Event-triggered Optimal Adaptive Tracking Control for Autonomous Vehicles Using RL-Based Credibility Filter under FDI Attack

This paper investigates the secure path-tracking control problem for autonomous vehicles under false data injection (FDI) attacks. A novel reinforcement learning (RL) filter with data credibility is designed to detect and mitigate malicious data, thereby enhancing the security and efficiency of the controller. The main contributions are summarized in two aspects: 1) A reinforcement learning-based particle filtering (RLPF) method with data credibility is established for secure state estimation under FDI attacks. The RL agent is designed to dynamically assess measurement credibility, enabling reliable estimation feedback. 2) An event-triggered ADP controller driven by RLPF based attacked-state estimation is proposed to address the communication burden, limited reliable samples, and safety risks caused by FDI attacks. A Bellman error-driven PER technique is designed to improve critic neural network update efficiency under sparse triggering conditions, and PPC without requiring exact initial error bounds is embedded to constrain possible tracking error increases caused by attacked measurements. The stability of the overall system is proven. Finally, high-fidelity Simulink simulations are conducted to verify the effectiveness of the proposed approach. Comparative studies with particle filter based MPC, resilient MPC, and other methods demonstrate the superiority of the proposed framework.