The Uber ATG Accident in Tempe, Arizona

In March 2018, an Uber self-driving vehicle struck and killed a pedestrian in Tempe, Arizona. The car was operating in autonomous mode with a human backup driver behind the wheel. The backup driver, Rafaela Vasquez, was looking at her phone rather than the road when the collision occurred. The vehicle's safety systems did not detect the pedestrian in time to brake.

This was the first known death caused by a fully autonomous vehicle in the United States. The incident raised when ready questions about how self-driving car companies train backup drivers, what they are required to monitor, and whether the role itself is realistic — a person cannot stay alert for hours while watching a machine that handles the driving.

Uber's self-driving division, Uber Advanced Technologies Group (ATG), was shut down in 2020. The company settled a lawsuit with the victim's family. Vasquez was charged with negligent homicide but the charges were later dropped after prosecutors determined the vehicle's safety systems bore primary responsibility.

Key Takeaways

  • The backup driver was distracted by her phone and did not see the pedestrian, but the vehicle's sensors also failed to detect the collision risk in time.
  • Uber's training and monitoring procedures for backup drivers did not prevent phone use or enforce continuous road attention during autonomous operation.
  • The incident exposed a fundamental problem: humans cannot reliably stay alert while supervising a machine that drives itself for extended periods.
  • This accident became a major case study in how self-driving car companies design safety systems and what role human oversight can realistically play.

What the Backup Driver Was Supposed to Do

Uber's backup drivers were hired to monitor the self-driving system and take control if something went wrong. They sat in the driver's seat while the car operated autonomously. The job description suggested they would watch the road and the vehicle's performance simultaneously.

In practice, this created a problem known as the "vigilance decrement." A person watching a system that works correctly most of the time will gradually stop paying attention. Their brain assumes the system is handling things and shifts focus elsewhere. After minutes or hours of nothing happening, staying alert becomes nearly impossible without active engagement.

Uber's internal documents later revealed that the company knew backup drivers were checking their phones, looking at dashboards, and doing other tasks while the car drove. The company did not have clear policies against these behaviors or consistent enforcement of monitoring requirements.

Why the Vehicle's Safety Systems Did Not Stop the Collision

Uber's self-driving system used cameras, radar, and lidar (light-based distance sensing) to detect objects on the road. In the Tempe accident, the system detected the pedestrian about 5.6 seconds before impact. However, the software classified her as a vehicle rather than a pedestrian for the first 4.1 seconds.

Because the system thought it was tracking a vehicle, it predicted the object would move out of the way. When the classification finally changed to pedestrian, the vehicle was already too close to brake in time. The system also had a feature that prevented sudden braking in certain situations, which may have delayed the emergency stop.

Investigators found that the vehicle could have avoided the collision if the safety systems had worked as intended. The backup driver could also have prevented it by watching the road. Both layers of protection failed.

How Backup Driver Contracts and Training Worked

Uber hired backup drivers as independent contractors, not employees. This classification affected their training, pay, and legal protections. Contractors typically received less formal training than employees and had fewer workplace safety requirements.

The backup drivers Uber hired came from various backgrounds. Some had commercial driving experience; others did not. Training lasted weeks rather than months. Drivers learned how to operate the vehicle's manual controls and how to take over in an emergency, but there was no standardized curriculum for monitoring an autonomous system or staying alert during long periods of inactivity.

Pay was modest — around $16 to $20 per hour in most markets. The job required sitting alert for hours with nothing to do, which made it difficult to retain workers. High turnover meant constant training of new people, which may have contributed to inconsistent safety practices.

What Changed After the Accident

The Tempe collision prompted investigations by the National Transportation Safety Board (NTSB) and the Arizona Department of Public Safety. Both found that Uber's safety culture and backup driver oversight were inadequate. The NTSB made specific recommendations about how self-driving car companies should monitor human operators and design safety systems.

Other self-driving car companies revised their backup driver programs. Some increased monitoring of what drivers were doing — using cameras to watch the driver's eyes and hands. Others added more frequent breaks, rotated drivers more often, or increased pay to reduce turnover. Some companies began questioning whether human backup drivers could ever be reliable enough for this role.

Uber itself decided to exit the self-driving car business in 2020, selling ATG to Aurora Innovation. The company cited the cost of development and the time required to build a safe system. The Tempe accident was not the only reason, but it was a significant factor in the decision.

The Broader Question: Can Humans Reliably Monitor Autonomous Systems?

The Tempe accident raised a question that remains unresolved: Is it realistic to expect a human to stay alert while watching a machine that drives itself? Research in human factors and attention shows that people cannot maintain high vigilance for extended periods, especially when the system they are monitoring rarely fails.

This is not a failure of individual backup drivers. It is a limitation of human attention. A person can stay alert during an active emergency, but not during hours of routine operation. Self-driving car companies have three options: design systems so reliable that human backup is unnecessary, use multiple human monitors with frequent rotation, or accept that backup drivers will sometimes miss problems.

Most companies developing self-driving vehicles have moved toward the first option — building systems that do not rely on human attention to stay safe. This requires more sophisticated sensors, better software, and more extensive testing before vehicles operate on public roads.

Frequently Asked Questions

Was the backup driver charged with a crime?

Rafaela Vasquez was initially charged with negligent homicide, but prosecutors dropped the charges in 2020. They concluded that the vehicle's safety systems bore primary responsibility for the collision, not the driver's distraction. The case was settled civilly with the victim's family.

Did Uber face legal consequences?

Uber settled a lawsuit with the victim's family for an undisclosed amount. The company also faced regulatory scrutiny and criticism from safety advocates. However, Uber was not criminally charged. The company's decision to exit the self-driving car business in 2020 was influenced by the accident but also by the high cost and slow progress of autonomous vehicle development.

Are backup drivers still used in self-driving car testing?

Yes, but with different approaches. Some companies use backup drivers with stricter monitoring — cameras that track the driver's attention, frequent breaks, and multiple people rotating shifts. Others have reduced reliance on human backup by improving their autonomous systems. The role has become more structured and the expectations clearer, but the fundamental challenge of maintaining human attention remains.

What did the NTSB recommend?

The NTSB recommended that self-driving car companies establish clear policies for backup driver monitoring, use technology to verify that drivers are paying attention, and design safety systems that do not depend on human intervention for basic collision avoidance. The board also suggested that companies should not rely solely on human backup as a safety measure.