Smart Traffic Lights System with Multi-agent Reinforcement Learning for Road Traffic Management

The significant environmental impact generated by transportation stimulates a constant search for solutions aimed at mitigating this issue. Numerous strategies are proposed in the literature, each with its own methodology. In this study, reference is made to Reinforcement Learning, an area of Machine Learning, which offers the possibility to explore solutions to address the problem. In particular, the Multi-Agent Deep Deterministic Policy Gradient algorithm was used, designed to facilitate the cooperative or competitive training of multiple agents in complex environments. This is the case with roads, where each agent can be associated with a traffic light, thus allowing coordination of these and the creation of an intelligent traffic light system. Some different scenarios were tested in a simulated environment, using the SUMO traffic simulator. The results show that the intelligent scheduling of traffic lights compared favourably with fixed scheduling, both in terms of reducing pollution and improving the flow of vehicles on the road.