Control of a Two-Dimensional Magnetic Positioning System with Deep Reinforcement Learning and Feedback Linearization

This paper presents a neuro-controller based on deep reinforcement learning to control the nonlinear dynamics of a two-dimensional magnetic positioning system. The feedback-linearized model of the magnetic positioning system is used to generate training data for the neuro-controller. The neuro-controller is trained using the Deep Deterministic Policy Gradient (DDPG) algorithm. The effectiveness of the proposed control strategy is verified with different desired set-points and trajectories, and diverse working conditions.