Traffic Light Control Using Reinforcement and Deep Learning: A Survey and Comparison

In recent years, the intelligent control of traffic lights has become a necessity for the efficient operation of a transportation system. Unlike traditional traffic lights with fixed timings, a smarter system would adjust light cycles in real time based on current traffic conditions. Reinforcement learning is emerging as a promising approach to achieve this dynamic traffic management. Nevertheless, none of the current research has yet tested real data. In this article, we first present the main learning techniques and approaches (RL, DL, DRL). Next, we will examine each of these techniques learning from their advantages, weaknesses, or shortcomings, to assess their effectiveness in urban traffic management. Additionally, this document will explore some of the simulation tools used for adaptive traffic control.