Multimodal iNtelligent Deep (MiND) Traffic Signal Controller

Abstract

Population growth around the world has led to a challenging level of demand for transportation. Constructing new infrastructure is not always the first option due to spatial, financial, and environmental constrains. Public transit is often considered to be a more affordable and sustainable option, as one transit vehicle can carry significantly higher number of passengers compared to regular vehicles. In urban cores, a considerable portion of travel time is spent waiting at traffic signals. Transit Signal Priority (TSP) methods has emerged over the years to reduce transit delays at traffic signals. Traffic signals are often optimized for regular traffic and TSP systems are added to adjust the background signal timing plans to provide priority for transit vehicles. Therefore, these two modes seem to constantly fight for the green signal, and improving one’s travel time leads to deterioration of the other’s. In this research we introduce a new multimodal traffic signal controller that explicitly considers both regular and transit vehicles and optimizes the throughput of people rather than vehicles, irrespective of what mode they are on. For this purpose, we use deep reinforcement learning to develop and test a Multimodal iNtelligent Deep (MiND) traffic signal controller.