Region & gateway mapping [Elektronische Ressource] : acquiring structured and object-oriented representations of indoor environments / Derik Schröter

Description

dd d dInstitut fu¨r Informatikd d d dd dd d dd d d d der Technischen Universit¨at Mu¨nchend d dRegion & Gateway Mapping:Acquiring Structured and Object-OrientedRepresentations of Indoor EnvironmentsDissertationDerik Schr¨oterInstitut fu¨r Informatikder Technischen Universit¨at Mu¨nchenRegion & Gateway Mapping:Acquiring Structured and Object-OrientedRepresentations of Indoor EnvironmentsDerik Schr¨oterVollst¨andiger Abdruck der von der Fakult¨at fu¨r Informatik der Technischen Universit¨atMu¨nchen zur Erlangung des akademischen Grades einesDoktors der Naturwissenschaften (Dr. rer. nat.)genehmigten Dissertation.Vorsitzender: Univ.-Prof. Alfons Kemper, Ph.D.Pru¨fer der Dissertation: 1. Univ.-Prof. Dr. Bernd Radig2. Univ.-Prof. Dr.-Ing. Darius Burschka3. Univ.-Prof. Dr. Wolfram Burgard,Albert-Ludwigs-Universit¨at FreiburgDie Dissertation wurde am 10.11.2005 bei der Technischen Universit¨at Mu¨ncheneingereicht und durch die Fakult¨at fu¨r Informatik am 29.03.2006 angenommen.AbstractState-of-the-art robot mapping approaches are capable of acquiring impressively accurate2Dand3Dmodelsoftheirenvironments. Tothebestofourknowledge, fewofthemrepre-sent structure or acquire models of task-relevant objects. In this work, a new approach tomapping of indoor environments is presented, in which the environment structure in termsofregionsandgatewaysisautomaticallyextracted,whiletherobotexplores.

Sujets

• Informatik

Informations

• Publié par: technische_universitat_munchen
• Publié le: 01 janvier 2006
• Nombre de lectures: 7
• Poids de l'ouvrage: 78 Mo

Extrait

dddddddddd
dddddddddd

Institutfu¨rInformatik
derTechnischenUniversita¨tMu¨nchen

Region&GatewayMapping:
AcquiringStructuredandObject-Oriented
RepresentationsofIndoorEnvironments

Dissertation

DerikSchr¨oter

Institutfu¨rInformatik
derTechnischenUniversita¨tMu¨nchen

Region&GatewayMapping:
AcquiringStructuredandObject-Oriented
RepresentationsofIndoorEnvironments

DerikSchr¨oter

Vollsta¨ndigerAbdruckdervonderFakulta¨tfu¨rInformatikderTechnischenUniversita¨t
Mu¨nchenzurErlangungdesakademischenGradeseines

DoktorsderNaturwissenschaften(Dr.rer.nat.)

genehmigtenDissertation.

Vorsitzender:Univ.-Prof.AlfonsKemper,Ph.D.
Pru¨ferderDissertation:1.Univ.-Prof.Dr.BerndRadig
2.Univ.-Prof.Dr.-Ing.DariusBurschka
3.Univ.-Prof.Dr.WolframBurgard,
Albert-Ludwigs-Universita¨tFreiburg

DieDissertationwurdeam10.11.2005beiderTechnischenUniversita¨tMu¨nchen
eingereichtunddurchdieFakulta¨tfu¨rInformatikam29.03.2006angenommen.

Abstract

State-of-the-artrobotmappingapproachesarecapableofacquiringimpressivelyaccurate
2Dand3Dmodelsoftheirenvironments.Tothebestofourknowledge,fewofthemrepre-
sentstructureoracquiremodelsoftask-relevantobjects.Inthiswork,anewapproachto
mappingofindoorenvironmentsispresented,inwhichtheenvironmentstructureinterms
ofregionsandgatewaysisautomaticallyextracted,whiletherobotexplores.Objects,both
in2Dand3D,aremodeledexplicitlyinthosemapsandallowforrobustlocalization.We
refertothosemapsasstructuredobject-orientedenvironmentrepresentationsorRegion
&GatewayMaps(RGMaps).TheprocessofbuildingsuchmapsiscalledRegion&
GatewayMapping(RGMapping).
RGMapsandRGMappingmakeseveralcontributionstotheeldofmapbuildingof
indoorenvironments.First,RGMappingautomaticallyrecoversthestructureoflarge
classesofindoorenvironmentsandrepresentsthemexplicitly.Therefore,novelalgorithms
havebeendevelopedforthedetectionandrecognitionofgateways,i.e.transitionsbetween
regions.Second,itdetectsrectangular2D/3Dobjectsusinglaserrangeaswellasimage
datathatareusedforgatewaydetectionandlocalization.Third,thesemanticdescription,
i.e.annotationofregionsandobjects,isobtainedfromhuman-machineinteractionin
thecontextoftaskassignment.Fourth,acompactRegion&GatewayGraphiseasily
extractedfromRGMaps,anditisusedforecientpathplanningontheglobalscale.
Itallowsreasoningaboutthefeasibilityofagivenpathandlearningoftheproperties
ofpathsegmentswhiletherobotmovesthroughtheenvironment.Fifth,duetothe
clusteringofmetricdataintoregions,region-basedlocalizationandpathplanningonly
needtoconsidertherespectiveregiondata.Thus,thecomplexityofthedataassociation
problemforlocalizationandthesearchspaceformetricpathplanningissubstantially
reduced.
TheRGMappingandNavigationSystemhasbeenfullyimplementedasadistributed
(module-based)system,andrunsinreal-timeonarealrobot.Duetoitsarchitecture,
thesystemcanbeeasilyportedtodierentplatforms,providedthatsimilarsensordata
isavailable.InordertosupportreliableexplorationofandnavigationwithinRGMaps,
thepresentedsystemfeaturesanovelapproachtocollisionavoidanceandlow-levelcontrol.
Thelow-levelcontrolallowsforverypreciseandfastpursuingofshorttrajectorysegments,
whichcanbechangedatanytime.Thecollisionavoidancegeneratestrajectorysegments
basedontheinterpretationofthecurrentsensordataandshort-distancetargetsfromthe
pathplanningprocess.Asaresult,thenavigationbehaviouroftherobotisreproducible
andpredictable,whichisaverydesirablefeatureforhigh-levelplanning.

Acknowledgements

TheopportunitytoconductresearchinroboticsandvisionintheIntelligentAutonomous
SystemsGroupattheTechnischeUniversita¨tMu¨nchenturnedouttobeanexcitingand
pleasantexperience.Thishasbeeneectedmainlybythefriendlyatmosphereandthe
scienticspiritinthedepartment.Needlesstomention,theworkpresentedinthisthesis
hasbenetedgreatlyfromtheinputandsupportofnumerouspeopleinsideandoutsideof
thegroup,andthereforeIwanttoexpressmydeepestgratitudetoeveryoneinvolved.
IwanttothankProf.BerndRadigforthecondenceandtrustheplacedinme,by
facilitatingmyresearchpositionandnancialsupport.Afterdeningthegeneraldirection
formyresearch,healwaysencouragedmetoformulatemyownideasandagendaand
gavemethemotivationandfreedomtoshapetheprojectasitprogressed.Withoutthe
unlimitedsupportofProf.MichaelBeetzImightbesittingherefacingonlyabunchof
goodideas.Inuncountedfruitfulandsometimesverylivelydiscussionshehelpedmeto
formmyresearch,andtherewasalwaystimeforfriendlyandbenevolentmotivations.Most
ofallhismanycommentsonmypapers,presentationsandnallymythesishavetaught
mehowtomakemywritingtellastoryandhowideascanformaresearchproject.Ihave
learnednotjusttoexplain,buttoconvince.Thankyou,Michael.
ThankstoProf.DariusBurschkaforservingasamemberofmydissertationcommittee,
aswellasProf.WolframBurgardwhohasbeenasourceofinspirationforbothmywork
andmywriting.Inparticular,Iwanttothankhimforreviewingmythesisandthetime
wesharedinseveraldiscussionsaboutrelatedsubjects.Likewise,mythanksgotoDirk
H¨ahnel,SteenGutmann,PatrikBeeson,MichaelThielscherandalltheothersIwasglad
tomeetatConferences,Workshopsandotheroccasions.Thanksforyourcriticalreections
andhelpfulcomments.IwanttoexpressspecialgratitudetoMariaFoxandBorisFlach
notonlyfortechnicaldiscussionsbutalsoforlongconversationsaboutlifeandresearchin
generalandthegoodredwinethatwentalongwithit.Theycontributedmorethanthey
possiblyknowtomymotivationandwell-being.
Ididnotmeantoforgetaboutmypastandpresentcolleaguesandstudentsinthedepart-
ment.Afterall,theyhavebeenatremendoushelpandsupportintheeverydaywork.First
andforemostIwanttothankFreekStulpandAndreasHofhauserforcriticallyreecting
onmyideasandconstantlyencouragingmetocontinue(andtonallynish).Theyhave
becomegoodfriendstome,andIsureliketorememberthetimeFreekandmeplayed
inthatgreatrockband(www.theunknownhost.com).Andreaswasamongthestudents
thatcontributedessentiallytotheimplementationoftheworkpresentedhere,othersare
JanBandouch,StefanPlafka,ThomasWeber,StefanHinterstoisserandAndreasIizuka.
Theirhelphasbeenverymuchappreciated.FurtherthanksgotoKajetanBerlingerand

iiiiii

HeikoGottschlingwhoseemtoknoweverythingaboutprogramming,scriptinganduseful
tools.ThanksalsogotoRobertHanek,ThorstenSchmitt,SebastianBuckandChristoph
Hansenforhelpandsupportthroughoutmy