Reverse Parking a Car-Like Mobile Robot with Deep Reinforcement Learning and Preview Control

摘要

This paper presents a control technique for reverse parking car-like vehicles based on deep reinforcement learning and preview control. The deep deterministic policy gradient (DDPG) algorithm is used for training a neurocontroller using a reward function defined in terms of the desired final state of the system. A preview control approach is employed to leverage knowledge of a known a priori reference input to generate a predictive control signal coupled into the neurocontroller output. Simulation results are presented to validate the proposed method. Moreover, these results show that incorporating a preview control signal improves the parking time.