PURPOSE: To provide an identification method for dynamics in the orbit scheme of an industrial manipulator.

CONSTITUTION: Since the power of each frequency component differs from an orbit interpolated by a polygon having been used for a conventional identification method and the power of high frequency components is smaller, the accuracy in the identification is reduced. The problems above are solved by using a pseudo white noise orbit in which the power of each frequency component is almost constant and the band is limited. Furthermore, since the change in a mass of a load grasped by a manipulator is not taken into account in a neural circuit network model used for a conventional orbit scheme, the model cannot cope with the orbit scheme for a general-purpose industrial manipulator, the mass of the load to be grasped is added to the input of the neural circuit network model to solve the problems that dynamics of the manipulator are learned by the model when the mass of the load is changed.