Adaptive Coordinated Traffic Control for Arterial Intersections based on Reinforcement Learning

Coordinated control of arterial roads is widely used to increase the efficiency of the main road. However, the traditional fixed-time control is difficult to simultaneously tackle the fluctuation of the initial queuing on the main road and the flow in all directions. To fill this gap, this research employs the reinforcement learning (RL) method to design adaptive coordinated traffic control for arterial intersections. Specifically, offsets and green splits of the arterial intersections are adaptively controlled by RL agent according to the perception of the traffic dynamics. This research deeply analyzes the boundaries, features and objectives of the problem and, accordingly, determines the training environment and state, reward, action, and timeline of the RL agent. 200 random traffic scenes are built, in which 50 percent are for the training and 50 percent are for the test. The results show that the algorithm proposed in this study can significantly outperform fixed-time signal scheme with optimization of offset and green-split. The stop number and delay are decreased by -11% and - 4% in the large-scale test set, which verifies the effectiveness of the proposed model.