Learning-based Extended Dynamic Mode Decomposition for Addressing Path-following Problem of Underactuated Ships with Unknown Dynamics

Path-following techniques of ships have received a lot of attention in recent years, to promote future autonomous ships and develop advanced autopilots. This paper deals with the path-following problem of underactuated ships without having prior knowledge regarding the hydrodynamic coefficients and ship parameters. A novel data-driven control strategy that combines Koopman operator theory and extended dynamic mode decomposition (EDMD) method and integrates with a model predictive control (MPC) framework is proposed. It makes use of data collected from experiments to learn the Koopman eigenfunctions of unknown ship dynamics via supervised learning, which are utilized as the lifting functions in the EDMD method to build a linear, lifted state-space model. The identified linear model acts as the predictor in the designed MPC controller, and a line-of-sight (LOS) algorithm is introduced as the guidance law for path-following. Simulation results show that the prediction model could provide sufficient prediction accuracy, and that it can be combined with MPC to achieve good path-following performance in a computationally efficient way.

Automated summary

This paper presents a data-driven control strategy for path-following of underactuated ships. By utilizing experiment data to learn the characteristics of unknown ship dynamics, a linear model is constructed and integrated with a model predictive control framework to achieve good path-following performance.