Dynamic risk assessment for autonomous vehicles from spatio-temporal probabilistic occupancy heatmaps

Area of Science:

• Robotics and Artificial Intelligence
• Transportation Engineering
• Computer Science

Background:

• Accurate collision risk assessment is vital for autonomous vehicle (AV) trajectory planning and safety evaluation.
• Current methods struggle with the uncertainties inherent in dynamic traffic environments.

Purpose of the Study:

• To introduce a novel Probabilistic Occupancy Risk Assessment (PORA) metric for enhanced collision risk evaluation.
• To improve the safety and real-time decision-making capabilities of autonomous driving systems.

Main Methods:

• Utilizes spatiotemporal heatmaps for probabilistic occupancy predictions of surrounding traffic participants.
• Estimates collision risk along planned AV trajectories considering potential vehicle interactions.
• Incorporates a Cox model to adjust risk based on relative motion between AVs and other vehicles.

Main Results:

• The PORA metric enhances the accuracy of collision risk assessment in dynamic traffic scenarios.
• Demonstrates improved safety for vehicle controllers and robust real-time decision-making.
• Outperforms other safety surrogate measures in characterizing collision risk through Monte Carlo simulations.

Conclusions:

• PORA provides a more accurate and robust framework for assessing collision risk in autonomous driving.
• The metric effectively accounts for uncertainties in traffic participant behavior.
• Enables safer and more reliable autonomous vehicle operation.