A decision-making method for autonomous vehicles based on simulation and reinforcement learning

There are still some problems need to be solved though there are a lot of achievements in the field of automatic driving. One of those problems is the difficulty of designing a decision-making system for complex traffic conditions. In recent years, reinforcement learning (RL) shows the potential in solving sequential decision optimization problems, which can be modeled as Markov decision processes (MDPs). In this paper, we establish a 14-DOF dynamic model of an autonomous vehicle and use RL to build a decision-making system for autonomous driving based on simulation. The decision-making process of the vehicle is modeled as an MDP, and the performance of the MDP is improved using an approximate RL. At last, we show the efficiency of the proposed method by simulation in a highway environment.