RoboCup-99 Team Descriptions Simulation League, Team Zeng99, pages 0–0
- 格式:pdf
- 大小:254.70 KB
- 文档页数:3
1
outlineofHiFICsystemisshown.Sectionexplainsmiddleandlowerlayer
configurationthosearemainpartofHiFICsystem.
2HierarchicalFuzzyIntelligentControlSys-
tem
1[Fig.1]
showsaconfigurationofthesystem.Thatsystemisdesignedforunstable
and/ornonlinearsystems.Socceragentshavenon-holonomicrestriction.
Thusthecontrolneedscomplicatedsequenceof(turn)and(dash)com-
mand.Inanotherhand,humanbeingscouldcontrolsuchanill-systemand
couldlearnhowtocontrolit.HiFICisbasedonthistypeofhumanbeings’
controlstrategy,thosemainpointisamacroscopicrecognition.Everypart
oflayeracceptsamesensoryinputs.Itreconstructsthemondemandof
decisionmaking.Eachlayeracceptsamacroscopictacticalcommandsfrom
upperlayer.Aninputfromupperlayerisinaformofmacrocommand,
andthatcann’tbeexecutedimmediatelyatlayerreceiveingit.Thatmacro
commandhassomefuzzinessofthatmeanings,consequentlyinterpretation
processisneededtoexecuteit.Theseinterpretationchainamonglayers
areremarkablecharacteristics.Torealizeacooperativebehaviorwithinthe
highestlayer,threetechnicalproblemshouldbethere;macroscopicrecog-
nitionofgamesituation,simulationbasedpredictionofgameanddecision
makinginfuzziness.Middlelayerandlowerlayer,whichareamainpartof
control,isexplainedindetailatnextsubsection
3Middlelayerandlowerlayer
Middleandlowerlayermakessureprimarymovements.Lowercontroller
interpretsabehaviorcommandsintoexecutablecommandsequence.
ßMiddlelayer:behaviorcontrolMiddlelayercontrolsabehavioroftactical
movesbymeansoffuzzylogicrules.Tacticalmovesconsistofaserver
commandsequence.Thissequenceisatacticalbehavior(2[Fig.2]).This
knowledgeastacticalrulesRCURL(sj)configuredfromthreekindofarts.
Rsji:ifCONDITIONisCithendoMi,i=1...nwhereCONDITIONisa
vectorofcontrolandgamesituationatthattime.Currentindexoftherule
usedatthattimeisincludedinthesituationCONDITIONvector,therefore
theycouldapplyrulessequentiallyaccordingtotheruleindexes.3[Fig.2
]showsthestatetransitionthroughrules.M
1Ref:http://./figs/fig1.html2Ref:http://./figs/fig2.html3Ref:http://./figs/fig2.html2
propertiesofplayeragentsinsoccerserver.Iftherewerenoerrorsin
sensoryinputsandservercommandresults,anymiddlelevelcommand
couldbetransformedintojustoneturnandonedash.However,errors
existinsensorandcommandexecution.Erroroccursaftercommandex-
ecution.Toavoidthaterror,thesystemneedsfeedbackloop.Amid-
dlelayermacrocommand“gotopointX”shouldberealizeasanfeed-
backalgorithm.“GoToPointX”1.Turntoheadingforthepoint
(turn$a*t)2.Dashtocomeclosedestination(dash$p(d))3.untilgetting
destinationX,goto1.andrepeat.Functionp(d)calculatespowertogo
throughddistance.“a”intheturncommandisafeedbackratioThelib-
sclientembeddedTcl/Tkshell,wecallitscwish,isformakingprototypeswithGUIfacility.Thescwithisd
nameport−numscsend“(servercommand)setmsg[receivemessage]scconnectwithservernameandport
5Summary
HierarchicalFuzzyIntelligentControlsystemisadoptedtosoccer
agents.Zeng99teambecomesatestbenchfordevelopinghowto
designcooperativebehaviorontheHiFICsystem.Itenableusto
makeacontrolrulesfromhumanknowledgesinceitacceptlinguistic
macroscopiccommandnotation.Thescwishthatembedlibsclientlibrarymakesiteasytoconstructpr
6Bibliography
JunjiNISHINO,AkihiroTAGAWA,HaruhikoSHIRAI,TomohiroODAKA,
andHisakazuOGURA.Hierarchicalfuzzyintelligentcontrollerfor
gymnasticbaractions.JournalofAdvancedComputationalIntelligence,
1999.tobeappeard.JensRasmussen.Skills,rules,andknowledge;
signals,signs,andsymbols,andotherdistinctionsinhumanperformance
models.IEEETrans.SMC,13(3):257--266,1983.