Dynamic path planning for bridge collision avoidance using trajectory clustering and BC-PPO fusion

This paper proposes an autonomous ship path planning and risk warning system for bridge zone waters to address navigation safety challenges in complex dynamic environments. The proposed method integrates AIS trajectory density analysis with a behavior cloning–proximal policy optimization (BC–PPO) reinforcement learning framework. Safe navigation zones for different vessel types are identified using hierarchical density-based spatial clustering, followed by kernel density estimation to extract safety boundaries and derive an optimal density-ridge-based route. A BC–PPO framework is then developed, where behavior cloning provides expert priors and PPO fine-tuning improves policy stability and convergence under a designed reward function, enabling vessels to follow the optimal route within safety zones. In addition, a risk detection module is designed to provide early warnings for vessels deviating from or approaching unsafe regions. Experimental results demonstrate that the proposed method effectively supports safe navigation and risk warning in bridge zone waters.