Loved the article and good point about human error in crashes. It reminds me why I don't subscribe to services like Progressive Snapshot even though it would lower my rates. Where I live (Chicago), drivers routinely run red lights, drive while smoking weed, speed through school zones, etc. In order to be on streets with irrational human drivers, you often have to take counter actions that aren't textbook to keep safe or avoid accidents. I expect the majority of future crashes with driverless cars to remain human caused.
This is very impressive and very heartening! I cannot wait for Waymo to start driving on freeways so I can use it more often.
For human drivers, do you know what percentage of airbag crashes are the driver's fault (as opposed to the fault of another driver)? I wouldn't have guessed that an 83 percent reduction in airbag crashes would be possible even with "perfect" driving. But perhaps I have a distorted intuition as to how many crashes involve a single car vs. two or more cars.
Do you think Waymo cars are somehow able to reduce the number of "other driver's fault" crashes (e.g. due to better defensive driving, quick reflexes & avoidance maneuvers, etc.?) Or would you attribute their safety record mostly to not causing crashes that are the Waymo's fault?
Final question: for the 83 / 84 percent reduction figures, how well matched are the conditions under which the Waymos are driving to the human baseline?
Something I've learned writing about automotive safety is that fault isn't a binary concept. Often crashes happen because both parties made a mistake (or at least drove in a less-than-ideal way). So a very conscientious driver can reduce crashes by more than 50 percent by not only eliminating the half of crashes that the average driver "causes," but also driving in a way that compensates for other people's bad driving.
This could just mean garden variety defensive driving where (for example) you leave plenty of room to stop if the car ahead of you stops suddenly. Or in more extreme cases it could mean taking "evasive maneuvers" when another car is about to crash into you.
With all that said, there probably is a limit to how many crashes even a "perfect" driver can avoid. There is not much you can do if you are stopped at a light, with other stopped cars in front of you, and another car crashes into you from behind.
As far as how good their matching is: I asked them about this back in September and I remember their answers being pretty convincing. But I don't remember all the details. Over the years I've talked to a few traffic safety experts about Waymo's baselines in studies like this and they've generally been pretty positive about them.
You can read a little bit about these methodological questions in my September story.
And if you want all the details, I believe the latest data release uses methodology for the human baseline that's similar to what's described in this 2023 study:
One thing I wonder is how close Waymo is to making an appreciable difference in overall crash volume in any of its service areas. Certainly replacing all the cabs in the Lower Manhattan congestion zone with Waymo should have a noticeable impact if their claims are right,but I don't know how close SF is to that.
Waymo estimates that it has prevented about 30 airbag crashes each in Phoenix and San Francisco, with most of those likely occurring in 2024. I don't know what the absolute number of serious crashes is in the San Francisco or Phoenix areas in a typical year, but I bet it's much larger than that.
I expect eventually they will license their technology to personally-owned vehicles, which will have a much bigger impact. But that's probably a few years away.
Thanks so much for reporting on this! It’s hard to patiently wait for Waymo to expand across the country and bring these safety benefits everywhere, although I realize their patient and deliberate approach is likely the reason their safety record is so good.
I love these safety updates. Using insurance data feels like an intuitive and irrefutable way to illustrate just how good robotaxis have gotten.
I took my 70yo mother on a Waymo ride in Austin this past weekend. Over the course of two short trips, she went from hushed nervousness to complete conviction that the days of human car ownership are numbered.
What I mean there is that I don't have a good way to talk to the other parties involved in the 38 crashes I'm discussing in that section. A handful of the most serious crashes were covered by local media, so I can get some information from those. But even when there is a media story it usually doesn't provide enough detail to draw any conclusions about whether Waymo or someone else was at fault.
There are of course other sources for crash statistics. I haven't tried to calculate my own human benchmark for Waymo's service territory, and I don't know of anyone else who has tried to reproduce Waymo's research. But over the years I've talked to several experts on automotive safety who have generally positive things to say about Waymo's methodology, so I'm inclined to give it some degree of credibility.
I think the resistance to self-driving cars is in part explainable by the regulatory (and, indeed, human) tendency to pin all bad events on a human somewhere along the way. It's impossible to punish inanimate objects (civil-forfeiture lawsuits where it's "State v. $12,963 in cash") notwithstanding), and difficult to punish companies without resorting to "death penalties" or "near-death experiences". People, on the other hand...
I have no doubt that self-driving cars can drive much better than humans can. But they aren't perfect, and people want someone (not something) to blame when they aren't perfect.
As my manager used to say, "whose throat do I choke?"
You have to remember that cars are status items for many people, especially those under age 35 or so. Advertisers build on the status theme.
So those fighting autonomous cars, which will most likely be fleet cars looking pretty much the same, are worried about losing another aspect of their individuality and people who work in advertising for the automotive sector ar eworried that they will lose their jobs.
I was making exactly the same point as far back as 2018. One lesson from this is that the advance of AI is commonly a lot slower than advocates expect. That's already evident with LLMs, which have been around for a few years with only modest real-world effects
The situation where the Waymo seems to be at fault doesn’t look super clear to me. Did the Waymo have time to avoid the crate or did it end up in the street suddenly? How much time did the scooter have to avoid it? I realize these answers aren’t knowable with available info, and it does seem like the Waymo would be primarily at fault, but idk. I asked Claude and it said that with California’s proportional fault system, “the Los Angeles driver in this scenario would likely be assigned approximately 60-80% of the fault for the accident.”
Yes, clearly whoever dropped that crate on the road bears some of the blame, and Waymo's degree of responsibility would depend on the specific circumstances.
Love to hear this. Car-related fatalities are the highest likelihood accidental causes of death, far greater than crime and guns (if you exclude gun suicides at least).
The insurance data here is compelling, despite it only covering injuries, since that is rightly the number one concern about robotaxis. Do we know if their insurance rates are also 80-90% lower (on a per mile basis) relative to human drivers, taxis, Ubers, etc.? If not, why not? Are they getting into a lot of minor accidents and scrapes that are still costly to repair, even if they are not causing injuries?
For years, I've been thinking about the stylized fact that humans are usually at fault in driverless car accidents. The implication is that autonomous cars are safer than human drivers, and that seems to be true, in Waymo's case. But it might still be the case that driverless cars behave in unexpected ways, which are legally and technically correct, but still lead to more collisions and rear-endings overall -- e.g., where the human driver expects the Waymo to go through a yellow or roll through a right turn on red, and instead it stops. If that's the case, driverless vehicles might reduce injuries and deaths, but not end up with vastly lower insurance or repair costs, which would remove a major financial incentive for investing in autonomy.
I don't know what their insurance rates are like. I would guess that their rates aren't really comparable to a consumer auto insurance product, since (for example) they can probably afford a much higher deductible than a normal company.
I do not think they are getting into an unusual number of minor accidents. But it's a little hard to say for sure because human drivers often don't report minor crashes, so we don't have a good baseline to compare to. But for example, Waymo says that it has gotten into 64 percent fewer police-reported crashes, which is a broader category than injury or airbag crashes.
It would be even more interesting (and probably revealing) to compare Waymo-caused crashes with human-caused crashes in the limited area where Waymo is currently in use. I don't follow car crashes, but I'm guessing that crashes in rural areas or on freeways are less common per mile than crashes in urban areas. If that's true, then Waymo's record might shine even more.
I've probably taken 10-15 trips in a Waymo and been pleasantly surprised at how safe they are. Obviously a small sample size but I also scan the news/Twitter/etc every week for 'Waymos behaving badly' and rarely do I see a situation where Waymo does something silly/that could lead to an accident. So this data definitely supports my anecdotal experience.
One thing I'd like to see you look into/cover in the future are all of the claims about Waymo's saving lives. The whole point of AVs is to be cheaper than ridehail so won't this induce more demand? Even if Waymo's get into 1/10 of the accidents, if the number of human driven miles doesn't decrease, we'd actually see an increase in injuries/deaths/etc.. Haven't fully fleshed this idea out/done the math but I think it will be a while before the absolute number of deaths (ie the 40k/year) starts to drop because of Waymo.
I do expect there will be some offsetting increase in miles traveled, but it seems unlikely we'll see close to a 10x increase in miles traveled. So I expect overall traffic deaths to decline significantly once driverless technology is widely used.
Loved the article and good point about human error in crashes. It reminds me why I don't subscribe to services like Progressive Snapshot even though it would lower my rates. Where I live (Chicago), drivers routinely run red lights, drive while smoking weed, speed through school zones, etc. In order to be on streets with irrational human drivers, you often have to take counter actions that aren't textbook to keep safe or avoid accidents. I expect the majority of future crashes with driverless cars to remain human caused.
Yes. The easiest way I’ve found to sell people on autonomous is to point out they there are no drunk autonomous drivers.
This is very impressive and very heartening! I cannot wait for Waymo to start driving on freeways so I can use it more often.
For human drivers, do you know what percentage of airbag crashes are the driver's fault (as opposed to the fault of another driver)? I wouldn't have guessed that an 83 percent reduction in airbag crashes would be possible even with "perfect" driving. But perhaps I have a distorted intuition as to how many crashes involve a single car vs. two or more cars.
Do you think Waymo cars are somehow able to reduce the number of "other driver's fault" crashes (e.g. due to better defensive driving, quick reflexes & avoidance maneuvers, etc.?) Or would you attribute their safety record mostly to not causing crashes that are the Waymo's fault?
Final question: for the 83 / 84 percent reduction figures, how well matched are the conditions under which the Waymos are driving to the human baseline?
Something I've learned writing about automotive safety is that fault isn't a binary concept. Often crashes happen because both parties made a mistake (or at least drove in a less-than-ideal way). So a very conscientious driver can reduce crashes by more than 50 percent by not only eliminating the half of crashes that the average driver "causes," but also driving in a way that compensates for other people's bad driving.
This could just mean garden variety defensive driving where (for example) you leave plenty of room to stop if the car ahead of you stops suddenly. Or in more extreme cases it could mean taking "evasive maneuvers" when another car is about to crash into you.
With all that said, there probably is a limit to how many crashes even a "perfect" driver can avoid. There is not much you can do if you are stopped at a light, with other stopped cars in front of you, and another car crashes into you from behind.
As far as how good their matching is: I asked them about this back in September and I remember their answers being pretty convincing. But I don't remember all the details. Over the years I've talked to a few traffic safety experts about Waymo's baselines in studies like this and they've generally been pretty positive about them.
You can read a little bit about these methodological questions in my September story.
https://www.understandingai.org/p/human-drivers-are-to-blame-for-most
And if you want all the details, I believe the latest data release uses methodology for the human baseline that's similar to what's described in this 2023 study:
https://arxiv.org/abs/2312.12675
One thing I wonder is how close Waymo is to making an appreciable difference in overall crash volume in any of its service areas. Certainly replacing all the cabs in the Lower Manhattan congestion zone with Waymo should have a noticeable impact if their claims are right,but I don't know how close SF is to that.
Waymo estimates that it has prevented about 30 airbag crashes each in Phoenix and San Francisco, with most of those likely occurring in 2024. I don't know what the absolute number of serious crashes is in the San Francisco or Phoenix areas in a typical year, but I bet it's much larger than that.
And I think the estimate is that Waymo has about 10% of the ride hail market, so if they replaced it all that would only be 300 crashes.
I expect eventually they will license their technology to personally-owned vehicles, which will have a much bigger impact. But that's probably a few years away.
It does, at this point, seem unlikely that my 7 year old will need to learn to drive for general travel.
Thanks so much for reporting on this! It’s hard to patiently wait for Waymo to expand across the country and bring these safety benefits everywhere, although I realize their patient and deliberate approach is likely the reason their safety record is so good.
I love these safety updates. Using insurance data feels like an intuitive and irrefutable way to illustrate just how good robotaxis have gotten.
I took my 70yo mother on a Waymo ride in Austin this past weekend. Over the course of two short trips, she went from hushed nervousness to complete conviction that the days of human car ownership are numbered.
"I only have Waymo’s side of the story. But it doesn’t seem like Waymo was at fault in any of them"
I am not dismissing your story but is there no other source around crash stats other than Waymo?
What I mean there is that I don't have a good way to talk to the other parties involved in the 38 crashes I'm discussing in that section. A handful of the most serious crashes were covered by local media, so I can get some information from those. But even when there is a media story it usually doesn't provide enough detail to draw any conclusions about whether Waymo or someone else was at fault.
There are of course other sources for crash statistics. I haven't tried to calculate my own human benchmark for Waymo's service territory, and I don't know of anyone else who has tried to reproduce Waymo's research. But over the years I've talked to several experts on automotive safety who have generally positive things to say about Waymo's methodology, so I'm inclined to give it some degree of credibility.
I think the resistance to self-driving cars is in part explainable by the regulatory (and, indeed, human) tendency to pin all bad events on a human somewhere along the way. It's impossible to punish inanimate objects (civil-forfeiture lawsuits where it's "State v. $12,963 in cash") notwithstanding), and difficult to punish companies without resorting to "death penalties" or "near-death experiences". People, on the other hand...
I have no doubt that self-driving cars can drive much better than humans can. But they aren't perfect, and people want someone (not something) to blame when they aren't perfect.
As my manager used to say, "whose throat do I choke?"
You have to remember that cars are status items for many people, especially those under age 35 or so. Advertisers build on the status theme.
So those fighting autonomous cars, which will most likely be fleet cars looking pretty much the same, are worried about losing another aspect of their individuality and people who work in advertising for the automotive sector ar eworried that they will lose their jobs.
I was making exactly the same point as far back as 2018. One lesson from this is that the advance of AI is commonly a lot slower than advocates expect. That's already evident with LLMs, which have been around for a few years with only modest real-world effects
https://crookedtimber.org/2018/01/19/bad-drivers-should-have-their-cars-driven-by-robots/
The situation where the Waymo seems to be at fault doesn’t look super clear to me. Did the Waymo have time to avoid the crate or did it end up in the street suddenly? How much time did the scooter have to avoid it? I realize these answers aren’t knowable with available info, and it does seem like the Waymo would be primarily at fault, but idk. I asked Claude and it said that with California’s proportional fault system, “the Los Angeles driver in this scenario would likely be assigned approximately 60-80% of the fault for the accident.”
Yes, clearly whoever dropped that crate on the road bears some of the blame, and Waymo's degree of responsibility would depend on the specific circumstances.
Love to hear this. Car-related fatalities are the highest likelihood accidental causes of death, far greater than crime and guns (if you exclude gun suicides at least).
The insurance data here is compelling, despite it only covering injuries, since that is rightly the number one concern about robotaxis. Do we know if their insurance rates are also 80-90% lower (on a per mile basis) relative to human drivers, taxis, Ubers, etc.? If not, why not? Are they getting into a lot of minor accidents and scrapes that are still costly to repair, even if they are not causing injuries?
For years, I've been thinking about the stylized fact that humans are usually at fault in driverless car accidents. The implication is that autonomous cars are safer than human drivers, and that seems to be true, in Waymo's case. But it might still be the case that driverless cars behave in unexpected ways, which are legally and technically correct, but still lead to more collisions and rear-endings overall -- e.g., where the human driver expects the Waymo to go through a yellow or roll through a right turn on red, and instead it stops. If that's the case, driverless vehicles might reduce injuries and deaths, but not end up with vastly lower insurance or repair costs, which would remove a major financial incentive for investing in autonomy.
I don't know what their insurance rates are like. I would guess that their rates aren't really comparable to a consumer auto insurance product, since (for example) they can probably afford a much higher deductible than a normal company.
I do not think they are getting into an unusual number of minor accidents. But it's a little hard to say for sure because human drivers often don't report minor crashes, so we don't have a good baseline to compare to. But for example, Waymo says that it has gotten into 64 percent fewer police-reported crashes, which is a broader category than injury or airbag crashes.
It would be even more interesting (and probably revealing) to compare Waymo-caused crashes with human-caused crashes in the limited area where Waymo is currently in use. I don't follow car crashes, but I'm guessing that crashes in rural areas or on freeways are less common per mile than crashes in urban areas. If that's true, then Waymo's record might shine even more.
That’s what these studies did.
I've probably taken 10-15 trips in a Waymo and been pleasantly surprised at how safe they are. Obviously a small sample size but I also scan the news/Twitter/etc every week for 'Waymos behaving badly' and rarely do I see a situation where Waymo does something silly/that could lead to an accident. So this data definitely supports my anecdotal experience.
One thing I'd like to see you look into/cover in the future are all of the claims about Waymo's saving lives. The whole point of AVs is to be cheaper than ridehail so won't this induce more demand? Even if Waymo's get into 1/10 of the accidents, if the number of human driven miles doesn't decrease, we'd actually see an increase in injuries/deaths/etc.. Haven't fully fleshed this idea out/done the math but I think it will be a while before the absolute number of deaths (ie the 40k/year) starts to drop because of Waymo.
I do expect there will be some offsetting increase in miles traveled, but it seems unlikely we'll see close to a 10x increase in miles traveled. So I expect overall traffic deaths to decline significantly once driverless technology is widely used.
Thanks for continuing to cover AVs!
People don't realize how important it is for the general populace to understand this.