Autonomous Vehicle Safety Lags Human Drivers by 1,000x, Study Finds

A newly released behavioral safety assessment reveals that tested autonomous vehicles experienced a crash rate approximately 1,000 times higher than that of average human drivers. The findings, detailed in a report released October 7, 2026, highlight critical safety gaps in current autonomous driving systems and raise significant questions about the timeline for large-scale deployment.

The assessment, conducted using Autoware.Universe – an open-source Level 4 Automated Driving System – involved both simulated and physical testing at the University of Michigan’s Mcity Testing Facility. Researchers evaluated the system against 14 defined behavioral competencies, finding that Autoware.Universe demonstrated proficiency in only six. More concerningly, the autonomous system registered a crash rate of 3.01x10^-3 crashes per mile during testing.

The study builds upon a previously proposed third-party testing initiative designed to systematically evaluate autonomous vehicle behavioral safety. This second part of the study validates the proposed framework through practical application, identifying numerous previously unknown unsafe scenarios within the tested system. These discoveries underscore the necessity of comprehensive behavioral safety evaluations prior to public release.

The results of this assessment arrive as the automotive industry prepares for potential rollout of fully autonomous vehicles by 2028. However, the substantial disparity in crash rates between autonomous systems and human drivers suggests that significant improvements in safety performance are required before public trust and regulatory approval can be secured. The study's focus on behavioral competencies represents a shift towards a more holistic evaluation of AV safety beyond purely technical metrics.