Intelligent agents in decentralized traffic control

Abstract

A multi-agent decentralized strategy to control an urban traffic network is presented. Each agent is in charge of managing the signals of an intersection. Local feedback information about an intersection state is obtained using lane sensors to implement an adaptive strategy for each agent. Cooperating behavior between agents is achieved by exchanging information, called "opinions", between adjacent agents to influence each other decisions. Results for this architecture are shown on a simulated environment modeling an area of the city of Pittsburgh, PA, and comparisons are made with other methods.