Ship path planning based on deep reinforcement learning and historical trajectories

A ship path planning method is proposed to reduce ship collisions caused by improper human operations or failure to react to obstacle risks in time during navigation, and to increase the safety of ship navigation. After pre-processing the ship's historical AIS data accordingly, the K-means clustering algorithm is applied to cluster the corresponding data and a characteristic trajectory is obtained as the global static path planning trajectory of the ship in the area. In addition, a local path collision avoidance method based on DDPG（Deep Deterministic Policy Gradient） and a local path regression method based on DON（Deep Q Network） are designed to enable the ship to respond effectively to dynamic local risks. The effect of the global static path is analysed and the local dynamic path planning method is also simulated and validated. The results show that the method can obtain a safer global path and can avoid the local dynamic risks during navigation, providing a certain reference for ships' safe navigation of ships.