AI learns ship navigation from experienced captains

An AI navigation system developed by Osaka Metropolitan University learned to steer a ship through crowded waterways by studying maneuvers made by an experienced captain—rather than by following manually written objectives and rules.

The challenge is especially acute in Japan’s Seto Inland Sea, where autonomous vessels must manage dense traffic, narrow channels, hundreds of islands and changing conditions while complying with maritime regulations. The research group, led by Assistant Professor Takefumi Higaki of the university’s Graduate School of Engineering, trained its model using operational data from Fukae-Maru, a training vessel operated by Kobe University. The study was published in Ocean Engineering.

Diffusion AI learns complex ship-handling behavior

Instead of predicting a single “best” action, the system uses diffusion AI to model a range of actions an experienced operator might take and generate a complete route. That approach is designed to capture ambiguous decisions and human judgment that can be difficult to express mathematically.

The researchers compared the system with two leading navigation AIs based on conventional machine-learning and imitation-learning methods. In realistic simulations, the new model handled arbitrary numbers of ships, coastlines, narrow waterways and speed control at the same time, including scenarios that confused the comparison systems.

Recommended reading

US computer-science enrolment falls for first time in 20 years

During ship-encounter tests, it consistently followed international collision-avoidance regulations and maintained a safe distance from other vessels. The model also developed behavior that had not been explicitly programmed: when navigating the Akashi Kaikyo Traffic Route, it consistently kept right within designated traffic lanes, reflecting a local navigation convention.

“The most distinctive feature of this study is that we did not explicitly balance multiple objectives such as collision avoidance, geographical constraints, navigation efficiency and compliance with maritime traffic rules; however, the AI achieved them.”

Higaki said the method allows an AI to learn sophisticated ship-handling skills directly from real-world operational data, without researchers having to define every element of correct behavior manually.

Autonomous shipping and maritime labor shortages

The researchers expect autonomous navigation systems to become more common as real-world vessel-operation data increases. They hope the technology will improve navigational safety and help address labor shortages in Japan’s maritime industry.

The research is detailed in “Diffusion route planner: Data-driven modeling of human ship navigation that implicitly balances safety, efficiency, and rule compliance under complex geographical constraints,” published in Ocean Engineering in 2026. DOI: 10.1016/j.oceaneng.2026.125656.