Intelligent Safety Decision-Making for Autonomous Vehicle in Highway Environment

Safe driving policies is the key technology to realize the adaptive cruise control of autonomous vehicle in highway environment. In this paper, the reinforcement learning method is applied to autonomous driving's decision-making. To solve the problem that present reinforcement learning methods are difficult to deal with the randomness and uncertainty in driving environment, a model-free method for analyzing the Lyapunov stability and H infinity performance is applied to Actor-Critic algorithm to improve the stability and robustness of reinforcement learning. The safety of taking an action is judged by setting a safety threshold, thus improving the safety of behavioral decisions. Our method also designs a set of reward functions to better meet the safety and efficiency of driving decisions in the highway environment. The results show that the method can provide safe driving strategies for driverless vehicles in both normal road conditions and environments with unexpected situations, enabling the vehicles to drive safely.