Uber unveiled the real reason why one of its self-driving cars hit and killed a pedestrian earlier this year in Tempe, Arizona.
According to the company, a glitch in the car’s detection system was the main reason the vehicle ignored the pedestrian even though its sensors had seen her.
According to The Information, the car is equipped with software that establishes which objects the car should ignore and which it should react to. That software was not trained to react quickly enough, which is what caused the accident.
Experts had known from the start that the crash could not have been generated by a faulty sensor due to the wide variety of vision systems installed on the autonomous car. Nevertheless, the car ignored Elaine Herzberg as she was crossing the street in front of its front-facing cameras.
Surprisingly, the car didn’t even tap the brakes or react in any other way after hitting the pedestrian. The backup Uber driver didn’t react either until it was too late for the woman to be saved.
Uber Admits that Car’s Software Was at Fault
Before Uber acknowledged that the software was at fault, experts had various theories. Some of them believed that the object recognition system somehow failed to recognize the woman as a pedestrian since she had a bike.
According to another theory, the autonomous car’s brain failed to identify the woman as an object to pay attention to. The system decides which objects should prompt the car to hit the brakes and which objects to be ignored.
Uber’s recognition system acts like the human brain when making decisions and prevents the car from stopping at every, let’s say, bicycle parked at the side of the road.
Uber confirmed the second theory: the system considered the woman a false positive and decided to simply ignore her.
Image Source: Wikimedia