Dynamic Route Planning Method Based on Deep Reinforcement Learning and Velocity Obstacle

Route planning is a key technology for unmanned surface vessel (USV) autonomous navigation. Traditional route planning algorithm usually has the shortcomings of complex calculation, long time and single algorithm function. In this paper, aiming at the shortcomings of traditional algorithm, the route planning strategy of USV based on deep reinforcement learning (DRL) and velocity obstacle (VO) is designed. Using electronic nautical chart to build visual environment model; Based on the kinematics of USV, Markov decision process is established, and combined with the advantages of VO method and DRL, a general reward mechanism is designed, so that USV can achieve fast and safe route planning strategy in the complex marine environment where dynamic obstacles and static obstacles exist at the same time. In order to prove our method, a simulation experiment is introduced, and the results confirm the correctness and effectiveness of the proposed method.