Intersections are the hubs of urban transport networks and the bottlenecks of transportation efficiency. Traditional fixed timing policies cannot be adjusted flexibly as traffic status changes. Adjusting traffic lights adaptively according to different traffic conditions would improve urban traffic efficiency. In recent years, with the development of machine learning, especially reinforcement learning technologies, these problems can be solved with the help of these advanced AI methods. Traffic signal control problems are just one kind of these. Therefore, in this paper, an adaptive model for controlling traffic signals based on asynchronous reinforcement learning algorithms is proposed. In order to test the effect of our policy, we selected queue length, average waiting time, and average speed of vehicles as evaluation indices. By contrast, we found that the policy based on asynchronous reinforcement learning performs better to the policy based on simple actor critic model in each index and fixed time plan.
展开▼
