Markov Decision Process for Traffic Light Control: A Comparative Approach

Abstract

Traffic congestion has steadily risen in the US, yet most of the traffic lights at intersections use old and outdated technology. In this paper, we go over the classical traffic light control (TLC) methods (pre-timed TLC and actuated TLC) and then introduce cutting edge adaptive TLC methods. In particular, we will zero in on three Markov Decision Process based TLC methods, which have implementations unique to this paper. These five mentioned TLC methods are accompanied with descriptions of simulations built in SUMO, which produce results that we are able to compare and analyze. Our simulation results suggest that the explored MDP TLC methods can cause significant time savings and reduction in stoppage count, especially in light traffic scenarios and in dense downtown districts. Our simulation results are also able to confirm some claims about how to best use the classical non-adaptive TLC methods.