Reinforcement learning in transportation research: Frontiers and future directions

The development of reinforcement learning (RL) provides innovative solutions for various decision-making problems in transportation, often pertaining to integrating advanced vehicular technologies such as connected and autonomous vehicles and electric vehicles. This paper reviews transportation research with RL-based methods over the recent decades. We start with a bibliometric analysis through extensive literature retrieval of 1030 papers from 1996 to 2023. We identify different research areas of RL in transportation, summarizing the most visited research problems. We find that, at the vehicle level, motion and route planning and energy-efficient driving problems have attracted the most attention. Meanwhile, adaptive traffic signal control and management have been the most visited problems at the network level. We discuss several potential future directions, including the migration of RL models from simulations to real-world cases, designing tailored control architectures for complex transportation systems, exploring explainable RL in transportation research to ensure transparency and accountability in decision-making processes, and integrating people and vehicles into transportation systems in a sustainable and equitable manner.