Deep learning-based sail-rudder coordinated waypoint tracking control for unmanned sailing vessels

To address the challenges of mutual control channel interference and over-conservative controller design arising from conventional sail-rudder separate control in waypoint tracking of unmanned sailing vessels, this study proposes a deep learning-based joint model predictive control (MPC) approach for sail-rudder waypoint tracking of unmanned sailing vessels.

First, a mathematical model describing the motion dynamics of the unmanned sailing vessel is established, along with a detailed analysis of the forces acting on the sail. Then, a prediction model is constructed using a nonlinear state-space discretization method. The prediction model is identified online using a deep neural network (DNN), and enhanced with multi-step prediction and output feedback correction techniques to improve state prediction accuracy. Subsequently, a composite objective function is formulated, incorporating both tracking error and vessel speed performance metrics. Using a cross-entropy optimization algorithm, the optimal control inputs for the sail angle and rudder angle are obtained within the prediction horizon, effectively addressing the limitations of separated controller design. Finally, the PyTorch deep learning simulation platform was used for simulation.

The simulation results demonstrate that, compared with the traditional separated PID control method for sail and rudder, the proposed method can significantly improve the waypoint tracking performance of unmanned sailing vessels under dynamic wind conditions, including variations in wind speed and direction. Additionally, it reduces the overall time required to complete the waypoint tracking task.

This method can provide a reliable theoretical support for enhancing waypoint tracking control in unmanned sailing vessels.