ReLight-WCTM: Multi-Agent Reinforcement Learning Approach for Traffic Light Control within a Realistic Traffic Simulation

Although traffic management methods are constantly evolving, traditional solutions are unable to adapt to the current dynamics of traffic. Machine learning methods provide promising results, but scientific dissertations usually work only on the theoretical basis of the models and do not take into account the legal requirements of traffic management. In the presented paper we propose a deep reinforcement learning-based multi-agent model called ReLight-WCTM that insists on maintaining reality at several points. We compared our model with the original signal setting of a real road network based on different metrics. According to the results, it can be stated that ReLight-WCTM exceeded the baseline settings in all parameters, presumably it can be an actual traffic management alternative.