A Flow-Efficient and Legal-by-Construction Real-Time Traffic Signal Control Platform

Abstract: Inefficiencies in traffic flow through an intersection lead to stopping vehicles, unnecessary congestion, and increased accident risk. In this paper, we propose a traffic signal controller platform demonstrating the ability to increase traffic flow for arbitrary intersection topologies. This model uses Model Predictive Control on a Mixed Logical Dynamical system to control the state of independently controlled traffic signals in a single intersection, removing constraints forcing the selection of signals from a set of phases. Further, we use constraints to impose a guarantee on the output of the system to be in the set of permissible actions under constraints including precise yellow timing, minimum inter-lane green transition timing, and selection of signal states with non-conflicting dependencies. We evaluate our model on a simulated 4-way intersection and an intersection in Denmark with true traffic data and a currently implemented timing schedule as a baseline. Our model shows at least 22\% reduction in time loss compared to baseline light schedules, and timing shows that this system can feasibly run online predictions at a frequency faster than 2s/prediction optimizing over a prediction horizon of 25s.