Heard a statistic on the news, coming from Waymo. They said their cars have 90% fewer serious injury accidents than cars with drivers. An obvious conclusion would be that the robotic driver is an order of magnitude safer, at least for serious accidents.
But there’s another force at work here. Cars with drivers always have at least one occupant. But Waymo cars sometimes have zero occupants, in which case they can launch themselves off a cliff and into a volcano without increasing that particular accident statistic.
There could be other similar influences. Sometimes the whole family piles into station wagon, such that one accident might possibly add a half dozen to the score. But perhaps the lone traveler choosing Waymo is a more common scenario. Similarly, people learning to drive, on a learner’s permit perhaps, are going to rack up their contribution with at least one other person in the car.
I think it’s a different but more useful statistic to ask what an individual’s risk is for each additional mile, in the Waymo versus ordinary case.
ETA: Sorry, should have added a TLDR: the author, a surgeon, says that the safety rate is so much better with self-driving cars that we should move to them as quickly as possible. Here’s a relevant quote:
In medical research, there’s a practice of ending a study early when the results are too striking to ignore. We stop when there is unexpected harm. We also stop for overwhelming benefit, when a treatment is working so well that it would be unethical to continue giving anyone a placebo. When an intervention works this clearly, you change what you do.
This isn’t going to be a significant factor. Something like 75% of miles in the US are single occupancy.
The bigger possible factor is the type of driving. Waymos are still primarily (entirely?) city driving. While accident rates are much higher in the city, serious injury rates are lower than highway driving at high speeds.
The useful comparison is driving under the same conditions. The article RitterSport linked mentions “crash statistics with miles driven that allow accurate comparison with human drivers in the same locations,” so those studies have been done although the data isn’t presented in that article.
This is a non-FQ response, but the improved safety makes total sense to me – these cars can see all around them all the time in all weather and never stop paying attention because they’re tired or to look at their phone or are fiddling with the radio or whatever. If they’re not sure what to do, I’m pretty sure they just stop.
I’m not sure I’ve ever seen one on a freeway, even in the city (I live and do most of my driving in San Francisco, one of the first cities to have these driving unattended, I think). If they’re not doing freeway driving, that would certainly affect the comparison with human driving.
I wonder if we need to limit the comparison to ride-sharing. I don’t know if ride-sharing miles safer than privately owned vehicle miles, but if we assume out the gate that Uber drivers generally aren’t drunk, it stands to reason.
My rationale is that Waymo isn’t going to replace Joe Commuter’s trip to work, even if it’s all city driving and even if it’s technically safer for Joe. Joe’s car is a sunk cost, and he’s not going to shell out for a Waymo when he’s got a car in the garage. So really the only miles Waymo can replace are Uber/Lyft miles.
Great point. A lot of accidents/injuries are caused by impaired drivers. The comparison of Waymos to sober drivers without impairments or distractions would be quite different.
But I’m not sure if that’s the right comparison to make. It probably makes more sense to compare it to the drivers who are actually on the road, not some ideal driver.
I’ll comment on this, it’s scary just to be in car in the Philippines with a human driver. Let alone have one controlling a car in the US over the internet.
They aren’t controlling the car. They respond to specific requests when the car encounters a situation it doesn’t know how to handle. The remote operators provide advice which the autonomous vehicle can incorporate in its decision. And the Philippines operators are not part of the event response team that handles emergencies and post-collision actions.
The rest of @steronz’ post you didn’t quote is what matters.
We should compare Waymo’s safety rate to the driving that it is already displacing and will displace if / when Waymo becomes national.
Which is by and large taxi / Uber driving. Not ordinary citizens driving themselves driving. And also not idealized super-drivers. So semi-screened people who are driving for income and whose performance affects their future income. And who are driving in the presence of unrelated witnesses (their passengers) who would be quick to pick up on, and report, impaired, aggressive, or hazardous driving.
Along with that, so far Waymo driving is all short range driving. If they do expand in SF or Phoenix to the point they’re covering the entire metro area including airports, stadiums, the burbs, etc., then the average trip length will grow as will the number of freeway miles driven. But not yet. So far it’s all urban core surface street driving.
Eh, I disagree. I don’t think anyone sees Waymo as the endpoint of autonomous vehicles. That’s the usage now, but it’s silly to think AVs stop at urban ride sharing.
For sure the eventual use case for AVs is displacing humans from virtually all driving. We’re in agreement there.
I suppose it depends on what question the OP is really asking. “How much are current 2026 Waymos improving safety in current e.g. Phoenix” versus “How much safety improvement could we get from hypothetical future AVs with hypothetical advances deployed hypothetically everywhere?”
If Waymos are materially safer than Ubers in benign suburban Mesa AZ, then it’s a darn good bet they’ll be vastly safer than Joe/Jane Average driver in any Metropolis USA.
No, you should compare Waymo’s record to the driving it is replacing AND factor in the stupid amount of miles those cars are driving around empty - waiting for fares, avoiding parking fees.
When you factor in all those useless miles on quiet side roads their record looks a lot less impressive.
The Waymos are everywhere out here. In fact, they tend to hang out in our neighborhood, probably because everyone has garages, so there is a lot of unused street parking available. Naturally, every time a Waymo gets in an accident (even if it was a human rear-ending them) someone makes a big deal about it, and posts it on Nextdoor. I’m obligated to point out that the same thing happens 10,000 times a day with human drivers…
I wave at them as I go by - figuring that my wave will be forever ensconced in some digital archive somewhere.
Then there’s the issue that if a human hits a Waymo, it still counts as a Waymo accident.
There was a comment on an article about Telsa’s robotaxi “high accident rate” and it included a list of half a dozen of the 19 accidents. One appeared to be the Tesla’s fault at 6mph it hit a vehicle, two or three were in parking lots but did not involve another vehicle (making me think it was confused about curbs), and one was obstacle in the road (not surprising, it does not know about potholes yet) and one a truck hit a stationary Tesla.
Another article described “bullying” where drivers realize a car is computer driven and do what they want, crowding and cut it off, etc. because they know the algorithm will do whatever it needs, including back off the space, rather than risk an accident even when not at fault.
I anticipate a future where all the cars are self-driving except for a few legacy vehicles. No need to stop for traffic lights, they zip through intersections between each other like stunt drivers, each car talking to others. They will warn each other “watch out for the red Corvette, it’s human driven!” so that driver will do what he wants and all the other cars get out of his way… until he runs head-on into the only other human driven car on the road.
I anticipate for people working in crowded cities, a Waymo service may end up being acceptable if it’s cheaper than parking downtown all day.
Plus, my experience with tesla full self drive, it was probably safer on the expressway everyone staying in their lane most of the time, than in complex downtown traffic.