Occlusion-aware Risk Assessment and Driving Strategy for Autonomous Vehicles Using Simplified Reachability Quantification

🧾 Introduction

• One of the key challenges in autonomous driving is safely navigating areas with occluded pedestrians and vehicles.
• Prior methods used phantom vehicle generation to estimate risk but were often overly conservative or ineffective in real time under heavy occlusion.
• To address this, we propose an efficient occlusion-aware risk assessment framework and a risk-adaptive speed control strategy.

⚙️ Method

• The proposed method models phantom agents in occluded regions using a simplified probabilistic reachability distribution.
• Based on the quantified risk of phantom agents, the system dynamically sets speed limits to enable safe yet efficient navigation.
• The approach maintains constant-time complexity and is computationally efficient, requiring less than 5 ms per decision.

✅ Result

• Simulation results show the proposed method increased intersection traversal time by 1.48×, but reduced average collision rate and discomfort score by 6.14× and 5.03×, respectively.
• The method achieves state-of-the-art time efficiency while significantly improving safety and ride comfort in occluded scenarios.