Get the top HN stories in your inbox every day.
Animats
throwyawayyyy
One question I am genuinely wondering about is whether a self-driving car _is_ cheaper than a human driver, once all of the externalities are priced in. In SF right now a Waymo is typically priced a little under an Uber (actually quite a bit under if you count that no one has got around to asking for tips for AIs yet). I am sure the running costs of each Waymo vastly exceed the costs of a human driver to Uber...
Animats
Good question. We know that Waymo has about one remote operator/customer service rep per 40 vehicles. Waymo also operates vehicle garages for charging, cleaning, and maintenance. Those probably all add up to roughly what it costs a rental car company to operate a car. Maybe more, because there's more complex maintenance, maybe less, because they park themselves.
There's also a huge sunk R&D cost and an ongoing R&D cost that probably dwarfs operating costs. But the per-car cost drops as more cars are deployed.
On the other hand, robot vehicles can have higher utilization than single-owner vehicles. They can be on the road as long as there are customers. Observation of their parking lots indicates most of the cars are out on the road about 12 hours a day.
dangus
Here’s the other thing: taxi/rideshare driving can essentially function as a jobs program. It’s a basic job anyone can do without significant training. From a government planning perspective, you can’t just assume that all your displaced labor force can skill up.
What I noticed about China is that they employ a lot of people to stand around in nice looking government (non-police) uniforms and do various menial work or not much at all.
The US does this to a perhaps a lesser extent with jobs like TSA agents.
Sure, I guess you can do UBI, but what if that’s less efficient overall?
Example given with made up numbers:
Status quo, an Uber driver makes $20/hour out of a revenue of $50/hour total covering vehicle operating costs and platform fees.
Self-driving cars: self-driving cars cost $40/hour to operate, UBI pays someone a wage of $20/hour since there’s no job available. This basically means that rideshares now cost $10/hour more to operate than before.
Or, maybe that person on UBI makes $10/hour instead of $20/hour and gets a worse job to cover the difference.
Obviously there are many flaws and assumptions with the way I present this scenario but it’s a really good question to bring up whether putting everyone out of work is actually going to be a net positive.
Regarding what you said about driver hours, it’s not unheard of to run multiple drivers on multiple shifts with the same vehicle. Not all rideshare drivers own the vehicle nor use it as a personal vehicle. But the other factor is that the drivers who do use personal vehicles effectively subsidize the fact that they can only drive it for a human-length shift. Waymo has to buy every car (more expensive than a normal car) and use it only for business purposes while an uber driver can just use the same used Toyota Prius they use to take their kids to soccer practice.
ieiee
Context is super important.
Unlike web services that giants like google provide (e.g search), waymo and other AVs essentially cannot fail. Like at all. It is suspectible to ‘randomness’ of nature that can be the difference between life and death.
A lot of so called ‘smart’ people are going to find themselves getting humbled by the real world.
Humans are able to make sense of the world around them through things like intuition. Machines do not possess this characteristic.
chung8123
Data should help transient situations as well.
svnt
> Google StreetView data is probably sufficient.
This is an extraordinary claim. Self driving cars just need 15 ft grid panoramic images that are months or years stale? What experience are you basing this claim on?
aaron695
[dead]
ra7
> The insight driving the program, Naga said, is that the limiting factor for AV development is no longer the underlying technology. “The bottleneck is data,” he said. “[Companies like Waymo] need to go around and collect the data, collect different scenarios. You may be able to say: in San Francisco, ‘At this school intersection, I want some data at this time of day so I can train my models.’ The problem for all these companies is access to that data, because they don’t have the capital to deploy the cars and go collect all this information.”
You can’t be the CTO of Uber wanting to do AVs, and get the data collection requirement shockingly wrong.
Waymo’s bottleneck has never been data. When they want data about a school intersection in SF at a certain time of day, they just... synthetically generate it and simulate: https://waymo.com/blog/2026/02/the-waymo-world-model-a-new-f...
Waymo is able to deploy with less (but targeted and high quality) data collection by having world class simulation capabilities. Not that they haven't collected huge amounts of data as it's no doubt important (I've heard their onboard storage is transferred and emptied every few days), it's just not a bottleneck. They have the most efficient operation in the AV industry.
The best example of why data collection isn’t the bottleneck is Tesla. They boast about billions of miles of data, yet they’re struggling to put out fully autonomous vehicles.
simmonmt
> When they want data about a school intersection in SF at a certain time of day, they just... synthetically generate it and simulate
I think it's more about detecting changes to the world. You need boots on the ground, so to speak, to see that new speed limit sign or the new lane paint. The Waymo vehicle can no doubt react to changes in the world when it encounters them, relaying them back to the mothership, but it's better to know about them in advance.
ra7
Most AVs, definitely Waymo vehicles, are self mapping. They can detect environment changes and relay it to the entire fleet. That's because they map using the same vehicles as the fleet.
delfinom
>You need boots on the ground, so to speak, to see that new speed limit sign or the new lane paint.
It'll shock you to know that you can simply get this from governments, some even provide this in API form
dmd
It probably won't shock you to know that those sources of data can be months to even years delayed from what's actually out in the world.
KaiserPro
no visual data, you need picture data for that. companies like NC tech do it for like $1m a city. or thereabouts.
paganel
> or the new lane paint.
I'd be surprised if this is a thing outside the biggest US (and European, for that matter) cities, judging from Google StreetView there are lots of streets in US cities/towns with almost no paint lines at all.
MagicMoonlight
That’s dumb then. It shows it’s just brute force rather than AI.
A human doesn’t need to be shown every single road that exists in order to drive.
simmonmt
That's true, but the human can do a much better job planning for the journey if they know what to expect along the way.
One example, from the end of the journey: knowing in advance where the actual entrance to the business is, or the specific curb cut that leads to the residence, makes it easier and far less error prone to decide exactly where the journey should end. Even humans have a hard time figuring out the right access point for a business or residence. This is a job for an offline process, fed by as many data sources as possible.
ThunderSizzle
Just a bunch of sophisticated if statements, I guess.
suddenexample
Yeah I'm not so sure this CTO is on the mark here, but to be fair, I do think some of this IRL long tail/edge case data is important for Waymo. The simulation software is super interesting to me - the real world can be so chaotic, and even if they could generate every possible real life case, there needs to be validation on whether the Waymo driver is responding in the optimal way. They certainly haven't solved this problem, you can see some of their growing pains in all of these articles - floods in Austin, more and more interactions with emergency vehicles that first responders seem to believe are getting worse, etc.
Tesla on the other hand has billions of miles of data, yet because there is a limit to camera-only techniques, that data isn't that useful is it? They have no ground truth data to evaluate their camera system on, which is why sometimes you see those Teslas driving around with lidar rigs mounted on them. Going camera-only is just asking for trouble.
ra7
I agree real world data is important for Waymo. I didn't mean to say it wasn't, so I've edited my comment to reflect that. It's just that data is not some magic bullet to achieve self driving like Tesla and others suggest.
Of course, Waymo still has much more room for improvement. But it's much more efficient to supplement less but higher quality IRL data with large amounts of synthetic data, than to run a million data collection vehicles 24x7 because most IRL data is boring and useless.
Waymo said 6 years ago they simulate 20 million miles every single day [1]. Clearly, it's working for them given their scale of deployment right now.
[1] https://waymo.com/blog/2020/04/off-road-but-not-offline--sim...
skybrian
Although most of the real-world data is probably boring, collecting more of it likely makes discovering rare edge cases more likely. But since they happen rarely, I imagine that after discovering them, they would then need to figure out how to simulate them.
KaiserPro
> The best example of why data collection isn’t the bottleneck is Tesla.
Exactly. plus any delivery company/dashcam company can provide a bunch of data where ever there is any sizeable population.
About 8 years ago, that data would have been really valuable, but at best its nice to have.
the only thing that is valuable is the breadth of different cars, but even then its not that much of a differentiator.
Sardtok
The biggest difference, is Uber has vehicles around the world. So there's more data from countries with different rules from the US. Signage is definitely different between the US and Europe.
iugtmkbdfil834
I.. am amused by the confidence on display, but I can't say that I am not concerned that people are confidently stating that real world data is not useful, because it can be just simulated. One would think that, by now at least, we know that simulation is at best an imperfect copy.
And I don't like the idea of even more data being harvested and used.. I just find the dismissal.. odd.
ra7
“Real world data is not the bottleneck” != “Real world data is not useful”
No one is suggesting the latter.
iugtmkbdfil834
Parent's post noted that it is not a bottleneck, because it can be readily simulated ( and thus not useful ). I am not sure if QED is too much in this case, but I stand by my amusement. Or are you arguing that real world data is somehow less useful than simulated data? It is very confusing. I would accuse of nitpicking, but I just noticed you are the parent:D You can certainly speak for yourself.
cogman10
> The best example of why data collection isn’t the bottleneck is Tesla. They boast about billions of miles of data, yet they’re struggling to put out fully autonomous vehicles.
Well, TBF, the tesla data was complete garbage with earlier vehicles. They had cheap and somewhat bad cameras in the earlier vehicles that was only somewhat recently updated. And even then, I don't think Tesla is at the end of their hardware journey. I think they don't think that either, which is why they've gone to a subscription only model for self driving vehicles.
Waymo, on the other hand, has gathered less data, but more high quality data. They do the expensive mapping of a city which is a big part of why their vehicles have early on been able to do some pretty impressive feats. The drawback is getting that high quality data takes a lot of time and resources.
kibwen
> And even then, I don't think Tesla is at the end of their hardware journey.
I dunno about that. Tesla seems completely adrift, pretending to pivot with random forays into humanoid robotics or whatever, to the point that I wouldn't be surprised if they exited the consumer vehicle space altogether within the next decade. They have no answer for Chinese competitors.
twobitshifter
I recently watched some videos related to the production of cybercab, which has now started public testing. They’ve still done some great engineering, to the point that the car is now assembled like a matchbox car. All the drive components are contained in a single package for a FWD configuration that the body just drops down on. The car now has no controls besides the screen and door pulls. The materials are all lower cost and they even found a way to skip painting the cars. All of this should help them cut costs significantly.
As far as the self driving, they may be far off still, it’s hard for me to get a read on that and this vehicle is a bet that they will be able to achieve it - right down to the braille in the cabin, so maybe that’s why they still fail. The thing I will say is that despite the PR disaster that the CEO is, which gives us that feeling that the company has lost its mind, it seems they are still quietly doing some advanced engineering.
cogman10
Well, let me rephrase, the previous stated goals of Tesla around self driving cars isn't complete with the current hardware.
gcheong
Didn't they need the data from the 200 million miles or so from actual driving before they could get to the generative model though? Data isn't everything, as you point out with Telsa (mainly because they decided to forego using lidar it would seem), but it is pretty fundamental.
ra7
IIRC, they had clocked 20 million real world miles before starting to scale their deployment. But they were also driving 20 million miles in the simulator every day: https://waymo.com/blog/2020/04/off-road-but-not-offline--sim...
ninjagoo
> before they could get to the generative model though?
Is that the right kind of model for this particular application?
whiplash451
Waymo might very well be missing specific kinds of data (e.g more incidents/accidents, near-collisions etc)
Also, Uber’s data might be useful for eval, not training (e.g « here is how Waymo would behave vs human drivers therefore it is safer »)
ra7
> Waymo might very well be missing specific kinds of data (e.g more incidents/accidents, near-collisions etc)
Accidents and near-collisions are exactly the kind of scenarios perfect for simulation. You don't test them out in the real world and risk injuries/deaths. You need to have confidence they're handled before you deploy.
pishpash
Again, how do you know you've handled it correctly without ground truth? Simulation without ground truth is a garbage in garbage out situation.
nerdsniper
I feel like they should have done this 6 years ago. Most AV companies already have tons of their own data today. But how would it work to install expensive LIDAR sensors on privately-owned vehicles?
Rebelgecko
FWIW, a large fraction of Uber drivers aren't actually driving their own personal cars, at least around me nowadays. They're either rented or some sort of fleet vehicle (complete with TCP #)
themanmaran
Exactly my take as well. This would have been the right diversification move a decade ago.
Uber did invest early in self driving back in 2015, but in 2018 there was a fatality which pretty much deleted their whole program. And looks like it's taken them way too long to try picking it back up.
mohsen1
I was working at Lyft 8 years ago and suggested this to the head of AV program then. They didn't listen.
reaperducer
Most AV companies already have tons of their own data today.
Real-world data spoils faster than a gas station banana.
If your AV company is relying on data from six years ago, you're going to kill someone.
samagragune
"Our goal is not to make money out of this data" is doing a lot of heavy lifting for a company that just committed $10B to robotaxis and is taking equity in the same AV companies it would be supplying. The actually interesting part isn't the sensor grid, that's years away and has real consent, compensation, and regulatory problems nobody's talking about. It's shadow mode: letting AV companies sim-run their models against millions of real Uber trips without putting a car on the road. That works today. That's the product. The sensor grid is the press release. Shadow mode is the business. Also completely absent from this article: do the drivers whose cars become "rolling data collection platforms" get anything? A cut? A notification? A commemorative badge? Uber has a rich history of finding creative ways to extract value from its driver network, so I'm sure they've thought carefully about this.
jdeibele
I'm old. Was anyone else's reaction to wonder what Uber was doing for audio-video companies?
The original title says "self-driving" and that's much more clear.
rjmunro
Sometimes AV is supposed to mean Anti Virus or Alternative Vote and that's really confusing because it really means Audio Visual. Anything else, no.
I saw the title and thought it can't be AV, they must mean AI and made a typo.
brendoelfrendo
Immediately after leaving this thread I saw a post on Bluesky where someone was discussing the GUARD act and used AV to mean "age verification." It's out of control!
darknavi
I was also picturing Uber drivers with a bundle of composite video cables in their hands.
philipov
AV obviously stands for Adult Video.
xp84
Yeah apparently JAV is a Toyota that drives itself now
whynotmaybe
Rumor has it that some adults video are filmed in a taxi so I guess it figures.
nickvec
Sorry, having “self-driving” in the title went over the HN title char limit, so I opted for AV (autonomous vehicles) instead.
undefined
darth_avocado
AV stands for anti vehicle.
ghstinda
Uber really did a lot of damage to the NYC's energy and personality, i won't miss them when they're gone. Both the delivery app and terrible ride share. The chinatown car services are still close to half the price of uber for those that don't know.
JumpCrisscross
How useful are these generic sensor inputs for AVs? Like, how much more valuable is a Waymo’s data for a Waymo than something Uber collects?
33MHz-i486
I remember Travis Kalanik spouting the talking points about self-driving in 2017, that after Waymo, Tesla had the advantage because they had the best data, that they were going to crack self-driving soon. Then I remember Dara scuttling Uber’s entire self driving division in 2019.
Self-driving is possible but it requires a massive sustained investment in custom hardware on the car, in real and simulation testing, in painstaking software developlment covering tens of thousands of scenarios, realtime remote control failsafes, fleet management capabilities in every city. Waymo is the only company that comes close to the right approach. All these other Elons, GM, Uber CEOs are just jangling shiny objects in front of investors. A moonshot on the financial model for what are otherwise mature stagnant businesses.
zitterbewegung
Isn’t this a pivot I always thought Uber wanted to automate their whole fleet instead of having to pay people ?
bastawhiz
Uber was working on self driving ten years ago. They had cars on the road loaded with cameras and sensors specifically to collect data. Then they negligently killed a woman crossing a street.
This isn't a pivot, this is them trying to sheepishly reenter the race they were dramatically ejected from.
dmix
The main reason Uber sold their self driving R&D unit was because they couldn't afford it. So they sold it to another company taking a 25% stake in Aurora and Uber CEO joined their board, the company is still operating automated trucking https://en.wikipedia.org/wiki/Aurora_Innovation?wprov=sfti1
They run trucks for Fedex in Texas and wants to offer an "Uber Freight network"
le-mark
This exactly; self driving was a large part of their valuation iirc.
mrweasel
The part I don't fully understand is what leverage this gives Uber over anyone else? Uber doesn't have the fleet management, mechanics, cleaners or even storage facilities. They do have the most used taxi app, but that seems like a very small edge.
There's nothing stopping the car makers from running their own taxi service and they already have networks of mechanics and cleaner as well as some level of storage. They'd need to scale up, but they don't need to start from zero.
Ubers success is in large part build on not having to own AND MANAGE their cars. With self-driving cars that advantage disappears, unless they're gaming that "drivers" will buy the cars and lease it to them.
LeoPanthera
I'm honestly surprised that Tesla never took advantage of all the cameras in all its cars to do some kind of mapping project. I always thought that was incredibly valuable data. Sort of an automatically crowd-sourced street view.
JumpCrisscross
> surprised that Tesla never took advantage of all the cameras in all its cars to do some kind of mapping
Don’t they [1][2]?
[1] https://www.privacyinternational.org/examples/1929/tesla-lea...
[2] https://electrek.co/2020/10/24/tesla-collecting-insane-amoun...
Animats
How much cellular uplink data does a Tesla vehicle transmit per minute? Anyone know? If they're collecting video it would have to be a lot.
rjmunro
Some people would tell me that they do, but only for training their internal self-driving AI.
I'm not sure about the privacy implications. You say "all its cars" but you actually mean "all its customers cars". The relationship between Uber and the cars/drivers is fairly different.
undefined
Get the top HN stories in your inbox every day.
> “The bottleneck is data.”
This seems to be wishful thinking on the part of Uber, and also Tesla. Google StreetView data is probably sufficient. Waymo's expansion into new cities does not seem to be delayed much by the need for more data.
Most of the reported problems with self-driving come from transient situations. More mapping data will not help with those.
China has the Beijing High-level Autonomous Driving Demonstration Zone, where traffic cams and other sensors let vehicles see beyond their own sightlines.[1] That's been going on since 2020. That's the ultimate in sensing - full real time road info.
The Beijing test area is getting some expansion. The new direction seems to be to focus on airports and railroad stations, so that driverless cars can be aware of congestion in detail. That makes sense.
[1] https://sinocities.substack.com/p/inside-chinas-connected-ve...