In context
Autonomous sailboats are emerging as low-cost, long-endurance platforms for ocean observation, but their trajectories are hard to predict because of highly variable marine conditions and limited onboard computing resources. By early 2025, researchers were seeking lightweight online learning approaches that could adapt in real time without the heavy data demands of deep learning.
What was reported
Researchers from the State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, proposed a path prediction method for the Seagull autonomous sailboat. The method combines a variable-forgetting-factor online sequential extreme learning machine (VFOSELM) with a streaming random patches (SRP) algorithm to compensate for prediction errors.
To handle noisy GPS data, a velocity-based online filter detects and corrects “drift points.” Wind and heading angles are preprocessed using sine-cosine decomposition and Gaussian blur filtering to avoid discontinuities around 0°/360°. The VFOSELM adapts its forgetting factor based on real-time prediction error, giving more weight to recent data when conditions change.
In sea trials in the South China Sea under sea state 3, the method reduced mean distance error by 77.4% and root mean square error by 78.3% compared with a fixed-forgetting OSELM. It also outperformed other advanced trajectory prediction algorithms in both training time and accuracy.
“The VFOSELM-SRP model demonstrates remarkable advantages in both training time and precision aspects.”
Why it mattered
The work shows that lightweight online learning with adaptive forgetting and error compensation can deliver high-accuracy path prediction on resource-constrained autonomous platforms. This is relevant for improving collision avoidance, energy management, and path tracking in autonomous sailboats and other marine robots operating in dynamic environments.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-03-15 · “一种基于VFOSELM-SRP的无人帆船路径预测方法”