Sim-to-Real RL for ASVs using SysID

This research presents a framework for training reinforcement learning policies for Autonomous Surface Vehicles (ASVs) using system identification. The approach enables zero-shot sim-to-real transfer without prior hydrodynamic or propeller information, relying instead on a CAD model and open-water trajectories to approximate and refine vehicle dynamics and thruster parameters. Real-world deployment on a BlueBoat ASV demonstrates successful performance in path following and station-keeping tasks.