VcycleBordeauxv13pdf智能交通世界大会ITS智慧城市社区人工智能AI物联网IT报告课件教案

VcycleBordeauxv13pdf智能交通世界大会ITS智慧城市社区人工智能AI物联网IT报告课件教案 22nd ITS World Congress,Bordeaux,France,59OctoberxxITS-2658An integral approach to autonomous and cooperative vehicles development and testing IgorPasschier*,Gwen vanVugt TASS International,Einsteinlaan6,2289CC Rijswijk,The Netherlands,*igor.passchiertassinternational. AbstractThe plexityin cooperative and automated vehicles increasesexponentially pared to traditionalvehicles.An integrated tool suitsupporting the full development V-cycle iscrucial toenable costand timeefficient development processes.This paperpresents such an approach.For development,State of the artsimulation platformsfor active safety and automated systems are required,ranging inscale fromplete traffietworks,down toponent levelsimulation tools.For testing,Hardware in the loop testing isnecessary forsensor andmunication systems,while a dedicated testenvironment forrapid,safe,and reproducibletesting of cooperative and automated vehiclesis required to testplete cooperative and automated systems.Finally,For validation and performancetesting,a test site forurban,interurban and highway is available.Keywords:cooperative systems,automated systems,developmentV-cycle,HIL testing,simulation models.Introduction Twomajor innovationdomains for the automotiveindustry areautomation andcooperation.In theautomation domain,vehicles beemore intelligentbased on their ownsensor input,and takesover moreand moretasks from the driver.This gradualprocess startswith(safety)warning systemslike lanedeparture warning,taking overtasks inspecial situations(emergency breaking,parking,driving intraffic jams),and couldend inpletely autonomouslydriving vehicles.A similargradual development is to be expectedfor cooperative systems,starting withsimple warningapplications(traffic jamahead)or drivingassistant applications(green lightspeed advise),via moresafety relatedapplications(electronic emergencybrake light)towards fullvehicle platooning.When takingthese twoAn integral approach to autonomous and cooperative vehicles development and testing2developments together,the ultimateconsequence isfully autonomousplatooning,possibly inany situation,where safety,fort,fuel and traffic efficiencyare alltaking into aount andimproved,paredto the currentsituation.See alsoFigure1for howthese twodevelopments arerelated toeach other.Figure1Development of autonomous and cooperative systems(from1)These innovationsintrinsically increasethe plexityof the systems themselvesand of the development andtestprocess.If theautonomy increases,the dependencyon the vehicle sensorsincreases.Multiple sensorsneeds to be binedto improvethe redundancyand reliability,and itneeds to be guaranteedthat theinformation remainscorrect alsoin more extreme scenarios:where somefalse warningsmight beaeptable for a driverwarning systems,such errorscould leadto unaeptablesituations ifthe vehiclewill takeautonomous action.Where traditionallythe vehiclecan beconsidered as“the pletesystem”,for cooperative systems alwaysmultiple vehiclesand/or roadsideunits areincluded insystem.In summary,for autonomous and cooperative systems,plexity increasesdue tohigher constraintson a)sensor auracy,b)sensor fusion,c)application auracy,d)support ofmoreextremescenarios,e)involvement ofmultiple vehicles,and f)inplete information,and g)an overloadof data.If thetraditional development andtest methodologies thathave workedfor manydecades aremaintained,the timeit takesfor newsystems toenter themarket willincrease,and innovationcould eto a plete standstill.New,more efficientdevelopment andtestmethodologiesare requiredto allow for aeptabletime scalesof theinnovation cycle.New tools for bothdesign anddevelopment on the onehand and testing on the otherare requiredto An integral approach to autonomous and cooperative vehicles development and testing3support theseprocesses.This paperpresents such a newand integratedapproach andtool suitefor supportingthe development and testing of cooperative and automatedsystem.This tool suite consistsof detailedsimulation tools,hardware in the looptesting facilities,closed andopen testtracks,and actualfield tests.The developmentV-model Awell-known model for system developmentis the V-model.Many differentvariants exist.They allconsist ofa leftside forthe designofa system,from thehighest level,down to the detailsof thesmallest ponents.On theright side,the ponentsare integratedand testedinto larger(subsystems),finally endingin thevalidation of the pletesystem.Figure2Development V-modelfor autonomous and cooperative systemsWe haveconstructed aV-model specificallyfor cooperative and autonomoussystems,see Figure2.At thehighest level,we definenot a single vehicle,but multiplevehicles together,identified astraffic.Especially for cooperativesystemsthis is essential,as everycooperative applicationinvolves alwaysmore actorsthan a single vehicle.These actorsinclude bothother vehicles(cooperative,non-cooperative,automated ornot),and roadsidesystems(cooperative ornon-cooperative).One levelbelow the vehicle andits applicationsare identified.These aremade ofsubponents,like munication units,application platforms,or ECUson whichthe functionsare implemented.At thesmallest detaillevel,we identifythe individualponents,like sensors and actuators.Important sensors are sensorsmonitoring theexterior of the vehicle,like radar,lidar,cameras,or provideinformation from the vehicleitself,like IMUsensors,wheel sensors,and allkinds of performance relatedsensors.An integral approach to autonomous and cooperative vehiclesdevelopment and testing4Traditionally,the designphase issupported withsimulation toolsto optimizethe behaviorof the system,subsystem,or ponents,while testingtools support thesystem validation.To supporta moretight integrationbetween designand testinghardware in the loop(HIL)simulation and testing can be used.In thisway,part of thesystemcan besimulated,while otherponents can be realizedby theactual hardwareimplementation.TASS Internationalhas developed,together withits partners,a pletetool suiteto supportdesign,development,testing,and validation,including HILtesting,forthe development forcooperative and automated vehiclessystems.In thefollowing sections,some examples of thesetools arepresented,and howthey areintegratedto allow forrapid productand systemdevelopment.Figure3The toolsuite developed by TASSand partnersto supportthe developmentand testingofcooperativeand automated systems SimulationTools Manysimulation toolsfor individualsensors on the ponentlevel arereadily available.For municationsystems,simulations toolslike ns-2or OPNETprovide verydetailed models.For opticalcamera systems,raytracing toolslike Zemax2can be used.Multiple optionsexist tointegrate thesetools intoa simulation toolsforthe subsystemand vehiclelevels.PreScan3isthephysics basedsimulation platformdevelopedbyTASS Internationalto supportthe developmentof Advanceddriver assistancesystems,automated vehiclesandcooperativesystems.It offerssimulation ofsensor technologieslike laser/lidar,radar,and cameras.Integration withexisting,state of the artsimulation toolsfor individualponents isrealized viathree differentmethods:An integralapproach toautonomous andcooperative vehiclesdevelopmentand testing5?The pluginfacility ofPreScan allows foradirect couplingbetween an external simulator and PreScan.In thisway,it is possible tohave PreScansimulate acooperative applicationand transmita messagefrom onevehicle andhand-off thepackage toan external simulator likens-3.The externalsimulator canthen determinewhether themessage willbe receivedby anothervehicle,and thetransmission delay.The messageis handedback againtoallthe receivingvehicles in PreScan.?Instead ofa pletecoupling,it isalso possible to extractasinglemodel froman existingsimulator,and integratesuchamodel directlyinto PreScan.This allows for betteroptimization,and canprevent integrationissues likeintegrating eventbased simulatorsin the(time based)simulator enginecontained in PreScan.?A thirdoption isto performsimulations with anexternalsimulator,and parameterizethe results.A dedicated sensor isthen implementedin PreScanthat istuned with the parametersfrom theexternal simulationrun.This methodologyise.g.used tointegrate Zemaxsimulations forthe opticsof camera systems into the detailedcamera modelwhich ispart ofPreScan.This methodologyallowsfora properbalance between simulation timeand level of detail.Furthermore,this methodologyallows to use thesame modelinPreScan,but extractthe parameterse.g.from actualmeasurements.Which ofthe threeoptions isthe mostappropriate,depends stronglyon theapplication orsubsystem that is beingdeveloped,the requiredlevel ofdetail,and theavailable externalsimulators.Therefore,we havedecided tosupport allmethodologies andhave madeseveral(example)implementations forall three.PreScan iswell suitedfor simulationswith upto dozensof vehicles.To simulatethe effectof automatedandcooperativesystems on the trafficflow asa whole,a dedicatedsimulationtoollike ITS Modeler4from TNO,Visum byPTV5,or SUMO6by DLRcan be used.To thisend,the simulationresults ofPreScan can be usede.g.as inputfor ITSModeler,and viceversa,ITSModelercan be coupled togenerate“background”traffic forPreScan.Via theintegration ofsimulation toolsat the different levels,consistent resultscan beobtained and the effects of changeson onelevelofdetails can be reliablypropagated to the higherlevels.This allowsfor reliableand rapidsystemdevelopment,but alsoallows demonstratingeffectsofcooperativeand automotive systemsto otherstakeholders in the suessof thesesystems,like roadoperators andpolicy makers.Test tracksand carlabs For testing and validation,TASS Internationalprovides aplete fleetof test vehicles.Four carlabs are available.The standardequipment includesa Densofront radar,Mobileye camera,GPS,ITS G5municationunit,andafail-safe CANgateway(read&write).A dedicatedAn integralapproach toautonomous andcooperative vehiclesdevelopmentandtesting6HMI(display+sound)is availableand aesstothestandard buttons,lever,pedals anddashboard isprovided.A real-time embeddedapplication platformcan be used forrapid prototypingandtestingnew applications.Optional equipmentinclude360degree viewbased onan additionalfront radar,2rear radarsandaside radarfrom Continental.Also sixIbeo laser scanners with aplete360degree aroundthevehicle isavailable.A dGPSsystem fromOXTS isavailable ifabsolute auracyin thecm rangeis required.These carlabs can be used to testnew applicationsand subponentsbased on the provided tools.Vice versa,it isalso possibleto addadditional(prototype)sensorsandto testthem eithervia integrationin theexisting systemsand applications,or byparing themtotheexisting sensors.Figure4Field ofview ofthe radarand lidarsystems onCarlab#1Fortesting and validatingcooperativeandautomatedsystems,independent observationsystems arerequired that can determinenot onlythe detailsofthesystems under test,but alsoofthesurrounding traffic.Furthermore,also roadsidesystemsarerequiredtobuild aplete cooperativeenvironment.To thisend,the DITCM Facilities test site7has beendeveloped onthe urban,interurban,andhighwayN270/A270between Helmond and Eindhoven,The Netherlands.Over50cameras observeall trafficonthe test siteand providedtrajectories ofall vehicleswitharesolution betterthan1meter andan updatefrequency of10Hz8.Real timeinformation isavailable fromfour intersectionsonthe test site,provided byImtech onbehalf ofthe cityof Helmond.The fulltest site is coveredwith ITSG5munication units.With theseunits,all cooperativemunication onthe test siteisreceived andlogged,and can be analyzed.Standards basedapplication platformsareavailableto provideexisting ornew roadsidecooperative applications,like greenwave optimumspeed advise,traffic jamahead warning,or roadworks warning.The roadsidesensors,munication unitsand applicationplatforms An integralapproach toautonomousandcooperative vehiclesdevelopmentandtesting7canbeflexibly usedto provideapplications,or tobeusedas anindependent observationsystem.All equipmentonthetest siteis connectedvia afiber opticsmunication work.The deployedIT architecturebased onhardware virtualizationand databuses betweenall systems,allowsforflexible useof informationin applications,logging,real timeanalysis,and visualization.Data fromtest vehicles,including thecarlabs describedbefore,canbeadded tothis system in realtime oroff-line forfurther analysis.TNO has developed astructured data analysis toolthat enablesdata cleaning,event detectionand calculationofperformanceindicators,thatiscapable toanalyze thepotentially largedata setsgenerated bycooperativeandautomatedsystemstests9.Figure5Overview ofthe DITCM Facilities test site betweenHelmondandEindhoven TheDITCM facilitiestest siteis situatedon public roads,guaranteeing arealistic trafficsituation thatcannot easilybe realizedonaclosed testtrack.The Dutchministry oftransport haseffectuated apolicy tosupportthe testingofautomated drivingsystems onthe Dutchroads(not limitedtothe DITCMFacilities testsite).This policyenables testingprototype systemsthat wouldnot beallowed tobe testedin normal traffic conditionsotherwise.If technologyreadiness andsafety doesnot allowtesting onpublic roads,it isstill possibleto use thetestsite bymeans ofspecial arrangementswith governmentthat allowsthe closureofthetestsitefor normaltraffic.These arrangementsand policiesallowforreuse ofthetestsite inall stagesofthedevelopment process,which reducescosts andimplementation time,as thesame testenvironment andtools canbeusedthroughout thedifferent stages.Hardware inthe looptesting Theprevious sectionsdescribed howthedifferentlevels inthe V-model canbe integratedto An integralapproach toautonomousandcooperative vehiclesdevelopmentandtesting8facilitate thedevelopmentprocess:vertical integration.Horizontal integrationcanberealized intwo differentways.The firstmethod istousemeasurement dataas inputin simulation tools.The secondmethod isby integratinghardware inthe simulationenvironment,or couplesimulationtoolswith realsystems.Examples ofboth willbe given.TASS Internationaland partnershave equippeda sensorreference vehiclefor automatic scene generation.This vehicleis equippedwith IBEOforward lookinglaserscanners,andadedicatedsensorontherear todetect road markings.A sensorplatform extractsinformation fromall roadusers(vehicles,pedestrians,bicycles,etc.),roadmarkingsandtrafficsigns.Forward andbackward trackingallows objectidentification atoptimum visibility,and extrapolatesthe objectparameters overthefullperiod theobject isinthescene10.The extracteddata isconverted automaticallyin scenesthatcanbeused inPreScan.Figure6Sensor referencevehicle,used forautomaticscenegeneration.A similartechnique isused basedonthe data obtainedfromtheDITCMFacilitiestestsite.Anonymous vehicleinformation andall trafficlight informationcanbecollected24/7,a hugedatabase isgenerated fromwhich it ispossibleto extractsituations thatrarely our(e.g.a nearaident ata mergingramp whena vehicleis overtakinga truck),andaplete scenariocanbeextracted fromthe datasetto testapplications inthese realisticscenarios.An exampleof hardware inthe looptestis wherePreScan isusedtosimulate acooperative application,like astranded vehiclewarning.It isthen possibleto notusethe munication modelprovided byPreScan(oranexternalsimulatoras describedearlier),but touse actualAnintegralapproachtoautonomousandcooperative vehiclesdevelopmentandtesting9munication unitsin binationwithachannel emulator.In thatcase,the datatobetransmitted isforwarded fromPreScan tothe municationunit,in binationwiththeabsolute vehicleposition(required forproper functioningofthegeoworking layer).At thesame time,the trafficand environmentalscene isprovidedtothe workemulator tool,which emulatesthe municationchannel betweenthemunication units.Such acoupling hasbeen realizedthe Car-to-X SignalStrength Emulatorof Qosmotec,and withthe GraceNetwork emulatorof TNO.The ITinfrastructure oftheDITCMFacilitiestestsite allowsforatight integrationbetweensimulationand realsystems.Based onthevehicledetections ofthe camerasystem,itispossibletosimulate acooperative platformfor an arbitrary number of non-cooperativevehicles.The roadsidemunicationunitscan transmitthe messagesthat wouldotherwise betransmitted bythe vehiclesthemselves ifthey werecooperative.In thisway,asingletestvehicleonthetestsitecan experiencethe cooperativemunication fromanarbitrarynumberofcooperativevehicles.The advantageof thisapproach paredtoatraditional simulationor hardwareintheloop approachwhere thesystem undertest wouldbecoupledtoafully simulatedenvironment,is that all vehiclesensors functionnormally(as thevehicleisdriving ona normalpublicroadand issurrounded withnormaltraffic),andcooperativemunication isconsistent withthe inputfrom othersensors.A lastvariant ofhardwareinthelooptesting,is whereboth scenarioinput iscollected inreal situations,and actualhardware isusedinthetestingphase:no simulationsare performedatall.TASS Internationalhas developedsuchasystem fortesting andvalidating ofpositioning andtiming,which isused bythe Car2Car CommunicationConsortium forsystemvalidation.With suchasystem,RF GPSsignals andposition andtime relevantCAN busdata iscollected indifferent scenarios,like urban,rural,highway,tunnel,or mountainscenarios.At thesame time,adedicatedsystem measuresthe groundtruth.During testing,the signalsare replayedagainst theasystem undertestandtheposition andtiming informationfromthesystemundertest canbe paredtothecollected groundtruth data.One ofthe advantagesis thatdata collectionhas tobe doneonly once,and cantake placeover anextended timeperiod andextended geographicalarea.Actual testingcanbeperformed quicklyand repeatedly,and systemtuning andretesting underexact samecircumstances isguaranteed.If specialsensorsaretobeincluded inthedataacquisition phase,then dataacquisition needstoberedone,but theadvantages ofreproducibility andquick testingremains.Closed loopcontrol,where theSystem UnderTest wouldinfluence thevehicle motion,is of