This paperproposesareal-timetrajectorygenerationforamasticatoryrehabilitationrobotbasedon surface electromyography(SEMG)signals.WeusedtwoGough-Stewartrobots.The first robotwasused as arehabilitationrobotwhilethesecondrobotwasdevelopedtomodelthehumanjawsystem.Thelegs of therehabilitationrobotwerecontrolledbytheSEMGsignalsofatele-operatortoreproducethe masticatory motioninthehumanjaw,supposedlymountedonthemovingplatform,throughpredicting the locationofareferencepoint.ActualjawmotionsandtheSEMGsignalsfromthemasticatorymuscles wererecordedandusedasoutputandinput,respectively.Threedifferentmethods,namelytime-delayed neural networks,timedelayedfastorthogonalsearch,andtime-delayedLaguerreexpansiontechnique, wereemployedandcomparedtopredictthekinematicparameters.Theoptimalmodelstructuresaswell as theinputdelayswereobtainedforeachmodelandeachsubjectthroughageneticalgorithm.Equa- tions ofmotionwereobtainedbythevirtualworkmethod.Fuzzymethodwasemployedtodevelopa fuzzy impedancecontroller.Moreover,ajawmodelwasdevelopedtodemonstratethetime-varying behaviorofthemusclelengthsduringtherehabilitationprocess.Thethreemodelingmethodswere capable ofprovidingreasonablyaccurateestimationsofthekinematicparameters,althoughtheaccuracy and training/validationspeedoftime-delayedfastorthogonalsearchwerehigherthanthoseoftheother two aforementionedmethods.Also,duringasimulationstudy,thefuzzyimpedanceschemeproved successful incontrollingthemovingplatformfortheaccuratenavigationofthereferencepointinthe desired trajectory.SEMGhasbeenwidelyusedasacontrolcommandforprosthesesandexoskeleton robots.However,inthecurrentstudybyemployingtheproposedrehabilitationrobotthecomplete continuous profile oftheclenchingmotionwasreproducedinthesagittalplane.

Keywords