Reinforcement learning-based finite-time cross-media tracking control for a cross-media vehicle under unknown dynamics and disturbances

Abstract

This study proposes a reinforcement learning-based finite-time cross-media tracking control approach for a slender body cross-media vehicle encountering unknown hydrodynamics, wind, and wave disturbances. Initially, a reinforcement learning framework consisting of the actor neural network and critic neural network is constructed. The critic neural network monitors the actions of the actor neural network and approximates the cost function, while the actor neural network estimates the unknown hydrodynamics and disturbances, minimising the cost function to optimise performance. Subsequently, the command filter featuring finite-time convergence is formulated, effectively managing the corresponding filter error through a proposed error compensating signal. By integrating these techniques, a reinforcement learning-based finite-time control strategy is developed, circumventing the singularity issue inherent in traditional finite-time backstepping strategies. Comparative analysis with existing methods demonstrates the strong robustness of the proposed scheme against unknown hydrodynamics and disturbances, ensuring finite-time convergence of the system's states and optimising controller performance. Finally, simulations confirm the effectiveness and superiority of the presented approach.