This is a big deal.

For the first time, Waymo is publishing data from its self-driving program in Phoenix. Mileage, crash stats, new information about its simulation program.

This is likely to shake up the AV landscape in a big way. https://twitter.com/verge/status/1322176465178398722

Here's some of the most important numbers:
From Jan-Dec 2019, Waymo drove 6.1M miles in Phoenix
From Jan'19-Sept'20, Waymo's fully autonomous cars drove 65K
Waymo was involved in 47 collisions during that period, 18 actual crashes, and 29 near-misses

None of the collisions resulted in serious injury, and "nearly all" where the fault of the other driver.

Here's what's interesting about those near-misses: the safety driver took control to avoid the crash. But Waymo ran the scenarios through simulation to see how it turned out

This is how Waymo generates counterfactual (what if?) situations that it can then learn from to better train its self-driving software.

"They’re really going beyond regular data. And that’s very new and very unique," @DanCrashMan tells me

What about these crashes? Eight of the 49 collisions were considered "severe" that resulted in airbag deployment. One (in simulation) likely would have resulted in injuries. Waymo lists all the ways human drivers violated traffic rules to create these crashes.

The conclusion? As long as Waymo's AVs share the road with error-prone human drivers, these crashes will keep happening. It's a "clear reminder of the challenges in collision avoidance," Waymo says.

In addition to this data, Waymo is also releasing its multilayer approach to AV safety. The hope is that this framework can serve as a foundation for a universally accepted standard for AV safety

Waymo says its goal is to "kickstart a renewed industry dialogue."

Anyway, there’s a lot to sift through here and I’m really curious how this will be received by Waymo’s competitors.

What say you @zoox @argoai @cruise @nurobots @aurora_inno?