Driving Risk Assessment for Intelligent Vehicles Based on Entropy-Informed Graph Neural Networks and Gaussian...

Area of Science:

• Intelligent Transportation Systems
• Autonomous Driving Safety
• Machine Learning for Risk Assessment

Background:

• Current autonomous driving risk assessment lacks comprehensive spatiotemporal modeling of vehicle interactions.
• Quantifying uncertainty in dynamic risk assessments remains a challenge for intelligent vehicles.

Purpose of the Study:

• To develop a novel framework for assessing intelligent vehicle driving risk in typical traffic scenarios.
• To address limitations in existing methods by incorporating spatiotemporal dynamics and uncertainty quantification.

Main Methods:

• Probabilistic modeling of vehicle speed and acceleration using Gaussian Distribution (GD).
• Application of entropy theory to quantify risk uncertainty.
• Development of a risk assessment model using Graph Neural Networks (GNNs) to capture multivehicle interactions.

Main Results:

• The proposed framework accurately quantifies collision risks in complex, multivehicle traffic scenarios.
• High accuracy and robustness were demonstrated across various driving situations including cruising, cut-ins, lane changes, and overtaking.
• The model effectively handles traffic with varying densities.

Conclusions:

• The entropy-informed graph neural network (EIGNN) framework provides accurate and robust driving risk assessment for intelligent vehicles.
• This approach offers significant technical insights and theoretical support for improving autonomous driving decision-making and safety.
• Integrating traffic risk analysis enhances the efficiency of autonomous driving systems.