Reinforcement Learning Approach to Adaptive Traffic Signal Control Using SUMO-RL

Traffic control in urban environments is a complex and critical challenge, with growing urbanisation and increasing vehicular congestion necessitating innovative solutions. Reinforcement learning has emerged as a promising approach to optimise traffic control and improve overall transportation efficiency, thanks to its ability to grasp the intricacies of complex issues without human involvement autonomously. In this paper, we propose a modified version of the SUMO-RL package and demonstrate its advantages in the context of our ongoing research regarding traffic control in the city of Zilina. The proposed modifications include support for custom reward functions for different agents and custom observation functions, representing a crucial design component when solving traffic signal control as a multi-agent reinforcement learning problem. We also present our preliminary results, which demonstrate the potential of reinforcement learning in addressing traffic control challenges.