SoylentNews is people
SoylentNews
from the it-is-everybody-else-you-have-to-watch-out-for dept.
More bad news for Uber: one of the ride-hailing giant's self-driving Volvo SUVs has been involved in a crash in Arizona — apparently leaving the vehicle flipped onto its side, and with damage to at least two other human-driven cars in the vicinity.
The aftermath of the accident is pictured in photos and a video posted to Twitter by a user of @FrescoNews, a service for selling content to news outlets. According to the company's tweets, the collision happened in Tempe, Arizona, and no injuries have yet been reported.
Uber has also confirmed the accident and the veracity of the photos to Bloomberg. We've reached out to the company with questions and will update this story with any response. Update: Uber has now provided us with the following statement: "We are continuing to look into this incident and can confirm we had no backseat passengers in the vehicle."
TechCrunch understands Uber's self-driving fleet in Arizona has been grounded, following the incident, while an investigation is undertaken. The company has confirmed the vehicle involved in the incident was in self-driving mode. We're told no one was seriously injured.
Local newspaper reports suggest another car failed to yield to Uber's SUV, hitting it and resulting in the autonomous vehicle flipping onto its side. Presumably the Uber driver was unable to take over the controls in time to prevent the accident.
Source: TechCrunch
(Score: 1, Interesting) by Anonymous Coward on Monday March 27 2017, @04:46PM
As has been posted here before, the required mileage to truly "prove out" an AI driver is enormous -- many times greater than combined efforts of all the various companies engaged in this research.
I wonder if they could mitigate that by putting telemetry on a whole bunch of human-driven cars and then just running the data through the AI's trainer. Obviously its not the same as having the AI in control, and you might end up training the AI with drivers' bad habits too. But maybe human-assisted classification to excise the worst instances would help, although that would be a boring AF job unless they figured out how to make it a video game or something.