In March, one of Uber’s self-driving vehicles hit and killed a pedestrian in Tempe, Arizona. Like other people with brains, you probably assumed that the accident was due to a software error, and according to The Information–by way of Fortune–you’re correct:
The issue seems to have been one of tuning. A car’s autopilot will regularly see objects in front of the car, but it must constantly decide whether or not the object requires sudden action. After all, you don’t want the car swerving because of something inconsequential lying on or floating over the road.
The Uber executives who spoke to The Information indicated that the car’s autopilot was overly inclined to dismiss objects in its path, leading to Herzberg’s death.
As a reminder, Uber was recently engaged in a long, drawn-out lawsuit with Waymo, which alleged that a former Waymo engineer had stolen the company’s trade secrets regarding autonomous vehicles and given them to his new employer, Uber. (The suit was settled in February.)
The irony is that Waymo has racked up as much or more data on self-driving vehicles as any company in the world. So you’d think that if Uber was using Waymo’s info, its software might’ve been more robust.