Development of Reinforcement Learning-Based Traffic Predictive Route Guidance Algorithm Under Uncertain Traffic Environment

There have been enormous efforts to develop a novel vehicle routing algorithm to reduce origin-to-destination (OD) travel time. Most of the previous studies have mainly focused on providing the shortest travel time route based on estimated traffic information. Few researches have considered the use of predictive information on traffic dynamics to improve the quality of route guidance algorithms. However, there is still uncertainty associated with future traffic conditions, particularly in non-recurrent traffic congestion caused by the abnormal event. For a reliable navigation service under uncertain traffic conditions, this research develops a reinforcement learning-based traffic predictive vehicle routing (RL-TPVR) algorithm. The proposed algorithm is designed to mitigate the variability of OD travel time by incorporating predictive state representation and prediction reward modeling in the reinforcement learning scheme. The RL-TPVR is evaluated in terms of OD travel time based on various traffic scenarios with different demand patterns. Several numerical studies including a performance gap analysis, case study, and comparative study are conducted using microscopic simulation experiments. The performance gap analysis demonstrates the superiority of the RL-TPVR with respect to traffic uncertainty, particularly in non-recurrent traffic congestion cases. In addition, the case study shows that the RL-TPVR exhibits a flexible and dynamic OD travel route depending on the given traffic situations. Furthermore, the comparative study verifies that the proposed algorithm outperforms other existing algorithms in both recurrent and non-recurrent traffic congestion cases. These findings suggest the RL-TPVR has great potential for providing the shortest travel time route under uncertain traffic conditions.