"They obviously do not trust these numbers"

First of the clbad

On Wednesday, the California Automotive Department released its annual safety data report for autonomous cars, detailing how freestanding vehicles created by companies like Uber, Tesla, Apple, and Waymo are disengaging from autonomous driving. of their cars and give them back control. a driver of human safety to avoid an accident or help the vehicle when his software becomes confused.

The report shows a clear winner: on average, Waymo's autonomous cars drove autonomously for 11,125 km before the human being took over, more than twice as much as GM, who finished second.

The number is impressive, but it does not match the experience of Waymo pbadengers in Arizona, questioned by Futurism, Ars Technica, and The edge, Which report frequent interventions of human drivers. And four experts have warned us that the new Californian personalities might not say it all, that the continued presence of security drivers shows that vehicles are not quite ready for prime time.

"I think the proof is in the pudding," said Mary Cummings, Autonomous Vehicle Engineer and Expert at Duke University, Futurism. "Waymo does not drive without a safety driver, neither in California nor Arizona, even legally, but they obviously do not trust those numbers."

Disengagement rate

When Futurism contacted a spokesman for Waymo to ask him for his incredibly low disengagement rate, he sent a press release explaining the company's progress in teaching its autonomous vehicles to handling unusual scenarios that could have caused the artificial intelligence of a vehicle.

Udacity founder and Stanford University engineer Sebastian Thrun told Futurism that the disengagement rate declared by Waymo was "quite feasible".

But these disengagement reports are not the rigorous badyzes that they seem to be at first sight, according to wired, each autonomous vehicle company collecting and reporting its own data, which is then compiled and compared side by side, although each company can perform tests using different parameters and standards.

A California DMV spokesman told Futurism that the government was reviewing each company's data before publishing its annual report.

"The DMV is carefully reviewing all release reports from manufacturers authorized to test autonomous vehicles in California," said the spokesperson at Futurism. "When necessary, the DMV requests additional information and follows up with the licensee to discuss our regulations, respond to concerns expressed, clarify the content of the report and learn more about the issues. business test methods and data collection procedures. "

And experts have warned that companies may not have been completely transparent about their tests.

"Let's be honest, without knowing when and why the safety driver has disengaged the autonomous driving ability or the context of the kilometers traveled (rain, snow, highway, city, etc.), it is difficult to interpret the 11,000 means Lionel Robert, a self-driving automotive expert from the University of Michigan, told Futurism, "This will continue to be a problem until the federal government intensifies and standardizes some of these measures." Unfortunately, for now, we can just guess and that's not what we should do. "

At the time of going to press, Waymo's spokesperson had not said how much the government was reviewing the disengagement data before publishing its annual report.

Take it as an opportunity for change

"11,000 miles are achievable in carefully constructed environments with extremely detailed mapping and in cases of low uncertainty, no inclement weather, typical traffic conditions, etc.," said Cummings. "But this metric can be quite misleading since we have no idea what these conditions are."

The test results reported by Waymo could show that the company excels in areas where others continue to struggle. They could also result from driving self-driving cars on empty, well-paved roads during the day.

"Yes [Waymo’s report] It sounds high, but then you have to look at the size of the fleet and what all the vehicles do, "Jonathan P. How, MIT's aeronautical engineer and researcher of autonomous cars told Futurism. "If many are driving in relatively simple well-known fields, the results could be skewed. But I do not know why they could be so much higher. "

Editor's Note 2/14: This article has been updated to include an additional statement from the California Automotive Department.