Time-consistent adaptive sampling RRT* for COLREG-compliant maritime collision avoidance

Safe and COLREGs-compliant collision avoidance in dynamic maritime environments remains challenging, especially with multiple ships and environmental constraints such as shallow water. This paper presents a complete and online collision avoidance system based on the Rapidly-exploring Random Tree ∗ (RRT ∗ ) planner, including estimation of the states of surrounding target ships, detection of collision risk and encounter type, and online replanning of the own ship route when needed. The work makes three main contributions. First, an adaptive potential field guided sampling strategy encodes COLREGs encounter rules and depth constraints from electronic navigational charts (ENC), and is updated online so that the sampling distribution adapts to the evolving encounter geometry during tree expansion. Second, continuous-curvature path generation is integrated directly into tree expansion, limiting abrupt heading changes, which improves the kinematic feasibility of the generated paths. Finally, a time-consistent rewiring strategy preserves the temporal coherence of the tree without requiring full tree repropagation. The system is validated in simulation across multiple-ship scenarios, restricted waters, and unexpected target ship maneuvers. Results show that the adaptive sampling strategy substantially reduces the number of iterations required to find feasible solutions compared with uniform sampling, while maintaining high success rates.