SINDy-based modeling of an uncrewed surface vessel with experimental validation

This paper introduces physics-based digital twins of uncrewed surface vehicles (USVs) using the Sparse Identification of Nonlinear Dynamical Systems (SINDy) framework. The identified governing dynamics are embedded into the digital twins to enable accurate simulation of USV behavior. Field experiments involving a range of maneuvers, both linear and curved trajectories, are conducted to collect operational data. To streamline the modeling process, the vehicle dynamics are decomposed into longitudinal and lateral subsystems, considering three degrees of freedom (DOF). Data from the onboard central computer are used to identify the system’s nonlinear dynamics matrix. Several candidate function libraries within the SINDy framework, including first- and second-degree polynomials and a physics-informed custom library, are tested for model training and validation. Root mean square error (RMSE) is employed as the performance metric. Results show that the proposed custom library captures the vehicle dynamics more accurately, achieving a 26.8% lower average RMSE with reduced standard deviation compared to the polynomial-based models.