Robust Koopman Operator-Based Data-Driven Predictive Control for 6-DOF AUVs

In this paper, we investigate a novel data-driven model predictive control framework integrating Koopman operator theory with extended state observer, which addresses the trajectory tracking challenge for Autonomous Underwater Vehicles (AUVs) in complex marine environments. The presented approach employs dynamic mode decomposition with singular value decomposition to extract a Koopman linearized model directly from state data, effectively circumventing the strong dependence on nonlinear dynamic parameters inherent in conventional physical modeling. Building upon this data-driven model, we develop a constraint-enhanced model predictive control architecture that systematically incorporates system state constraints and actuator saturation limits into the optimization objective. Numerical simulations validate the superior trajectory tracking performance and disturbance rejection capabilities of the proposed framework under various marine disturbances, demonstrating marked improvements over conventional model-based control strategies.