Agent-Based Traffic Signal Control Using a Reinforcement Learning Approach

The constant growth in road traffic, especially in large cities, leads to a significant increase in the cost (travel time, fuel consumption) spent by users to complete their trips, as well as to an increase in harmful emissions into the atmosphere and a deterioration in the environmental situation. One of the possible solutions to improve this situation is the optimization of the existing transport infrastructure in order to increase the efficiency of its use. Among many new techniques developed for traffic control recently, traffic signal control remains one of the most popular problems. In this paper, we evaluate an agent-based traffic control model, in which each intersection is operated by one agent. We use a reinforcement learning approach to estimate the traffic signal control policy taking into account the traffic signal phase and the observed information about vehicle movements. Experimental study of the developed algorithm is carried out in a multi-agent environment for large-scale city traffic scenarios CityFlow. Experimental results confirm that the proposed model is superior to classical approaches.