Another factor is highway vs surface miles. Highway miles would seem significantly safer on a per-mile basis (but I can't find any numbers to back that up), and with Waymo excluding highways for now they'd be all surface miles. I can't find data that breaks down crash rate by street type, but I suspect surface traffic has a much higher crash rate than highway miles.
[1] https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...
Freeway is about 0.3 per 100 million, and the aggregate total is 3x higher than that around 1-1.2 per 100 million.
In other words, I don't think this is a concern at all. It's similar to humans already and will only get better.
I think economics and scaling are the main challenges and unknowns for Waymo. I hope they will soon finish phase 1 (perfect self-driving in a small area) and start pahse 2 (expand as fast as possible).
I wonder if a huge amount of the driving are things like OTR trucking and have a low accident rate compared to the average civilian commuter.
> The majority of incidents in our report fall into this category of typically unreported minor collision or contact.
It looks like in 2018 Arizona had 1.4 crashes per million miles https://azdot.gov/sites/default/files/news/2018-Crash-Facts.... and Waymo released data that they had 2.9 crashes per million miles, over twice as many. 7.7 if they hadn't had human intervention, almost at another order of magnitude than the average driver. That's rough.
I'm really not interested if the other driver is almost always at fault, that's not the number to get down. Their mid-term goal seems to be to drive on public roads with mostly human drivers, so they need to get better results. If they're getting bad results over lots of data, it's not just bad luck: the way they're driving is causing more crashes, even if they're not at fault. Additionally, fault classification is worrisome, since there is the potential for bias.
Various factors might make this remark wrong
- "Crashes per mile" isn't the right metric: injury, fatality, or similar should replace or augment it.
- I compared data from Arizona to data from Waymo's zone, which might be more crashy.
- Crashes in the doc I linked underestimate total crashes, since many go unreported.
- I am not comparing apples-to-apples -- new commercial minivans might be far more crash-prone than the average vehicle.