On Uncertainty Quantification Impact for Autonomous Driving Decisions at Intersections

Autonomous driving technology represents a pivotal direction in the future of intelligent transportation. Despite significant advancements in recent years, this technology faces numerous challenges, particularly in making robust and trustworthy intelligent decisions amidst highly uncertain environments. This paper addresses the issue of uncertainty quantification in autonomous driving decision-making, focusing on the roles of aleatoric uncertainty and epistemic uncertainty. It analyzes the impact of different uncertainty thresholds on the accuracy, safety, and reliability of driving decisions, ultimately determining significant threshold ranges of uncertainty affecting decision performance under varying conditions. Experimental results underscore the critical importance of considering uncertainty in autonomous driving systems. The findings provide theoretical insights and parameter guidance for developing more robust and trustworthy autonomous driving systems.