01openalex2026-04-21Takefumi Higaki, Hitoshi Yoshioka, Nobukazu Wakabayashi, Hirotada HashimotoDOI ↗ (opens in a new tab)
Diffusion route planner: Data-driven modeling of human ship navigation that implicitly balances safety, efficiency, and rule compliance under complex geographical constraints (opens in a new tab)
The paper introduces a diffusion route planner (DRP) that models human ship navigation using 181 days of AIS–radar-integrated data from a training ship. In tests, the DRP followed COLREGs Rules 13–17 in one-on-one scenarios and maintained safety margins in complex, multi-ship real-sea conditions. Unlike other data-driven methods, it handled realistic constraints better due to offline training on fused sensor data. The approach also showed potential to incorporate additional rules and local regulations without explicit programming.