Real-Time Safety Optimization of Connected Vehicle Trajectories Using Reinforcement Learning

Area of Science:

• Intelligent Transportation Systems
• Traffic Engineering
• Artificial Intelligence in Transportation

Background:

• Traditional highway speed advisories aim to mitigate traffic conflicts and delays.
• Extending dynamic speed advisories to signalized intersections is crucial for connected vehicle environments.
• Real-time traffic parameters are essential for effective intersection speed management.

Purpose of the Study:

• To develop and evaluate a dynamic speed advisory system for connected vehicles at intersections.
• To enhance intersection safety and efficiency by guiding vehicles to optimum speeds.
• To reduce traffic conflicts and delays through real-time speed recommendations.

Main Methods:

• Utilized real-time safety evaluation models incorporating traffic volume and shock wave characteristics.
• Developed a hybrid algorithm combining a rule-based approach with Deep Deterministic Policy Gradient (DDPG) reinforcement learning.
• Tested the algorithm on two distinct intersections using real-world traffic data.

Main Results:

• Achieved an average reduction in traffic conflicts between 9% and 23% across tested intersections.
• Demonstrated effectiveness even at lower market penetration rates (MPR) of connected vehicles.
• Validated the algorithm's performance under varying traffic volumes and intersection physical characteristics.

Conclusions:

• The proposed dynamic speed advisory system effectively reduces rear-end traffic conflicts at signalized intersections.
• The algorithm offers a low-cost, computationally efficient solution for improving intersection safety.
• Real-time speed advisories are a viable strategy for enhancing connected vehicle traffic flow and safety.