Intelligent Ships Collision Avoidance and Navigation Method Based on Deep Reinforcement Learning

Abstract

In order to improve the intelligence of the ship and ensure the safety and reliability of the ship during navigation. Based on the deep Q-network algorithm in the field of deep reinforcement learning, this paper studies of ship intelligent collision avoidance and navigation to enable the ship to autonomously carry out collision avoidance operations and reach the goal without manual operation. According to the characteristics of actual waters, several simulated water environments are designed to study the collision avoidance and navigation training effects of ships based on algorithms. Combining collision avoidance and navigation problems with ship motion characteristics, a reasonable state space, reward function, and Q-value neural network structure are optimized. It enables the ship to choose appropriate actions according to the environmental conditions, conduct static and dynamic obstacle avoidance and navigation. There are multiple ships in the experimental environment, including different ship encounter situations, which has certain practical significance. Simulation experiments have proved the effectiveness of the algorithm. After training, the ships can smoothly complete collision avoidance and navigation operations, which can effectively reduce the risk of ship collision.