Cyber-Physical Security and Safety of Autonomous Connected Vehicles: Optimal Control Meets Multi-Armed Bandit Learning

摘要: Autonomous connected vehicles (ACVs) rely on intra-vehicle sensors such as camera and radar well inter-vehicle communication to operate effectively which exposes them cyber physical attacks in an adversary can manipulate sensor readings physically control the ACVs. In this paper, a comprehensive learning framework is proposed thwart ACV networks. First, optimal safe controller for ACVs derived maximize street traffic flow while minimizing risk of accidents by optimizing speed inter-ACV spacing. It proven that robust aim at making systems unstable. Next, two data injection attack (DIA) detection approaches are address their impact system. The leveraging stochastic behavior use multi-armed bandit (MAB) algorithm. shown that, collectively, DIA minimize vulnerability against maximizing system’s robustness. Simulation results show outperforms current state art controllers robustness attacks. also approaches, compared Kalman filtering, improve security ultimately