sonar-based mapping of large-scale mobile robot environments using em

《sonar-based mapping of large-scale mobile robot environments using em》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、Sonar-BasedMappingWithMobileRobotsUsingEMWolframBurgard1DieterFox2HaukeJans1ChristianMatenar1SebastianThrun21DepartmentofComputerScience2ComputerScienceDepartmentUniversityofBonnCarnegieMellonUniversityD-53117Bonn,GermanyPittsburgh,PA15213Abstractbeestimatedfromdata.Unfortunately,errorsinodomet

2、ry(wheelencoders)amplifyovertime.Forexample,smallrotationalerrorcanhaveahugeeffectontherobot'sx-y-Thispaperpresentsanalgorithmsforlearn-locationlaterintime.Thismakesitdifficulttoestimateingoccupancygridmapswithmobilerobotsposeswhilelearningmaps.equippedwithrangefinders,suchassonarsen-sors.Ourappr

3、oachemploystheEMalgorithmInthepast,theproblemofmaplearninghasbeentack-tosolvetheconcurrentmappingandlocalizationledbyagreatnumberofresearchers(seee.g.,[CL85,problem.ToaccommodatethespatialnatureofCho96,CKK95,Elf87,Elf89,KB91,Mat90,Mor88,rangedata,itreliesonatwo-layeredrepresenta-Ren93,SK97]).So

4、merecentmethods,suchasthemet-tionofmaps,whereglobalmapsarecomposedricapproachesdescribedin[GN97,LM97,TGF+98],havefromacollectionofsmall,localmaps.Toavoidsuccessfullybeenappliedforlearningmapsupto80bylocalminimaduringlikelihoodmaximization,a25m2insize.However,theyarecloselymarriedtolasersoftmaxv

5、ersionoftheMstepisproposedthatisrangefinders,ahighlyaccurateandquiteexpensivesen-graduallyannealedtotheexactmaximum.Ex-sortechnology.Othermethods,suchasthetopologicalap-perimentalresultsdemonstratethatourapproachproachesdescribedin[Sha98,TFB98],relyonlandmarksiswellsuitedforconstructinglargemaps

6、oftypi-tobuildmaps.Theyworkwithlessaccuratesensors(suchcalindoorenvironmentsusingsensorsasinaccu-assonars),buttheyareforcedtothrowawayalmostallsen-rateassonars.sordata,exceptforextremelyscarcelandmarkdata.Forexample,theycannotexploitthewidthofacorridortodis-ambiguatedifferentcorridors.Someofthe

7、largesttopolog-1Introductionicalmaps(80by25m2)havebeenlearnedusingthealgo-rithmin[TFB98].However,ithasonlybeendemonstratedLearningmapswithmobilerobotshasfrequentlybeenrec-toworkwithpre-definedlandmarks.Itisremarkable,how-ognizedaso