Transit Signal Priority for Arterial Road with Deep Reinforcement Learning

Transit signal priority (TSP) is an effective measure to reduce the delay of public transit and improve transit service reliability by prioritizing buses to move through signalized intersections. This paper develops the multi-intersection TSP strategy at the arterial road based on multi-agent deep reinforcement learning. Agents would consider the current states and choose the traffic signal's best actions to reach the maximum expected rewards. We record the information of buses from conflicting directions in the state to make agents consider multiple priority requests and use invalid actions masking method to consider constraints of the traffic signal. Micro-simulation results of an arterial road by SUMO show that the proposed strategy significantly reduces the person delay of buses compared with fixed time signals. The proposed TSP strategy easily handles conflicting requests and incorporates traffic signal constraints into RL methods for the arterial road with multiple signalized intersections.