A Cooperative Multiagent System for Traffic Signal Control Using Game Theory and Reinforcement Learning

Agent-based systems are widely used in urban traffic network modeling. Cooperation is an important feature of multiagent systems (MASs), especially those that are used in a dynamic complex network. Game theory can be employed in interactive decision making for MASs. The focus of this article is on developing traffic signal controllers for multiple intersections using MASs. A two-mode agent architecture is proposed to effectively control the traffic congestion problem through independent and cooperative procedures. Each intersection is controlled by an agent using Q-learning in the independent mode. In the cooperative mode, game theory is employed to determine how cooperation between the agents can dynamically control traffic signals at multiple intersections. Experimental results indicate that the proposed method can effectively reduce the average delay time in different traffic demand scenarios. In addition, cooperation between the agents prevents the network from becoming saturated; therefore, collaboration is essential to traffic signal controllers, especially in areas that have congested traffic.