大范围环境下移动机器人同步定位和地图创建研究

上海交通大学硕士学位论文大范围环境下移动机器人同步定位和地图创建研究姓名：董海巍申请学位级别：硕士专业：控制理论与控制工程指导教师：李柠;陈卫东20080201VSLAMSLAMBayesSLAMEusticeSLAMThrunBayesSLAMSLAMSLAMSLAMExtendedInformationFilterSLAMEIF-SLAMVISLAMSLAMSLAMSLAMFrontier-SIFTSLAMICNNSLAMABSTRACTVIISIMULTANEOUSLOCALIZATIONANDMAPPINGINLAGREAREAABSTRACTRecentresearchconcerninginformationsparsificationforSimultaneousLocalizationandMapping(SLAM)hasbecomequitepopular.Byclassifyinglandmarksintodifferenttypes,variousBayestopologicalnetworkstructuresarebuiltwhichcouldmakeinformationmatrixexactlysparse.However,althoughtheaccuracyremains,theefficiencyofSLAMalgorithmisruinedduringlandmarkclassification.Comparedtotheresearchworkshownbefore,ournewalgorithmwhichisbasedonEustice'sworkismuchmoreefficientandpracticalwhilemaintaininghighaccuracy.Focusingonsparsinginformationmatrix,geometricalmeaningofExtendedInformationFilterSLAM(EIF-SLAM)andrelatedformulaareanalyzedrespectively.Theconclusionisdrawnthatnormalizedinformationissparse.MoreoverthestructureofinformationmatrixinEIF-SLAMisputforwardforthefirsttime.Accordingtothestructureofinformationmatrix,animprovedalgorithmbasedonEIFisintroduced.Owningtothehighefficiencyofthespecialrulesinsparseoperation,thenewalgorithmsolvesSLAMveryefficientbysparsinginformationmatrixdirectly.Theerrorsthatcomefromsparsificationdecreaseapparentlyduetoloop-closure.Inaword,thenewalgorithmrealizesefficiencywithconsistentestimateofSLAM.Atlast,therelationshipbetweensparsificationandSLAMaccuracyisanalyzedtheoretically.Thenewalgorithmissimulatedinalargescaleenvironment.Informationmatrixsparsification,algorithmefficiency,relocalization,errorandcovarianceareanalyzedrespectivelyaftersettingupparametersofmotionmodelandobservationmodel.Indoortwo-wheelrobotwithcameraABSTRACTVIIIandoutdoorfour-wheelrobotwithlaseraretakenintoaccountinexperiments.Thetwoexperimentsareofgreatdifference.ExperimentoneaccomplishesbyFrontier-IIrobotwhichisdevelopedbyourlab.PicturesrepresentlandmarksthatareassociatedbyScaleInvariantFeatureTransform(SIFT).ExperimenttwoadoptsstandardCarParkDatasetwhichisfamousinSLAMfield.Theenvironmentofexperimenttwoisoutdoorcarparkandtherobotisafour-wheelmobilecarmountedwithlasersensor.CylindricalobjectsrepresentlandmarksthatareassociatedbyIndividualCompatibilityNearestNeighbor(ICNN).Theresultsofsimulationinlargescaleenvironmentaswellasindoorandoutdoorexperimentstestifyvalidityofthealgorithm.Keywords:MobileroboticsSLAMExtendedinformationfilterInformationmatrixSparsificationXI2.1...................................................................................................72.2...........................................................................................................................72.3....................................................................................................................................102.4ICNN..........................................................................................................................142.5JCBB..........................................................................................................................162.6.............................................................................................................................172.7ICNN..........................................................................................................................182.8JCBB..........................................................................................................................193.11-D,1-DOFEIF-SLAM............................................................................................263.22-D,2-DOFEIF-SLAM................................................................................................273.3.............................................................................................293.4.....................................................................................................................293.5EIF-SLAM.................................................................................................303.62-D,2-DOFSLAM.............................................................................................313.7............................................................................................................................................313.8.....................................................................................333.9SLAM.........................................................................................................333.102WB,2-D,3-DOFSLAM...........................................................................353.112WB,2-D,3-DOFSLAM................................................................................374.1BAYES........................................................................................384.2BAYES................................................................................404.3BAYES........................................................................................404.4.................