Cooperative multi-agent system for coordinated traffic signal control

Abstract

A novel, distributed, cooperative multi-agent approach employing multiple, interacting, autonomous hybrid agents to provide effective signal control for real-time traffic management is presented here. This multi-agent system uses advanced cooperative behaviours to improve individual agents' learning process and adaptability. Each agent cooperates with other agents within its cooperative zone, the size of which changes dynamically according to the changing needs of the agent. The performance of the proposed cooperative ensemble multi-agent system is tested on a large, complex traffic network and compared against two other approaches. For the 6 h extreme scenario with two peaks, the proposed approach reduces the total mean delay by 35.6 when compared to the GLIDE benchmark, while for the 24 h extreme scenario with multiple peaks, the reduction is 75. The results demonstrate the efficacy of the cooperative multi-agent approach in dealing with the approximated version of an infinite horizon dynamic problem.