39 Comments

Computer hasn't yet figured out how to intelligently create non-misleading URLs :)

https://www.understandingai.org/p/new-data-shows-waymo-crashes-a-lot

Expand full comment
author

That's funny.

Expand full comment

Terrible PR apart I wouldn’t be surprised if Cruze crashed a lot less than human drivers. Just paying attention all the time probably avoids a huge number of crashes and the AI is always paying attention.

Expand full comment

And they're never under-rested, and they're not participating in America's favorite pastime (driving drunk).

Expand full comment

It's striking that in the three injury-causing crashes described here, each seems to have been primarily the fault of a human driver, not the Waymo vehicle.

Expand full comment
author

Yes, though I think there's a little more nuance than just saying the other driver was at fault. I'd be interested to know in the July incident if the branch fell down as Waymo was driving, necessitating a rapid deceleration, or if Waymo just didn't notice it until the last minute. If it's the latter, I'd say Waymo bears some responsibility, though of course the two cars behind also bear some fault for tailgating and would probably be found at fault in a court of law.

Similarly, the start-stop behavior in the August incident doesn't seem ideal. Again, the driver behind would probably be at fault legally but I wouldn't say Waymo was blameless.

The October incident does sound like it was completely the fault of the other driver.

Overall it's a very impressive record.

Expand full comment

Another way of looking at it: if you replace the human driver(s) in the crashes, do the crashes still happen? I don't know the answer, but I'm very curious. I'm guessing Waymo doesn't tailgate the same way humans do, and can better handle decelerations in cars ahead of it.

Expand full comment
author

Yeah, I think it's very unlikely that the crash would have happened if both (or all three) vehicles in these crashes were Waymo cars.

Expand full comment

I don't believe there is any doubt that autonomous cars, compared on an equal basis, will show a vastly safer profile when compared against human driven cars.

Let's get autonomous cars on the highways ASAP. Even if they wound up killing 1000 people in a year, that would be 40000 less than what human drivers cause.

Expand full comment

I think we can already be pretty sure Waymo will have less fatalities. Consider that most fatal accidents are caused by drunk driving, driver inattention, and speeding and breaking other traffic laws, all of which are basically not applicable to autonomous cars. I'm very much looking forward to autonomous cars becoming common so that less people die, basically.

It's fascinating that Waymo is the only company so far to have successfully made it to production. Uber, Lyft, and a bunch of other companies were in the running a few years ago, but they all dropped out. I think there's an opening now for 1-2 other major competitors. I wonder if any of the legacy car companies are going to get in the game.

Expand full comment
author

Besides Cruise, the other two companies in the running are Zoox (owned by Amazon) and Motional (joint venture by Aptiv and Hyundai). They have not begun fully driverless operations yet so I would guess they are 3-5 years behind Waymo. I think it would take a decade to build a new AV company from scratch so I don't think it would make sense for anyone new to get into the market unless they have an idea for greatly accelerating the development process.

Expand full comment

If we assume that a company could get meaningful market share and that self-driving is strongly preferable in a lot of cases, even a long timeline shouldn't deter companies in principle. But in practice, what CEO wants to spend money on something that only their successor will benefit from? The market share question is also worth considering. It would be disappointing to finish such a massive development effort and end up the Bing to Google's Waymo.

Even so, if I were in charge of a major OEM like Toyota, I would look at buying a company like Luminar and turning it into a full-fledged self-driving operation.

Expand full comment
author

Yeah, market share is a big question. If Google spends $10 billion to get 90 percent of the global taxi market, that's probably a good deal. If somebody else spends $10 billion and winds up with the other 10 percent because Waymo is already dominant, that doesn't seem like such a good deal.

But also there's a lot of uncertainty about all of this. Maybe the technology fundamentally isn't ready and it'll take another 10 or 20 years for advances in AI or sensors or whatever to make this commercially viable. Maybe they can only get it working for taxis and the taxi market isn't that big. Maybe Waymo will get shut down as soon as they get their first deadly crash. The longer it takes Waymo to reach commercial viability, the more wary you'd expect other companies to be to start on the same journey.

Expand full comment

Maybe we have different assumptions about all this. I think self driving is a big enough advantage that it will take over most of the car market over time, not just what's currently defined as ride sharing or taxi. I also expect driving volume to grow faster as the time cost of driving goes away. So I think it'll be an enormous market. The car market is currently $2 trillion per year. The average American also spends several hundred hours a year driving. If say half the market paid $5/hour not to drive, that would be another couple of hundred billion a year. I don't know that it would monetize quite like that, but there's a lot of value there.

I'm not sure how the technology wouldn't be ready since it's already in limited production.

I also expect self-driving capabilities to get much cheaper over time, both to develop overall and per unit. It's still a massive undertaking, but I would bet it goes from $5-10 B for Waymo to $1-2 B in say another 5 years (to get to the same level).

Expand full comment
author

No, I agree with all of that. It's just that I also thought all of this six years ago and things have progressed much slower than I expected back then. So I'm not so confident that I'd bet my life savings on it.

The technology is in limited production today but it's wildly unprofitable. We know Cruise was intervening every few miles—it seems quite possible Waymo is doing the same. Probably they'll be able to get that down over time, but if they don't then it might not be any chaper to operate than an Uber or Lyft, which would greatly limit the market size. Or again an early misstep could sour regulators on the whole project and lead to a decade-long "self-driving winter" where nobody is allowed to test their technology.

Again I'm not predicting any of this. I'm just saying that if you're starting a new project that's 6-8 years behind Waymo you've got to be really, really confident there's a pot of gold at the end of the multi-billion-dollar rainbow.

Expand full comment

Aren't human operators intervening remotely, though? Or was that just Cruise? "Driverless" seems misleading if periodic remote intervention is required, since that would be extremely difficult to scale out to the whole country.

I also strongly suspect that there are plenty of human drivers who are reliably safer than Waymo. Admittedly anecdotal, but the people I know who got into one fender bender or similar tend to get into several.

Expand full comment
author

Yes there are remote operators who provide the vehicles with guidance when the cars aren't sure how to proceed. I agree that this makes the economics challenging, and they may need to reduce the frequency of these interventions for the service to be commercially viable.

Expand full comment

Does anyone know if Waymo cars drive in all kinds of weather? How do accident rates in port weather compare between humans and AVs?

It’s encouraging news that Waymo is safer than human drivers, but significantly less encouraging if the only reason is that they don’t drive when it’s raining or icy.

Expand full comment
author

Waymo cars are limited to San Francisco and Phoenix, two cities that aren't known for icy road conditions. However, they are being compared to human drivers in those same cities, so the human drivers also aren't dealing with icy roads.

Expand full comment

Sounds like all three of the "serious" Waymo crashes were not even Waymo's fault.

Expand full comment

This is such an interesting case of technological adoption. What is the bar for Waymo? Be honest - humans are terrible, with cell phones, alcohol, etc. But we each trust ourself more. And we will absolutely sue if our self driving car crashes…

Expand full comment

It is fascinating technology, that is not eating, drinking coffee, checking their hair, and texting when driving. I saw that in one car just last week!!!

Expand full comment

Two things occurred to me.

First, Waymo causing injuries at one sixth the rate of human drivers is apparently true, but not necessarily representative. Driver training in the US is shambolic. In Germany, speaking from personal experience, it is very rigorous. The German fatality rate is one third that of the US, which *might* mean that properly trained drivers are only twice as likely as Waymo to be involved in an injury accident.

Second, the vehicle populations are very different. Waymo's AV's are as state of the art as possible. So, perhaps the best comparison is against contemporary HDV's (or whatever the acronym is). Why? Because they have lane departure and blind spot warning; cruise control with speed dependent following distance, speed limit tracking, and corner predictive speed reduction. Also predictive and emergency braking. Oh, and lane following.

All of which amounts to Level 2 automation.

It is likely that Level 2 cars have much lower mishap rates than Level 0 — otherwise, there's no point to all that stuff. Since nearly all vehicles are Level 0, the population of HDV's is very much different than AV's. Consequently, limiting the HDV mishap rate to just Level 2 cars would give a better idea of the differential possible with AV's over the long term.

Shambolic driver training aside.

Expand full comment

I don’t suppose there’s any point even comparing Waymo’s data to Tesla? It’s an apples-to-oranges comparison, but I’m curious just how far in the lead Waymo is.

Expand full comment

The one thing people have going for them is diversity. You're never going to have a leap second bug* crashing all the meat driven cars simultaneously.

* Although given Waymo's Google heritage, they are probably inoculated against leap second bugs in particular.

Expand full comment

The list should include this incident in which a Waymo killed a dog in San Francisco: https://www.kron4.com/news/bay-area/waymo-vehicle-hits-and-kills-small-dog-in-sf/ It’s not the same, but it seems less than fully forthright to omit it. Their encounters with public safety incidents also deserve a look: https://missionlocal.org/2023/08/cruise-waymo-autonomous-vehicle-robot-taxi-driverless-car-reports-san-francisco/

They also have not posted disengagement reports for this period and I will be interested to see those: https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/disengagement-reports/ Are they safer because of frequent human intervention, or is that a minor factor?

None of these are my real concerns, and in general I agree from watching them drive that Waymos seem safer than Cruise cars. (My real concern is that they will increase VMT in the same way carshare companies have, but worse.)

Expand full comment
author

This week's study was specifically for fully driverless miles. The dog strike happened while there was a safety driver behind the wheel, so it wouldn't be in-scope even if you consider it an injury.

Waymo has done significantly more safety driver miles (around 40 million) so if they included those miles that would be a totally different analysis. Waymo did include data from these test miles in their September study looking at insurance claims.

Expand full comment

That's fair. You're right.

Expand full comment

Tim wrote a bunch about the public safety incidents a few months ago.

Expand full comment

I missed that. Thanks.

Expand full comment

Why are you concerned with the number of Vehicle Miles Traveled increasing? Is it CO2 emissions? Because all their vehicles are fully electric and while neither AZ or CA have fully renewable grids yet, autonomous vehicles are probably better short term grid storage devices than human piloted EVs.

Expand full comment

We devote far too much public space and resources to cars, and transportation oriented around cars as the primary mode prevents density, which in turn is harmful for the climate. You’re right the current AVs are all electric, but that also makes them heavier than ICE cars, which implies both more road maintenance and worse survivability for the collisions that do occur. If we reduce car VMT and shift to more sustainable modes we get a lot of other benefits. If instead being in a single-passenger vehicle is much easier than it is today, we’ll wind up increasing space and resources devoted to that mode at the expense of many other potential uses. (This is a quick sketch of the arguments; I think there are a bunch of articles published on this idea that lay out a better case than I have here.)

Expand full comment
author

The main thing that prevents density is parking mandates and local voter concerns about parking shortages. If self-driving cars promote a shift from private car ownership to taxi rides, as I expect they will, then that should significantly reduce demand for parking on a per-VMT basis. Widespread use of self-driving cars should also make it easier to implement congestion pricing, which should help prevent over-use of roads.

Expand full comment
Dec 21, 2023Liked by Timothy B Lee

I think this is a "yes and" situation. Self-driving cars are a big plus, and public transportation and smart use of space will still be important. However, looking at the US, I expect our progress in those areas to continue to be a bit haphazard.

Expand full comment

Oh, I don't agree with that at all.

I agree parking (mandates and defenders of) is a huge issue. However, I don't agree it's the main thing preventing density. We devote an enormous amount of space to roads, which will be needed whether cars are autonomous, electric, or otherwise. If we remove *all* parking (an exaggeration of your argument to make a point), then AVs would have to be in continuous motion, increasing VMT; if we make AV parking too scarce, service time goes down and the possibility of AVs as human-driven car replacements goes down, too.

We would get far greater density, and far more benefits from it, by reducing the amount of road space devoted to cars at all, parked or otherwise, and instead shaping cities to support car-free life. That is the case for existing, dense cities with stronger support for car-free life (see https://www.nytimes.com/interactive/2022/12/13/climate/climate-footprint-map-neighborhood.html?unlocked_article_code=1.Hk0.71yw.MwyVjJSfoMfr&smid=url-share), and would be even more so if we went further than those cities have so far (the 15 minute city argument).

If AVs substantially replace human-driven cars, we're encouraging more sprawl by making it easier for people to live further from dense areas with lots of jobs. That drives up VMT, and requires more allocation of public space to support that sprawl (highways, bridges, etc etc) even if the AVs involved never park.

I'm a fan of congestion pricing, but your mention of it sounds like you're saying, "We should deploy AVs more broadly and then price them so only richer people can use them." If your appreciation of AVs is only about safety, I don't think that gets to your goal.

Expand full comment
author

You're assuming that the creation of AVs would force the creation of more roads, but I don't see why that would be true. A big obstacle to densification is that people in developed neighborhoods with low to medium density worry that allowing townhomes or apartment buildings will increase the number of drivers without increasing the number of parking spaces. So most cities have minimum parking rules that effectively make it illegal to increase density. And even when these rules aren't formally in place, there is often a discretionary review process that has the net effect of vetoing high-density projects in existing neighborhoods because existing residents rally to stop the project based partly on parking concerns.

If self-driving cars make taxis cheaper, then at the margin you'll see trips shift from personally-owned vehicles to taxis and that will reduce the demand for parking per resident. And this will make it politically easier to repeal minimum parking rules and for city governments to approve higher-density projects.

A good congestion pricing regime should only apply in areas where congestion is a significant problem, which in most cities is only a relatively small area near downtown. So an ideal AV+congestion pricing regime should make it cheaper and easier for people to take taxis outside the central business district, while giving people going to the CBD a financial incentive to take transit. You can use the congestion charge to improve the transit system, which should be a win-win for everyone.

I agree that at the margin self-driving cars will also increase demand for land in the outer reaches of large metro areas, since people will be willing to tolerate longer commutes. I think this won't be a problem once the vehicle fleet is electrified. But if you don't find that convincing the solution is to have a higher carbon tax, urban growth boundary, or some other policy to directly deal with the externalities involved, not to slow development of a technology that can save a lot of lives.

Expand full comment

I agree completely with your points on parking minimums and congestion pricing (specifically, that a well-crafted congestion pricing scheme can create good incentives, and also that it can be used to fund alternative modes).

On roads, I don’t know that either of us can confidently predict how AVs would change road demand—I’ll just say I worry about it a lot. I hope you’re right and I’m wrong.

I haven’t proposed slowing development of AV tech. I don’t think that’s a realistic approach; the technology is here and driving around my neighborhood today. Instead, I very much think policy discussions around permitting and funding of AV are worthwhile for every city (and some answers will differ by city). Questioning the future implications of these technologies is critical. The US is in a situation where we have allowed cars to dominate public space in harmful ways—including road violence, the focus of your post. You’re arguing that AVs can help on that problem, and it seems likely that’s true, but I am arguing that is not the only solution to that problem, nor is road violence the only problem with widespread use of cars. You could, for instance, dramatically reduce road violence by creating car-free areas or low-traffic neighborhoods, and do so without any technological developments at all.

If your response is, “We haven’t done that so far, why would we now?” I would reply, climate change should motivate us to look at much more drastic solutions than simply removing drivers from cars. Other cities around the world (e.g. Paris) have been doing that with great results. If *public funds* are going to go to addressing road violence, I have far more faith in those solutions than any form of car we could invent.

Expand full comment