Uber driverless software ‘ignored’ pedestrian in fatal crash
Flaw reportedly meant that on-board systems detected woman but decided not to react
Uber believes a software fault may have been responsible for one its driverless cars fatally wounding a pedestrian in March, according to company insiders.
However, the insiders said that as a result of how the system had been “tuned”, the software decided it didn’t need to react when Herzberg walked into the road, in Tempe, Arizona, the site reports.
According to Engadget, programmers of driverless software tune the sensitivity of the system’s sensors so that the vehicle “only reacts to true threats”. These include “solid objects” such as a brick wall, and in most cases, human beings.
However, the way the system had been tuned in Uber’s driverless Volvo XC90 reportedly meant that the car identified Herzberg as a “false positive” and therefore ignored her.
Although the vehicle had a “safety driver” behind the wheel to intervene in the event of an emergency, the driver failed to respond quickly enough, according to TechCrunch.
Prior to The Information’s report, Tempe police chief Sylvia Moir told the San Francisco Chronicle that video footage from on board the vehicle suggested that the accident, on 18 March, “would have been difficult to avoid”.
But that claim was refuted by Bryant Walker Smith, a University of South Carolina law professor who studies autonomous vehicles. Smith told the Associated Press that the car’s on-board radar systems “should have detected and classified” Herzberg.
Uber has yet to comment on the latest report, but said in a statement that it has “initiated a top-to-bottom safety review”.
The company added: “Our review is looking at everything from the safety of our system to our training processes for vehicle operators, and we hope to have more to say soon.”