A Decoupled Global-Local Planning Framework for COLREGs-Aware Maritime Autonomous Surface Ships in Coastal Waters

Autonomous navigation of Maritime Autonomous Surface Ships (MASS) in complex coastal waters requires reliable global path generation and stable local collision avoidance. Existing planners often couple path search, path smoothing, and collision-avoidance reactions, increasing global search burden and reducing reference-path usability. This paper proposes a decoupled hierarchical framework for International Regulations for Preventing Collisions at Sea (COLREGs)-aware MASS navigation. The framework separates global planning from local collision avoidance and further decomposes the global layer into connectivity search and path-usability refinement. Adaptive Angular-Domain Potential Field (AAPF)-based waypoint expansion generates feasible raw paths through multi-scale, multi-angle candidate search, while Farsighted Sweep Trajectory Optimization (FSTO) refines them through forward visibility scanning and offset-based trajectory adjustment. At the local layer, the Environment-Adaptive Dynamic Window Approach (EA-DWA) integrates Distance to Closest Point of Approach (DCPA)- and Time to Closest Point of Approach (TCPA)-informed risk evaluation, COLREGs-aware velocity-window constraints, and dynamic target guidance to coordinate path tracking and collision avoidance. Simulations in virtual scenarios and real coastline-based environments show that the staged framework reduces global search burden and waypoint count while supporting COLREGs-consistent local avoidance and reference-path recovery. Cross-map and sensitivity results further reveal the need for limited environment-dependent adjustment of the guidance weights and the performance trade-offs associated with key parameters.