Traffic light optimization is known to be a cost-effective method for reducing congestion and energy consumption in urban areas without changing physical road infrastructure. However, due to the high installation and maintenance costs of vehicle detectors, most intersections are controlled by fixed-time traffic signals that are not regularly optimized. To alleviate traffic congestion at intersections, we present a large-scale traffic signal re-timing system that uses a small percentage of vehicle trajectories as the only input without reliance on any detectors. We develop the probabilistic time-space diagram, which establishes the connection between a stochastic point-queue model and vehicle trajectories under the proposed Newellian coordinates. This model enables us to reconstruct the recurrent spatial-temporal traffic state by aggregating sufficient historical data. Optimization algorithms are then developed to update traffic signal parameters for intersections with optimality gaps. A real-world citywide test of the system was conducted in Birmingham, Michigan, and demonstrated that it decreased the delay and number of stops at signalized intersections by up to 20% and 30%, respectively. This system provides a scalable, sustainable, and efficient solution to traffic light optimization and can potentially be applied to every fixed-time signalized intersection in the world.

With the rapid development and deployment of vehicle sensing and communication technology, vehicle trajectory data is becoming increasingly available for urban traffic network applications. However, it is difficult to use raw trajectory data points generated from global navigation satellite system (GNSS) coordinates without matching them to traffic networks. Real-world trajectory data is also prone to noise and errors. This paper proposes a trajectory data processing pipeline to serve different urban traffic network applications. The steps of the pipeline include matching the trajectory points to a well defined network representation, splitting them into different movements, and extracting distance information from their GNSS coordinates. Smoothing and filtering algorithms also reduce the influence of noise and errors. Based on the processed trajectory data, this paper also proposes algorithms for calculating different mobility performance indices including vehicle delay, number of stops, space-mean speed, and coordination measurements. These performance indices provide comprehensive evaluations of urban traffic network from different perspectives. Our case study uses real-world trajectory data collected from the Ann Arbor Connected Vehicle Test Environment. Different mobility performance indices are calculated and visualized. The proposed methods and algorithms are efficient, robust, and scalable, and could be applied to large-scale urban traffic networks.

Previous studies have shown that the max pressure control is a throughput-optimal policy that can stabilize the store-and-forward traffic network when the demand is within the network capacity. Most of the existing studies on the max pressure control do not consider the loss of capacity associated with phase switching, which will undermine the stability of the network. This work proposes a novel framework that utilizes reinforcement learning algorithms to optimize a max pressure controller considering the phase switching loss. We extend the max pressure control by introducing a switching curve and prove that the proposed control method is throughput-optimal in a store-and-forward network. Then the theoretical control policy is extended by using a distributed approximation and position-weighted pressure so that the policy-gradient reinforcement learning algorithms can be utilized to optimize the parameters in the policy network including the switching curve and the weight curve. Simulation results show that the proposed control method greatly outperforms the conventional max pressure control. The proposed framework combines the strengths of the data-driven method and the theoretical control model by utilizing reinforcement learning algorithm to optimize the max pressure controller, which is also of great significance for real-world implementations because the proposed control policy can be generated in a distributed fashion based on local observations.