Autonomous Vehicle Disengagement Reports 2017

dmv.ca.gov

10 points·Shank··6 comments

Reading the reports is a bit disconcerting. Waymo appears to be about 60x better at these measurements than their competition.

The Tesla report says "these are not autonomous vehicles nor have they been driven in autonomous mode". This might surprise some Autopilot users.

In Waymo’s most recently reported month, November 2017, they reported over 30,000 miles driven with only a single disengagement. It’s hard to dispute that they’re years ahead of the competition at this point.

I don't doubt they are ahead, but on the other hand it appears there are about 1.25 fatalities per 100 million miles from human drivers. If you equate a "disengagement" to a human releasing all the controls of a vehicle and covering their eyes in panic, then I think it's reasonable to extrapolate the Waymo figure to about 3300 fatalities per 100 million miles. Which is about 2600 times worse than the human average.

This seems a bit odd, considering the people in other threads about self-driving cars vehemently arguing that we must switch to self-driving cars immediately to save lives.

On the bright side, once Waymo works out the bugs, we should have enough data to say their self-driving cars are probably comparable to humans in about...270 years of no failures (100M/30K/12).

Can anyone point me to article, or provide an argument, as to why we should trust these self-reported numbers?

If these estimates are not accurate, what process produces them?

a process which favors looking good to the general public when there is no consequence of fudging the numbers.

Reply on news.ycombinator.com