A Model Predictive Control Approach for USV Autonomous Cruising via Disturbance Learning

Unmanned surface vehicles (USVs) are widely applied in ocean exploration and environmental protection. To ensure efficient execution of tasks, the motion control of USV is essential and critical. However, the hydrodynamics disturbances from ocean environment are commonly highly nonlinear, time-variant and impractical to be modelled, which renders control extremely challenging. Therefore, in this paper we propose a learning-based model predictive control (MPC) approach for USV course-keeping subject to disturbances and uncertainties. A relatively simplified dynamics model is augmented by a long short term memory (LSTM) residual model, which can capture complicated hydrodynamics effect and eliminate model mismatch. The resulting formulation is incorporated in MPC framework to achieve optimal real-time control. Further, the proposed approach is verified through simulation experiments.