Predefined-time data-driven distributed neural formation control for underactuated AUV systems

This paper proposes a predefined-time data-driven distributed neural sliding mode formation tracking controller for autonomous underwater vehicles (AUVs) with unknown dynamics. Although existing data-driven methods provide model-free solutions, their black box learning limits interpretability and lacks a precise specification of convergence time. Therefore, we develop a novel predefined-time-stable data-driven modeling framework that reconstructs nonlinear dynamics online from input and output data, reducing the dependence on accurate model parameters, and rigorously show that the upper bound of convergence time is prescribable via Lyapunov analysis. Furthermore, the reconstructed dynamics are embedded into the sliding-mode controller, forming a closed-loop framework with a prescribed convergence time upper bound. Bio-inspired neural shunting dynamics alleviate high frequency chattering. Lyapunov analysis substantiates practical predefined-time stability of the closed-loop system, and simulations verify the effectiveness and superiority of the proposed method.