The proposedworkintroducesaneuralcontrolstrategyforguidingadaptationinspikingneuralstructures acting asnonlinearcontrollersinagroupofbio-inspiredrobotswhichcompeteinreachingtargetsina virtual environment.Theneuralstructuresembeddedintoeachagentareinspiredbyaspecific partofthe insect brain,namelyCentralComplex,devotedtodetect,learnandmemorizevisualfeaturesfortargeted motor control.Areduced-ordermodelofaspikingneuronisusedasthebasicbuildingblockfortheneural controller. Thecontrolmethodologyemploysbio-inspired,correlationbasedlearningmechanismslike Spike timingdependentplasticity with theadditionofareward/punishment-basedmethodexperimentally found ininsects.Thereferencesignalfortheoverallmulti-agentcontrolsystemisimposedbyaglobal reward, whichguidesmotorlearningtodirecteachagenttowardsspecific visualtargets.Theneural controllers withintheagentsstartfromidenticalconditions:thelearningstrategyinduceseachrobottoshow anticipated targetingactionsuponspecific visualstimuli.Thewholecontrolstructurealsocontributesto make therobotsrefractoryormoresensitivetospecific visualstimuli,showingdistinctpreferencesinfuture choices. Thisleadstoanenvironmentallyinduced,targetedmotorcontrol,evenwithoutadirect communication amongtheagents,givingrobots,whilerunning,theabilitytoperformadaptationinreal- time. Experiments,carriedoutinadynamicsimulationenvironment,showthesuitabilityoftheproposed approach. Specific performanceindexes,likeShannon'sEntropy,areadoptedtoquantitativelyanalyze diversity andspecializationwithinthegroup.
Titolo: | Spiking neural controllers in multi-agent competitive systems for adaptive targeted motor learning |
Autori: | |
Data di pubblicazione: | 2015 |
Rivista: | |
Abstract: | The proposedworkintroducesaneuralcontrolstrategyforguidingadaptationinspikingneuralstructures acting asnonlinearcontrollersinagroupofbio-inspiredrobotswhichcompeteinreachingtargetsina virtual environment.Theneuralstructuresembeddedintoeachagentareinspiredbyaspecific partofthe insect brain,namelyCentralComplex,devotedtodetect,learnandmemorizevisualfeaturesfortargeted motor control.Areduced-ordermodelofaspikingneuronisusedasthebasicbuildingblockfortheneural controller. Thecontrolmethodologyemploysbio-inspired,correlationbasedlearningmechanismslike Spike timingdependentplasticity with theadditionofareward/punishment-basedmethodexperimentally found ininsects.Thereferencesignalfortheoverallmulti-agentcontrolsystemisimposedbyaglobal reward, whichguidesmotorlearningtodirecteachagenttowardsspecific visualtargets.Theneural controllers withintheagentsstartfromidenticalconditions:thelearningstrategyinduceseachrobottoshow anticipated targetingactionsuponspecific visualstimuli.Thewholecontrolstructurealsocontributesto make therobotsrefractoryormoresensitivetospecific visualstimuli,showingdistinctpreferencesinfuture choices. Thisleadstoanenvironmentallyinduced,targetedmotorcontrol,evenwithoutadirect communication amongtheagents,givingrobots,whilerunning,theabilitytoperformadaptationinreal- time. Experiments,carriedoutinadynamicsimulationenvironment,showthesuitabilityoftheproposed approach. Specific performanceindexes,likeShannon'sEntropy,areadoptedtoquantitativelyanalyze diversity andspecializationwithinthegroup. |
Handle: | http://hdl.handle.net/11570/3148488 |
Appare nelle tipologie: | 14.a.1 Articolo su rivista |