自动驾驶论文1556 Robust SRIF-based LiDAR-IMU Localization for Autonomous Vehicles.pdfVIP

2021IEEEInternationalConferenceonRoboticsandAutomation(ICRA2021)

May31-June4,2021,Xian,China

RobustSRIF-basedLiDAR-IMULocalizationforAutonomousVehicles

KunLi,ZhanpengOuyang,LanHu,DayangHaoandLaurentKneip

Abstract—Wepresentatightly-coupledmulti-sensorfu-Anincreasinglypopularsolutiontotheurbancanyonprob-

sionarchitectureforautonomousvehicleapplications,whichlemisgivenbydroppingGNSSmeasurementsaltogether,

achievescentimetre-levelaccuracyandhighrobustnessinandbyrelyingonaLiDARthatmeasures3Dscansofthe

variousscenarios.Inordertorealizerobustandaccurate

point-cloudfeaturematchingweproposeanovelmethodenvironmentandalignsthemwithaprior,geo-registered

forextractingstructural,highlydiscriminativefeaturesfrom3Dmap.Thepresentpaperproposessuchaframework.

LiDARpointclouds.ForhighfrequencymotionpredictionandHowever,mostLiDAR-basedlocalizationsolutionswith

noisepropagation,weuseincrementalon-manifoldIMUpre-priorpoint-cloudmaps[8],[17],[25]assumethattheroad

integration.Wealsoadoptamulti-frameslidingwindowsquarescenesarerelativelyconstant,whilenewconstructions,road-

rootinversefilter,sothatthesystemmaintainsnumerically

stableresultsunderthepremiseoflimitedpowerconsumption.sidevegetation,partialocclusionsbydynamicobjects,and

Toverifyourmethodology,wetestthefusionalgorithminmul-adverseweatherconditionsmayseverlycompromiserobust-

tipleapplicationsandplatformsequippedwithaLiDAR-IMUness.

system.