自主构建分层任务网络人机协作布拉德利蒂1slam关键.pdfVIP

2016IEEEInternationalConferenceonRoboticsandAutomation(ICRA)

Stockholm,Sweden,May16-21,2016

AutonomouslyConstructingHierarchicalTasworksfor

nningandHuman-RobotCollaboration

BradleyHayesandBrianScassellati

—Collaborationbetweenhumansandrobotsre-

quiressolutionstoanarrayofchallengingproblems,luding

multi-agentnning,stateestimation,andgoalinference.

Therealreadyexistfeasiblesolutionsformanyofthese

challenges,buttheydependuponhavingrichtaskmodels.

InthisworkwedetailanoveltypeofHierarchicalTask

NetworkwecallaClique/ChainHTN(CC-HTN),alongsidean

algorithmforautonomouslyconstructingthemfromtopological

propertiesderivedfromgraphicaltaskrepresentations.Asthe

presentedmethodreliesonthestructureofthetaskitself,

ourworkimposesnoparticulartypeofsymbolicinsightinto

motorprimitivesorenvironmentalrepresentation,makingit

applicabletoawidevarietyofusecasescriticaltohuman-robotFig.1:Collaborationreliesoncomplexnningandin-

inction.Wepresentevaluationswithinamulti-resolutiontentionrecognitioncapabilities.Formanytasks,especially

goalinferencetaskandatransferlearningapplicationshowingwithintheassemblyandcookingswebaseour

theutilityofourapproach.evaluationson,hierarchicalmodelsarerequiredtoidentify

I.INTRODUCTIONandsolvetractablesub-problems.

Combininguser-friendlyskillacquisitionandrobust

task-levelnning