Robust Trajectory Tracking for UVMS via Fully Actuated System Theory and Liquid Neural Networks

Underwater vehicle-manipulator systems (UVMS) are extensively utilized in ocean exploration, deep-sea operations, and autonomous underwater robotics. However, achieving high-precision trajectory tracking poses a significant challenge due to the complexities of nonlinear dynamics, external disturbances. Traditional control methods, such as adaptive sliding mode control, have limitations in addressing time-varying disturbances, which complicates the balance between stability and smooth control. Additionally, existing approaches often introduce high-frequency chattering and struggle to adapt to variations in external disturbances. To overcome these challenges, this paper presents a robust trajectory tracking control method based on Fully Actuated System Theory (FAST) and Liquid Neural Network (LNN). The FAST controller linearizes the pseudo-linear system and optimizes the feedback gain matrix to ensure that the eigenvalues of the closed-loop system remain in the left half-plane, thus achieving rapid convergence. Concurrently, the LNN utilizes data-driven learning for real-time prediction of external disturbances, optimizing network weights to enhance disturbance rejection capabilities. Simulation results indicate that, in comparison to sliding mode control, the proposed method achieves superior trajectory tracking accuracy, improved disturbance suppression, and faster convergence. Unlike adaptive sliding mode control's passive compensation, the proposed approach utilizes LNN for disturbance prediction and compensation, ensuring smoother control inputs, reduced chattering, and improved robustness.