移动机器人环境建模与定位问题研究(SLAM)

UDC20084Researchonmapping&localizationforamobilerobotDissertationSubmittedtoInstituteofAutomation,ChineseAcademicofSciencesinpartialfulfillmentoftherequirementsforthedegreeofDoctorofEngineeringbyWangHong-mingControlTheoryandControlEngineeringDissertationSupervisor:ProfessorTanMinProfessorHouZeng-guang:::SLAM1°2°3°SLAMSLAMSLAMSLAMSLAMEKF-SLAMSLAM4°SLAMSLAMSLAM-i--ii-Researchonmapping&localizationforamobilerobotAuthor:HongmingwangSupervisor:MinTanAbstractMappingandLocalization,whichserveasthebasisforautonomouslynav-igatingamobilerobot,havebeenextensivelystudiedformorethan20years.Howevertherearestillsomeproblemsunsolved,especiallyonmappingandlocal-izationbyuseofsonar.Withwidebeam,sonaractsasarange-onlymeasurementequipmentinsomesense.Consequentlymappingandlocalizationwithsonardataarewithmorechallenges.Inthisthesis,westudytheproblemofmappingandlocalizationbyuseofsonar.Themaincontributionsofthisthesisincludefollowingissues:1°Consideringthatsonarcandetecttwotypesoftarget(pointandlinefeature)inindoorenvironments,weproposeamodelcalledthreemeasurementsas-sociationmodeltoassociatesparsesonardatatospecificfeatures,andbasedonthemodelweuseaniterativeleastsquareestimatortoestimatethepa-rametersofthefeaturesfromassociatedsonardataandbuildasonarfeaturemapfinally.Ourapproachavoidstheproblemcausedbyimplementingdataassociationfromdensesonardata,satisfytheneedofreal-timeapplication.2°Beforeimplementationofrobotlocalizationbasedonthelandmarkorfea-turemap,thereisacommonquestionthathowmanylandmarksorfeaturesareenoughtolocalizetherobot.Inthisthesis,withsomeimportantresultsfromobservabilityanalysisonthelocalizationsystemswithrang-onlymea-surement,weanswerthequestionabove.Furthermore,basedontheresults,weproposeanapproachforheadingestimationusinganglehistogramanddevelopaMCLmethodbasedonoptimalproposaldistribution.3°SLAMproblemattractsmoreandmoreattentionsinmobilerobotcommu-nity.However,thereisnoaffirmativeansweronwhetherSLAMproblemcanbesolved.WithsomeimportantresultsfromobservabilityanalysisonthreetypesofSLAMsystemsandSLAMsystemswithrang-onlymeasure-ments,wedrawaconclusionsthatnoneoftheSLAMsystemsweanalyzediscompletelyobservable.Furthermore,basedontheseresults,weaccountfortheestimationinconsistencyinEKF-SLAMimplementationandproposetheconditionsforimplementingSLAMbyaddingknownlandmarks.4°Formappingdynamicenvironments,aGaussianmixturemodelwhichcannotonlyrepresentthepositionoftheobjects,butalsothestateoftheobjectsischoseninourapproach.Inaddition,wededucedafactoredformulationofBayesianposteriorforSLAMproblem,baseonwhichwedevelopan-iii-approachforSLAMindynamicenvironmentswithGMMtomodeltheenvironmentandparticlefiltertolocalizethemobilerobot.KeyWords:mobilerobot,mapping,localization,sonar-iv-11.1..................................11.2...................11.2.1..............21.2.2..............41.3.............71.3.1..................81.3.1.1.....................91.3.1.2..................111.3.2.....................121.3.2.1...........151.3.2.2.............151.3.2.3...........161.3.3SLAM..........................171.3.4.............................181.4............191.5.....................21232.1..................................232.2.................242.2.1...........................242.2.2...........................252.3........................272.4........................302.4.1...........................302.4.2..........................312.5............................322.5.1........................322.5.2........................342.6.........................362.6.1........................362.6.2........................372.7.............................402.8...............................422.9..................................422.102.1.............................45-v-473.1..................................473.2....................483.3.....................493.3.1..........................503.3.1.1....................513.3.1.2Grid-based................523.3.1.3EKF.........................533.3.1.4Grid-based..............543.3.1.5.....................553.3.2..............563.3.2.1......................563.3.2.2....................573.4...................583.4.1.....................583.4.2.....................593.5...................613.5.1....................623.5.2.............633.5.2.1.............633.5.2.2.............643.5.3............683.5.3.1..................693.5.3.2............703.5.3.3....................723.5.4..............733.5.4.1............753.5.4.2..................763.5.4.3......................773.6..................................77SLAM794.1..................................794.2SLAM.......................804.2.1EKF...........................824.2.1.1....................824.2.1.2....................834.2.1.3....................834.2.2FastSLAM........................84-vi-4.2.2.1FastSLAM................844.2.2.2FastSLAM................864.2.2.3....................894.2.2.4....................894.2.3.........................904.2.3.1ScanMatching.............904.2.3.2EM.......................914.2.3.3......................924.3