With the development of detection and computation techniques, the use of reinforcement learning (RL) in traffic signal control problems is widely discussed. After formulating diverse isolated RL agents to control one intersection design, most existing studies tend to directly duplicate isolated agents for large-scale coordinated control problems. However, two questionable challenges are 1) the coordinated control commonly differs from isolated control owing to different objectives; 2) the coordination necessity varies under different traffic demand. Thus, a naive duplication or aggregation of isolated RL agents seems unsatisfactory. In this paper, we focus on the classical arterial control problem to investigate an appropriate coordination strategy. Inspired by green-wave control, a knowledge-combined RL controller is proposed that can predict the potential opportunity of creating non-stop traffic through an artery by matching an ego intersection's phase selection and upstream historical states. Relying on realistic detection, the potential coordination cases can be recorded and rewarded, which can enhance the controller to catch opportunities to create a green wave in further learning. A simulation experiment was conducted to systematically compare the existing coordinated RL methods. According to the results, a promised performance of the proposed method was observed under various traffic conditions.
WaveLearner: A Knowledge-Combined Reinforcement Learning to Understand Coordinated Traffic Signal Control along Urban Arteries
2022-10-08
585754 byte
Conference paper
Electronic Resource
English
Adaptive Traffic Signal Control for Urban Corridor Based on Reinforcement Learning
Springer Verlag | 2024
|Reinforcement learning-based traffic signal control
IEEE | 2024
|