Description
In an innovative venture, the research team embarked on a mission to redefine urban traffic flow by introducing an automated way to manage traffic light timings. This project integrates two critical technologies, Deep Q-Networks (DQN) and Auto-encoders, into reinforcement learning, with the goal of making traffic smoother and reducing the all-too-common road congestion in simulated city environments. Deep Q-Networks (DQN) are a form of reinforcement learning algorithms that learns the best actions to take in various situations through trial and error. Auto-encoders, on the other hand, are tools that help simplify complex data, making it easier for the DQN to understand and make decisions. To enhance the accuracy of these decisions, the research team chose average vehicle speed as a crucial indicator of traffic flow and employed HyperOPT, a method for fine-tuning the system’s hyper-parameters. The team put their method to the test in three different traffic scenarios: controlling a single intersection, managing multiple intersections, and overseeing protected left-turn signals. The results were clear and promising. The innovative system significantly improved traffic conditions by either reducing the average wait time atlights or increasing the overall speed of vehicles passing through intersections. This research not only presents a leap forward in traffic management but also offers a glimpse into a future where road congestion could be significantly alleviated. By employing cutting-edge AI and data processing techniques, the project stands as a testament to the potential for smart cities where traffic flow is optimized, making commutes faster and safer for everyone.
Publication Date
6-2025
Publication Type
Report
Topic
Transportation Engineering, Transportation Technology
Digital Object Identifier
10.31979/mti.2025.2322
MTI Project
2322
Mineta Transportation Institute URL
https://transweb.sjsu.edu/research/2322-Urban-Traffic-Cognestion-Deep-Learning-Approach
Keywords
Traffic signal timing, Left turns, Advanced traffic management systems, Optimization, Vehicle to everything communications
Disciplines
Computer-Aided Engineering and Design | Transportation
Recommended Citation
Tairan Liu. "Addressing Urban Traffic Congestion: A Deep Reinforcement Learning-Based Approach" Mineta Transportation Institute (2025). https://doi.org/10.31979/mti.2025.2322
Research Brief
