文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 电子/电路 > The Next Best Touch for Model-Based Localization
侵权投诉

The Next Best Touch for Model-Based Localization

2013 IEEE International Conference on Robotics and Automation (ICRA)Karlsruhe, Germany, May 6-10, 2013

TheNextBestTouchforModel-BasedLocalization

PaulHebert,ThomasHoward,NicolasHudson,JeremyMa

JetPropulsionLaboratory,CaliforniaInstituteofTechnology

Pasadena,CA,91109,USA

JoelW.Burdick

CaliforniaInstituteofTechnologyPasadena,CA,91124,USA

Abstract—Thispaperintroducesatactileorcontactmethodwherebyanautonomousrobotequippedwithsuitablesensorscanchoosethenextsensingactioninvolvingtouchinordertoaccuratelylocalizeanobjectinitsenvironment.Themethodusesaninformationgainmetricbasedontheuncertaintyoftheobject’sposetodeterminethenextbesttouchingaction.Intu-itively,theoptimalactionistheonethatisthemostinformative.Theactionisthencarriedoutandthestateoftheobject’sposeisupdatedusinganestimator.Themethodisfurtherextendedtochoosethemostinformativeactiontosimultaneouslylocalizeandestimatetheobject’smodelparameterormodelclass.ResultsarepresentedbothinsimulationandinexperimentontheDARPAAutonomousRoboticManipulationSoftware(ARM-S)robot.

I.INTRODUCTION

Wewouldlikeautonomousroboticstoperformdexteroustaskssuchasgraspingscrewdriversandhouseholdobjectssuchascups,insertingcablesintopoweroutlets,andusingkeystoopenandunlockdoors[1]–[3].Allofthesetasksrequirepreciselocalizationofkeyobjectsintherobot’sframe,withtaskssuchaskeyinsertionrequiringlocalizationontheorderofamillimeter.Whilemanyofthesetaskscouldbedonewithhighlypreciseroboticsystems,thereisabroadinterestinperformingthesetaskswithinexpensivesystems,whichmaylacktherequiredprecision.Eveninknownenvironments,withprecisevisualsensors,theprecisionofaclosedchainroboticmechanismmaybeinsuf cienttoexecutemanyofthesetasksopenloop[4].Moregenerally,errorstypicallyintroducedbyvisualposeestimationand/orinternalkinematicerrorsareoftentoolargeforprecisegraspingandmanipulation.Instead,estimationalgorithmscanre nethelocalposeestimateduringtaskexecution,whenphysicalinteractionwiththeobjectoccurs[5]–[7].Thispaperbuildsupontheseestimationalgorithms,andprovidesanautomatedmethodtoselectroboticmanipulationactionswhichwillprovideincreasedknowledgeaboutthesystemstate.

IntherecentDARPAAutonomousRoboticManipulationSoftware(ARM-S)competition,teamswererequiredtocreateautonomousmanipulationsystemstoexecutevarioustasks.Inparticular,teamswererequiredtounlockandopenadoorautonomously.Aninterestingobservationisthatalmostallteams,regardlessoftheunderlyingapproach,createdactionsthat rstre-localizedthedoorhandle(byeithertouchingthedoororthehandle)beforeproceedingwithkeyinsertion.Inthispaper,theactionselectionalgorithmautomaticallydecideswhichprobingmotionstomake,

The Next Best Touch for Model-Based Localization

as

Fig.1:Exampledoorhandletouchingactions.opposedtousingprede nedprobes,motionsorpolicies.Therearesituationsinwhichtherobotmaybepresentedwithanovelversionofaknownobjectclass(e.g.imaginearobottasktopickupapriorinewscrewdriver.Therobot’sperceptionsystemmaybeabletoeasilyrecognizethescrewdriverclass,buttheprecisedimensionswillnotbeknown.).Asaconsequence,itwouldbeadvantageoustoselectinteractingactionsthatwouldsimulataneouslylocalizetheobjectandestimatecharacteristicparametersofthemodelthatmayimpactlocalization.Similarly,arobotmayberequiredtomanipulateanobjectofwhichitisunabletodiscerntheobject’sclassorcategory.Thismayoccurfromvisualocclusionorviewpointdependency.Thispaperpresentsamethodtoselectactionsthataimstobothlocalizeanddeterminetheobjectmodel.

Here,anactionisde nedasthecombinationofatrajec-tory,andasetofendconditions.Thetrajectoryencodestheplannedopenloopexecutionoftherobotmotion,andtheend-conditionsspecifywhenanactionterminates(e.g.uponacertainforcethreshold,orwhencontacthasoccurred).Ouractionselectiontheorychoosesactionswhichareexpectedtominimizetheentropyoftherelativerobot-environmentbelief.Accuratelocalizationrequiresminimaluncertaintyaboutthepose.Inordertominimizeuncertainty,theentropyofthestatedistributionshouldbeminimized,asdistributionswithlowentropyarepeakedandcertain.Equivalently,theinformationgaincanbeviewedasthechangeinrelativeentropycausedbyundergoinganactionandupdatingthebeliefbyacquiringanewmeasurement.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email