Neural Network Based Perturbation Identification Approach for High Accuracy Tracking Control of Robotic Manipulators

This paper proposes an approach for high accuracy tracking control of a robotic manipulator in the presence of model perturbations. The proposed approach designs a neural network for estimation and compensation of the modeling errors, also referred to as perturbations. Experiments are carried out on a five-axis direct-drive robotic manipulator for automated pick-place operations in semiconductor manufacturing applications. It is shown that the proposed approach can substantially improve the robot tracking and settling performance of the original computed torque algorithm implemented.