The self-driving Uber test car that hit and killed a woman in Arizona in March did spot the victim, but failed to recognise her as human according to an internal investigation by the taxi company.
Tech website The Information has spoken to insiders at the company with knowledge of the report – the indications are that the software was desensitised to reduce false positives in order to improve the riding experience for passengers.
Early testing showed that the cars were liable to brake unnecessarily for items such as plastic bags, leading to a ‘detuning’ that seems to have gone too far. Uber is working with the National Transport Safety Board, a federal body, as well as local police as the investigation continues.
The Volvo XC90 test car hit 49-year-old Elaine Herzberg on 18 March as she crossed a road in Tempe, Arizona. Local police released onboard video from the car shortly after the crash, which seemed to show the safety operator not paying attention to the road.
Uber CEO Dara Khosrowshahi recently gave an interview to US outlet CBS News and hinted that the company had been conducting its own investigation into the incident: ‘We want the NTSB to move forward with their investigation. It is their job to determine who or what was at fault. We don't want to get in their way.'
'Ultimately, our job is to allow the authorities to do their job. And what I'm doing is a top-to-bottom audit of our procedures, training, software, hardware, what our practices are so that I can be comfortable and our board of directors can be comfortable that when we get back on the road, we get back in a responsible and safe manner.’
The company has suspended all of its self-driving test programmes since the incident, as it tries to find out exactly what went wrong and what can be done to stop it happening again.