版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
第1题Duringthe8thFive-Yearperiod____,China’sfirstautonomousvehicle____isdeveloped.正确答案::1991-1995正确答案::ATB-1第2题DriveGPTcombines____andautonomousdrivingtodoend-to-endlearning.正确答案::GPT第3题Exceptfordrivingontheurbanroads,somecompaniesaretryingtosolvetheproblemofautonomousdrivingincampus,airportandsoon.()第4题SimilartotheGrandandUrbanchallengesheldbyDARPA,thefirst“IntelligentVehicleFutureChallenge()”washostedbyNationalNaturalScienceFoundationofChina()in2009.()第5题Manyfamouscontrolmethodsforautonomousdriving,likeStanleymethod,weredevelopedinDemo’97.()第6题Manyresearchersinautonomousdrivingcommunitymentionthe1939WorldFairasthefirstexhibitofself-drivingcars.()第7题Theautonomousvehiclenamed“Stanley”builtbytheStanforduniversitywonthechampionshipinthesecondGrandchallenge.()第8题Forglobaldecisionmakingandpathplanning,thesystemneedstodeterminethedestinationandthenfindawaytothedestinationfromthecurrentposition.()第9题Theend-to-endlearningmethodstrictlyfollowsthreetasks:environmentalperception,decisionmakingandplanning,motioncontrol.()第10题Level4meansnoautonomy,thevehicleshouldbecontrolledtotallybythedriver.()第11题AccordingtoSAE,thelevelofautonomycanbeclassifiedinto6levels,fromlevel0tolevel5.()第12题Theconstructionoftestroadsforintelligentvehiclesprovides()AdifferentscenariosBadvancedfacilitiesCcontrollablerisksforthetestDverificationofintelligentvehicletechnologies正确答案:ABCD第13题Avehicleforfullyautonomousdrivingmaynothave()AcentralcomputerBbreakpedalCacceleratorpedalDsteeringwheel正确答案:BCD第14题Forenvironmentalperception,theautonomousvehicleusedifferentsensorsasitseyes,like()ARadarBLidarCGPUDCameras正确答案:ABD第15题Thearchitectureforatypicalautonomousvehiclehasthreecomponents()AEnvironmentalperceptionBDecisionmakingandplanningCBrakingsystemDMotioncontrol正确答案:ABD第16题Thelevelofautonomydependsonhowmanydriverinputsareinvolvedduringdriving.Driverinputsinclude()AthecontrolofsteeringwheelBthebrakepedalCtheacceleratorpedalDLidar正确答案:ABC第17题InDemo’97,autonomousdrivingtechnologiesfor()wastested.AsinglevehiclesBmultiplevehiclesCmultipleplatformsDcooperativevehicleinfrastructuresystem第18题TheDefenseAdvancedResearchProjectsAgency()intheUSAlaunchedthefirstGrandChallengein()A2003B2004C2005D2006正确答案:DAB第19题Anautonomousvehicleisabletofindapathtochangelane.Thisabilitybelongsto()AglobaldecisionmakingBglobalpathplanningClocaldecisionmakingDlocalpathplanning第20题At()inSAE,thevehicleisinchargeofalldrivingtasksandcanoperateinallenvironmentswithouthumaninterventions.ALevel5BLevel4CLevel3DLevel2第1题Theprocessofmachinelearningis:()AGatheringdata——Preparingthatdata——HyperparameterTuning——Choosingamodel——Training——Evaluation——PredictionBChoosingamodel——Training——Evaluation——HyperparameterTuning——Gatheringdata——Preparingthatdata——PredictionCGatheringdata——Preparingthatdata——Choosingamodel——Training——Evaluation——HyperparameterTuning——Prediction第2题InPython,the‘break’statementisusedtoterminatethecurrentloopandresumeexecutionatthenextstatementaftertheloop.()第3题Pythonsupportsbothsingle-lineandmulti-linecomments.()第4题Pythonisaninterpretedlanguage,whichmeansthatthesourcecodeisexecutedlinebyline.()第5题The‘len()’functioninPythonreturnsthelengthofasequence.()第6题Pythoniscompiledprogramminglanguage.()第7题Theactivationfunctionisveryhelpfulforimprovingthemodel'snonlinearexpressionability.()第8题InCNN,differentconvolutionkernelswillextractdifferentfeaturedata.()第9题ThepoollayerinCNNwillperformadownsamplingoperationalongthespatialdimensions.()第10题Machinelearningisasubsetofdeeplearning.()第11题TheK-meansalgorithmneedstomanuallyspecifythenumberofclusters.()第12题Thedataofunsupervisedlearninghavenolabels,andtheunsupervisedlearningmodelfindsthecharacteristicsofdataandclusterthesedataintodifferentcategories.()第13题Classificationisatechniqueforidentifyingsimilaritygroupswithindata.()第14题Whenthemodelhasconverged,theoveralllossstopschangingoratleastchangesextremelyslowly.()第15题Classificationalgorithmscanpredictonlybetweentwocategories()第16题Thetwo-classclassificationalgorithmtriestofindaboundarythatcanclassifythetrainingdataintotwocategories.()第17题Standardconvolutionalneuralnetworksgenerallyconsistofaninputlayer,(),andanoutputlayer.AaconvolutionallayerBapoolinglayerCarandomforestLayerDafullyconnectedlayer正确答案:ABD第18题ActivationfunctionsinANNinclude:()AlinearfunctionBsigmoidfunctionCRELUfunctionDtanhfunction正确答案:BCD第19题Whatisthepurposeofthe‘range()’functioninPython?()AGeneratealistofnumbersBPerformarithmeticoperationsCIterateoverasequenceofnumbers第20题Whatwillbetheoutputoftheabovecode?()A7B3C123.8Exercises第1题Formodel-basedRL,beforeupdatethevalueandpolicyusingtherealexperiencecollectedfromtherealworld,wecanuseaworldmodeltodotheupdateandinteractionformanytimes.()第2题DeeplearningintheDQNisoftenusedasavaluefunctionapproximator.()第3题Ifthelearningrateistoolarge,wemaytakeaverylongtimetogettothebottom.()第4题Theoptimalactionvaluefunctionisindependentofthepolicybeingfollowed.()第5题Thestate-valuefunctionisusuallyrepresentbyq,sothestatevaluefunctionisusuallycalledQfunction.()第6题ThestatetransitionofMDPistotallydeterminedbythepresentstate,historicalstateshavenoeffectsonthetransition.()第7题Usingthegreedypolicywillleadtothemaximizedvalueaccordingtocurrentknowledgeofthesystem,thus,itshouldbelargerthanoratleastequaltotheoriginalvalue.()第8题TheproblemsthatcanbesolvedbyDynamicprogrammingshouldhaveanoptimalsubstructure,meansoptimalsolutioncanbedecomposedintosubproblems.()第9题Thedifferencebetweenstate-valuefunctionandaction-valuefunctioniswhethertheactionisinvolvedornot.()第10题SARSAisanoff-policyalgorithm.()第11题Markovdecisionprocessisthebasisofreinforcementlearningalgorithms.()第12题TheinputtothedeepneuralnetworkinDeepQ-Learningis()AstateBpolicyCactionDreward第13题Inepsilongreedypolicy,weoftenchoose()withprobability1-AthegreedyactionBanactionatrandomCthegreedystateDthegreedyreward第14题AteachtimestepinTDlearning,wereceiveanimmediate(),andthenwecombineitandtheestimationforthestatevalueofthesuccessorstatetogettheestimatedreturn.AactionvalueBstatevalueCrewardDlearningrate第15题()estimatehowgooditisfortheagenttobeinagivenstate.ALossfunctionBRewardfunctionCState-valuefunctionDAction-valuefunction第16题Formally,()isgoinglikethis.Ateachiteration+1,Forallstatesinthestateset,Updatefrom,usingtheupdaterule.ApolicyinitializationBpolicyupdateCpolicyevaluationDpolicyimprovement第17题A()isadistributionoveractionsgivenstates,mappingfromstatestoprobabilitiesofselectingeachpossibleaction.AenvironmentBagentCpolicyDreward第18题Q-learningalgorithmusesQtabletoexpressvaluefunction.()第19题ThestatetransitionmatrixisnotusefulinthecalculationofMDP.()第20题Forallthestate-actionpairs,usetheupdateruletoupdateQtable.Utilconverge.Oncetheoptimal()isfound,theoptimalpolicycanbefoundthereafter.AQBVCrewardDstate4.6Exercises第1题Learningbasedmethodscannotintroduceglobalsemanticinformationsuchasmirrorpriorsandreflectionpriorstomakematchingmorerobust.()第2题Feature-basedmethodsrelyonidentifyingdistinctivefeaturesintheenvironment,suchascornersoredges,andmatchingthembetweendifferentsensordataframes.()第3题Comparedwithdirectcameraposeprediction,generatingmultiplecameraposeassumptionscanultimatelyimprovetheaccuracyofposeprediction.()第4题ResFlowNetcannotonlycorrecterrorsinthepredictionofmovingobjects,butalsocorrecttheimperfectresultsofthefirststage.()第5题ThemathematicaldescriptionofSLAMproblemisusuallydividedintomotionequationandobservationequation.()第6题Thebehaviorofpedestriansishighlyuncertainandeasilyaffectedbythesurroundingenvironment.()第7题Inmixedtrafficscenarios,cyclistsoftenhaveflexiblesteeringandspeedadjustmentbehaviors,which,duetotheirstrongrandomness,willposeagreatrisktotheneighboringautonomousvehicle.()第8题Boundingboxisaboxaroundthedetectedobject,whichisusuallyusedtodescribethespatiallocationofanobject.Thepositionoftheboundingboxisdeterminedbyitstop-leftcornerandbottom-rightcorner.()第9题BeiDousatellitenavigationsystemconsistsof35satellites,whichcanprovidepreciselocalizationinformationandhasbeenwidelyusedforvehiclenavigationincities.()第10题ThereareproblemsandchallengesinexistingBA:().ASensitivetoinitialization,luminosityerrorincreasesnon-convexity.BSensitivetochangesincameraexposureandwhitebalance.CSensitivetooutliers,suchasmovingobjects.DSensitivetobuildsensegraph.正确答案:ABC第11题Typically,AConvolutionalNeuralNetworkhas().AinputlayerBconvolutionallayerCpoolinglayerDfullyconnectedlayerandoutputlayer正确答案:ABCD第12题Oneconvolutionalkernelcorrespondstoonechannel.()第13题KalmanFilterisusuallyusedforcombiningGPSandIMU.()第14题(),oneofthemostchallengingplacerecognitiondatasetstodate,consistsoffourtimesynchronizedvideosoftrainjourneysthroughNorway.ATheWaymodatasetBTheNordlanddatasetCTheKITTIdatasetDTheAIRdataset第15题()measuresdistanceswithwell-definedglobalminimumsandsmoothedbasins.AInputsBRGBCC3DF3第16题()calculatestheaverageEuclideandistancebetweeneachpointinthepredictedtrajectoryandthecorrespondingpointintherealtrajectory.ALSTMBADECCNNDRNN第17题Wecanmodelthetrajectorypredictionproblemasa()problem.AregressionBclusterCplanningDcontrol第18题()canbeusedtodealwiththelinearlyinseparableproblem,byfindingahyperplanethatcandividedifferentkindsofdata.ASVMBLSTMCCNNDGNN第19题Wemodelthebehaviorrecognitionproblemasa()problem.AclusterBclassificationCplanningDcontrol第20题Wecanuse()toextractskeletoninformationfromthecroppedimages.AcontrolBplanningCOpenPoseDSLAM5.5Exercises第1题InPOMCP-DPWbasedlane-changingdecisionmakingsystem,themeanandstandarddeviationoftheparticlesamplingatthenextmomentneedtoberedesignedbasedontheaccuracyoftheparticle'sestimationofthesurroundingvehicle'spositionatthepreviousmoment.()第2题POMCP-HDPWderivesfromtheoriginalversionofPOMCP-DPWwithheuristicallypruningalgorithm,avoidinguselessexploration.()第3题InPOMDPbasedlane-changingdecisionmakingsystem,Gaussiannoiseisaddedtotherealstatetoconsidertheuncertaintyofperceptionerror.()第4题InthestatespaceofMDP,theunobservablestateofsurroundingvehicleincludesreferencespeedanddesiredtimegapthatcontrolthefuturemotionofsurroundingvehicle.()第5题Atpresent,thedesignofautonomousdrivingsystemhastwodistinctroutes,includingmodulardesignideasandend-to-enddesignideas.()第6题Thepredictionunitofsocialpreferenceisbuilttoestimatesocialpreferenceineachtimestep.()第7题Insimpleovertakingscenarioonstraightroadmodeledinourcourse,motionprimitivesaretrajectoriescombinedwithcontrolsequences.()第8题IntheovertakingdecisionmakingmethodundertheframeworkofHRL,thealgorithmselectssuitableoptionsprovidedbythemotionprimitivelibraryindifferentovertakingphases.()第9题TheNGSIMdatasetconsistsofvariousvehicletrajectorydatafrommultiplehighways,itcanbeobtainedforafeefromtheirofficialwebsite.()第10题AfterinstallingwiththecompliedversionofCARLA,package“carla”hasbeenbuiltintothefolder“PythonAPI”.()第11题InCARLA,apythonscriptcancollectthevehicleinformationfromtheclientandsendthecontrolcommandstoserver.()第12题CARLAisbasedontheUnrealEngine4,whichprovideshighrenderingqualityandrealisticphysics.()第13题Thestatetransitionprocessofstatemachineisbasedonpredefinedrules.()第14题InthestatespaceofMDP,theobservablestateofvehicleincludes()ApositionBspeedCheadingangleDall第15题Inatypicalautonomousdrivingtask,thedecisionmakingandpathplanningshouldworkinorder()①Preparingthemapinformation,②Localpathplanning,③Globalpathplanningusingthemapinformation,④DecisionmakingA①③②④B①③④②C②①③④D①④③②第16题InovertakingscenariowithsocialpreferencemodeledunderthearchitectureofRLinourcourse,thestatespaceincludesthe()ofegoandovertakenvehiclesinthelatticespace.ApositionsBvelocityCaccelerationDsocialpreferencelabel正确答案:AB第17题InHierarchicalreinforcementlearningproblemformulatedinovertakingdecisionproblem,trafficefficiencyandtrafficcomfortareevaluatedseparatelyby().AtraveltimeBaveragelateralaccelerationCaveragelateralvelocityDminimumsafedistance正确答案:AB第18题Whatisnotthedisadvantageofrule-baseddecisionmakingmethodinfollowingoptions()AoverrelianceonexpertknowledgeBdifficulttoconsiderinteractionandscenariouncertaintyCpoorreal-timeperformanceDdifficulttodesignrulesincomplexscenarios第19题()isgeneralizedfromMDPwheretheactionscantakeavariableamountoftimetocomplete.AHMMBMRPCPOMDPDSMDP第20题TheinformationacquisitionmoduleinCARLAcanbeusedtoobtaintheinformationofthe()AspeedBpositionCangleoftheactorDall6.6Exercises第1题Inthecar-followingscenario,wehaveanegovehiclecontrolledbyourselves,anda()controlledbyanotherdriver.AfrontvehicleBmainvehicleCfollowingvehicle第2题IntheRLmodeloflearningfromhumandrivers,thestatecanbethedifferencebetweenyawangleoftheautonomousvehicleandtheyawangleofahuman-drivenvehicle.()第3题Experiencedhumandriversalwayshaveabetterchoicethanthealgorithms,especiallyinthecomplexdrivingenvironment.()第4题Traditionalautonomousdrivingsystemssufferfromlowefficiencyinthedynamictrafficenvironmentbecauseoftheirrigidplanningandcontrolmodules.()第5题Forthepathtrackingproblem,atwodimensionalstateset,xandtheta,issufficienttofullydescribethepathtrackingproblem.()第6题ForusingMPC,wedonotneedtoknowthevehicledynamicmodel.()第7题Steeringcontroloutputstheyawangleateachtimestepofthevehicle.()第8题Lanekeepingreferstotheabilityoftheautonomousvehicletomaintainitselfinthelanebyadjustingitselfwhenitapproachestheboundariesofthelane.()第9题Rewardisdefinedasfunctionofstateandaction.()第10题Pathplanningispartoflongitudinalcontrolinautonomousdrivingsystems,responsiblefordeterm
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 小区物业服务保洁合同
- 宜宾房产交易合同
- 2026湖北鄂州人才集团招聘派往中核湖北鄂城新区联合投资有限公司6人备考题库附答案详解ab卷
- 2026陕西西安电子科技大学附属中学招聘备考题库附答案详解(培优)
- 2026北京大学电子学院招聘劳动合同制工作人员1人备考题库附答案详解(达标题)
- 2026江苏宿迁市苏州外国语实验学校招聘事业编制教师8人备考题库附答案详解(培优a卷)
- 宽带物业合作合同
- 2026新疆铁门关市人民医院招聘事业单位人员36人备考题库附答案详解(培优)
- 2026浙江金华市联创智绘大数据科技有限公司工作人员招聘3人备考题库及答案详解(历年真题)
- 2026山东司法警官职业学院招聘42人备考题库含答案详解(巩固)
- DB31-T 1433-2023 扬尘在线监测技术规范
- 【MOOC】融合新闻:通往未来新闻之路-暨南大学 中国大学慕课MOOC答案
- JGJT46-2024《施工现场临时用电安全技术标准》条文解读
- (高清版)TDT 1013-2013 土地整治项目验收规程
- 一年级数学下册 期中综合模拟测试卷(人教浙江版)
- 银行客户经理考试:建行对公客户经理考试题库考点
- 初中八年级数学课件-一次函数的图象与性质【全国一等奖】
- GB/T 7969-2023电缆用纸
- 内分泌科慢性肾上腺皮质功能减退症诊疗规范2023版
- 《世界名画蒙娜丽莎》课件
- 春小麦田间管理子肥水控制(春小麦栽培课件)
评论
0/150
提交评论