Council touts domestically developed AI vessel system

• By Jake Chung / Staff writer, with CNA

• • Add TT as Preferred Source

  • 

  • 

  • 

A domestically developed system combining artificial intelligence (AI)-powered autonomous navigation and federated learning has completed autonomous patrol tests using a medium to large-sized uncrewed surface vessel, the Ocean Affairs Council said yesterday, adding that it would help protect sensitive maritime data and improve vessels’ object-recognition capabilities.

The system is the result of the council’s “AI Navigation and Federated Learning for Maritime Object Recognition: Feasibility and Reliability Testing for Scalable Unmanned Surface Vessels” project, conducted under its Marine Technology Project Program.

The project was part of the council’s efforts to accelerate marine technology development and the application of smart maritime systems in Taiwan, it said.

A composite image shows navigation information, and medium-sized and large uncrewed surface vessels off the coast of the Port of Kaohsiung.

Photo courtesy of the Ocean Affairs Council

Maritime imagery has been collected mainly as snapshots of individual incidents, while national and port security concerns limited access to broader datasets, restricting the amount of data available for AI training, the council said.

The use of federated learning allows AI models to be trained using data from multiple vessels without requiring ships to share sensitive data or imagery, it said.

Previously, uncrewed vessels trained their own AI models and could share the algorithms used in the training process, but not the underlying data, the council said.

The new system allows vessels across the fleet to improve their models collaboratively while keeping sensitive information and imagery on individual platforms, significantly improving their ability to recognize objects, it said.

To prevent “catastrophic forgetting,” in which AI models lose previously learned skills when acquiring new ones, the research team incorporated elastic weight consolidation.

The technology allows uncrewed vessels to learn to recognize new objects, such as newly introduced maritime warning buoys, without losing their ability to identify commercial vessels and fishing boats, the council said.

The vessels underwent stress tests in waters off Fengbitou (鳳鼻頭) near Kaohsiung Harbor.

Under real-world conditions, including sea states of force 2 or higher and challenging lighting, the vessel completed a series of demanding tasks, including point-to-point obstacle avoidance, hexagonal navigation, Z-shaped patrols and large-area “lawnmower” searches, the council said.

The system achieved a mean average precision of more than 90 percent for object detection, an inference speed of at least 15 frames per second and a forgetting rate of less than 10 percent, it said.

The project also established Taiwan’s first standard operating procedures for operating medium and large-sized uncrewed vessels in open waters, the council said.

The vessels are equipped with an emergency system that allows human operators to take control within three seconds of a communications failure, ensuring safe navigation, it said.

All of the system’s algorithms were developed domestically, the council said, adding that the project was designed to reduce reliance on foreign technologies subject to export controls and bolster Taiwan’s resilience in marine research and maritime security.