資源描述：
《Neural network - trajectory generation》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、302IEEETRANSACTIONSONSYSTEMS,MAN,ANDCYBERNETICS—PARTB:CYBERNETICS,VOL.31,NO.3,JUNE2001NeuralNetworkApproachestoDynamicCollision-FreeTrajectoryGenerationSimonX.Yang,Member,IEEE,andMaxMeng,Member,IEEEAbstract—Inthispaper,dynamiccollision-freetrajectorysiblepathsinthew
2、orkspace,whichnormallydealwithstaticgenerationinanonstationaryenvironmentisstudiedusingenvironmentonlyandarecomputationallyexpensivewhenthebiologicallyinspiredneuralnetworkapproaches.Theproposedenvironmentiscomplex.Somesearchingbasedmodels(e.g.,neuralnetworkistopolo
3、gicallyorganized,wherethedynamics[12],[23],[50])sufferfromundesiredlocalminima,i.e.,theofeachneuronischaracterizedbyashuntingequationoranadditiveequation.ThestatespaceoftheneuralnetworkcanberobotsmaybetrappedinsomecasessuchaswithconcaveeithertheCartesianworkspaceort
4、hejointspaceofmulti-jointU-shapedobstacles.SeshadriandGhosh[47]proposedanewrobotmanipulators.Thereareonlylocallateralconnectionspathplanningmodelusinganiterativeapproach.Howeveramongneurons.Thereal-timeoptimaltrajectoryisgeneratedthismodeliscomputationallycomplicate
5、d,particularlyinthroughthedynamicactivitylandscapeoftheneuralnetworkacomplexenvironment.LiandBui[28]proposedafluidwithoutexplicitlysearchingoverthefreespacenorthecollisionpaths,withoutexplicitlyoptimizinganyglobalcostfunctions,modelforrobotpathplanninginastaticenvir
6、onment.Ongwithoutanypriorknowledgeofthedynamicenvironment,andandGilbert[38]proposedanewmodelforpathplanningwithwithoutanylearningprocedures.Thereforethemodelalgorithmpenetrationgrowthdistance,whichsearchesovercollisioniscomputationallyefficient.Thestabilityoftheneur
7、alnetworkpathsinsteadofsearchingoverfreespaceasmostothermodelssystemisguaranteedbytheexistenceofaLyapunovfunctiondo.Thismodelcangenerateoptimal,continuousrobotpathscandidate.Inaddition,thismodelisnotverysensitivetothemodelparameters.Severalmodelvariationsarepresente
8、dandinastaticenvironmentonly.Orioloetal.[39],[40]proposedathedifferencesarediscussed.Asexamples,theproposedmodelsmodelforreal-timemapbuild
