Nonlinear Model Predictive Control for Ship Collision Avoidance Using Gaussian Process Learning

From the perspective of automatic navigation and autonomous collision avoidance, establishing a more precise and robust ship maneuvering model and designing an appropriate collision avoidance algorithm are key to achieving autonomous ship driving.In this paper, we will study the use of Gaussian processes to establish a comprehensive mathematical model of ship motion and combine it with model predictive control to address the issue of autonomous ship collision avoidance. Gaussian processes, as a non-parametric regression model, combined with model predictive control, have shown excellent results. We use Gaussian process regression to model unknown hydrodynamic functions and combine them with a nominal model to establish a more comprehensive and accurate mathematical model of the ship. Model predictive control handles various constraints on inputs and states to ensure that physical and performance limitations are met during the autonomous collision avoidance control process of the ship. By considering multiple factors, we integrate Gaussian processes into model predictive control. The Gaussian process can serve as the predictive model within model predictive control. We have validated that the established ship model can achieve intelligent collision avoidance for ships.