Perceptual Interaction-Based Path Tracking Control of Autonomous Vehicles Under DoS Attacks: A Reinforcement Learning Approach

Abstract

This paper considers the distributed optimal path tracking control problem of autonomous ground vehicles (AGVs) with matched uncertainties under denial-of-service (DoS) attacks. First, to solve the uncertain system parameters, an inverse reinforcement learning (RL) technique is PROPOSED to seek the optimal controller instead of directly solving the robust control problem, which can ensure that the designed controller is globally optimal with respect to the proposal performance function. An off-policy RL algorithm is further proposed to learn the optimal control solution online without requiring any knowledge of the vehicle dynamics. Then, to lower the communication load among vehicles, a perception-based optimal resilient controller and its corresponding perception-based static deadband condition are developed, respectively. Moreover, the perception-based dynamic deadband control mechanism is further exploited to consider the secure path tracking control problem. The proposed algorithms verify that the RL-based optimal path tracking control can be solved, and the occurrence of the Zeno phenomenon can be avoided. Finally, a numerical simulation is demonstrated to verify the effectiveness of the proposed approach.