An End-to-End Traffic Signal Control Approach Driven by Fixed-Location Detector Occupancy Patterns Under Demand Uncertainty

This study proposes an end-to-end traffic signal control approach that seamlessly integrates dynamic traffic state information and real-time signal control decisions into a unified framework. This framework establishes a mapping function between traffic detector patterns and optimal adaptive control policies. The adjustments in the adaptive signal plan are cycle-based and driven by real-time traffic detector data represented by binary strings. We rely on traffic detection data from first principles rather than on estimations of traffic arrivals or residual queues as external inputs. The adaptive signal control problem is formulated into a two-stage stochastic program, explicitly addressing control variables including cycle time, base green time, offsets, and adaptive policies. A gradient descent algorithm is developed, aiming to derive optimal adaptive policies tailored to detector patterns. To validate the framework, we initially test it on a simple junction to illustrate the mapping properties and the selection of critical detector patterns. Subsequently, we extend the study to a corridor of three connected intersections to demonstrate versatility of the framework, encompassing intricate traffic configurations such as mixed-lane and pocket lane setups. This investigation contributes new insights into urban traffic control and management through a novel adaptive signal control modelling approach.