Thispaper investigatesthegeneticbasedre-planningsearchstrategy,usingneural learnedvibrationbehavior forachievingtolerancecompensationofuncertainties inroboticassembly. Thevibrationbehaviorwascreated fromcomplexrobotassemblyofcoggedtubeovermultistageplanetaryspeed. Complexextensiveexperimental investigationswereconductedfor thepurposeoffindingtheoptimumvibrationsolutionforeachplanetarystage reducerinordertocompletetheassemblyprocessindefinedreal-time.However, tuningthoseparametersthrough experimentaldiscoveringforimprovedperformanceisatimeconsumingprocess.Neuralnetworkbasedlearning wasusedtogeneratewiderscopeofparameters inorder toimprovetherobotbehaviorduringeachstateof the assemblyprocess.Asanovelmodellingformalismofreactivehybridautomata,weproposetheWormholeModel withbothlearningandre-planningcapacities(WOMOLERE).Forourapplication, thestatesofhybridautoma tonincludeamplitudesandfrequenciesofrobotvibrationmodule. Thetransitionactionisafunctionofminimal distanceanduncertaintyeffectsduetojammingduringtheassemblyprocess.Theresultssuggestthatthemethod ologyisadequateandcouldberecognizedasanideafordesigningofrobotsurgeryassistancemethods,especially insoft-robotics.