A Cooperative Merging Method for Mixed Traffic Based on Enhanced Graph Reinforcement Learning with Vehicle...

Area of Science:

• Intelligent Transportation Systems
• Artificial Intelligence
• Robotics

Background:

• Cooperative perception and decision-making are vital for connected and autonomous vehicles (CAVs) in mixed-traffic environments.
• Existing methods inadequately model dynamic interactions among heterogeneous agents, risking congestion and accidents in ramp merging.
• A robust system is needed to manage complex interactions for safe and efficient autonomous driving.

Purpose of the Study:

• To propose an Enhanced Graph Reinforcement Learning algorithm (VCG-EGRL) for cooperative merging decisions in mixed-traffic ramp scenarios.
• To develop a Vehicle Collaboration Intensity (VCI) model for quantifying vehicle interaction dynamics.
• To create a local-global cooperative graph integrating vehicle-to-vehicle and vehicle-to-infrastructure communication.

Main Methods:

• Designed a Vehicle Collaboration Intensity (VCI) model to capture interaction dynamics.
• Constructed a local-global cooperative graph using VCI, V2V, and V2I communication relationships.
• Employed a Graph Convolutional Network with Kolmogorov-Arnold Networks (KANs) for feature extraction (GKAN).
• Optimized the Graph Reinforcement Learning strategy using graph mutual information maximization.

Main Results:

• The VCG-EGRL algorithm demonstrated superior performance in mixed-traffic ramp merging scenarios.
• Experimental results showed significant improvements in merging success rate, efficiency, and robustness compared to baseline models.
• The method effectively models complex vehicle interactions and driving behaviors.

Conclusions:

• The proposed VCG-EGRL algorithm enables effective cooperative merging for CAVs in challenging mixed-traffic environments.
• The integration of VCI, graph neural networks, and reinforcement learning offers a promising approach for intelligent traffic control.
• This research contributes to safer and more efficient autonomous navigation in complex traffic scenarios.