在动态环境中强大图形大满贯带有移动地标lingzhu xiang1 zhile详解.pdfVIP

2015IEEE/RSJInternationalConferenceonIntelligentRobotsandSystems(IROS)

CongressCenterHamburg

Sept28-Oct2,2015.Hamburg,Germany

RobustGraphSLAMinDynamicEnvironments

withMovingLandmarks

LingzhuXiangZhileRenMengruiNiOdestChadwickeJenkins

—Recentdevelopmentsinhuman-robotinteraction

bringsabouthigherrequirementsforrobotnavigation.Existing

SimultaneousLocalizationandMapping(SLAM)algorithms

faceopenchallengesfornavigationincomplexdynamicen-

vironmentsduetopresumptionsofstaticenvironmentsor

exceedingcomputationallimitations.Inthispaper,wepro-

posearobustgraphSLAMformuationexploringExpectation

izationalgorithmstocharacterizelandmarkmobility

whileestablishingtheestimationsofrobottrajectoryandthe

map.WeevaluatetheperformanceofexistingrobustSLAM

algorithmsasbaselines,andvalidatetheimprovementofour

newframeworkagainstdatasetsofdynamicenvironmentswitha)iSAM/g2oN=2b)iSAM/g2oN=5

movinglandmarks.

I.INTRODUCTION

Simultaneouslocalizationandmapping(SLAM)isacen-

tralproblemforautonomousrobotstonavigateandperform

mobilemanipulationtasks.GraphSLAMformulatesitas

aninferenceproblemonafactorgraph.Inthefactorgraph,

spatialmeasurementsaretheobservedfactornodesascon-

straintsbetweenvariablenodesoflandmarklocationsand

robotposes.Thegoaloftheinferenceproblemistoobtainc)mobility-robustN=2d)mobility-robustN=5

theumlikelihoodestimateofthejointprobabilit