Intelligent traffic signal controller for heterogeneous traffic using reinforcement learning

Abstract

A traffic signal controller is an essential part of a signalized intersection to alleviate congestion and pollution by ensuring safety. However, the available research solutions are focused on homogeneous traffic scenarios, whereas heterogeneous traffic is the reality in most countries. Hence, a traffic signal control scheme suitable for heterogeneous traffic conditions is proposed in the current study using Reinforcement Learning. A novel reward function with an objective to reduce the traffic residual is defined and a combination of exploration and exploitation optimal policy is applied which made the system learn quickly. The proposed scheme can choose the appropriate phase sequence with optimal signal lengths based on traffic demand on each approaching road. The simulation results proved that the proposed model is well-suited for heterogeneous traffic conditions and its performance against the actuated traffic signal controller is significant in reducing the green time wastage and mean waiting time at the intersection.

Keywords

• Machine learning
• Reinforcement learning
• Multi-agent
• Signalized intersection
• Congestion optimization
• Traffic light control