This paper discusses and verifies a design method of reward functions in reinforcement learning algorithm. Reinforcement learning is a type of machine learning algorithm. The characteristic is that the agent can accumulate experience in repeated experiments and learn from experience to complete the target task. The core idea of this method is trial and error. In the reinforcement learning algorithm, the design of the reward function is very important. This paper discusses a method of designing a reward function and verifies it in an autonomous driving simulator called carla.