夏水流萤A.Cardonetal./RoboticsandAutonomousSystems33(2000)179–190
187
Fig.6.Representationofthecontract-netprotocolinanagent.
Amachinecanacceptmanycontractscorrespond-ingtothesamejob,butnegotiatedbydifferentNAswhichhavedifferentstrategies.Then,thecontractideologyis“thestrongestwins”sinceacontractcanbebrokenbybothpartners[21].However,wecanproposetwotypesofagents:aforwardplacingandabackwardplacingagent.Moreover,therearesomeworkshopconstraintssince,wehavegenericma-chines,groupsofmachinesandspecializedmachines.Wecanconsiderthatwehave“groupofmachines”agentswhichwillproposethedifferentjobsto
themachinestheyrepresent.Atthislevel,wecanoperateaparallelcomputationonthemachinesandso,ontheGanttdiagram(Fig.7).Nevertheless,forthemoment,weconsiderthattheDAJcanonlyprocessonejob.Anintermediateagentinchargeofchoosingthenum-berofjobscanbeproposed.Resultscorrespondingtothisapproacharegiveninthefollowingdiagram(Fig.8)whichgivesthevalueoftheeconomicfunc-tion(minimizationofthetardinessandtheadvance)accordingtothenumberofagentsandthenumberofgeneticoperationsusedbyagents.
9.GoingdeeplyintotherelationshipsofGAandMAS
TheuseofGAinMASisthebeginningofwhatcanbeaninterestingresearcharea.Thereareclearlytwokindsofapproaches,the rstiscentralized,inotherwords,someofthegeneticisoutsidetheagent.Thefunctionofselectionisagoodexampleofsuchfeatureout-of-the-agent[18,55].
However,webelievethatifonewantstocompletelymergetheGAandMAS(Fig.9),wemustmaketheagentacompletelyautonomousgeneticentity.Bythis,wemeanthatnotonlythegeneticpatrimonymustbe“onboard”butalsothefunctionsofselectionandcrossing.Anagentmustchoosewhichotheragentitwantstoreproducewith[55].Thelocationofthefunctionofmutationisnotclearlyknownsinceitis
Fig.7.Contract-netstructurebetweenthedifferentlevelrepresen-tations.
