Deep reinforcement learning based neuro-control for a two-dimensional magnetic positioning system

This paper presents a control scheme based on deep reinforcement learning for a two-dimensional positioning system with electromagnetic actuators. Two neuro-controllers are trained and used for controlling the X-Y position of an object. The neuro-controllers learning approach is based on the actor-critic architecture and the deep deterministic policy gradient (DDPG) algorithm using the Q-learning method. The performance of the control system is verified for different setpoints and working conditions.