Sliding Mode Control with Gaussian Process Regression for Underwater Robots

Abstract

Sliding mode control is a very effective strategy in dealing not only with parametric uncertainties, but also with unmodeled dynamics, and therefore has been widely applied to robotic agents. However, the adoption of a thin boundary layer neighboring the switching surface to smooth out the control law and to eliminate the undesired chattering effect usually impairs the controller’s performance and leads to a residual tracking error. As a matter of fact, underwater robots are very sensitive to this issue due to their highly uncertain plants and unstructured operating environments. In this work, Gaussian process regression is combined with sliding mode control for the dynamic positioning of underwater robotic vehicles. The Gaussian process regressor is embedded within the boundary layer in order to enhance the tracking performance, by predicting unknown hydrodynamic effects and compensating for them. The boundedness and convergence properties of the tracking error are analytically proven. Numerical results confirm the improved performance of the proposed control scheme when compared with the conventional sliding mode approach.