Collision avoidance of USV by model predictive control-aided deep reinforcement learning

This paper proposes a model predictive control (MPC)-aided deep reinforcement learning (DRL) algorithm to address the collision avoidance problem of unmanned surface vessel (USV). By combining with MPC, the complexity of the learning model will decrease, and the obstacle avoidance path will be safer. In this control and learning synthesized algorithm, the DRL network learns and generates velocity, and the MPC controller adjusts USV to navigate by tracking this velocity. The simulation indicated that the proposed algorithm could successfully avoid obstacles with steady and safe status.