Published Paper Details:

Conventional traffic optimization solutions expect that the graph structure of street systems is static and missing for further traffic flow optimization. We are interested in optimizing traffic flows as a latest type of graphbased issue, where the graph structure of a street network can adjust to traffic conditions in real time. Specifically, we focus on the dynamic setup of traffic-lane directions, which can offer assistance in adjusting the utilization of traffic paths in opposite directions. The rise of connected autonomous vehicles offers an opportunity to apply this kind of dynamic traffic optimization on a huge scale. The existing procedures for optimizing lane directions are however not appropriate for dynamic traffic situations due to their high computational complexity and the static nature. In this paper, we propose an effective traffic optimization solution, called Coordinated Learning-based Lane Allocation (CLLA), which is appropriate for dynamic setup of path directions. CLLA comprises of a two-layer multi-agent architecture, where the bottom-layer agents use a machine learning technique to discover a appropriate arrangement of lane-directions around street crossing points. The lane-direction changes proposed by the learning agents are at that point coordinated at the next level to diminish the negative impact of the changes on other parts of the street network. Our test results show that CLLA can decrease the average travel time significantly in congested street networks. We accept our method is common enough to be applied on other types of networks as well.