Deep Reinforcement Learning for Traffic Signal Control

The increasing demand for mobility in cities and beyond presents challenges for traffic engineering. As a solution to these challenges, we can use the power of artificial intelligence and reinforcement learning. These scientific branches can enable us to use the current infrastructure more efficiently, thus reducing the environmental impact and increasing the comfort of drivers. In this paper, using the Ingolstadt and Cologne benchmarks from RESCO, we compare our agent, which is based on Perceiver with IDQN (Independent Deep Q-Network), to the conventional method of Max-pressure and random times for green signals.