Intel Responds to Fatal Uber Self-Driving Car Crash

Experience, transparency, and redundancy are key, says Intel’s autonomous-car boss.

byStephen Edelstein|
Intel Responds to Fatal Uber Self-Driving Car Crash
Share

0

The recent fatal crash involving an Uber self-driving car has put more scrutiny on other companies testing autonomous vehicles. In a blog post, Professor Amon Shashua, head of Intel's self-driving car program and CEO of Intel-owned sensor supplier Mobileye, said this is an important time to have "substantive" conversations about the safety of autonomous vehicles.

Shashua said the crash, in which one of Uber's autonomous Volvo XC90s struck pedestrian Elaine Herzberg while she was pushing a bicycle across a street in Tempe, Arizona, shows the difficulty of interpreting sensor data. He said recent developments in artificial intelligence have led some to believe that this process is easier than it actually is, and that the experience of more established "computer vision experts," like those who help design existing driver-assist systems, should not be discounted.

Experience forms one pillar of self-driving car safety, Shashua said, and transparency forms another. He pointed to Responsibility Sensitivity Safety, a recent initiative by Intel to use mathematical models to ensure that self-driving cars operate safely, as an example of how his company is trying to earn the public's trust.

Shashua's third point involved redundancy. He said an autonomous car should pull information from cameras, radar, and lidar independently, rather than fusing the information provided by all three into one data stream. He said Mobileye has built independent "end-to-end" systems for all three sensors types.

While he did not say so explicitly, Shashua seemed to implicitly criticize Uber and its approach to self-driving cars. It's impossible to know if things would have turned out differently if Elaine Herzberg had crossed paths with a car operated by Intel instead of Uber, but it's hard to argue with Shashua's final point: a call for more public discussions of self-driving car safety.

"We invite automakers, technology companies in the field, regulators, and other interested parties to convene so we can solve these important issues together."

stripe