Optimal Actuator Attacks on Autonomous Vehicles Using Reinforcement Learning

Recently, there has been a growing focus on the security and safety issues associated with these vehicles. Due to focus on the stealthiness of the attacks, which is crucial given the high reliance of autonomous vehicles on software and that modern autonomous vehicle systems are equipped with communication systems, they are vulnerable to different advanced attack detectors. Second, the design of their secure types of attacks, as shown in Figure 1, which may lead to severe controller is based on specific FDI attack and training data accidents [3]. Attacks on autonomous vehicles are typically obtained through RL, limiting its generalization to different categorized into those targeting actuators and those targeting types of attacks. These limitations motivate our research.