Toward Robust Tracking Control of Unmanned Surface Vehicles in Complex Aquatic Environments: Online Data-Driven Model Predictive Control via Sparse Gaussian Process Regression

Abstract Recently, nonlinear model predictive control (NMPC) exhibits its superiority for time-critical tracking in the autonomous navigation of unmanned surface vehicles (USVs), while strictly adhering to operational and actuator constraints. Nevertheless, the complexity of hydrodynamic forces in harsh marine environments, coupled with time-variant system dynamics pose significant challenges to achieving high-precision tracking. These challenges arise from the deterministic model assumption and the omission of external disturbances. To address these issues, this paper proposes a Gaussian process regression (GPR)-based online system identification and data-driven model predictive control scheme for enhanced trajectory tracking accuracy and robustness. Given that the true dynamic model is inaccessible, it is decomposed into nominal and uncertain components. A sparse Gaussian process regression (GPR) framework, employing fully independent training conditional (FITC) approximation, captures the nonlinear uncertainty in the system dynamics, ensuring computational tractability within the receding-horizon optimal control scheme. To validate the approach, experiments were conducted on USV tracking of lemniscate and arbitrary trajectories in simulation environment and real-world waters. Compared with one of the popular offline method, sparse identification of nonlinear dynamics (SINDy), the proposed online identification and control scheme show potential in accuracy, robustness, promisingly supporting rapid deployment across various USVs in complex environments.