Improved Dynamic Window Approach for Vessel Dynamic Obstacle Avoidance in Restricted Waters

Vessel navigation in restricted waters is challenged by dense obstacles and frequent dynamic interactions. The traditional Dynamic Window Approach (DWA) often falls into local optima in such environments and fails to achieve timely dynamic obstacle avoidance. To overcome this limitation, this study proposes an improved DWA-based method for vessel dynamic obstacle avoidance. The method employs a dual-layer prediction structure that combines long-horizon coarse integration with single-period fine integration, allowing earlier risk identification and avoidance of potential collisions. Furthermore, a dynamic obstacle distance term is added to the evaluation function, integrating relative motion features with the International Regulations for Preventing Collisions at Sea (COLREGs) for effective recognition and avoidance of dynamic targets. Simulation results show that the proposed method improves adaptability and safety in restricted waters with both static and dynamic obstacles, and significantly outperforms the traditional DWA in collision avoidance.