Optimizing unmanned surface vehicle control: A data-enabled learning approach

• Data-enabled learning is proposed to optimize USV motion control. • Gradient-based tuning requires no explicit vehicle model knowledge. • Special experiments use tracking errors to estimate gradients accurately. • Experiments on a USV prototype show rapid convergence and improved tracking. Unmanned surface vehicles (USVs) have gained significant attention recently for applications such as delivery and trash removal. However, accurately modeling these vehicles is difficult due to their inherent underactuation and complex dynamics, which often result in inaccurate tracking. To address this challenge, we propose a data-enabled learning approach to fully exploit the abundant data available for achieving enhanced control performance. The core concept is that suboptimal motion generates a substantial amount of data, specifically related to surge, yaw rate, and control inputs. This rich information can enable an efficient learning process to enhance motion control. In this work, we use data collected from experiments to optimize planar motion control in an underactuated vessel. The optimization algorithm allows for efficient tuning of the control gains for a predefined controller, with quick convergence. Importantly, the gain optimization does not require knowledge of the vehicle model. Simulations and experiments conducted on a vessel prototype demonstrate improved controller performance and efficiency in learning.