Experienced captain vs. traditional AI: Setting a new course for autonomous navigation

Teaching an AI to steer a ship is much more difficult than teaching it to drive a car. In busy waterways like Japan’s Seto Inland Sea, ships must safely navigate dense traffic, narrow straits, hundreds of islands, and ever-changing conditions while adhering to the rules of navigation.

A research group led by Assistant Professor Takefumi Higaki of the Osaka Metropolitan University Graduate School of Engineering has adopted a different approach to developing autonomous ship navigation AI.

Instead of training the model based on goals and rules, the researchers used maneuvers performed by ships. FukaemaruKobe University’s training ship. The researchers focused on “diffuse AI,” which makes decisions and creates an overall trajectory based on a variety of actions an experienced human might take, rather than trying to predict the best course of action as traditional AI does. As ships navigate the crowded Seto Inland Sea, operators must make decisions that involve ambiguity and human judgment, sometimes making decisions that are difficult to explain mathematically.

The researchers evaluated the AI ​​against two major navigation AIs built using traditional machine learning techniques based on imitation learning. Their AI was found to be able to effectively handle situations that would confuse other models. Realistic simulations could handle any number of ships, coastlines, narrow channels, and speed controls simultaneously. When conducting vessel encounter tests, it consistently complied with international collision avoidance regulations and maintained a safe distance from other vessels.

Once the model was trained, it began exhibiting unexpected behavior, such as performing local navigation customizations that were not programmed. When traveling on Akashi Kaikyo Traffic, it is local custom to drive on the right side within designated lanes. Although this was never explicitly programmed into the AI, it consistently steered the vessel into the correct lane.

“The most distinctive aspect of this study is that it did not explicitly balance multiple objectives such as collision avoidance, geographic constraints, navigational efficiency, and compliance with maritime traffic rules, but the AI ​​achieved them,” Dr. Higaki said. “Our approach allows AI to autonomously learn advanced ship handling skills directly from real operational data, rather than having researchers manually define what constitutes correct behavior.”

As more real-world ship operation data becomes available and its use continues to expand, autonomous ship navigation systems will also become more common, similar to the recent rise in AI-powered vehicles. The researchers hope their work will improve navigation safety and help solve the growing labor shortage in the maritime industry, particularly in Japan.

This study marine engineering.

###

About OMU

Founded in Osaka as one of Japan’s largest public universities, Osaka Metropolitan University is committed to shaping the future of society through the fusion of knowledge and the promotion of world-class research. For more research news, please visit https://www.omu.ac.jp/en/ and follow us on social media. ×Instagram, and LinkedIn.