High-precision model predictive control and experiment of an unmanned surface vehicle with Gaussian process-based error model

Model Predictive Control (MPC) is commonly employed for trajectory tracking in control systems. However, Unmanned Surface Vehicle (USV) systems frequently encounter disturbances and model inaccuracies, resulting in mismatches between predicted and actual behaviors. This paper proposes a Gaussian Process Model Predictive Control (GP-MPC) framework to address the challenges of trajectory tracking in USVs. The MPC framework comprises a nominal model based on kinematic analysis and a Gaussian process error model. The latter compensates for system inaccuracies using sampled data. Simulations and real-world tests are conducted to compare GP-MPC with conventional MPC. The results demonstrate that GP-MPC outperforms the conventional MPC in accurately guiding the USV along the desired trajectory, effectively mitigating environmental disturbances and measurement uncertainties, and enhancing the accuracy and stability of trajectory tracking.