimingproblemsassociatedwiththebalancebetweenfalse alarmsandmisseddetections(DeWinteretal.,2008;Navarroetal., 2010).ItisinterestingthattheContRFcondition,which didnot functionabove130km/h,didnotshowloweracceptancethanthe threeotherconditions.TherelativelyhighacceptanceforContRF mayhavebeencausedbythefactthatittooktheguidanceawayin agradualmanneryetclearlycommunicatedtheADASfunctionality andavailability.Alternatively,participantsmayhavethought Con-tRFwashelpfulforthereasonthatitpreventedthemfromdriving excessivelyfast.
4.8. Temporaleffectsofbehavioraladaptation
Duringthisstudy,participantswereexposedtoeachcondition forabout13min,henceonlymeasuringtheinitialandshort-term BAeffects.Previousresearchsuggeststhatthedegreeof experi-encewiththesystemisimportantinassessingBA(Martensand Jenssen,2012;Panouetal.,2007;seealsoMarsetal.,2014bwho foundthat drivingspeed increasedwiththeamountof experi-ence).Considering that trustin ADASand mental modelsgrow overperiodsofweeksormonths(Beggiatoetal.,2015),itis plau-siblethatthespeeddifferencebetweenContandManualmaybe evenlargerthan7km/hinthelongrun.Thepresentparticipants, manyofwhomwererecruitedfromthetechnicaluniversity com-munity,maybemorefamiliarwithintelligentvehicletechnology, andmorelikelytounderstandtheworkingmechanismsofsuch technology,thanthegeneralpublic.Thiswasexemplifiedbythe factthat only3 of24 participantshadnot heardoftheGoogle DriverlessCarbefore,whichisconsiderablylowerthanKyriakidis etal. (2015)who reportedthat48% of4845respondents inan internationalcrowdsourcedsamplehadnotheardoftheGoogle Driverlesscar.Thus,itremainstobeinvestigatedhowourBA find-ingsgeneralizetothegeneralpopulation,whomayneedalonger periodtogrowaccustomedtohapticguidancetechnology.The cor-relationsbetweenthedrivingexperiencevariablesandthemean speedwere0.25atmaximumand soexplainedupto6%ofthe variance,suggestingthatthedegreeofBAofadriverisnoteasily predictablefromaperson’slevelofdrivingexperience.Atpresent, therelationbetweenshort-andlong-termBAeffectsisunknown (DeWinterandDodou,2011),anditisrecommendedtoobtaina betterunderstandingofthistopicbymeansoflongitudinalstudies.
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 383
Inourstudy,BAwasoperationalizedasincreaseddrivingspeed, anddrivingperformancewasquantifiedintermsoflane-keeping performance and safety margins, which are measures that are causallyrelatedtosafety.Wealsomeasuredworkload,andfound that Contyielded lower scoresonthe Mental Demand item of NASA-TLXthan ManualandBand.Thismaybebecausethe lat-tertwoconditionsrequiredthedrivertokeepvisuallyattentive, whereastheformerprovidedcontinuousassistanceonthe steer-ingwheel.Researchershavepointedoutthatreducedattentiveness itselfshouldalsoberegardedasamanifestationofBA(e.g.,Elvik, 2013).Indeed,itispossiblethatlowmentalworkloadassociates withlow attentiveness and complacency, poor performance in reclaimingmanualcontrol,andaninclinationtowards perform-ingnon-drivingtasksbehindthewheel(DeWinter andDodou, 2011;YoungandStanton,2002).Futureresearch,conductedacross multipledaysor months,shouldestablishtheoptimalrange of workload(seealsoDeWaard,1996).Thatis,ontheonehand,it canbearguedthatareductionofworkloadleavesmentalspare capacityforscanningobjectsintheenvironmentandforenhancing situationawareness(e.g.,McDowelletal.,2008),whereasonthe otherhandtheaforementionedmentalunderloadisariskfactor too(YoungandStanton,2002).
4.9. Drivingsimulatorsversuson-roaddriving
Thelane-keepingperformanceobtainedinthisstudyisslightly betterthanfoundin realcars,witha standarddeviationof lat-eralposition[SDLP]ofabout0.10m,whereasvaluesof0.15–0.20m aretypicallyobservedinon-roadexperiments(seeVeldstraetal., 2015;VersterandRoth,2011).Thisdifferencemaybecausedbythe factthatourstudyfeaturedanarrowroadandreal-timefeedback onperformance.
Participantsinourstudyadoptedhighmeanspeedsof110km/h despitethefactthattheroadwasnarrowandcontainedvarious curves.Therelativelyhighspeedin adrivingsimulator maybe explainedbythefacttherewerenoroaduserssharingtheroad withtheparticipant,aswellasbyincorrectspeedperceptionand lowperceivedriskwhendrivinginasimulator(DeWinteretal., 2012;Wallisetal.,2007).Driversinasimulatordonotexperience arealriskofcrashing,whichisadownsideofusingadriving sim-ulatorasopposedtoarealcar,especiallybecausesubjectiverisk isconsideredtobeanimportantdeterminantofBA(Näätänenand Summala,1974;Wilde,1998).
Nevertheless,simulatorsofferimportantadvantagescompared totestsinrealcars.Forexample,itwouldbetechnically,legally, andethicallychallengingtoexposeparticipantstoasharpcurve onarealroadinacontrolledmanner.Regardingethicalandlegal implications,itmaybedebatedwhethertheContRFsystem,which removeshapticguidancewhen drivingfaster than125km/h so thatdriversare‘ontheirown’,isanacceptabletypeofdriver sup-portin realtraffic.Here, valuable lessons maybelearnedfrom existinglane-keepingassistancesystems,whichallhavevarious functionallimitations.Forexample,currentlane-keepingsystems donotfunctionabove200km/h,whenlanemarkersareabsent, insharpcurves,oronnarrowroads(e.g.Daimler,2013;VolvoCar Corporation,2015).
Although simulators have various limitations, their relative validity(i.e.,theeffectsizesbetweenthepairwisecomparisons) maystillbehigh,aswasillustratedrecentlybyKlüveretal.(2016). Theseauthorsfoundthatparticipantshadsubstantiallydifferent SDLPvaluesbetweenfixedbasesimulators,movingbase simula-tors,andarealcarbuttheeffectsizesasafunctionofsecondary taskconditionsweresimilarforthesethreehardwareconditions. Inourstudy,thelackofsubjectiveriskwastriedtobeaccountedfor byprescribingataskthatpenalizesriskybehavior(i.e.,‘minimize thenumberofconehits’).Nevertheless,thefactremainsthatthe environmentinthecurrentstudywasrelativelyuncomplicated, featuringasingle-laneroadandnootherroadusers,and there-forefurtherresearchshouldinvestigatetheexternalvalidityofthis simulator-basedresearch.
4.10. Forced-pacedversusself-pacedexperimentaldesigns
Finally, this research indicates that future ADAS developers shouldtakeintoaccounthumanadaptationswhenassessingand designingnewsystems.Thespeedadaptationfoundinthisstudy showedthatdrivingata fixedspeed, asisdoneinmuch ADAS research,maygiveadistortedviewofasystem’sbenefits.Although fixingthespeedisconvenientbecauseithomogenizesthe chrono-logicaltimingofeventsamongparticipants,italsorestrictsdrivers fromadaptingtoasysteminarealisticway.
4.11. Conclusionsandrecommendations
Inadrivingsimulatorexperiment,threedesignsofhaptic guid-ance were compared to unsupported driving, with the aim of quantifyingtheinfluence ofhapticsteering guidancedesign on speeding. Ashypothesized,continuoushapticsteering guidance sufferedfromBA,intermsofanincreasedspeedof7km/h com-pared to the Manual condition. We tested two fundamentally differentremedialtechnologies(Band&ContRF):Bandonly pro-videdfeedbackwhendrivingoutofbound,resultinginasystem thatwasidenticaltoManualfor84%ofthedrivingtime,whereas ContRFprovidedcontinuousfeedbackin95%oftime.Thesetwo differentapproachesbothsuccessfullypreventedBAwithsimilar lane-keepingperformanceandself-reportedacceptanceas contin-uousguidance.Participantsinourexperimentdrove13minineach ofthefourconditions;futureresearchshouldaddressreal-world andlong-termeffects,andestablishwhichcognitivemechanisms areatplayregardingthecausesandremediesofBA.
Acknowledgements
The authors would like to thank Sjors Oudshoorn, Dimitra Dodou,andBastiaanPetermeijerfortheirusefulsupportand well-foundedcomments.
AppendixA. –CorrelationMatrices
For eachcondition,theSpearmancorrelationwascalculated betweendependentmeasures(with anadditionalmeasure, the meanabsolutesteeringspeed[deg/s]).Thecorrelationmatricesare showninthetablesbelow.
TableA1
Spearmanrank-ordercorrelationmatrixfortheManualcondition(N=24).
Mean speed Percentage time above 125 km/h Percentage time off-road Mean absolute lateral error Max absolute lateral error Mean lane return time Median TLC NASA-TLX Mean steering reversal rate Mean absolute driver torque Mean steering speed DBQ violations Mileage Meanspeed 1.00
Percentagetimeabove125km/h 0.68 1.00 Percentagetimeoff-road 0.33 0.29 1.00 Meanabsolutelateralerror 0.40 0.44 0.93 1.00
Maxabsolutelateralerror 0.55 0.29 0.66 0.72 1.00 Meanlanereturntime 0.21 0.13 0.67 0.64 0.42 1.00 MedianTLC −0.68 −0.68 −0.72 −0.77 −0.71 −0.47 1.00
NASA-TLX 0.03 0.00 0.23 0.37 0.37 0.27 −0.27 1.00
Meansteeringreversalrate 0.20 0.24 −0.16 −0.10 0.10 −0.21 −0.39 0.21 1.00 Meanabsolutedrivertorque 0.92 0.67 0.30 0.37 0.51 0.23 −0.75 0.10 0.45 1.00 Meanabsolutesteeringspeed 0.37 0.40 −0.04 0.05 0.23 −0.14 −0.52 0.20 0.93 0.62 1.00 DBQviolations 0.08 0.16 −0.07 −0.10 −0.02 −0.14 −0.03 0.09 0.06 0.19 0.12 1.00
Mileage 0.25 0.04 0.03 0.05 0.04 0.18 −0.11 0.35 0.01 0.23 0.08 0.13 1.00
Weeklydriving 0.20 −0.02 0.07 0.08 0.10 0.17 −0.11 0.26 −0.06 0.26 0.05 0.22 0.81 Note:p<0.05for||≥0.41,p<0.01for||≥0.52andp<0.001for||≥0.63.
TableA2
Spearmanrank-ordercorrelationmatrixfortheContRFcondition(N=24).
Mean speed Percentage time above 125 km/h Percentage time off-road Mean absolute lateral error Max absolute lateral error Mean lane return time Median TLC NASA-TLX Mean steering reversal rate Mean absolute feedback torque Mean absolute driver torque Mean steering speed Usefulness scale Satisfaction scale Meanspeed 1.00
Percentagetimeabove125km/h 0.57 1.00 Percentagetimeoff-road 0.19 0.20 1.00 Meanabsolutelateralerror 0.31 0.23 0.86 1.00 Maxabsolutelateralerror 0.00 0.13 0.83 0.59 1.00 Meanlanereturntime 0.26 0.16 0.66 0.44 0.50 1.00 MedianTLC −0.54 −0.53 −0.42 −0.46 −0.50 −0.28 1.00 NASA-TLX 0.11 0.21 0.33 0.40 0.15 0.22 −0.13 1.00 Meansteeringreversalrate 0.15 0.16 −0.07 −0.08 0.09 0.05 −0.66 −0.17 1.00 Meanabsolutefeedbacktorque 0.19 −0.13 0.73 0.76 0.59 0.33 −0.51 0.24 0.23 1.00 Meanabsolutedrivertorque 0.77 0.47 0.06 0.14 0.05 0.20 −0.57 0.28 0.31 0.10 1.00 Meanabsolutesteeringspeed 0.29 0.18 −0.01 0.06 0.06 0.09 −0.69 −0.14 0.96 0.34 0.37 1.00 Usefulnessscale −0.11 −0.19 −0.06 −0.20 0.15 −0.02 0.11 −0.24 −0.10 −0.03 0.13 −0.11 1.00 Satisfactionscale 0.19 0.19 0.13 0.08 0.38 0.05 −0.30 −0.25 0.05 0.05 0.11 0.06 0.54 1.00 DBQviolations 0.25 0.23 −0.22 −0.17 −0.05 −0.20 −0.30 −0.03 0.27 0.04 0.31 0.32 0.32 0.35 Mileage 0.01 0.25 −0.11 0.08 −0.10 −0.06 −0.03 0.39 −0.08 −0.28 0.05 −0.09 −0.32 0.04 Weeklydriving −0.04 0.07 −0.02 0.27 −0.05 −0.14 −0.08 0.30 −0.12 0.03 −0.06 −0.07 −0.37 −0.07 Note:p<0.05for||≥0.41,p<0.01for||≥0.52andp<0.001for||≥0.63.
T.Melmanetal./AccidentAnalysisandPrevention98(2017)372–387 385
TableA3
Spearmanrank-ordercorrelationmatrixfortheBandcondition(N=24).
Mean speed Percentage time above 125 km/h Percentage time off-road Mean absolute lateral error Max absolute lateral error Mean lane return time Median TLC NASA-TLX Mean steering reversal rate Mean absolute feedback torque Mean absolute driver torque Mean steering speed Usefulness scale Satisfaction scale Meanspeed 1.00
Percentagetimeabove125km/h 0.77 1.00 Percentagetimeoff-road 0.10 0.15 1.00 Meanabsolutelateralerror 0.14 0.09 0.87 1.00 Maxabsolutelateralerror 0.32 0.41 0.77 0.63 1.00 Meanlanereturntime 0.10 0.08 0.89 0.71 0.67 1.00 MedianTLC −0.65 −0.68 −0.55 −0.39 −0.73 −0.55 1.00 NASA-TLX 0.33 0.16 0.41 0.39 0.52 0.31 −0.37 1.00 Meansteeringreversalrate 0.27 0.32 −0.22 −0.44 0.13 −0.05 −0.44 0.04 1.00 Meanabsolutefeedbacktorque 0.36 0.35 0.89 0.81 0.73 0.83 −0.77 0.35 −0.04 1.00 Meanabsolutedrivertorque 0.83 0.58 −0.04 −0.09 0.16 0.03 −0.63 0.28 0.54 0.28 1.00 Meanabsolutesteeringspeed 0.35 0.43 −0.10 −0.30 0.25 0.05 −0.53 0.08 0.97 0.09 0.56 1.00 Usefulnessscale 0.04 0.16 −0.11 −0.09 −0.08 −0.16 −0.01 −0.32 −0.07 −0.07 −0.13 −0.04 1.00 Satisfactionscale −0.24 −0.07 −0.15 −0.07 −0.22 −0.25 0.31 −0.62 −0.28 −0.20 −0.43 −0.28 0.63 1.00 DBQviolations 0.07 −0.06 −0.03 −0.09 0.06 −0.03 −0.15 0.04 0.36 −0.07 0.10 0.37 −0.04 −0.04 Mileage 0.20 0.26 0.16 0.02 0.25 −0.05 −0.35 0.37 0.02 0.19 0.23 0.01 0.02 −0.25 Weeklydriving 0.20 0.12 0.17 0.04 0.11 0.07 −0.35 0.11 −0.03 0.21 0.26 −0.05 0.11 −0.13 Note:p<0.05for||≥0.41,p<0.01for||≥0.52andp<0.001for||≥0.63.
TableA4
Spearmanrank-ordercorrelationmatrixfortheContcondition(N=24).
Mean speed Percentage time above 125 km/h Percentage time off-road Mean absolute lateral error Max absolute lateral error Mean lane return time Median TLC NASA-TLX Mean steering reversal rate Mean absolute feedback torque Mean absolute driver torque Mean steering speed Usefulness scale Satisfaction scale Meanspeed 1.00
Percentagetimeabove125km/h 0.80 1.00 Percentagetimeoff-road 0.22 0.38 1.00 Meanabsolutelateralerror 0.11 0.31 0.93 1.00 Maxabsolutelateralerror 0.41 0.36 0.79 0.67 1.00 Meanlanereturntime 0.16 0.17 0.73 0.65 0.70 1.00 MedianTLC −0.49 −0.57 −0.60 −0.53 −0.56 −0.51 1.00 NASA-TLX −0.08 −0.13 0.28 0.26 0.24 0.16 −0.02 1.00 Meansteeringreversalrate 0.03 0.26 0.16 0.18 0.04 0.29 −0.62 −0.15 1.00 Meanabsolutefeedbacktorque 0.45 0.60 0.84 0.79 0.66 0.60 −0.86 0.04 0.40 1.00 Meanabsolutedrivertorque 0.90 0.71 0.32 0.19 0.42 0.35 −0.64 −0.05 0.21 0.59 1.00 Meanabsolutesteeringspeed 0.33 0.49 0.30 0.27 0.22 0.36 −0.77 −0.21 0.92 0.60 0.51 1.00 Usefulnessscale 0.21 0.09 −0.44 −0.47 −0.31 −0.30 0.10 −0.50 −0.23 −0.20 0.22 −0.13 1.00 Satisfactionscale 0.29 0.20 −0.28 −0.28 −0.12 −0.28 −0.06 −0.49 −0.09 −0.07 0.18 −0.01 0.70 1.00 DBQ 0.11 0.09 −0.14 −0.15 −0.12 −0.19 −0.01 −0.19 0.00 −0.03 0.17 0.06 0.37 0.37 Mileage 0.12 0.00 0.21 0.28 0.19 0.10 −0.09 0.44 −0.23 0.18 0.18 −0.08 −0.20 −0.41 Weeklydriving −0.06 −0.13 0.27 0.35 0.15 0.20 −0.06 0.35 −0.15 0.18 0.06 −0.06 −0.24 −0.44 Note:p<0.05for||≥0.41,p<0.01for||≥0.52andp<0.001for||≥0.63.
References
Aarts,L.,VanSchagen,I.,2006.Drivingspeedandtheriskofroadcrashes:areview. Accid.Anal.Prev.38,215–224,http://dx.doi.org/10.1016/j.aap.2005.07.004. Abbink,D.A.,Mulder,M.,2009.Exploringthedimensionsofhapticfeedback
supportinmanualcontrol.J.Comput.Inf.Sci.Eng.9,011006,http://dx.doi.org/ 10.1115/1.3072902.
Abbink,D.A.,Mulder,M.,VanderHelm,F.C.T.,Mulder,M.,Boer,E.R.,2011. Measuringneuromuscularcontroldynamicsduringcarfollowingwith
continuoushapticfeedback.IEEETrans.Syst.ManCybern.PartB:Cybern.41, 1239–1249,http://dx.doi.org/10.1109/TSMCB.2011.2120606.
Abbink,D.A.,Mulder,M.,Boer,E.R.,2012.Hapticsharedcontrol:smoothlyshifting controlauthority?Cogn.Technol.Work14,19–28,http://dx.doi.org/10.1007/ s10111-011-0192-5.
Assum,T.,Bjørnskau,T.,Fosser,S.,Sagberg,F.,1999.Riskcompensation—thecase ofroadlighting.Accid.Anal.Prev.31,545–553,http://dx.doi.org/10.1016/ S0001-4575(99)00011-1.
Banks,V.A.,Stanton,N.A.,Harvey,C.,2014.Whatthedriversdoanddonottell you:usingverbalprotocolanalysistoinvestigatedriverbehaviourin