Reinforcement learning based overtaking decision-making for highway autonomous driving

In this paper, we develop an intelligent overtaking decision-making method for highway autonomous driving. The key idea is to use reinforcement learning algorithms to learn an optimized policy via a series of simulated driving scenarios. A vehicle model based on data fitting of real vehicles as well as a traffic model is established to simulate driving scenarios and validation tests of obtained policies. Human driving experiences are considered in designing the reward function. A reinforcement learning method called the Q-learning algorithm is used to learn overtaking decision-making policies. Simulations show that our method can learn feasible overtaking policies in different traffic environments and the performance is comparable or even better than manually designed decision rules.