Improved reinforcement learning for collision-free local path planning of dynamic obstacle

The rapid development of artificial intelligence has driven the transformation of traditional industry. The proposal for an intelligent ship reflects the development of the shipping industry in the direction of intelligence. Safe maritime transportation is an essential branch of the intelligent ship industry. In this context, applying various intelligent algorithms in ship dynamic obstacle avoidance has attracted the attention and discussion of scholars. Due to the complexity and diversity of ship navigation, environments cannot be described by a definite mathematical model. Reinforcement learning has certain advantages in solving path planning in a complex environment. Traditional mathematical algorithms and swarm intelligence algorithms need mathematical models to constrain boundary conditions when solving complex problems. However, some complex dynamic environments in the real world cannot define mathematical models. This paper optimizes the parameter update and exploration strategies of the actor-critic algorithm in reinforcement learning to improve the algorithm's performance. Under complex meteorological conditions, improved reinforcement learning is applied to the dynamic ship offshore channel simulation environment to verify the algorithm's performance.