Dynamic Parameter Identification of Robots Using a Neural Net

Abstract

This paper addresses issues of dynamic parameter identification of robot manipulators. A new identification approach with neural network based compensation of uncertain dynamics is proposed. The parameter identification process is divided into two steps. The first step is to determine unknown dynamic parameters using inverse dynamics of the robot manipulator and pseudo-inverse matrices. The second step is to establish a dynamic compensator by neural network and learning method for improving accuracy of the dynamic model with parameters given in the first step. A Direct Drive (DD) SCARA type industrial robot arm AdeptOne is used as an application example for the parameter identification. Simulations and experiments are carried out. Comparison of the results confirms the correctness and usefulness of the proposed identification method.