Distributed fault-tolerant control for multi-agent pursuit-evasion games under communication link faults

This study investigates fault-tolerant control (FTC) strategies for multi-pursuer single-evader pursuit-evasion (PE) games subject to communication link faults. To address the challenges posed by time-varying and unknown topological weights caused by these faults, a distributed adaptive observer with dynamic gain adjustment is designed to enable each pursuer to estimate the evader's state accurately, even under unreliable or corrupted links. Based on the estimated states, performance index functions are constructed by incorporating relative state errors and control efforts, thereby formulating the game dynamics. To approximate the solution of the underlying Hamilton-Jacobi-Isaacs (HJI) equation, which characterizes the Nash equilibrium of the differential game, a single-critic neural network (NN) is employed for real-time pursuit strategy updates. This approach significantly reduces computational complexity while ensuring strategic optimality. A Lyapunov-based analysis rigorously establishes the convergence of the observer estimation errors and guarantees that the relative state between each pursuer and the evader converges to zero, ensuring successful capture despite communication faults. Finally, simulation results based on a nonlinear unicycle dynamics model validate the effectiveness and robustness of the proposed control framework in the presence of communication link faults.