英文文献-足球机器人

介绍足球机器人的

FuzzySetsandSystems127(2002)

49–64

http://www.77cn.com.cn/locate/fss

Strategy-baseddecisionmakingofasoccerrobotsystemusinga

real-timeself-organizingfuzzydecisiontree

Han-PangHuang ,Chao-ChiunLiang

DepartmentofMechanicalEngineering,RoboticsLaboratory,NationalTaiwanUniversity,

Taipei,Taiwan10674,ROC

Abstract

Inasoccerrobotgame,theenvironmentishighlycompetitiveanddynamic.Inordertoworkinthedynamicallychangingenvironment,thedecision-makingsystemofasoccerrobotsystemshouldhavethefeaturesof exibilityandon-lineadaptation.Thispaperproposedastrategy-baseddecision-makingsystemforasoccerrobotsystem.Thedecisionsystemisbasedontheproposedself-organizingfuzzydecisiontree(SOFDT).Duetoevent-drivenstrategies,theproposeddecision-makingsystemhastheadvantageof exibility.TheproposedSOFDTpossessessimple,apparent,andfastgeneration=reasoningprocesses.Theresultantsystemisaself-organizingstrategy-baseddecision-makingsystem.Accordingtodecisionresults,SOFDTcanon-linemodifyitsparametersandstructureforachievingbetteradaptabilityandimprovement.Theperformanceoftheproposedsystemswasveriÿedbyapplyingthemtothesimulationandexperimentof3-to-3robotsoccergames.The

c2002ElsevierScienceB.V.Allrightsreserved.resultsaresatisfactory.

Keywords:Fuzzylearningdecisiontree;Self-organizing;Strategy-baseddecisionmaking;Soccerrobots

1.Introduction

Inasoccerrobotgame,theenvironmentishighly

competitiveanddynamic.Inordertoworkinthedynamicallychangingenvironment,thedecision-makingsystemofasoccerrobotsystemshouldhavethefeaturesof exibilityandon-lineadaptation.Atypicalmulti-agentsoccerrobotsystem,suchasNTUFormosabuiltbyourlaboratory,isshowninFig.1[1,5–9,11,14,15].ThisstudyusedNTU-Formosaastheplatformtodevelopaself-organizingstrategy-baseddecision-makingsystem.NTU-Formosa

Correspondingauthor.Tel.=fax:+886-2-2363-3875.

E-mailaddress:hanpang@ccms.ntu.edu.tw(H.-P.Huang).

consistsofmultiplemobilerobots,avisionsystem,awirelesscommunicationsystem,andahostcomputer.InFig.1,theimageinformationoftheentiresoc-cerÿeldiscapturedbyaCCDcameraandsenttothecorrespondinghostcomputer.Thehostcomputerthenanalyzestheimageinformationtodeterminethesitu-ationofthesoccerÿeld.Accordingtothedeterminedsituation,thehostcomputerdecidesastrategyandplansthemotionmodesandthecorrespondingveloc-itycommandsforeverysoccerrobotofthesameteam.EachsoccerrobotofNTU-Formosathenreceivesve-locitycommandfromthehostcomputerandregulatestherotationalvelocitiesofitsrightandleftwheels.Adecisiontreehasseveralnodesarrangedinahier-archicalstructure.Itimplementsdecisionsinasimple,

c2002ElsevierScienceB.V.Allrightsreserved.0165-0114/02/$-seefrontmatter

PII:S0165-0114(01)00152-X

介绍足球机器人的

50H.-P.Huang,C.-C.Liang/FuzzySetsandSystems127(2002)

49–64

Fig.1.GraphicaldiagramoftwocompetingteamsofNTU-Formosa.

apparent,multistagemanner.Furthermore,sinceeachnodeofadecisiontreeusesonlyasimplesplittingruleandasmallsubsetofallfeatures,theentiredeci-sionprocessisfastande cient[2,3,10,13].Thisstudyproposedaself-organizingstrategy-baseddecision-makingsystemforNTU-Formosatomodifystrategiesanddecidewhichstrategyshouldbeappliedtothecur-rentsituation.Thedecisionsystemisbasedonthepro-posedself-organizingfuzzydecisiontree(SOFDT).Duetoevent-drivenstrategies,theproposeddecision-makingsystemhastheadvantageof exibility.ItcanbeshownthattheproposedSOFDTpossessessimple,apparent,andfastgeneration=reasoningprocesses.Togenerateapracticaldecisiontree,numeroustrainingdataarenecessary.Thefuzzylearningdeci-siontreeadoptedinthisstudyusesfuzzystatistics[12,16]togenerateseveralfuzzysetsfromthetrainingdata.Thismethode ectivelycompressesthetrainingdataandmeaningfullyrepresentsitsstatisticaldistri-bution.Thishastheadvantagesofincreasingcompu-tationspeed,savingstoragememory,andmaintainingsatisfactoryperformance.Givenasetoftrainingdata,anoptimaldecisiontreeisinherentlydi culttoob-tain[10,13].Tocopewiththisdi culty,theadoptedfuzzylearningdecisiontreeusesasub-optimalsplit-tingruleateachnodetogenerateadecisiontree.Thisisacost-e ectivemethodthatquicklyobtainsadecisiontreewithsatisfactoryperformance.

Inordertoadapttothechangingenvironment,theproposedSOFDTon-linetunesitsparametersandadaptsitsstructuretoimproveitsperformanceintermsofdecisionresultsandtheinputdata.Toincreaselearningspeed,theproposedSOFDTmodiÿesonlyasmallsubsetofthedecisiontreebyalocallearningmechanismineachlearningcycle.Ifthedecisionre-sultiscorrect,SOFDTtunesitsparameterstore ectthenewstatisticaldistribution.Ifthedecisionresultisincorrect,SOFDTtunesitsparametersforsingle-modalcaseandadaptsitsstructureformulti-modalcase.Ifthedecisionresultisunknown,SOFDTadapts

介绍足球机器人的

H.-P.Huang,C.-C.Liang/FuzzySetsandSystems127(2002)49–6451

itsmodelstructuretocopewiththemulti-modalandnewSOFDTcases.andhasaFinally,setofrulesifthetoinputdeÿnedataandisunreasonable,identifythemperformthenapplications.on-lineignoretheminlearningprocess.SOFDTcanposedto-3systems,Torecognitiontheexaminesimulationtheandperformancelearninginreal-timeandexperimentsofthepro-strategy-basedTherobotorganizationsoccergameswillbeconducted.of3-Formosadecision-makingofthispaperissystemasfollows.Theself-organizingisdevelopedinSection2.InSectionofNTU-itsfuzzydecisiontreeisproposed3,andaappliedperformanceinof3-to-3Sectiontotheisanalyzed.TheproposedSOFDTis4.Instrategy-basedSection5,simulationdecision-makingsystemFinally,severalsoccerconcludingrobotgamesremarksaregivenandareandexperimentdrawn.discussed.2.NTU-Formosa

Astrategy-baseddecision-makingsystemoflectingThestrategy-baseddecision-makingsystemforse-Therentdecision-makingstrategiesisimplementedsystembydecisiontrees.lyingsituationAinwhichofthezoneandsoccerunderÿeld,ÿrstrecognizesthecur-whichincludingtheballsituation,strategyisthenselected,accordingteam’stothecontrol.Formosaconstructstotocoordinatethemultiple …… 此处隐藏：34176字，全部文档内容请下载后查看。喜欢就下载吧 ……