A Learning-Based Estimation and Fault-Tolerant Control Scheme for Uncertain Heterogeneous Multi-Agent Systems

The learning-based cooperative control problem is investigated for the heterogeneous multi-agent system affected by unknown system uncertainties, actuator faults, and external disturbances. A learning-based estimation scheme is proposed to simultaneously identify unknown uncertainties, actuator faults, and disturbances. Exponential convergence of estimation errors is guaranteed by the designed learning laws without requiring persistent excitation. Based on the time-varying system matrix estimation, a constrained matrix observer is developed to satisfy specified matrix inequalities, thereby guaranteeing the stability of the closed-loop system. Furthermore, a distributed cooperative controller is designed for virtual agents to achieve synchronization with the leader while preventing fault propagation. Subsequently, a fault-tolerant controller is constructed for the actual heterogeneous agents, ensuring convergence between the faulty agents and the leader. Theoretical results are validated through simulations of connected vehicle platoons incorporating realistic vehicle dynamics and fault scenarios. Finally, the superiority of the proposed method regarding transient performance and control efficiency is highlighted through comparisons with existing adaptive learning techniques.