- Uber’s self-driving-car software has been routinely fooled by the shadows of tree branches, which it would sometimes mistake for real objects, insiders say.
- The revelation comes as part of a Business Insider investigation into the death of an Arizona pedestrian who was hit by an autonomous Uber vehicle in March.
The human brain is an extremely complex pattern-recognition machine.
Among the hundreds of millions of other daily tasks, there’s one that’s extremely important for driving a car: distinguishing a shadow from a physical object.
For weeks during regular “triage” meetings of Uber’s Advanced Technologies Group, the division that handles self-driving cars, where engineers and a vice president decided what issues to prioritize in their workflow, tree branches and their associated shadows were routinely mentioned, a former employee told Business Insider.
Uber’s software “would classify them as objects that are actually moving, and the cars would do something stupid, like stop or call for remote assistance,” one engineer told Julie Bort as part of her deep dive into the killing of an Arizona pedestrian by an Uber self-driving car in March.
An Uber representative denied that the car stopped for tree-branch shadows, adding that the car would stop for actual tree branches in the road.
Other employees speaking with Business Insider said that piles of leaves would often confuse the car and that the group had to place a concerted effort into recognizing foliage.
Uber’s Pittsburgh-based Advanced Technologies Group employees some 1,100 employees. Sources tell Business Insider the division is burning through $600 million a year, a gap Uber is reportedly trying to fill ahead of its hotly anticipated IPO next year.
Earlier this month, Uber self-reported third-quarter financial data that showed widening losses and slowing growth at the ride-hailing giant.
Source: Business Insider