A new report from the National Transportation Safety Board (NTSB) in the United States has found that an Uber self-driving test vehicle, which hit a woman in Arizona in March 2018, had software flaws which meant that it was unable to identify her as a pedestrian crossing the street. The report comes ahead of a meeting planned for November 19 which will determine the probable cause of Elaine Herzberg’s death as she was wheeling a bicycle across the street at night.
From September 2016 to March 2018, Uber’s self-driving vehicles had been involved in 37 crashes, with 33 of those involving another vehicle. After the incident involving Herzberg, Uber suspended all its self-driving car operations until December. During the break it revised the software on-board, adding new restrictions and safeguards.
According to the NTSB, the vehicle didn’t recognise the bicycle as an imminent collision until 1.2 seconds before impact so there was not enough time for the crash to be averted. The NTSB said that Uber’s software did not take jaywalking pedestrians into account. With the new software, Uber’s simulations show that the car would have noticed Herzberg 4.5 seconds before impact with the car having 4 seconds to engage the brakes.
While Arizona prosecutors did not hold Uber criminally liable, the police did investigate what the safety driver was doing when the crash happened. The police consequently said that the crash was “entirely avoidable” and that the driver was watching a TV show leading up to the crash.