Predictive Control of a Robot Manipulator with Deep Reinforcement Learning

This paper tackles the problem of trajectory following of a two-link rigid robot manipulator. The proposed controller bases its operation on the idea behind preview control in which the control law is divided in two parts: a feedback component that depends only on the present state of the system, and a predictive component that only uses future values of the reference trajectory. In this sense, the designed controller uses for training and control both present and future states of the system. Simulation results when following a test trajectory are presented to validate the proposed method and to show that the proposed controller exhibits better performance with respect to a neurocontroller that does not use a predictive component neither for training nor control.