空调清洗机器人 论文的外文翻译Word格式文档下载.docx

空调清洗机器人 论文的外文翻译Word格式文档下载.docx

《空调清洗机器人 论文的外文翻译Word格式文档下载.docx》由会员分享，可在线阅读，更多相关《空调清洗机器人 论文的外文翻译Word格式文档下载.docx（15页珍藏版）》请在冰豆网上搜索。

班级08机设Q1

学生姜晓亮

学号20083006058

指导教师张冰

二〇一二年四月八日

AutonomousIndoorMobileRobotNavigationbydetectingFluorescentTubes

FabienLAUNAYAkihisaOHYAShin’ichiYUTA

IntelligentRobotLaboratory,UniversityofTsukuba1-1-1Tennoudai,Tsukuba,Ibaraki305-8573JAPAN{launay,ohya,yuta}@roboken.esys.tsukuba.ac.jp

Abstract

Thispaperproposesanindoornavigationsystemforanautonomousmobilerobotincludingtheteachingofitsenvironment.Theself-localizationofthevehicleisdonebydetectingthepositionandorientationoffluorescenttubeslocatedaboveit’sdesiredpaththankstoacamerapointingtotheceiling.

Amapofthelightsbasedonodometrydataisbuiltinadvancebytherobotguidedbyanoperator.Thenagraphicuserinterfaceisusedtodefinethetrajectorytherobotmustfollowwithrespecttothelights.Whiletherobotismoving,thepositionandorientationofthelightsitdetectsarecomparedtothemapvalues,whichenablesthevehicletocancelodometryerrors.

1Introduction

Whenawheeltypemobilerobotnavigatesonatwodimensionalplane,itcanusesensorstoknowitsrelativelocalizationbysummingelementarydisplacementsprovidedbyincrementalencodersmountedonitswheels.Themaindefaultofthismethodknownasodometryisthatitsestimationerrortendstoincreaseunboundedly[1].Forlongdistancenavigation,odometryandotherdeadreckoningsolutionsmaybesupportedbyanabsolutelocalizationtechniqueprovidingpositioninformationwithalowfrequency.

Absolutelocalizationinindoornavigationusinglandmarkslocatedonthegroundoronthewallsissometimesdifficulttoimplementsincedifferentobjectscanobstructthem.Thereforeanavigationsystembasedonceilinglandmarkrecognitioncanbethoughtasanalternativetothisissue.

Thenavigationsystemwedevelopedconsistsintwosteps.Inthefirststep,thevehicleisprovidedwithamapoftheceilinglights.Buildingsuchamapbyhandquicklybecomesaheavytaskasitssizegrows.Instead,therobotisguidedmanuallyundereachlightandbuildsthemapautomatically.Thesecondstepconsistsindefininganavigationpathforthevehicleandenablingitspositionandorientationcorrectionwheneveritdetectsalightrecordedpreviouslyinthemap.

Sincethemapbuiltbytherobotisbasedonodometrywhoseestimationerrorgrowsunboundedly,thepositionandorientationofthelightsinthemapdonotcorrespondtothereality.However,ifthetrajectorytobefollowedbythevehicleduringthenavigationprocessisdefinedappropriatelyabovethisdistortedmap,itwillbepossiblefortherobottomovealonganydesiredtrajectoryintherealworld.AGUIhasbeendevelopedinordertofacilitatethismap-basedpathdefinitionprocess.

Weequippedamobilerobotwithacamerapointingtotheceiling.Duringthenavigationprocess,whenalightisdetected,therobotcalculatesthepositionandtheorientationofthislandmarkinitsownreferenceandthankstoamapofthelightsbuiltinadvance,itcanestimateitsabsolutepositionandorientationwithrespecttoitsmap.

Wedefinetheposeofanobjectasitspositionandorientationwithrespecttoagivenreferential.

2Relatedwork

Theideaofusinglightsaslandmarksforindoornavigationisnotnew.Hashino[2]developedafluorescentlightsensorinordertodetecttheinclinationanglebetweenanunmannedvehicleandafluorescentlampattachedtotheceiling.Theobjectivewastocarryoutthemainpartoftheprocessbyhardwarelogiccircuit.

Insteadoflights,openingsintheceilingforaerationshavealsobeenusedaslandmarkstotrack.Ootaetal.[3]basedthistrackingonedgedetection,whereasFukuda[4]developedamorecomplexsystemusingfuzzytemplatematching.Hashibaetal.[5]usedthedevelopmentimagesoftheceilingtoproposeamotionplanningmethod.Morerecently,Amatetal.[6]presentedavisionbasednavigationsystemusingseveralfluorescentlighttubeslocatedincapturedimageswhoseabsoluteposeestimationaccuracyisbetterthanaGPSsystem.

Oneadvantageofthesystemproposedhereisitslowmemoryandprocessingspeedrequirementsthatmakeitsimplementationpossibleonarobotwithlimitedimage-processinghardware.Moreover,ournavigationsystemincludesalandmarksmapconstructionprocessentirelybasedontherobot’sodometrydata.ThedevelopmentofaGUIenablestheconnectionbetweenthelightsmapproducedduringtheteachingprocess,andtheautonomousrobotnavigation,whichresultsinacompletenavigationsystem.Thisisthemaindifferencewiththepreviousworkswhicheitherassumetheknowledgeoftheceilinglandmarks’exactposethankstoCADdataofbuildingmaps,orrequiretheabsolutevehicleposetobeenteredmanuallyandperiodicallyduringthelandmarksmapconstruction

soastocancelodometryerrors.

Figure1:

Targetenvironmentconsistingoflightsofdifferentshapesincorridorsexposedtoluminosityvariationsduetosunning.

3Lights’mapbuilding

Inordertocancelodometryerrorswheneveralightisdetected,therobotneedstoknowinadvancetheposeinagivenreferentialofthelightsunderwhichitissupposedtonavigate.

Sinceweareaimingatlongdistanceautonomousindoornavigation,thesizeofthelandmarksmapis

unbounded.Buildingsuchamapmanuallybecomesaheavytaskfortheoperatorandwebelievethatanautonomousmobilerobotcancopewiththisissue.

Duringthelearningprocess,thevehicleequippedwithacamerapointingtotheceilingisguidedmanuallyundereachlightandaddslandmarkinformationtothemapwheneveranewlightappearsaboveitspath.Thishumanassistedmapbuildingisthefirststepofourresearchconcerninglandmarksmapbuilding.Wewanttochangeittoafullyautonomousmapbuildingsystem.Astheimage-processinginvolvedduringthelearningprocessisidenticaltotheoneusedduringthenavigation,wewillpresentthefeatureextractionmethodinsections5and6.

Oncetheteachingphaseiscompleted,therobotholdsamapofthelightsthatcanbeusedlaterfortheautonomousnavigationprocess.

4Dealingwitharobot-mademap

4.1Odometryerror’sinfluenceonthemap

Askingtherobottobuildamapimpliesdealing\withodometryerrorsthatwilloccurduringthelearningprocessitself.Astherobotwillbeguidedundernewlights,becauseoftheaccumulationofodometryerrors,theposeofthelandmarksrecordedinthemapwillbecomemoreandmoredifferentfromthevaluescorrespondingtotherealworld.

SeveralmapsoftheenvironmentrepresentedinFig.1aregiveninFig.2.Theodometrydatarecordedbytherobotduringthelearningprocesshasalsobeenrepresentedforoneofthemaps.

4.2Usageofthemap

Onlyonemapisneededbytherobottocorrectitsposeduringthenavigationprocess.Whenevertherobotdetectsalightlearntpreviously,itcorrectsitsabsolutepose1byusingthelandmark’sinformation

recordedinthemap.Sincethemapcontentsdon’tcorrespondtothevaluesoftherealworld,thetrajectoryoftherobothastobespecifiedaccordingtotheposeofthelightsinthemap,andnotaccordingtothetrajectorywewanttherobottofollowinitsrealenvironment.

Forexample,ifthemobilerobot’staskistonavigaterightbelowastraightcorridor’slights,therobotwon’tberequestedtofollowastraightlinealongthemiddleofthecorridor.Insteadofthissimplemotioncommand,therobotwillhavetotraceeverysegmentwhichconnectstheprojectiononthegroundofthecenteroftwosuccessivelights.ThisisillustratedinFig.3whereazoomofthetrajectoryspecifiedtotherobotappearsindottedline.

AGUIhasbeendevelopedinTcl/Tkinordertospecifyeasilydifferenttypesoftrajectorieswithrespecttothemaplearntbytherobot.ThisGUIcanalsobeusedon-lineinordertofollowtheevolutionoftherobotinrealtimeonthelandmarksmapduringthelearningandnavigationprocesses.

Figure2:

SeveralmapsoftheenvironmentrepresentedFig.1builtbythesamerobot.Rectanglesandcirclesrepresentlightsofdifferentshapes.

5Fluorescenttubedetection

5.1Fluorescenttubemodel

Itisnaturaltothinkoffluorescenttubeasanaturallandmarkforavision-basedprocessaimedatimprovingthelocalizationofamobilerobotinanindoorenvironment.Indeed,problemssuchasdirt,shadows,lightreflectionontheground,orobstructionofthelandmarksusuallydonotappearinthiscase.

Oneadvantageoffluorescenttubescomparedtootherpossiblelandmarkslocatedontheceilingisthatoncetheyareswitchedon,theirrecognitioninanimagecanbeperformedwithaverysimpleimageprocessingalgorithmsincetheyaretheonlybrightelementsthatarepermanentlyfoundinsuchaplace.

Ifa256greylevelsimagecontainingafluorescenttubeisbinarizedwithanappropriatethreshold0≤T≤255,theonlyelementthatremainsafterthisoperationisarectangularshape.Fig.4.ashowsatypicalcameraimageoftheceilingofacorridorcontainingafluorescentlight.Theaxisofthecameraisperpendiculartotheceiling.Shownin（b）isthebinarizedimageof（a）.Ifwesupposethatthedistancebetweenthecameraandtheceilingremainsconstantandthatnomorethanonelightatatimecanbeseenbythecameralocatedonthetopoftherobot,afluorescenttubecanbemodeledbyagivenareaS0inathresholdedimageoftheceiling.

Figure4:

（a）Sampleimageofafluorescentlight,（b）binarizedimage.

5.2Fluorescentlightdetectionprocess

Usingodometry,therobotisabletoknowwhenitgetsclosetoalightrecordedinitsmapbycomparinginacloseloopits