Deep reinforcement learning-based path planning of underactuated surface vessels

Abstract

This paper is dedicated to the issues of path planning for the underactuated surface vessel (USV) in unknown environments with obstacles. Aiming at the usability problem caused by the complicated control law of the traditional method, a deep deterministic policy gradient (DDPG)-based path planning algorithm is proposed with the powerful actor-critic architecture. The main contributions of this paper are as follows: (1) The DDPG-based path planning method is the decision control system of USV whose output is continuous. (2) A complete reward system is specially designed for target approaching, speed control and attitude correction. (3) Since the reward system is independent, it is highly customisable and can be changed according to the task and the control model. Through the simulation, the results show that the perfect path can be automatically generated under unknown environmental disturbance, thus providing the proposed DDPG-based path planning scheme effectiveness and applied meaning.