PROMPTER: Probabilistic Inference for Motion Planning in Ship Collision Avoidance Within Restricted Waterways

Navigating restricted waterways is widely considered one of the most stressful phases for ship operators due to the confined navigable areas and the uncertainty of encounter situations. Most existing ship collision avoidance methods are developed for open-water navigation and do not adequately address the unique constraints and uncertainties associated with restricted environments. This paper proposes PROMPTER, a probabilistic motion planning framework designed to assist pilots in navigating restricted waterways. PROMPTER transforms motion planning into an inference task by integrating ship maneuverability, situational awareness, and path planning within a Bayesian factor graph, allowing optimal policies to be derived through posterior inference over control. Within the graph, operational restrictions are encoded as prior knowledge to constrain and guide the control inference process. Furthermore, we theoretically prove the closed-form solution and convergence of the constrained control inference. To ensure that PROMPTER is applicable in diverse operational conditions, we consider scenarios with both reliable and unreliable communication, leading to the development of cooperative and non-cooperative collision avoidance strategies. Experiments conducted in both synthetic and real-world restricted waterways demonstrate that PROMPTER outperforms existing methods by generating collision-free and kinematically feasible paths.