Thegeneralizedregressionneuralnetwork-onekindofRBFneuralnetwork,ischosentoconstructtheinverse-kinematicsmodelfortheshotcreterobotwhichhasredundantdegree-of–freedom.Theinverse-kinematicsmodeloftheobjectistrainedbythegenerallearningmethod.Inconstructingmodelprocess,differentpartitionmethodsistriedtopidethejointspaceanddifferentdiffusioncoefficientvaluetotraintheneuralnetwork.Theinfluenceofthespreadcoefficienttotheapproachabilityisalsostudied.Thesimulationmethodisadoptedtotesttheperformanceoftheneuralnetwork.Thesimulationresultturnsouttobesatisfactory.