CompletedSeptember 2021 - June 2022
Traffic Signal Control System for Scramble Intersections Using Reinforcement Learning
Developed a system that dynamically controls traffic signals at scramble intersections using deep reinforcement learning. Through simulation-based training, the system optimizes traffic flow and demonstrates reduced waiting times and fewer accidents.
Overview
This research developed a system that optimizes traffic signal control at scramble intersections using deep reinforcement learning. Various traffic situations were simulated, and the agent learned optimal signal switching timings. Compared to conventional methods, the system demonstrated reduced waiting times and accident risks.
Technologies & Methods
Deep reinforcement learningTraffic simulationPythonPyTorch
Results & Achievements
- Reduced waiting time by 76% compared to conventional methods
- Confirmed reduction in traffic accident risk
- Presented at a domestic conference and received a student encouragement award
Related Publications
“Traffic Signal Control at Scramble Intersections by Deep Reinforcement Learning [Japanese]”
Takuya Ogami, Shunpei Norihama, Tetsuo Akimoto, Yoshimasa Tsuruoka.
Presented · JSAI 2022 (June 2022) Student Encouragement Award