Learning-Based Control of
Huarong Zheng, Chenguang Liu · 2025
Due to the universal marine environmental disturbances and the uncertain USV hydrodynamics, it is in general difficult to obtain an accurate USV model. In this chapter, the model free control methods for USVs are discussed, with emphasize on the emerging powerful learning based approaches. Overall, model free based methods do not need to obtain the complex USV hydrodynamics accurately, which is in line with the practical situations. A new hybrid physics-learning model based predictive control (PL- MPC ) method is proposed that utilize both the USV identified and learned models. The learning dynamics are modeled with the Deep Neural Networks ( DNN ). The performance of the proposed model free control approach is demonstrated with theoretical analysis and comprehensive simulation results.