New Study: Waymo is reducing serious crashes and making streets safer

It will only be cheaper if you don't drive much anyway, and you are the type who would never be seen in a car more than 3 years old.

If you drive a lot (like the person in the countryside) the car that is there when you want to is worth owning vs a shared car that you might have to wait for. Plus by owning the car you can just leave your golf clubs in the trunk.

If you can stand being in a used car you will discover that shared cars are all more expensive just because at the first sign of cosmetic wear they get rid of it while seats that have been sat in a few times are still good enough for many more years. (unless you almost never drive anyway)

Because of the above I don't see much growth in the shared car market. There will be some because there are people who don't have parking, people who don't drive much, and people who demand a new car that they don't otherwise care about. However the vast majority of people will still own their own car.

Cost is only one dimension of renting a car. As long as the car rental companies keep making it as painful as possible, it won't be a solution for usual usage. Hopefully Waymo goes after them too.

Even weekend vacation homes will go up in value. It's no big deal to have a vacation home on a lake 10 hours drive away if I can sleep in the car overnight on Friday and Sunday.

You've clearly never been to the US.. Late-stage capitalism will find ways to make it cost exactly as much as driving your own car.

OTOH, if I'm in a decent-sized car (minivan?) for 45 minutes+, I can get work done. I can then stay less time at work.

> congestion pricing will become the norm.

not with the current set of voters it wont

Probably not, if they're all computer-controlled, and can communicate with each other. I posit traffic jams is mostly caused by idiot and reckless drivers...

Well sure but drones won't shit, that's why we need the organic piece. I guess they could drop rotten fruit in lieu of shit, but then we need a supply chain to restock the rotten fruit in the drones.

It's so frustrating. These things are trivially solved. There's basically a 50/50 shot, every time the light cycles, that someone will illegally take a right on red on the street outside my house. All you need is a single cop sitting there and watching. Or just one camera! Argh.

But they won't be implemented with a privacy preserving architecture. They'll be outsourced to a third party with unknown privacy and security, and eventually be treated as a revenue generator, leading cities to implement rule changes that enhance revenue at the cost of privacy and safety.

Health insurance companies should pay for it. Their costs would come down if they subsidize the full R&D cost of this system for all car manufacturers. It would work in their favor.

They're probably too stupid to think like that, though.

They won't have to if they use sliding power doors (the geely cars have them). They should have had them for the jaguar generation :)

Dooring people aside, what do you do if someone just leaves the door open when they leave their ride?!

It's already there, my new fancy car has it. Push the lever to open with electronic help, pull the lever twice for mechanical release. The electronic help version checks for safety first (as long as you do it with a timeout from when the car was running/ready) We'll have to see how the fancy car does over time, but I did get one with handles on the outside that don't disappear.

If "the Googs" knows so much about me, why do they keep showing me ads for products I not only have zero interest in, I'm also not even the target demographic for?

When I met my ex and we linked over what was then Google+, we found that I had been auto labeled in a photo of a protest from years before we met. They've got a lot of info that they don't surface...

True; I should have said metadata and not just data since you're right that the volume of raw images would be too big to store indefinitely. It's way more feasible to process the raw images and store the inferences, like number of persons visible in last 5 seconds, or dates and times a person who looks like me has been seen by a Waymo while my particular Android phone is nearby, or dates and times they have seen [my OCRed car number plate].

Flock is the same way. For example here's the Flock privacy policy from one of SF's fine local shopping centers: https://www.stonestowngalleria.com/en/visit/lpr-privacy-poli...

> Video Clips captured by the LPR system will automatically be deleted after 30 days; although Images are deleted when no longer needed, the data obtained from the Images may be retained indefinitely. Should any information from the LPR Dashboard be needed to assist with a security or law enforcement matter, it may be retained indefinitely, in paper and electronic form, as part of the security file until it is determined it is no longer needed; in addition, it may be shared with local law enforcement who may retain it in accordance with their own retention policy.

If anyone can share a link to a similar IRL privacy policy for Waymo I would love to read it. The one on their website is conspicuously labeled Waymo Web Privacy Policy lol

It's Google we're talking about. In no way do I trust them to take pictures of me using their city-wide camera network and not use face/body recognition to keep track of where I go, for the purposes of targeting advertisement.

> But car crashes are the third highest cause of death in the US (https://www.cdc.gov/nchs/fastats/leading-causes-of-death.htm).

No, that says “Accidents (unintentional injuries)” as a category are collectively the third leading cause of death, and that category contains a lot of things.

CDC “Underlying Cause of Death” dataset sez… https://wonder.cdc.gov/ucd-icd10-expanded.html https://i.imgur.com/4PB0xyC.jpeg

- “Person injured in unspecified motor-vehicle accident, traffic” is the 50th leading cause of death at 0.4% of deaths.

- “Person injured in collision between other specified motor vehicles (traffic)” is the 108th leading cause of death at 0.2% of deaths.

I will be sure to think of that the next time I see a human driver whose car has 29 dash cams, 5 lidars, 4 radars, and is using four H100s to process all of its realtime images of me: https://thelastdriverlicenseholder.com/2024/10/27/waymos-5-6...

What bothers me regarding surveillance and self driving cars is that an executive sympathetic to the surveillance state could build a system that allows arbitrary surveillance of vehicles or housing by license plate or location. Eric Schmidt was a regular visitor of The Pentagon and Billionaires simply live in a different world than we do. So while some driver could happen to capture me and upload it to YouTube, Waymo could, if someone wanted, have a secret operations center which allows surveillance of all sorts of people, locations, and vehicles. The same way that AT&T had a secret NSA closet that split off major fiber pipes, some data pipeline could have an invisible filter that duplicates data matching certain variables and delivers it to a surveillance partner.

> Scale that to 10s of millions of cars.

Don’t they currently have less than 1000 cars? I don’t think they will keep every recording forever but technically, they still could at the current scale.

And not just with law enforcement but internally for memes.

Point is, there is flexibility in how different brands implement it. I think it will be the same as eco mode / sports mode / track mode.

On the other hand, I wonder why insurance companies haven't led to the ubiquity of dashcams. I thought by now every vehicle sold would have one built in.

And my suspicion is that insurance companies don't push for you to get one because it prevents them from fighting claims that they would've won had there been no evidence.

Maybe it's similar for self-driving or whatever we're talking about here (sensors?).

120 traffic deaths per day in the US - not sure where you’re getting your data, but you’re off by more than an order of magnitude

>Then I found out in 2019, the Texas legislature outlawed the use of red-light cameras.

>Muh freedum.

To be fair you have progressives arguing against red light cameras as well, on the basis that it's a regressive tax on poor people, and that it causes more accidents through drivers slamming on the brakes on a yellow.

Maine does not have red light cameras and this isn't a problem.

The variability of how much people crash in various parts of the US is fascinating, like at least one entire order of magnitude difference in otherwise comparable cities.

So the answer is that we clearly need that car also to be a Waymo. Problem solved! Hah.

> they've got their settings tuned in favour of safety stats is the whole point

For now, yes. My point is that there's very often big gap between "how safely does it work in a lab when the people running it are trying to play up its safety" versus "how safely will X actually work once we start using it everywhere."

Manually-driven vehicles could be a lot safer if they were being prototyped under strict guidance as well!

If we want self-driving cars to retain the same safety later, there needs to be something which prevents humans from flicking the safety-versus-speed dial a little bit over and over in order to make quarterly earnings projections.

Yep that was it, an even bigger gap than I remembered. Thanks for linking it.

Waymo ( and self driving programs as well) have been careful not to go for public large scale deployments in any city with difficult weather and for good reason focused on cities like Austin, Phoenix, Los Angeles, SF[1] so far with easy driving weather.

There have been promising progress and there have been hints of a New York trial soon, but it it well known that self driving cars have not done large scale trial in cities with bad weather.

[1] Yes, I am aware SF gets a bit of bad weather with fog and rain but not nearly not as much to make driving quite unsafe like somewhere that gets a feet of snow in 24 hours in winter, and likely promixity to engineering HQs and favourable regulatory climate influenced the SF choice.

> 4.6 cab crashes, 3.7 livery car crashes, and 6.7 crashes with private cars.

> Waymo 2.1 crashes

The numbers become much less 80+% plus claim in the article as you remove factors. It comes closer to 30% with professional drivers.

Livery car is still not always well maintained a high sitting SUV with better visibility[1], perhaps with all these factors included if it is 20% better it is impressive technical achievement for sure, but not going to create headlines anywhere.

The point is the methodology is not as objective as it could be, and this is biased/selective claim, not that self driving cannot be better than humans.

[1] Also there is major difference in the price point between Waymo and Livery cars, I cannot say how it will influence rates but the different rates means different class of clients using at different times of day/night to different locations that needs to be normalized for.

Not the same cars though, people are not driving newish mid range SUV with professional maintenance all the time.

Also there is some sort of bias not accounted for: People drive when most people drive and most people are stuck in the most dangerous area: traffic. Waymo driving at night on empty streets is not a good indicator for accident prevention when measured against the average human, who is stuck mostly in traffic.

It may not be option to most people in the US (most self driving use case is built around the US market)

In America driving is a economic necessity, from going to work to even the grocery shop needs cars and dependency increases inversely with affluence [1]

Mass public transit is non existent barring very few regions.

So car (for commute) and flight (for long distance) are the only two viable transit options .

People cannot choose to not work because weather is bad, and remote work / work from home applies to only some jobs.

[1] food and other service deserts are more likely less affluent neighborhoods meaning you will need to drive and for longer for food , pharmacy or any other services if you are low income .

Higher willingness to pay does not matter unless the market is segmented. So if the tech gets cheaper, unless they explicitly make an expensive "oft-cleaned" tier and a less expensive "less-oft-cleaned" tier, what matters is average willingness to pay.

"Once the tech becomes cheap, expect the car quality and cleanliness to go down" -- here you were positing that the _average_ willingness to pay for cleanliness will go down enough to affect things, barring any market segmentation. And I disagree:

1. You're assuming that the reason _why_ Waymo's cars tend to be cleaner than your typical Uber / Lyft is to satisfy a wealthy clientele. This elides a big reason for the existing gap: Uber / Lyft drivers aren't professionally managed. You don't directly pay for your Uber driver's interior car cleaning when you buy a ride, but the salary of folks managing Waymo's fleets is factored directly into pricing. Even if Waymo's clientele were less wealthy, you have to clean your cars and pass the cost on to all users. Additionally, interior cameras are pretty motivating to not mess up cars!

2. You're assuming that the cars today don't get maximum utilization, and that with more utilization you'd see dirtier cars. This is a pretty bad assumption - the cars are being utilized about as heavily as you can hope. In SF for most of Waymo's existence demand has outstripped supply. And the cars are still very clean :)

So if the usage is the same, and peoples' expectations for cleanliness are the same, why would rate of cleaning change as time goes on for the service?

The only thing I can think is if no reasonable alternatives to Waymo arise - in that case, cleanliness could go down but it has less to do with the clientele / willingness to pay, and more to do with competition / monopoly.

As another note, I just don't see how cleaning-based market segmentation would make good operational sense. Is cleaning the car slightly less frequently really gonna help the bottom line? Is the price differential big enough at single-ride scale? Do even rental car companies do this for their fleets - the cheap ones still seem to clean their cars.

I totally agree that grade-separated bike and pedestrian lanes would be better. What I’m arguing against is that this behaviour by Waymo is less safe than San Francisco’s human drivers given the rules as they are.

Merging into the bike lane before turning right is safer for cyclists. Drivers are supposed to do shoulder checks when changing lanes (I mean they're also supposed to do them when turning but you know) so they likely won't hit a cyclist when merging into the lane.

Once the car is in the bike lane, any bike going straight is forced to remain safely behind the car until the car completes its turn.

That is what happens now. Most drivers do not cause accidents from speed most of the time. This is a red herring.

The speed limits on most residential roads are too low from a variety of factors, including tradition, policies, and desire for revenue or to generate PC to pull over drivers.

You don't drive 25 on residential roads, because you know this to be true. Neither do I, nor does anyone else.

It is also the case that Waymo will be dramatically better than all humans in ice because it is going to take the aviation approach and stay in the depot, rather than fooling itself into believing it is competent at driving on ice.

You still need to be in neutral as the ABS/stability control doesn't control the engine. 2wd makes no real difference other than rear wheel drive tends to be the most susceptible to this issue, but any wheel can lose traction.

I think it's often about when you're on rural single lane roads and perfectly bad ice/snow/slush conditions where you need to keep your speed very low, and sometimes even very moderate braking can cause your vehicle to veer off the road and into a snow bank, not necessarily the need to avoid something in front of you

They mean go into neutral and lightly brake. In a 2WD car, the braking force applies more strongly to the non-powered wheels. Since those were traditionally the back wheels, this meant when going down a hill the front wheels would have more traction. Those are also your steering wheels, which means them locking up is bad. Again, this is the sort of advice that is germane in highly restricted environments which become folk knowledge and later mis applied by humans in ways that reduce safety.

(You also only get into this scenario when your stopping distance is shorter than your reaction time and perception length. Something automated drivers can manage better than humans.)

I remember being taught the neutral trick for emergency braking back in the 90's, and it had nothing to do with traction in poor conditions. It was simply to remove any engine power that might extend your braking distance. It's definitely bad advice in any modern car.

When you let off the throttle your wheels start driving the engine and slowing you down. In snow/ice that engine breaking alone can be more braking force than is safe and so you go into a skid. Shifting to neutral removes engine braking and allows for more controlled slowing down.

I think the OP means in a manual transmission. Auto will decouple at low speeds / engine braking won't be as severe. In a manual car, in 1st gear, I could possibly see it.

As someone who grew up BEFORE ABS, drove in the winter (in Canada), including first winter owning my own car with sport tires because I couldn't afford winter tires, spun / slid a few times even with top-of-the-line winter tires, etc.

ABS is a game changer in the snow. I used to go to an empty parking lot every winter during early snowfalls to play around and skid, start/stop, etc. Even EARLY ABS ('94 VW) means that 98% of the time (IMHO), the answer even in snow/ice is "slam on the brakes". Sure, you might have a few percent longer stopping distance than an expert who can do threshold braking - are you an expert? And the fact that you don't lose control of the steering is a huge advantage.

If you are in a situation where your car is sliding, like down a slick hill. You dont hit breaks, you put the car into neutral and steer into the way your car is sliding to try and keep the car on the road.

I am in upstate new york, unless the situation is extremely bad, just dont drive isnt a valid answer for 5 months a year.

Almost all of these tricky situations are avoided by slowing down and giving more space around hazards.

what about when there's an idiot slightly ahead and in the next lane who decides to randomly swerve into my lane?

do you also maintain a long following distance when there's a car right next to you in the next lane? I try to, because I don't want to stay in someone's blind spot, but sometimes it's not really possible to fall back.

Interesting!

When I was much younger, I worked for a couple companies that had (what I would consider) large fleets of vehicles, and they all were insured through an insurance company. I guess I just assumed that's how it was. I wasn't aware self-insuring was a possibility. Thanks.

I don't think Waymo needs a significant amount of vehicle support staff now, and will need even less (per vehicle) in the future. The ratio will probably be something like the number of elevator repair people to number of unattended elevators operating smoothly.

I emphasized that Waymo staff does not drive vehicles remotely, because this persists as a common misconception.

> If you are basing something on 10 incidents but it's 50% of the total, it's still terrible accuracy.

The ratio of 3.2 trillion to 56.7 million, which is already incredibly generous to Waymo's position, is 5 orders of magnitude in difference. So any calculations from Waymos data are going to be insanely inaccurate and not something you can extrapolate from.

The main, and most obvious case, evidenced by this, is Waymo does not operate where snow falls. Human beings do.

We're missing so much of the picture I don't think you can say Waymo's are 75% less accident prone, or 80% less likely to hit a pedestrian. Those are just nonsense numbers.

This! (Thank you for the comment). There’s a reason a 1000 random samples is adequate to reasonably estimate what’s common metrics in a population the size of USA or India (or infinitely large).

It reminds me of a large sample in the Reader’s digest poll: https://en.wikipedia.org/wiki/The_Literary_Digest#Presidenti...

size along is not enough. The sample may be biased and it certainly is in Waymo case e.g., roads/vehicle conditions.

If an only if every drunk person decides to take one instead of driving themselves home.

Drunk people being known for having exceptionally poor judgement and self awareness.

It suggests that they _could_ cut fatalities by that much. Then again, a whole new mode of accident, where the inebriated decide to step out of a moving vehicle and injure themselves that way.

This is a dynamic system where human decisions are never fully removed from the loop.

Our safety research team is interested in this topic, too! In a previous study, they've tried to model it:

> Building on that, the Collision Avoidance Benchmarking paper presents a novel methodology to evaluate how well autonomous driving systems avoid crashes. The study, which to our knowledge is the first of its kind, introduces a reference model that represents an ideal human state for driving—the response time and evasive action of a human driver that is non-impaired, with eyes always on the conflict (NIEON). Put simply, unlike an average human driver, NIEON is always attentive and doesn’t get distracted or fatigued¹. The data showed that the Waymo Driver outperformed the NIEON human driver model by avoiding more collisions and mitigating serious injury risk in simulated fatal crash scenarios.

(From https://waymo.com/blog/2022/09/benchmarking-av-safety)

AIUI (I'm not on that team), a major challenge is getting good baseline data. Collision reports may not (reliably) capture that kind of data, and it's clearly subjective or often self-reported outside of cases like DUI charges.

I think I was pretty clear.

> You do not get to counter-argue: “What matters is the rate of cases per mile driven” without actually presenting that number with supporting evidence. Otherwise the only sound conclusion is the default presumption of non-safety.

I then pointed out how Waymo does present such evidence. But, if you applied that argument to Cruise you would be wrong. That demonstrates how that argument (when not presenting the numbers) can be used to support both good and bad and is thus a bad argument.

The correct argument when somebody points to anecdotes of bad outcomes is to present statistically sound data of good outcomes, not argue they did not present statistically sound data of bad outcomes thus you get to assume it is good.

I think I was quite clear on my position.

> In the case of Waymo, we have some tentative supporting evidence from this and other studies Waymo has run. However, that is still insufficient, even ignoring the lack of audits by non-conflicted parties, to strongly conclude Waymo is safer than a human. The evidence is promising, but it is only prudent to wait for further confirmation.

You are not making a distinction between concluding unsafe and not being able to conclude safe. It is standard practice to not presume safety and that positive evidence of safety must be presented. Failure to demonstrate statistically sound evidence of danger is not proof of safety. Failure to disprove X is not proof of X. This is a very important point to avoid fallacious conclusions on these matters.

To discuss your specific points. Yes, the data is promising, but it is insufficient.

Traffic fatalities occur on the order of 1 per 60-80 million miles. Waymo has yet to reach even one expected traffic fatality yet. They appear to be on track to doing better, but there is not enough data yet.

The reports Waymo present are authored by Waymo. Even the Swiss Re study is in cooperation with Swiss Re, not a independent study by Swiss Re. The studies are fairly transparent, they point to various public datasets, there are fairly extensive public reporting requirements, and Waymo has not demonstrated clear malfeasance, so we can tentatively assume they are “honest”. But we have plenty of examples of bad actors such as Cruise, cigarette companies, VW , etc. who have done end-runs around these types of basic safeguards.

Waymo operational domain is not equivalent to standard human operational domain. They attempt to account for this in their studies, but it is a fairly complex topic with poor public datasets (which is why they cooperated with Swiss Re) so the correctness of their analysis has not been borne out yet. When Waymo incorporates freeways into their public offerings this will enable a less complicated analysis which would lend greater confidence to their conclusions.

Waymo is still in “testing”. As their processes appear to be good, we should assume that their testing procedures are safer than should be expected out of actual deployment or verification procedures. That is not a negative statement. In fact, it would be problematic if their “testing” procedures were less or even equal in safety to their deployment procedures. That is just how testing is. You can and must apply more scrutiny to incomplete systems in use and prevent increased risks especially while under scrutiny otherwise you are almost certainly going to be worse off in deployment where there is less scrutiny. We have yet to see how this will translate out to deployment, so we will need to wait and see if safety while under test will appropriately apply to safety while in release. This is analogous to improved outcomes for patients in medical studies even if they are given the placebo because they just get more care in general while in the study.

Anyways, Waymo appears to be doing as well and honestly as can be determined by a third party observer. I am optimistic about their data and outcomes, but it is only prudent to avoid over-optimism in safety-critical systems and not accept lazy evidence or arguments. High standards are the only way to safety.

Key point: the root cause investigation is not driven by the company involved. The FAA and the NTSB investigate and can by law demand documentation from the company.

My intent was to have some real data. Do you only write comments to persuade others? Form your own opinion, because now you have data to inform one.