Gaussian Process Regression for System Identification of Autonomous Surface Vessels

Autonomous surface vehicles (ASVs) have gained significant attention across a range of applications, yet a primary challenge lies in developing accurate mathematical models to describe their complex dynamical behavior. Given the partial submersion of surface vessels in water, deriving a first-principles description proves difficult. In general, data-driven approaches, particularly black-box and gray-box models, are increasingly employed to avoid the need of structural first-principle models. Among the range of supervised learning approaches, Gaussian Process Regression (GPR) models stand out due to their simplistic and nonparametric nature. This paper presents an approach to modeling the dynamics of surface vessels using GPRs. The work outlines the process of generating synthetic data, training the GPR model, and applying it to the vessel maneuvers of path-following and dynamic positioning.