LLM-assisted adaptive dynamic programming and prescribed performance control for USV trajectory tracking under...

To address the challenges of stochastic disturbances, model uncertainties, and the inherent difficulty in tuning multi-parameter controllers for Unmanned Surface Vehicles (USVs) in complex maritime environments, this paper proposes a backstepping control strategy that integrates Prescribed Performance Control (PPC) and Adaptive Dynamic Programming (ADP). Furthermore, a Large Language Model (LLM) is innovatively introduced to provide intelligent guidance for parameter tuning. Initially, a prescribed performance function is employed to constrain the transient and steady-state performance of the tracking. Subsequently, a hierarchical ADP architecture based on the backstepping method is constructed. Within the dynamics loop, a dual Deep Neural Network (DNN) framework is designed to approximate unknown disturbances and model uncertainties, as well as to solve the stochastic Hamilton-Jacobi-Bellman (HJB) equation effectively. To mitigate the issues of parameter coupling and the low efficiency associated with manual tuning, an LLM-assisted parameter tuning mechanism is designed. Utilizing multimodal information incorporating both natural language descriptions and response curves, this mechanism analyzes the characteristics of the closed-loop system and the internal correlations within the control parameter set to guide hyperparameter optimization. The Semi-Global Uniform Ultimate Boundedness (SGUUB) of the tracking errors is proved using Lyapunov stability theory. Extensive simulation results under diverse operating conditions demonstrate that the proposed method outperforms traditional baseline methods in terms of trajectory tracking accuracy. Moreover, the LLM-assisted tuning mechanism provides insightful suggestions for parameter adjustment, significantly improving tuning efficiency and offering a novel perspective for parameter tuning in complex control systems.