| Abstract: | This paper reports on a study that developed a next‐generation Transit Signal Priority (TSP) strategy, Adaptive TSP, that controls adaptively transit operations of high frequency routes using traffic signals, thus automating the operations control task and relieving transit agencies of this burden. The underlying algorithm is based on Reinforcement Learning (RL), an emerging Artificial Intelligence method. The developed RL agent is responsible for determining the best duration of each signal phase such that transit vehicles can recover to the scheduled headway taking into consideration practical phase length constraints. A case study was carried out by employing the microscopic traffic simulation software Paramics to simulate transit and traffic operations at one signalized intersection along the King Streetcar route in downtown Toronto. The results show that the control policy learned by the agent could effectively reduce the transit headway deviation and causes smaller disruption to cross street traffic compared with the existing unconditional transit signal priority algorithm. |
