A toolbox for automated driving on the STISIM driving simulator

(1)

Delft University of Technology

A toolbox for automated driving on the STISIM driving simulator

Eriksson, A.; de Winter, Joost; Stanton, Neville A.

DOI

10.1016/j.mex.2018.08.003

Publication date

2018

Document Version

Final published version

Published in

MethodsX

Citation (APA)

Eriksson, A., de Winter, J., & Stanton, N. A. (2018). A toolbox for automated driving on the STISIM driving

simulator. MethodsX, 5, 1073-1088. https://doi.org/10.1016/j.mex.2018.08.003

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)

Method

Article

A

toolbox

for

automated

driving

on

the

STISIM

driving

simulator

Alexander

Eriksson

a,c

,

Joost

de

Winter

b,

*

,

Neville

A. Stanton

a

a_{Transportation}_Research_Group,_Faculty_of_Engineering_and_Environment,_University_of_Southampton, BoldrewoodCampus,SouthamptonSO167QF,UK

b

DepartmentofBiomechanicalEngineering,FacultyofMechanical,MaritimeandMaterialsEngineering, DelftUniversityofTechnology,Delft,TheNetherlands

c

TheSwedishNationalRoadandTransportResearchInstitute,Box8072,SE-40278Göteborg,Sweden ABSTRACT

Drivingsimulatorshavebeenusedsincethebeginningofthe1930stoassistresearchersinassessingdriver behaviourwithoutputtingthedriverinharm’sway.Thecurrentmanuscriptdescribestheimplementationofa toolboxforautomateddrivingresearchonthewidelyusedSTISIMplatform.Thetoolboxpresented inthis manuscriptallowsresearcherstoconductﬂexibleresearchintoautomateddriving,enablingindependentuseof longitudinalcontrol,andacombinationoflongitudinalandlateralcontrol,andisavailableasanopensource downloadthrough GitHub.Thetoolboxallowsthedrivertoadjustparameterssuchassetspeed (in5kph increments)andtime-headway(instepsof1,1.5,and2s)aswellasautomationmodedynamically,whilelogging additionalvariablesthatSTISIMdoesnotprovideout-of-the-box(time-headway,timetocollision).Moreover,the toolboxpresentedinthismanuscripthasgonethroughvalidationtrialsshowingaccuratespeed,time-headway, andlanetracking,aswellastransitionsofcontrolbetweenmanualandautomateddriving.

AtoolboxwasdevelopedforSTISIMdrivingsimulators. Thetoolboxallowsforautomateddriving.

Functionalityincludestrackingofspeed,headway,andlane.

ARTICLE INFO Methodname:Softwaretoolbox

Keywords:Drivingsimulator,Automateddriving,Toolbox,Humanfactors,Adaptivecruisecontrol,Highlyautomateddriving, STISIM

Articlehistory:Received10August2017;Accepted6August2018;Availableonline15August2018

*Correspondingauthor.

E-mailaddresses:alexander.eriksson@vti.se(A.Eriksson),j.c.f.dewinter@tudelft.nl(J.deWinter),n.stanton@soton.ac.uk

(N.A. Stanton).

https://doi.org/10.1016/j.mex.2018.08.003

ContentslistsavailableatScienceDirect

MethodsX

(3)

SpeciﬁcationsTable

SubjectArea Psychology Morespeciﬁcsubjectarea HumanFactors Methodname Softwaretoolbox Nameandreferenceoforiginalmethod Notapplicable,seepaper

Resourceavailability https://github.com/he1y13/Toolbox-for-automated-driving-in-STISIM

Methoddescription

In thispaper, we describeaset ofalgorithms developedfor theSTISIM drivingsimulator

platform. The goal of the algorithms was to enable dynamic human-automation

interactionthroughcustomsoftwareusingtheSTISIMV3Build3.07.04OpenModuleinVisual Basic6 (VB6). Althoughthis implementation ofautomated driving is platform-speciﬁc, the toolbox can be implemented on other platforms that offer an API or SDK by translating thesubroutinesintotheprogramminglanguagesupportedbythesimulatorinquestion,with therequirement that the lead-vehiclecanbe identiﬁedandqueried forinformation suchas speedthroughthe API/SDK.

OpenModuleisapluginfeatureoftheSTISIMplatformthatallowsresearcherstoimplementtheir modulesusing unmanagedcode(e.g., VB6or C++).One ofthe functionsof OpenModule is the ‘Update’-functionwhich is called once every simulation frame, just before the graphics display updates.The‘Update’-functionallowstheresearchertodirectlycontrolthebehaviourofavehiclevia pedalandsteeringinput, afunctionalitythat wasutilisedin developingourtoolbox.Thetoolbox consistsofseveralsubroutines,eachresponsibleforapartofthevehiclecontrol,allowinglateraland longitudinalautomationtobeusedseparatelyor inconjunction.Thefunctionalityofthetoolbox algorithmsisdetailedbelow.

Algorithms

Inthissectionthealgorithmsaredescribedintwoparts:(1)longitudinalcontroland(2)lateral control.Thisstructureenablesthesimulationofdifferentlevelsofautomateddriving,rangingfrom manualdrivingandACC (i.e.,automatedlongitudinalcontrol) tohighly automateddriving(i.e., automatedlongitudinalandlateralcontrol)asshown inFig.1.Themanualmodeisvoidof any automatedfeatures,meaningthattheoperationofthevehicleisdependentonthehumandriver only.

TheACCmodeshowninFig.1controlsthevehicle_’sspeedbyprovidingcontrolsignalstothe throttleandbrakeinceptorstodrivethevehicleatatargetspeed.Thismodeisbrokendownintoits constituentsinFig.2.Thetargetspeedissetbythedriverorisdictatedbythespeedofaslowervehicle withinsensorrange(thisrangemaybechangedinthesourcecodetosimulateradarswithhigheror lowerrange).Additionally,thereisanoptiontouncommentasectionofthecodein‘OM_Module.cls’ whichwillsetthemaximumspeedtothatofthespeedlimit.ACCisanintegralpartofachievinghighly automateddrivingandisnowcommonlyavailableinproductionvehicles.ACConproductionvehicles utilisesaradarunitattachedtothefrontofthevehiclethatkeepstrackofanyleadingvehicles,feeding thecruise controlalgorithmwiththedistancetotheleadvehicle,whichis usedtocomputethe desiredspeedtomaintaintheselectedtime-headway.

ThehighlyautomateddrivingmodeincorporatesthefunctionalityoftheACCfeature,withthe additionofautomatedlateralcontrol.Thehostvehicle(hostvehiclereferstothevehicleequipped withthedescribedalgorithm)automaticallyfollowstheroadcurvature.Inaddition,lanechangesmay beperformedin responsetodrivercommands,byforexample_ﬂickingtheindicatorstalkin the directionofthelane-changewheninhighlyautomateddrivingmode,muchlikeafunctionofthemost recentadditionofautomationonthemarket,theTeslaMotors[1]AutopilotLanechangefunctionality (however,initscurrentstatetheautomationdoesnotassessthetrafﬁcintheadjacentlanebefore

(4)

changing lane).Thelongitudinalandlateral automationsubroutinesarefurtherexplained inthe sectionsbelow.

Longitudinalautomation

Longitudinalautomationfunctionalityandstates

Theprimaryfunctionandfundamentalrequirementofalongitudinalcontrolsystemistocontrol andadaptspeedtoleadingvehiclesanddriversettings(thisisreferredtoastargetspeed,vtarget). Furtherrequirementsandassumptionsforalongitudinal(i.e.,ACC-based)controlsystemarethat:

Fig.1.Afunctionalblockdiagramoftheautomateddrivingtoolbox.ASreferstoautomatedsteeringcontrol.Hostvehiclerefers tothevehicleequippedwiththedescribedalgorithm.

(5)

1 thesystemcannotbeengagedwhilethevehicleisdrivinginreverse[2]. 2 thedriverhastheabilitytooverridethesystem.

3 thesystemmaintainsthespeedsetbythedriverintheabsenceofaslowleadingvehicle. 4 thesystemusesaccelerationthresholdstoensurecomfortabledrivingduringnormaloperating

conditions.

5 thesystemslowsdownthevehicletothespeedofaslowermovingleadvehicleandmaintainsthe desiredheadway.

6 thesystemignoresdecelerationthresholds(intermsofcomfort)whensuchathresholdhinders bringingthevehicletoasafesystemstatethroughslowingdownorstoppingcompletely. 7 thesystemhandsbackcontroltothedriverwhentheoperationallimitsareapproached.Suchlimits

mayincludesensorfailure,geographicalconstraints,orexternalfactorsleadingtodegradedsystem performance(thiscanbesimulatedthrougha shutdowneventspeciﬁedinaneventﬁleinthe automationtoolbox).

Inthealgorithm,theleadingvehicleisconsideredtobeavehicledrivinginthehostvehicle’slane (itisinthehostvehiclelanewhenitscentreofmassiswithinthelaneboundary)withintherangeof thesimulatedradar.ThecontrollerworksinaccordancewithSAEJ2399[3]onallpointsexceptthat ourcontrollerdoesnotallowforsettingaminimumspeed,andthatitdoesnotactivateanauditory, visual,and/orhapticalerttoinformthedriverthathe/sheneedstotakebackmanualcontrol.The formerlackscompliancebecauseourcontrollerworks asastop&gosystem. Thelatterlimitation comesdowntothedesignandimplementationofahuman-machineinterface,whichisnotcoveredin thecurrentmanuscript.However,itispossibletoaddaninterfacebyenablingthesocketconnection describedelsewhereinthispaper.

WebelievethatthiscontrollerforACCworkswellenoughformostHumanFactorsapplications. However, normally, ACC systems on the market include more complex control algorithms for longitudinalcontrol,suchasgainschedulingproportionalintegralcontrol(GSPI),gainscheduling linear quadratic control (GSLQC) [4] and nonlinear model predictive control (NMPC) [4]. The implementationof such control algorithms is straightforward: Should one wish to use a more advancedcontrolalgorithm,thecontrollerusedinthetoolboxcanbereplacedwithanothertypeof control,eitheranimplementationofoneoftheabovementionedoravehiclemanufacturer’sversion. Itmust,however,benotedthatVisualBasic6doesnotsupportmultithreading,andthus,allthecode intheOpenModuleareexecuted inthesamethreadasthesimulation,sotheadditionofmore complexfunctionalitiesmayhaveaneffectonthesimulatorperformance(i.e.,framedrops,lagetc.). Thealgorithmtoaccessvehicledataforthelead-vehicleisdescribedinErikssonandStanton[5]. Parametersrelatedtothelead-vehiclearedenotedwiththesubscriptlead.Thelongitudinalcontrol algorithmisdesignedasaFiniteStateMachine(FSM),containingthreestates(cruise,follow,and adapt),eachwithitscontrollercharacteristics.SeveralconditionsneedtobefulﬁlledbeforetheFSM cantransitionfromonestatetoanother.Theprocessofdeterminingcontrollerstatesareshownin Pseudocode1.Theﬁxedparametersinthebelowpseudocodeweretunedmanuallytoensuresmooth switchingbetweenthecontrolstatesoftheFSM.

(6)

Pseudocode1:thealgorithmtodeterminethestageofthelongitudinalcontrolsubroutine.Note:^ isthelogical‘AND’operator.The1.15multiplierforthetime-headwayconditionensuresthatthereis nosuddenswitchingoftime-headwayandspeedbasederrorterms.‘Toofast’referstotheleadvehicle travellingfasterthanthedesiredspeedsetbythedriver

Each state has a Proportional-Integral-Derivative (PID)controller and depending onthe state, differentgainsforthedifferentparametersareused.ThetransferfunctionofaPIDcontrolleristhesumof theoutputsofthreesub-controllers:aproportional,anintegral,andaderivativecontroller.Theerror signal undergoesprocessing ineach controller(i.e., theproportional,integral, andderivative sub-controllers),theresultingsignalsareaddedandconstitutethetotaloutputfromthePIDcontroller.

Inthecaseofautomation,oneoftheinputstothecontrolsystemisthetargetspeed,andtheoutput isanumberrepresentingthevirtualpedalposition.Positiveoutputvaluesaresignalssenttothe virtualthrottlepedal.Fornegativesignals,theirabsolutevaluesrepresentthevirtualbrakepedal position.

Becausetheenvironmentisinherentlydigital,discretemathematicsappliestothecomputations. Hence,thetemporalresolution(

D

t)islimitedtoasinglesimulationframe,i.e., 1/30sforafrequencyof 30Hz.Thecontroller’soutput(thebrake/throttleposition)isgovernedbyEq.(1),whereei¼

D

vi¼

vtarget;iviistheerrortermrepresentingthedifferencebetweenthesetspeedandvehicle’scurrent

speedatthecurrentinstantoftime,i. npedal¼KPeiþKI

D

t Xi j¼0 ejþKD eiei1

D

t ð1Þ

(7)

The PID controller for car automation in a discrete simulation environment.j=start time of the controllercycle,i=currenttime-stepinthecontrollercycle.Kpistheproportionalgainonthecontroller.Ki istheintegralgainofthecontroller.Kdisthegainofthederivativecomponentofthecontroller.

The K-coefﬁcientsof thesub-controllers representthecontroller gains andhave a signiﬁcant impactonthebehaviourofthesystem,astheirrelativeandabsolutevaluesdeterminetherisetime, overshoot,anddampingcharacteristics.Therefore,differentcontrolstatesrequiredifferentsettings. Followingstate

TheFollowstateaimstomaintainaconstanttime-headwaytotheleadvehicle.Maintaining headwayisamorechallengingtaskthanmaintainingspeed,asthedistanceiscontrolledbythe hostvehicle speed relativetothe leadvehicle. Thetime-headwayisset by thedriver(in the followingincrements:1,1.5,and2s,thesevalues canbechangedinthesourcecodeintheﬁle ‘OM_Module.cls’underthe‘cycleTHW’function)andisdeﬁnedast¼dlead=Vhost.Additionally,the

algorithmwillslowdowntoastopifthereisacrossingvehicleinitspath,orifthereisavehicle approachingheadon.It will,however,notexecuteanyevasivelateral manoeuvres initscurrent implementation.

Adaptstate

TheAdaptstateisusedforsmoothspeedadjustmentstomeetthedesiredtargetspeed.Thespeed errorsignalintheAdaptstateisdeﬁneddifferentlythaninotherstates.Theultimatetargetspeedis stilleitherthespeedsetbythedriverortheexternallylimitedspeed (i.e.,comingfromaslower leadingvehicle).However,toattainasmoothmanoeuvreandspeedadjustment,theerrorsignalsrefer toinstantaneoustargetspeed,whichcomesfromlinearinterpolationfromthevehiclecurrentspeed andthetargetspeed.

The interpolatedspeed is calculated using Beziercurves. Bezier curvesare frequently usedin computergraphicstorenderanimationsorvectorgraphics.TheBeziercurvesareusedtodrawsmooth curvesthatcanbescaleddynamicallyandindeﬁnitely.InanimationBeziercurvescanbeusedtocontrol speedovertimeofanimatedobjects.ThesecharacteristicsmakeBezierfunctionswellsuitedforusein trajectoryplanningandinterpolation.Bezierfunctionshavebeenproposedasawayofplanningand traversingtrajectoriesinatwo-dimensionalspacebyChoietal.[6].Suchanalgorithmwouldbedivided intotwoparts,trajectoryplanningandtrajectoryinterpolation[6].Inthecurrentimplementationofthe controlalgorithmforlongitudinalcontroltheBezierfunctionsareusedtointerpolatethespeedofthe hostvehicletoasetspeedoraleadingvehicle’sspeedtoensuresmoothaccelerationanddecelerationby modellingthetargetspeedusingaﬁrst-orderBeziercurve(seeEq.(2)).

1t

ð ÞP0þtP1;t2½0;1 ð2Þ

The equation fora ﬁrst-orderBezier curve. WhereP0is the host vehiclespeed atthe start of the interpolation,andP1isthetargetspeed.t0isthenormalizedstartpointoftheinterpolationandt1isthe endpoint.ThemanoeuvretimeiscalculatedandnormalizedaccordingtoEqs.(3)and(4)below.

Toplanthespeedtrajectory,amanoeuvredurationmustbecomputedtomatchthehostvehicle speed withthe target speed taking a “comfortable acceleration” thresholdas shown in Eq. (3). Followingthecomputationofthemanoeuvreduration,thetimeintervalneededisrescaledtoavalue between0and1takingthesimulatorframerateintoaccountthroughEq.(4).

Tmanoeuvre¼

D

vi

acomfortable ð3Þ

Theformulausedforﬁndingmanoeuvredurationusedforinterpolation. t¼ TcurrTinterp:;start

ðTmanoeuvreHzsimulationÞ ð4Þ

RescalingofTManoeuvretoascaleof0–1basedonthefrequencyofthesimulator.Tcurrreferstothecurrent time, Tinterp., startrefers to the start of the interpolation time, Tmanouvre refers to the manoeuvre time calculatedinEq.(3).

(8)

Whenthemanoeuvreduration hasbeendeterminedand scaledtotheappropriaterange,the currentspeed,targetspeed,andtimeareintroducedtoEq.(2)tocreatethetrajectory.Followingthe creationofthetrajectory,thecontrollersetpointinterpolatesalongthetrajectoryuntilthetarget speedisreached.Thisapproachensuresthattheaccelerationthresholdisneverexceeded.

Asthespeediscomputedateachdiscretestepofthesimulation,theerrorsignalsforthePIDis signiﬁcantlysmallerthanastepinput(i.e.,from60kphto100kph),whichismoremanageablebythe PIDasthelikelihoodofanovershoot,oraggressiveaccelerationisavoided.TheerrorsignalforthePID isgivenby:ei¼

D

vi¼vivcurrwhichresultsinsmootheraccelerationanddeceleration.

Cruisestate

Thecruisestateisusedwhenthevehicledoesnotneedtoadjustitsspeedmorethan3.5m/s(i.e., whensmalladjustmentstothethrottleoutputarerequiredtomaintainthesetspeed,whenpassing throughhillyareasorcurves),andwhenthereisnoleadvehicleoraleadvehiclefasterthantheset speed.ThecruisestatecontrolsthespeedinaccordancewithEq.(1).TheerrortermusedforthePID controlleriscalculatedas:ei¼vtargetvcurr.

Lateralautomation

Thelateralcontrolisresponsibleforsteeringthecarandcontrollingitspositioninthedesiredlane. Thisisachievedbycontrollingthevehicle'slateralpositionwithrespecttotheroadcentrelineandthe centreofthedesiredlane.Thetargetpositionistypicallytheexactcoordinateofthecentreofthelane with nolook-ahead function. Thus, theimplemented controller for lateral control is somewhat rudimentary,andothercontrollersarereportedintheliterature,asintheworkofHessburg and Tomizuka[7]wherea‘fuzzy’controllertakesinconsiderationroadgeometrytosteerthevehicle.As STISIMdoesnotaffordalookaheadfunctionforroadwaygeometry,thistypeofcontrollerwasnot possibletoimplement.However,theimplementedcontrollerdoesaccommodateforvehiclespeedto someextentthroughamodiﬁcationofthegainforthesteeringPIDcontroller.

Avehicle’strajectoryisdependentonbothsteeringangleandvehiclespeed.Withthisinmind,the PIDcontrollerwasmodiﬁedtovarytheproportionalgainofthecontrolsignalasafunctionofcurrent vehiclespeed(i.e.,atypeofgainscheduling)(Eq.(5)).ThiswasalsodonetocompensatefortheSTISIM vehicledynamicsmodelthathasgotsomeundersteerathigherspeeds:ahighersteeringanglemust beproducedathigherspeedstofollowtheroad_’scurvature.Thecontrollerisabletokeepthevehicle initslaneinmostconditions,butinsituationswherethecurveradiusissmall,andthespeedishigh, theundersteeringofthedynamicsmodelcausesthevehicletogooutofthelane.

nsteering¼ðK1PþK2P

viK3PÞeiþKI

D

t Xi j¼0 ejþKD eiei1

D

t ð5Þ

PIDcontrollerforthelateralcontroller.K1pisthemainproportionalscalingfactor,K2pisthesecond scalingfactorfortheeffectofvehiclespeedonsteeringoutput,Viisthecurrentspeedofthehostvehicle,eiis theerrorterm(thedifferencebetweencurrentlanepositionandthelanecentre).KIistheintegralscaling factor,ejistheintegratederrorterm,andKDisthederivativescalingfactor.

Algorithmperformance

Anumberoftestswerecarriedouttodemonstratetheeffectivenessoftheautomateddriving toolbox.Thetestscenariowas10kmlongandcontainedanumberofcurvesandcut-insituations.This scenariowasextensivelyusedinErikssonandStanton[8]andproducesthesamevehiclebehaviouron repeatedtests.Thetestsaredetailedinthesectionsbelow.

Carfollowing

Toassesslongitudinaldrivingperformance,thealgorithmswererunthroughamotorwaydriving scenariowhereanumberofcarsmovedintothehostvehicle’slane,aswellascut-insaspartofdouble lanechanges.

(9)

Fig.3showsthespeedpro_ﬁleofthehostvehicleinrelationtothesetspeed,whereasFig.4shows thetime-headwayofanyvehiclesinfront,inrelationtothesettime-headway(1.5s).AsFig.3shows, thehostvehicleslowsdownbelowthelead-vehiclespeedtoaccommodatethelargeneedforsudden decelerationtoachievethedesiredtime-headwaywhenthereisalargedifferenceinspeedbetween thehostandthelead-vehiclecausedbythesuddencut-ins.AsshowninFig.4,thehostvehiclecloses thegapbetweentheleadandhostvehicledowntothedesiredtime-headwayandthenmaintainsthe desiredtime-headwayconsistently. Whenthelead-vehicleisnolongerdetectedthevehiclethen returnstotheoriginalsetspeed.

Lanekeeping

Thesamemotorwayscenariowasusedtoassesstheautomatedlateralcontrolofthealgorithm. Fig.5showsthelateraldeviationfromthelanecentre.Itispossibletoidentifywherethevehicle encounteredaturnbasedonthedeviationdata.However,thelateraldeviationisatmost15cmfrom vehiclecentretolanecentre,indicatinggoodlateralvehiclecontrol.

Scenario

Thefollowingsectioncontainsthescenarioparametersrequiredtoreproducethedriveusedinthe assessmentsectionsabove.

(10)

Fig.4.Time-headwayproﬁleofthecar-followingbehaviourduringmotorwaydriving.Thegapsintherecordedtime-headway signalarecausedbytheleadvehicleleavingthehost-vehiclelane.THW=time-headway.

(11)

Behaviouralvalidity

The software toolbox presented in this manuscript has already been used in research into automateddriving witha STISIMdriving simulator[8,9].Eriksson andStanton[8]assessedthe process of driver transitions between automated and manual vehicle control in non-urgent scenarios(SAELevel4Automation[10]).TovalidatetheﬁndingsfromErikssonandStanton[8],an on-roadstudywasdesignedwithamatchedsampletoassessthecorrelationofthetimeittook driverstotransitionbetweenautomatedandmanualcontrolinthesimulatorandontheroad.The resultsshowedthatdrivers’averagetransitionstimesfromautomatedtomanualcontrol,andvice versa,wereabout30%fasterfor on-roaddrivingthanfor simulator-baseddriving;however,the shapeofthedistributionsoftransitiontimeswashighlysimilarbetweensimulator-basedand on-roaddriving[11].ThestudybyErikssonetal.[11]concludedthattherewasanindicationofrelative behaviouralvaliditywhenthealgorithmspresentedinthismanuscriptandusedinErikssonand Stanton’s[8]simulatorstudywerecontrastedwithon-roaddrivingbehaviourinavehicleoffering contemporaryautomateddriving.

Theseﬁndingsshowthatthesimulator producesresultscorrespondingtothatofon-road conditions. Consequentially, it lends preliminary validity to the use of the algorithms presentedinthismanuscriptforuseinresearchintoautomatedvehiclesbeingconductedin simulators.

Limitations

Thealgorithmoutlinedabovehassomelimitationswhenitcomestointeractingwithcertainroad environmentsandroadusers.Thevehicleisunabletonavigateintersections(requiringturning),and

Fig.5. Lanekeepingperformance(12ftlanewidth)duringamotorwaydrive.Lanepositionreferstothevehiclelateralposition initscurrentlane.

(12)

roundaboutswhilstinautomateddrivingmode. Additionally,thesoftware,initscurrentform,is unable toaccount for vulnerable roadusers such ascyclists, pedestrians, and motorcycles. This functionalitymaybecreatedbyaccountingfortheseroadusersinthelateralandlongitudinalcontrol algorithms,shouldtheneedarise.Asoursoftwaretoolboxwasoriginallyintendedforresearchon automotive automation onnon-urban roadways,cyclists, pedestrians, and motorcycles werenot implemented.

Howtousethetoolbox

Thetoolboxisabletorunout-of-the-boxwithlittle set-upandfull accesstoitssourcecode (https://github.com/he1y13/Toolbox-for-automated-driving-in-STISIM) where researchers can makeedits and recommitthem toGitHub for use byotherresearchers. Thesubsections below describehowtosetupthetoolboxtoberunfromthepre-compiledDynamicLinkLibrary(DLL)ﬁle andsourcecode.

Usingthepre-compiledDLLﬁle

Tousethetoolboxusingthepre-compiledDLLﬁle,anumberofstepsmustbefollowed. 1CreateafolderontheC:/driveofthecomputerthatrunsSTISIMandnameitSTISIM(fullpathof

folder:C:/STISIM/)

2Movethe“OM_Automation.dll”,the“ButtonAssignment.txt”andthe“shutdownevents.txt”ﬁlesto C:/STISIM/

3Openthestartmenuonthecomputer,andtypein_‘cmd’,rightclickontheshortcutandclickrunas administrator

a Enterthecommand:cdC:/Windows/SysWOW64/

b Enterthecommand:regsvr32C:/STISIM/OM_Automation.dll 4OpenSTISIMtoedittheconﬁgurationﬁle

a Openthetab“DataCollection”andtickthebox“Collecttimetocollisiondata”

b Openthetab“OpenModule”andaddthefollowingpathtothe“OpenModuleDLLﬁle”box:C:/ STISIM/OM_Automation.dll

cOpenthetab“SimulationControl”andchangethedesiredframerateto20framespersecond (thisvaluecanbeeditedinthesourcecodetohighervalues)

5OpenC:/STISIM3/Tools/CalPot32.exeandselectthecontrollerbeingused a Openthetab“TestControls”andclickDriverInputs

b OpentheﬁleC:/STISIM/ButtonAssignment.txtandmapthebuttonsonthecontrollerbeingused withthecorrespondingfunctionalityshowninTable1

Table1

FunctionmappingintheButtonAssignment.txt.

Lineinﬁle Functionassociatedwithbuttonvalue 1 Cycletime-headway

2 IncreaseACCspeed 3 DecreaseACCSpeed

4 ActivateAdaptiveCruiseControl 5 ActivateHighlyAutomatedDriving 6 Deactivateautomateddriving 7 Leftlanechange

(13)

Whentheabovestepshavebeencompleted,theusermaystartanyscenario(itmustbenoted thattheautomationcannothandleintersectionsandroundabouts)andpresstheactivatebutton onthecontrollerdesignatedinthe“ButtonAssignment.txt”file.Moreover,iftheresearcherwishes thattheautomateddrivingfeatureshouldbecomeunavailableatasetpointduringadrive,this behaviour maybe specified in thefile “shutdownevents.txt”.Thisfile contains a single event wheretheresearchermayspecifya distancedown theroad(inpositivefeetdowntheroad,a negativevaluemeansitisignoredbythesoftware),thetimefromtheeventbeingtriggeredtothe eventoccurring (in seconds), the time after the event occurringuntil the automated driving featurebecomesavailable again(in seconds)inthefollowingmanner:“500;5;2500 (theevent countdown occurs 500 feet down the roadcounting down for 5s after which all automated featuresareunavailabletoengagefor25s).

Datacollection

Tosavetheadditionaldatageneratedbythetoolboxsoftware(suchassettime-headway,setspeed andlevelofautomationetc.)anumberofparametershavebeenpre-settobesavedintotheBSAVdata filenormallygeneratedinSTISIM(theresearchermustaddvariable49tobeloggedasoneofthe collectedparametersintheBSAVeventinthescenariodefinitionlanguage).Theparameterssavedto thedataoutputfileareshowninTable2.

Table2

Parametersfromthetoolboxcollectedfordataloggingpurposessavedinparameter49oftheBSAVevent.Theunitsreportedin thetableareintheunitssuppliedintheOpenModulevariables.

Parameter Typeofdata Units

1 Distancedowntheroad feet

2 Levelofautomation(manual/ACC/Highlyautomated) integer(0,1,2) 3 AdaptiveCruiseControlstate(Cruise,Follow,Adapt) integer(1,2,3)

4 Desiredtime-headway seconds

5 Currenttime-headway seconds

6 Desiredspeed ft/s

7 Currentspeed ft/s

8 Currenttimetocollision seconds

9 Optimallaneposition feet

10 Currentlaneposition feet

Table3

Pre-speciﬁeddatabeingsentoverthesocketconnectionwhenenabled.

Parameter Typeofdata Units

1 Laterallaneposition feet

2 Currentlane integer

3 Selectedtime-headway seconds 4–8 Distancetothefollowingvehicles:–leadvehicleinhostlane–trailingvehiclerightofhost

lane–leadingvehiclerightofhostlane–trailingvehicleleftofhostlane–leading vehiclerightofhostlane

feet

9 Numberoflanesoncurrentsectionofroad Integer

10 Vehiclespeed Ft/s

11 EngineRPM RPM

12 Automationmode Integer(0,1,2) 13 Take-overrequestcountdown Integer(1–X) 14 Distancedowntheroad feet

(14)

Socketconnection

The toolbox also allows sending data to an external device, for example, a human-machine interface.This canonly bemade available through re-compilingthe toolbox’s sourcecode after uncommentinganumberoflinesinthe_ﬁle_{‘Open_module.cls’.}

Theuserneedstouncommentlines700and702toenablethesocketconnection,andalsospecify thedesiredIPaddressandporttoreceivethedatapacketsandlines305–308toenabledatatobesent overthesocket.Thedatabeingsentisspeci_ﬁedinTable3:

Compilingfromsourcecode

Thereareafewadditionalstepsrequiredtoruntheautomationtoolboxwheneditsneedtobe madeinthesourcecode(examplesofthiswouldbetoadddata-outputoverTCP/IPortore-tunesome ofthecontrollersforlongitudinalorlateralcontrol).Tomakeeditstothesourcecodeacomputerwith theMicrosoftVisualBasic6.0editorinstalledmustbeused.Wheneditstothesourcecodehavebeen made, the new DLL must be compiled;this is donethrough the drop-down menu “File>Make OM_Automation.dll”intheVisualBasic6.0editor.

Summary

In this manuscript, we described a generic set of algorithms for automated drivingresearch implementable on any simulation platform that allows access to internal variables relating to surroundingtrafficthroughanAPI/SDK.Wethendescribedanimplementationofthesealgorithmson theSTISIMdrivingsimulatorplatformforSTISIMV3.07.04withaccompanyingperformancemetrics andadescriptionofvalidationwork.WethenprovidedastepbystepguideonhowtosetupSTISIMto accessthistoolboxforuseinresearchusingtheOpenSourcereleaseversionofthesoftware.Whilst thisimplementationisprimarilyintendedforusewithSTISIM,itispossibletoimplementthesame controlfunctionsandfinitestatemachineinothersimulators(possiblyinadifferentprogramming language),providedthattheysupportinformationacquisitionfromsurroundingtraf_ficinorderto providethecontrollerswithdata.

Additionalinformation

The topicofautomated drivingreceives an increasinglevelof attentionfrom HumanFactors researchers.Untilrecently,automateddrivingtechnologyrequiredintermittentdriverfeedback,for examplebytouchingthesteeringwheel,thusmaintainingalevelofdriverengagementsimilarto manualdriving[12].However,recentamendmentstotheViennaConventiononRoadTrafﬁcenable driverstobefullyhands-andfeet-freeaslongasthesystemcanbeoverriddenorswitchedoffbythe driver[13].Thisamendmentallowsdriverstobe‘out-of-the-loop’forprolongedperiodsoftime,yet driversarestillexpectedtoresumecontrolwhentheoperationallimitsoftheautomateddriving systemareapproached[10].

Theavailabilityofthesehighlyautomateddrivingsystemsmayfundamentallyalterthedriving task,andcouldgiveriseto‘ironies’and‘surprises’ofautomationsimilartothoseproposedby Bainbridge[14]andSarteretal.[15]inthecontextofprocesscontrolandaviation.Indeed,several empiricalstudieshaveshownthatdriversofhighlyautomatedcarsoftenrespondslowlywhen manualinterventionisnecessary[16–20].Inlightofthis,intermediateformsofautomationhave beendeemedhazardous becausedrivers arerequired tobeableto regaincontrol atall times [21,22]. To study these psychological phenomena and develop effective Human-Machine Interfaces for supportingdrivers of future automatedcars, the driving simulator is seenas a viableoption[11,23].

(15)

Simulators

Driving simulatorshavebeenused sincethebeginning of the1930s [24] andHuman Factors researchintoautomateddrivinghasbeenongoingsincethemid-1990s[25,26].Comparedtoon-road testing,drivingsimulatorsallow driver reactionstonewtechnologytobemeasuredin a virtual environment,withoutphysicalrisk[27–31].

Furthermore,drivingsimulatorsofferahighdegreeofcontrollabilityandreproducibility,and provideaccesstovariablesthataredif_ﬁculttoaccuratelydetermineintherealworld[32],suchas lanepositionanddistancetoroadwayobjects[33,34].Mostdrivingsimulatorsofferﬂexibilityin designingcustomplug-insthroughAPIs.WithOpenSourcesoftwareeffortsindrivingsimulation, suchasOpenDS[35],itislikelythattheuseofdrivingsimulatorswillcometogrowinthecoming years.

STISIM

STISIMisapopulardrivingsimulatorthatisusedforresearchpurposes[36–41].TheSTISIMdriving simulator software comes with an ‘Automated Driving’ feature accessible through its Scenario DeﬁnitionLanguage(SDL)[42,43]. TheSDL-basedautomation allowstheresearchertoenableor disableautomatedlateral and/orlongitudinalcontrolthroughthe‘ControlVehicle’(CV)eventby specifyingadistancedowntheroadatwhichpointtheeventshouldtrigger,andwhatmodechange shouldoccur(e.g.,thescript‘2000,CV,speed,20_initiates_automated_control_of_both_steering_and_speed

whentheparticipanthastravelled2000malongtheroad).TheSTISIMdocumentationstatesthatthis automateddrivingfeatureisintendedfordrivertraining[44],anapproachalsotakenbyotherdriving simulatormanufacturers(e.g.[45]).Thatis,byenablingautomatedcontrolofspeed,thedrivercan fullyconcentrateonlearninghowtosteer,orviceversa,byenablingautomatedcontrolofsteeringthe learnerdrivercanconcentrateonhow toaccelerateand stopthecar.Thistypeof automationis sufﬁcientwhenitcomestoresearchwheretheresearcherdoesnotwantthedrivertobeableto(dis) engagetheautomationorchangetheautomationmodes.TheCVeventhasbeensuccessfullyusedin thismanner(asdescribedby[46,47];andpresumedlyalsoinsimilarstudiesusingSTISIM:[40,48– 51]). However, if the researchaims to understand how drivers interact with automated driving systems,asinKircheretal.[52],ErikssonandStanton[8]andErikssonetal.[11],thistypeof hard-codedautomationisnotsufﬁcient.

Acknowledgment

ThisresearchwasfundedbytheEuropeanMarieCurieInternationalTrainingNetworkprojecton theHumanFactorsofAutomatedDriving(PITN-GA-2013-605817).

References

[1]TeslaMotors,ModelSsoftwareversion7.0,(2016).https://www.teslamotors.com/presskit/autopilot.

[2]S.Vakili,Designandformalveriﬁcationofanadaptivecruisecontrolplus(ACC+)System.MScThesis,McMasterUniversity, 2015.

[3]SAEJ2399,Adaptivecruisecontrol(ACC)operatingcharacteristicsanduserinterface,SaeJ2399_201409,SAEInternational, 2014.

[4]P.Shakouri,A.Ordys,D.S.Laila,M.Askari,Adaptive CruiseControlsystem:comparinggain-scheduling PIandLQ controllers,IFACProc.Vol.44(2011)12964–12969.

[5]A.Eriksson,N.A.Stanton,he1y13/TrafﬁcQuery_STISIM3:atrafﬁcqueryingclassforstisimdrivesimulationkernel–Build 3.07.04,(2017).

[6]J.W.Choi,R.Curry,G.Elkaim,Pathplanningbasedonbeziercurveforautonomousgroundvehicles,Wcecs2008:Advances inElectricalandElectronicsEngineering-IaengSpecialEditionoftheWorldCongressonEngineeringandComputer Science,(2009),pp.158–166.

[7]T.Hessburg,M.Tomizuka,Fuzzylogiccontrolforlateralvehicleguidance,IEEEControl.Syst.14(1994)55–63. [8]A.Eriksson,N.A.Stanton,Takeovertimeinhighlyautomatedvehicles:noncriticaltransitionstoandfrommanualcontrol,

Hum.Factors59(2017)689–705.

[9]A.Eriksson,N.A.Stanton,Drivingperformanceafterself-regulatedcontroltransitionsinhighlyautomatedvehicles,Hum. Factors59(2017)1233–1248.

(16)

[10]SAEJ3016,Taxonomyanddeﬁnitionsfortermsrelatedtodrivingautomationsystemsforon-roadmotorvehicles, J3016_201609,SAEInternational,2016.

[11]A.Eriksson,V.A.Banks,N.A.Stanton,Transitiontomanual:comparingsimulatorwithon-roadcontroltransitions,Accid. Anal.Prev.102(2017)227–234.

[12]F.Naujoks,C.Purucker,A.Neukum,S.Wolter,R.Steiger,Controllabilityofpartiallyautomateddrivingfunctions–doesit matterwhetherdriversareallowedtotaketheirhandsoffthesteeringwheel?Transp.Res.PartFTrafﬁcPsychol.Behav.35 (2015)185–198.

[13]T.Miles,CarscoulddrivethemselvessoonerthanexpectedaftereuropeanpushRetrievedfrom,(2014).https://www. reuters.com/article/us-daimler-autonomous-driving/cars-could-drive-themselves-sooner-than-expected-after-european-push-idUSKBN0DZ0UV20140519.

[14]L.Bainbridge,Ironiesofautomation,Automatica19(1983)775–779.

[15]N.B.Sarter,D.D. Woods,C.E.Billings, Automationsurprises,in:G. Salvendy(Ed.),HandbookofHumanFactors& Ergonomics,2nded.,Wiley,1997.

[16]J.C.F.deWinter,R.Happee,M.H.Martens,N.A.Stanton,Effectsofadaptivecruisecontrolandhighlyautomateddrivingon workloadandsituationawareness:areviewoftheempiricalevidence,Transp.Res.PartFTrafﬁcPsychol.Behav.27(2014) 196–217.

[17]A.H.Jamson,N.Merat,O.M.J.Carsten,F.C.H.Lai,Behaviouralchangesindriversexperiencinghighly-automatedvehicle controlinvaryingtrafﬁcconditions,Transp.Res.PartCEmerg.Technol.30(2013)116–125.

[18]N.A.Stanton,M.S.Young,Vehicleautomationanddrivingperformance,Ergonomics41(1998)1014–1028.

[19]N.A.Stanton,M.S.Young,B.McCaulder,Drive-by-wire:thecaseofmentalworkloadandtheabilityofthedrivertoreclaim control,Saf.Sci.27(1997)149–159.

[20]M.S.Young,N.A.Stanton,Backtothefuture:brakereactiontimesformanualandautomatedvehicles,Ergonomics50 (2007)46–58.

[21]S.M.Casner,E.L.Hutchins,D.Norman,Thechallengesofpartiallyautomateddriving,Commun.ACM59(2016)70–77. [22]B.D.Seppelt,T.W.Victor,Potentialsolutionstohumanfactorschallengesinroadvehicleautomation,RoadVehicle

Automation3,Springer,2016,pp.131–148.

[23]E.R.Boer,M.D.Penna,H.Utz,L.Pedersen,M.Sierhuis,Theroleofdrivingsimulatorsinevaluatingautonomousvehicles, PaperPresentedattheDrivingSimulationConference(2015).

[24]B.D.Greenshields,Reactiontimeinautomobiledriving,J.Appl.Psychol.20(1936)353–358.

[25]L.Nilsson,Safetyeffectsofadaptivecruisecontrolincriticaltrafﬁcsituations,PaperPresentedatthetheSecondWorld CongressonIntelligentTransportSystems:StepsForward(1995).

[26]N.A.Stanton,P.Marsden,Fromﬂy-by-wiretodrive-by-wire:safetyimplicationsofautomationinvehicles,Saf.Sci.24 (1996)35–49.

[27]O.Carsten,A.H.Jamson,Drivingsimulatorsasresearchtoolsintrafﬁcpsychology,in:B.Porter(Ed.),HandbookofTrafﬁc Psychology,vol.1,AcademicPress,2011,pp.87–96.

[28]J.C.F.deWinter,P.VanLeeuwen,P.Happee,Advantagesanddisadvantagesofdrivingsimulators:adiscussion,Paper PresentedattheProceedingsofMeasuringBehavior,(2012).

[29]J.Flach,S.Dekker,P.J.Stappers,Playingtwentyquestionswithnature(thesurpriseversion):reﬂectionsonthedynamicsof experience,Theor.IssuesErgon.Sci.9(2008)125–154.

[30]L.Nilsson,Behaviouralresearchinanadvanceddrivingsimulator-experiencesoftheVTIsystem,PaperPresentedatthe ProceedingsoftheHumanFactorsandErgonomicsSocietyAnnualMeeting,(1993).

[31]G.Underwood,D.Crundall,P.Chapman,Drivingsimulatorvalidationwithhazardperception,Transp.Res.PartFTrafﬁc Psychol.Behav.14(2011)435–446.

[32]S.T.Godley,T.J.Triggs,B.N.Fildes,Drivingsimulatorvalidationforspeedresearch,Accid.Anal.Prev.34(2002)589–600. [33]J.Santos,N.Merat,S.Mouta,K.Brookhuis,D.deWaard,Theinteractionbetweendrivingandin-vehicleinformation systems:comparisonofresultsfromlaboratory,simulatorandreal-worldstudies,Transp.Res.PartFTrafﬁcPsychol.Behav. 8(2005)135–146.

[34]W.vanWinsum,K.A.Brookhuis,D.deWaard,Acomparisonofdifferentwaystoapproximatetime-to-linecrossing(TLC) duringcardriving,Accid.Anal.Prev.32(2000)47–56.

[35]OpenDS,Opends:theﬂexibleopensourcedrivingsimulation,(2017).https://www.opends.eu/.

[36]D.R.Large,G.Burnett,A.Bolton,Augmentinglandmarksduringthehead-upprovisionofin-vehiclenavigationadvice,Int. J.MobileHum.Comput.Interact.(IJMHCI)9(2017)18–38.

[37]D.R.Large,E.Crundall,G.Burnett,C.Harvey,P.Konstantopoulos,Drivingwithoutwings:theeffectofdifferentdigital mirrorlocationsonthevisualbehaviour,performanceandopinionsofdrivers,Appl.Ergon.55(2016)138–148. [38]R.C.McIlroy,N.A.Stanton,L.Godwin,A.P.Wood,Encouragingeco-drivingwithvisual,auditory,andvibrotactilestimuli,

IEEETransactionsonHuman-MachineSystems47(2017)661–672.

[39]C.Neubauer,G.Matthews,D.Saxby,FatigueintheautomatedvehicleDogamesandconversationdistractorenergizethe driver?PaperPresentedattheProceedingsoftheHumanFactorsandErgonomicsSocietyAnnualMeeting(2014). [40]C.Neubauer,G.Matthews,D.Saxby,Theeffectsofcellphoneuseandautomationondriverperformanceandsubjective

stateinsimulateddriving,ProceedingsoftheHumanFactorsandErgonomicsSocietyAnnualMeeting56(2016)1987– 1991.

[41]G.D.Park,R.W.Allen,T.J.Rosenthal,Novicedriversimulationtrainingpotentialforimprovinghazardperceptionand self-conﬁdencewhileloweringspeedingriskattitudesforyoungmales,PaperPresentedattheProceedingsoftheEighth InternationalDrivingSymposiumonHumanFactorsinDriverAssessment,TrainingandVehicleDesign,(2015). [42]R.W.Allen,T.J.Rosenthal,B.L.Aponso,G.Park,Scenariosproducedbyproceduralmethodsfordrivingresearch,assessment

andtrainingapplications,PaperPresentedattheProceedingsoftheDrivingSimulationConference,(2003). [43]W.Allen,T.Rosenthal,B.Aponso,Z.Parseghian,M.Cook,S.Markham,Ascenariodeﬁnitionlanguagefordevelopingdriving

simulatorcourses,PaperPresentedattheDrivingSimulationConference,(2001).

[44]R.Allen,T.Rosenthal,J.Hogue,F.Anderson,Low-costvirtualenvironmentsforsimulatingvehicleoperationtasks,Paper Presentedatthe78thAnnualMeetingoftheTransportationResearchBoard(1999).

(17)

[45]J.C.F.deWinter,P.A.Wieringa,J.Kuipers,J.A.Mulder,M.Mulder,Violationsanderrorsduringsimulation-baseddriver training,Ergonomics50(2007)138–158.

[46]G.J.Funke,Theeffectsofautomationandworkloadondriverperformance,subjectiveworkload,andmood.doctoral dissertation,UniversityofCincinnati,2007.

[47]G.Funke,G.Matthews,J.S.Warm,A.K.Emo,Vehicleautomation:aremedyfordriverstress?Ergonomics50(2007)1302– 1323.

[48]C.Neubauer,Theeffectsofdifferenttypesofcellphoneuse,automationandpersonalityondriverperformanceand subjectivestateinsimulateddriving,UniversityofCincinnati,2011.

[49]C.Neubauer,G.Matthews,L.Langheim,D.Saxby,Fatigueandvoluntaryutilizationofautomationinsimulateddriving, Hum.Factors54(2012)734–746.

[50]D.J.Saxby,G.Matthews,E.M.Hitchcock,J.S.Warm,Developmentofactiveandpassivefatiguemanipulationsusinga drivingsimulator,ProceedingsoftheHumanFactorsandErgonomicsSocietyAnnualMeeting51(2016)1237–1241. [51]D.J.Saxby,G. Matthews,E.M.Hitchcock,J.S.Warm,G.J.Funke, T.Gantzer,Effectofactiveandpassive fatigueon

performanceusingadrivingsimulator,ProceedingsoftheHumanFactorsandErgonomicsSocietyAnnualMeeting52 (2008)1751–1755.

[52]K.Kircher,A.Larsson,J.A.Hultgren,Tacticaldrivingbehaviorwithdifferentlevelsofautomation,IEEETrans.Intell.Transp. Syst.15(2014)158–167.

A toolbox for automated driving on the STISIM driving simulator

Delft University of Technology