Related Experiment Video
Updated: Aug 5, 2026
11:41
Evaluation of an Exclusive Spur Dike U-Turn Design with Radar-Collected Data and Simulation
Published on: February 1, 2020
Dynamic Traffic Signal Control for Isolated Intersections: Enhanced SARSA Reinforcement Learning with Expectation
Yuhong Gao1,2, Wenyuan Sun1, Meiying Jian1,2
1Department of Transportation Engineering, Transportation Institute, Inner Mongolia University, Hohhot 010070, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces a novel two-branch framework for intelligent traffic signal control using reinforcement learning (RL). The enhanced approach improves efficiency and robustness in adaptive signal timing for intersections.
Area of Science:
- Intelligent Transportation Systems
- Artificial Intelligence in Traffic Management
- Reinforcement Learning Applications
Background:
- Traditional traffic signal control methods face limitations.
- Existing reinforcement learning (RL) algorithms for traffic signals have drawbacks like bias, instability, and slow convergence.
- Intelligent signal timing control using RL shows significant potential.
Purpose of the Study:
- To develop an improved reinforcement learning framework for adaptive traffic signal control at isolated intersections.
- To address the limitations of existing RL algorithms, including estimation bias, robustness, training time, and convergence speed.
- To enhance the efficiency and reliability of real-time adaptive signal control.
Main Methods:
- Developed a two-branch cooperative signal control framework integrating Expected SARSA and SARSA(λ).
- Constructed a traffic-adaptive Markov Decision Process (MDP) dynamic decision model using real-time traffic flow data.
- Utilized Expected SARSA to reduce policy variance and Q-value overestimation.
- Employed SARSA(λ) with eligibility traces for efficient multi-step temporal difference error backpropagation.
Main Results:
- The proposed framework demonstrated superior performance compared to baseline algorithms in quantitative evaluations.
- Experiments confirmed the generality and robustness of the developed framework across various scenarios and parameters.
- The integrated Expected SARSA and SARSA(λ) algorithms showed significant improvements in sample utilization efficiency and model robustness.
Conclusions:
- The developed two-branch cooperative framework offers an efficient and reliable solution for real-time adaptive signal control of isolated intersections.
- The integration of Expected SARSA and SARSA(λ) effectively overcomes the limitations of traditional and complex RL algorithms.
- This research contributes to the advancement of intelligent transportation systems through improved traffic signal management.
