Optimal Actuator Attacks on Autonomous Vehicles Using Reinforcement Learning