Autonomous sailing against the wind

Reinforcement Learning (RL) is a branch of Artificial Intelligence (AI) that can optimize a strategy through interactions with a dynamic environment. One advantage of RL lies in its ability to learn a strategy without requiring explicit knowledge. Once learned, the strategy can be applied to automate a sequential decision-making process. Such techniques can have various applications in maritime domain: improving ship operational design, crew training, optimizing ship navigation routes, and enhancing overall vessel control and safety in various operations such as course-keeping in waves, collision avoidance and docking.

AI SAIL Project

The aim of the AI SAIL project is to explore together with the sector the possibilities and limitations of RL for maritime applications. We did it with an intuitive case study: sailing a small sailboat (the Optimist) autonomously against the wind in MARIN’s Offshore basin. The RL agent was trained in interaction with a fast time-domain simulation providing an environment for an easier development, a speed-up training and a safer learning. When applied in the real world, the strategy learned in the simulated environment might perform poorly, as the simulation model might not capture the complex physics of the real world phenomena. We addressed this issue by adopting various sim-to-real transfer techniques ensuring that the strategies learned in simulation generalize well to the real world. On 24 November 2023, we gave a public demonstration of the project in the Offshore basin at MARIN.

Reference: Kiki Bink, Autonomous Sailing with Sim-to-Real Reinforcement Learning, MSc thesis, Delft University of Technology, 2024

Contact

Fanny Rebiffe

Applied Data Scientist

Download

AI Sail –Sailing against the wind autonomously (pdf)