ChatGPT Participates in Traffic Control as a Traffic Manager Assistant

This paper uses ChatGPT to assist traffic manager in traffic signal control and verifies the possibility of applying ChatGPT in traffic control scenarios. We believe ChatGPT has four capabilities, including: knowledge acquisition, data analysis, decision support, and programming interface support. Proper use of these capabilities can assist traffic manager in their traffic strategy optimization efforts. We constructed a simulation environment and used ChatGPT to assist human traffic manager in performing signal optimization efforts. The process of evaluating the application potential of ChatGPT consists of four steps. First, we conducted a test on traffic related issues to test ChatGPT’s knowledge acquisition ability. Second, the road network structure topology analysis test was conducted to help traffic manager quickly understand the road structure. Third, we performed traffic flow characterization and corresponding optimization suggestions to help traffic manager determine optimization policies. Finally, the policy deployment code modification is done by interacting with ChatGPT through natural language, which can improve the efficiency of traffic policy deployment compared with manual programming. The simulation results show that ChatGPT can help traffic managers to complete traffic policy optimization faster. Also, the traffic control optimization under this process improves the average passing speed of vehicles compared with that before the policy change.