Accident and death statistics are often reported per passenger mile, which would account for empty driving. Accidents can be further differentiated by type of road (urban/rural, arterial, collector, local, highway.)
Whether these are all being differentiated for AVs, I can’t say.
Well, they can probably see through the snow, which human drivers can’t. And, if they can’t see, they’ll probably just stop, unlike human drivers.
Waymo operates in very specific urban areas that are well mapped out, though.
I took two waymo rides in Phoenix. One was suburban (i went home from a party, from one suburban home to another) and the other was a trip to the airport.
That’s fair. Humans get into not-at-fault accidents, too. And also, better drivers have fewer not-at-fault accidents.
That’s a hard question to answer because so many more cars have a person in the driver’s seat than in other positions, and also, there’s no record of exactly who sits where else when. There are excellent stats in the US about (almost) every fatal auto accident (FARS), but that’s the numerator. The denominator is only known for drivers.
I felt very safe in the two Waymo rides i took. Far safer than when my friend, the Tesla fanboy, has his car drive us around Chicago. But Phoenix traffic is more benign than Chicago traffic. Also, the Waymo was slow. At least, it took me a lot longer to get to the airport than my host estimated.
It navigated the airport beautifully, though, because it knew exactly where it wanted to merge from one road to another.
One unexpected bonus of the Waymo is that i didn’t feel socially awkward with the cab driver.
Waymos will not exceed the speed limit. So yeah, slower than basically any human driver.
They also didn’t take the freeway until recently. That doubled the time to get back from the airport.
Here is Waymo’s summarization of their report:
Aside from the graphs and links to their data, they have a FAQ that covers some of the discussion points here (including the above point about using the same locations). It’s obviously biased from their perspective, but it’s transparent enough to be peer-reviewed.
@LSLGuy had a good point about how it matters exactly what question the OP is asking.
Waymo scopes the question as:
This comparison answers the research question, “what is the effect of Waymo’s driving on the status quo.”
This eliminates the need, for better or worse, to resolve a lot of the differences we discussed in this thread (about ride service vs. personal use and why they are counting the miles when the vehicle is empty). Part of that is self-serving, but part of it is that there isn’t the right data to do apples-to-apples.
I think these differences would matter more if the improvements were smaller, but reductions of 80-90% are large enough to overlook many of the variables.
Certainly you could deploy self-driving tanks that only travel the safest side-streets at a max of 10 mph and your stats would be great, but that’s not the case here.
The report is quite interesting; I didn’t know they were up to 5 cities.
On the one hand, i feel a little bad about a massively subsidized product taking jobs from people. On the other hand, I’m eagerly looking forward to self-driving vehicles becoming economically viable in most places, and most new industries need some degree of subsidy. I paid $100 for my first LED lightbulb. (Well, really my third, but my first bright one. Also, it rapidly burned out.)
I’m really curious how you figure that.
Presumably a car which crashes into fixed obstacles with no passengers contributes neither to the numerator of injured people aboard, nor to the denominator of passenger miles accumulated.
As to two cars colliding with one another, whether Waymo or not, there’s an inherent ambiguity. One which the standard stats solve somehow, although I don’t know the specifics.
Namely that we ought to include all the occupants of both cars, and all the mileage of both cars. But do they? And does that count as one collision event, or two vehicle-collision events? I can make a coherent argument for both.
The stats are Rider-only (RO) which means there are humans in the car.
They track accidents regardless of fault, crashes above and below DeltaV of 1 mph (below indicates a minor collision), injuries reported (adjusted by 32% based on typical underreporting), airbags deployed. and they separate stats by collision with car, pedestrian, bike, and motorcycle. Stats are independent for each of the 4 cities as well as aggregated.
They compare their metrics against research of human-driver accidents. This research is based on state police reports and is broken down by county. They do not have data to compare on a street level so they compare by counties.
The Waymo link above goes into the data, the methodology, and FAQ. It’s really quite interesting.
DOT publishes their methodology but I’m not familiar with the ins and outs.
I am not so sure. From one of the Waymo papers linked in their methodology, they built datasets for human-driven vehicles in the regions they are operating in. These datasets combine crash reports and fatality reports, then map those against general government estimates of miles driven in those regions.
Similarly NHTSA collects and publishes the raw data from self-driving crashes (not aggregated per mile) here.
NHTSA has issued a Standing General Order (the General Order) requiring identified manufacturers and operators to report to the agency certain crashes involving vehicles equipped with automated driving systems or SAE Level 2 advanced driver assistance systems.
This has been amended to include Level 3 and Level 4 (Waymo’s level).
Calculating statistics per mile adds more complexity and variables than just reporting stats per individual car, but the self-driving companies do it because it normalizes the data based on length of use to give a more accurate picture while making self-driving look much better.
I had my first Waymo experience when I had a business trip to Atlanta last month. I was an instant convert. I now get unreasonably annoyed when I have to deal with a human driver.
I was in an area with Waymo coverage a couple of weeks ago so I compared a hypothetical ride against Uber. Waymo would have cost $30+ versus approximately $15 for Uber.
An advantage of living in SoFL is about 80% of uber/lyft drivers speak negligible English. Its more like 99% don’t do English for taxicabs.
Can have adverse consequences in oddball situations, but the lack of pressure to converse or have to listen to Faux news radio is a plus. 
All my Waymo rides were through Uber.
When i was in Phoenix, i price checked Waymo to Uber, and it was slightly cheaper. But the Waymo price is fake. Google just picks something that they think will be competitive, or look good, or set up the market for later, or… I don’t believe they are close to breaking even.
This is from a year and a half ago, but i think this is still where they are:
This is more recent, and had information about where they go and what the vehicles cost and who has invested, but doesn’t have any estimate of profitability:
I mean, the Uber price is fake, too, since they manage to completely avoid things like depreciation of vehicles.
Does anywhere in the US have privately-owned self-driving cars?
I saw some article where an Uber driver was complaining - they used to be doing well, but then everyone got into the business, and now the business is tougher, too many drivers chasing too few customers.
On top of that - I haven’t had to get a taxi for a looooong time, but I recall that on busy times, like New Years Eve, the report was that waiting times could exceed an hour or two. How many Waymo’s need to be in the system to handle the peak demand? What if your Waymo gets into an accident - gonna wait twice as long…
I kind of wonder how private Waymo would work? Musk is touting his RoboTaxis, he seems to think they’ll sell like hotcakes - but there’s been no details how that would work, how the summon and ride allocation works. Imagine being an owner of an Uber-only version of self-drive taxis, no guarantee of revenue; competing, possibly, against the vehicles owned by the service itself, which suggests you’d be at the bottom of the list of vehicles summoned. All you are doing is supplying capital for a billionaire’s business.