Uber's self-driving car killed a pedestrian — she was classified as 'other'

On March 18, 2018, an Uber autonomous test vehicle struck and killed Elaine Herzberg, a 49-year-old pedestrian, in Tempe, Arizona. It was the first fatality involving a self-driving car in autonomous mode.

The investigation found that the vehicle's software detected Herzberg 5.6 seconds before impact but classified her as a 'false positive' — an object to be ignored. The software's classification system cycled through 'unknown,' 'vehicle' and 'bicycle' labels but never settled on a category that would trigger braking. The safety driver, who was supposed to monitor the road, was watching a television show on her phone.

Uber suspended its self-driving program and settled with Herzberg's family. The NTSB found that Uber had disabled the Volvo's built-in emergency braking system to avoid 'false positives' and had reduced the safety driver staffing from two to one. The case illustrated how the pressure to demonstrate progress in autonomous driving can lead to safety compromises, and how a system that classifies a human as a 'false positive' has a fundamental design problem.