← Back to incidents
Uber Self-Driving Car Kills Pedestrian in Tempe, Arizona
CriticalAn Uber self-driving test vehicle struck and killed 49-year-old pedestrian Elaine Herzberg in Tempe, Arizona, marking the first documented fatality involving a fully autonomous vehicle. The car's AI perception system detected Herzberg but repeatedly misclassified her and failed to initiate emergency braking.
Category
Safety Failure
Industry
Technology
Status
Resolved
Date Occurred
Mar 18, 2018
Date Reported
Mar 19, 2018
Jurisdiction
US
AI Provider
Other/Unknown
Model
Custom Perception/Planning System
Application Type
embedded
Harm Type
physical
People Affected
1
Human Review in Place
No
Litigation Filed
Yes
Litigation Status
settled
Regulatory Body
Yes
autonomous_vehiclepedestrian_fatalityperception_failure
Full Description
On March 18, 2018, at approximately 9:58 PM, an Uber Advanced Technologies Group (ATG) autonomous test vehicle struck and killed 49-year-old Elaine Herzberg as she crossed Mill Avenue in Tempe, Arizona, while walking her bicycle. The modified Volvo XC90 was operating in autonomous mode with safety driver Rafaela Vasquez behind the wheel when it struck Herzberg in a poorly lit area outside of a designated crosswalk. This incident marked the first recorded fatality involving a fully autonomous vehicle and immediately drew national attention to the safety of self-driving car technology. Herzberg died from her injuries at a local hospital shortly after the collision.
The National Transportation Safety Board (NTSB) investigation revealed critical failures in Uber's perception and decision-making systems. The vehicle's lidar and radar sensors detected Herzberg approximately 5.6 seconds before impact, but the AI system repeatedly misclassified her, cycling between categorizations as an "unknown object," "vehicle," and "bicycle" without ever properly identifying her as a pedestrian crossing the roadway. The system's path prediction algorithms failed to anticipate her trajectory across the street, and crucially, Uber had disabled the Volvo's factory-installed automatic emergency braking system to prevent what the company described as erratic vehicle behavior that could cause uncomfortable rides for passengers. The vehicle was traveling at 38 mph in a 35 mph zone and did not slow down before impact.
The incident resulted in immediate and severe consequences for Uber's autonomous vehicle program. Arizona Governor Doug Ducey suspended Uber's testing permits in the state indefinitely, effectively ending the company's operations there after previously maintaining one of the most permissive regulatory environments for autonomous vehicle testing. Uber faced intense regulatory scrutiny from federal agencies including the NTSB and National Highway Traffic Safety Administration (NHTSA), along with widespread public criticism and calls for stronger oversight of autonomous vehicle testing. The company's stock valuation and public reputation suffered significant damage, with several other cities and states reconsidering their autonomous vehicle policies.
Uber responded by immediately suspending all autonomous vehicle testing nationwide and conducting an internal safety review. The company ultimately settled a wrongful death lawsuit with Herzberg's family for an undisclosed amount and reached a separate settlement with Maricopa County prosecutors to avoid criminal charges, though the company pleaded guilty to a lesser charge. Safety driver Rafaela Vasquez was charged with negligent homicide after investigators determined she had been watching "The Voice" on her mobile phone for 42 minutes before the crash, looking up from the screen only 0.5 seconds before impact.
The Tempe incident fundamentally altered the autonomous vehicle industry's approach to safety testing and regulatory compliance. Multiple states implemented stricter safety requirements for autonomous vehicle testing, including mandatory safety driver attention monitoring systems and enhanced reporting requirements for incidents. The incident exposed significant gaps in how AI perception systems handle edge cases and ambiguous objects, leading to industry-wide improvements in sensor fusion algorithms and safety protocols. Federal legislators introduced multiple bills to establish comprehensive autonomous vehicle safety standards, though comprehensive federal regulation remains pending.
The case also highlighted the complex liability issues surrounding autonomous vehicles, establishing important legal precedents for how responsibility is allocated between technology companies, safety drivers, and AI systems. Uber eventually sold its autonomous vehicle division to Aurora Innovation in 2020, marking the end of its independent self-driving car ambitions. The incident continues to be cited in ongoing debates about the readiness of autonomous vehicle technology for public deployment and the adequacy of current safety testing protocols across the industry.
Root Cause
Uber self-driving test vehicle struck and killed pedestrian Elaine Herzberg in Tempe, Arizona. The vehicle perception system detected Herzberg 6 seconds before impact but repeatedly misclassified her, cycling between unknown object, vehicle, and bicycle. The system could not predict her path. Emergency braking had been disabled by Uber to provide smoother rides. The human safety driver was watching a video on her phone.
Mitigation Analysis
Comprehensive provenance logging of the perception system decisions, classification confidence levels, and planning module outputs would have created an audit trail enabling post-incident reconstruction and pre-incident pattern detection. If systematic misclassification patterns had been logged and analyzed, the dangerous failure mode might have been identified before a fatality.
Lessons Learned
Safety-critical AI systems require redundant safety mechanisms that cannot be overridden for comfort or convenience. Human safety operators must be actively monitored. Perception system confidence logging and threshold-based interventions are essential.
Sources
Collision Between Vehicle Controlled by Developmental Automated Driving System and Pedestrian
NTSB · Nov 19, 2019 · regulatory action