Multi-agent cooperative control for traffic signal on geographic road network

Urban traffic efficiency is one of the key factors affecting urban productivity and is also a crucial topic in the process of smart city construction. With the development of computer technology, artificial intelligence, especially reinforcement learning, plays an increasingly important role in traffic signal control. Currently, traffic signal control based on reinforcement learning is mainly used for the optimization for simple scenarios, such as single road intersection or urban arterial road, not yet for regional coordinated control on an urban geographic road network. This paper is motivated to fill this gap by proposing a two-layered agent cooperative control approach based on reinforcement learning. The first layer implements a coarse-tuning training at a single intersection, where the agents make the single intersection non-blocking by observing the queue length for each lane; In the second layer, the coarse-tuning-trained agent models are put into the geographic network to execute the cooperative fine-tuning training at multiple intersections. This paper conducts the optimization-orientated traffic coordination through a case study of a middle school area in Ningbo. The results show that our control approach is superior to the traditional fixed timing scheme in terms of the passage efficiency.