Zoox (Amazon Subsidiary) should be rolling out soon. They've already done successful test rides and should be launching in Las Vegas next year.

Saw at least one zoox scooting around Vegas last week at Reinvent

Saw one driving around SF recently too. Not the SUVs they've been testing here for years – but the full custom-built cube-shaped passenger taxi.

> Who else is even close to being able to provide it?

Baidu/Apollo, maybe. There are deployed robo-taxis in China. Reviews of driving quality indicate "meh", they are sometimes directly teleoperated to get out of problems, and difficult intersections have fixed cameras the cars can see through.

From what I've read China is forging ahead very successfully, but I assume there would be a lot of friction if they tried to enter the US market, to say the least.

Reading from intersection-mounted fixed cameras seems like a valid solution.

Has the issue with all cameras, which is that they don't work when they can't see.

Of course, if they're actually a more general kind of sensor but called a camera then it's fine.

I don't see how Waymo is going to survive the onslaught that is CyberCab. Cheaper to build and operate, more efficient, much faster to build, and deployable across all cities without any manual configuration. The clock is ticking on Waymo, they have 2 years to figure out how to compete otherwise they're cooked.

It's very similar to legacy space and internet, the existing players saw SpaceX coming years out, did nothing and got crushed.

Isn’t it kinda the other way around? I feel like the clock starts the first time a CyberCab gives a paid passenger ride, and that seems like quite a ways out — potentially much more than 2 years.

And Waymo has that whole time to bring down their costs of construction, learn how to solve problems of scale, and so on.

I agree that a CyberCab would have structural advantages if the service were at parity w/ Waymo, but that is a very big hypothetical!

More than that, even -- Waymo is already transitioning to having partners that do the operations & maintenance work while they focus on the tech. I read this as a confidence in the tech timeline, strategy, and maturity. Meanwhile, Tesla's strategy is "hype, ???, deal with it."

I know which one I'd back.

So I assume you've put your money where your mouth is and own GOOG and not TSLA.

Confidence in the timeline is one read. Lack of confidence and finding partners to shoulder the financial burden at the cost of lower upside is another.

Or, identifying core competencies.

Yeah, I could be wrong and they're just looking for the lowest risk profile. But, riders seem to be pleased and accident rate has been impressively low.

I would so love to do a bet with you, easy money. I’ll even take 2 to 1 odds

"Tesla has a software problem, Waymo has a hardware problem" - Andrej Karpathi

Is it easier to ramp to 1M vehicles a year and cut BOM costs by 80%, OR, install more GPUs and collect more data?

If you have to install the GPUs in each car, then you also have a hardware problem.

Tesla has been mass producing bespoke AI hardware for four generation.

It's not a hardware problem for Tesla. They are a trillion dollar vertically integrated manufacturing powerhouse.

Not an insurmountable problem for sure. If Level 4 self driving requires more compute in the car, someone has to physically put more compute in the car, which may require redesigning something to do retrofits, etc. A "FSD" car today may not actually have Full Level 4 Self Driving when Tesla finally rolls it out.

Thus it's a hardware problem.

You pose your comment like the answer is obvious, but I don’t know that it is.

If Tesla can solve FSD with only cameras, then yes, their ability to scale is much better than Waymo’s.

But that is a stupendously large “if”!

I’ve seen the Karpathy quote before, and I think it’s kind of hilarious to compare one company which does X0,000 robotaxi rides per day, to another company which has a robotaxi maybe entering production in two years, and say, “well look, they both have problems!”

Bit of a misleading simplification, no? Tesla boosters assume they will crack FSD, it’s just a matter of time. That seems like a big leap to me.

It's "hardware is hard" versus "scaling laws".

Andrejs point is that software problems are easier to scale your way out of. I think we can both agree there's plenty of precedent for this line of thinking.

A company cannot as easily scale their way out of lowering material costs by an order of magnitude or go from 1000 cars a year to 1M cars a year.

But the implicit assumption is that Tesla’s problems can be solved by scaling laws. That seems to me like a big assumption, no?

I’ve read The Bitter Lesson, I understand the argument “the human eyes are just cameras, and we do just fine”; I’m not totally unsympathetic to the idea that Tesla is better-positioned than some people think.

But I still think it’s a big leap to suggest Tesla will definitely solve FSD by scaling / software, and Waymo is fundamentally screwed and has no way to innovate out of its own scale problems.

Companies lower material costs by an order of magnitude all the time! Look at battery/solar over the last 15 years — in many areas we see two OOMs of cost reduction. It’s a challenge; but one that seems on par with creating software to replicate the optical capabilities of the human brain.

Waymo (and my lived experience with FSD) is proof the end goal is obtainable.

We're surrounded by AI miracles, this website is full of examples, yet people can't conceive of Tesla reaching this finish line. It's pretty odd.

I've clocked 20k miles on FSD. In the brief period I've used it I've seen step improvements. I guess in my mind it's a foregone conclusion with no breakthroughs required. All I need to believe is on the Tesla earning reports, which is Tesla data center build progress (training) and vehicles sold (data)

Its literally the opposite. The only reason Enron is so desperate to pump this vaporware Cyber Cab is because of Waymo running away with robotaxis. 14k paid journeys per day.

In the two years you're referring to, regulating a Waymoe monopoly is going to be the headline development - not the shitco currently being neglected by a drug addicted pedo.

There is no proof that tesla can make FSD work with the level of safety they would need for a robitaxi service. Especially with camera only.

If they do pull that off, or are able pivot to a fused sensor architecture then I do think their hardware potential will make them a good competitor.

I'm not up to speed on the area of autonomous driving at all. Does anyone know if Tesla is basing its next moves off just what the cameras see at the current moment that gets fed into the model? Or is there also a "mapped" component, where it looks up previously mapped information from other Teslas that have traversed this area before? That is what I understand Waymo and GM Supercruise do at least some of the time. There are only certain pre-mapped routes that Supercruise will work on.

https://electrek.co/2024/02/15/gm-nearly-doubles-map-super-c...

They have non-heuristic training. Maybe the AI model is making those connections on the hyper graph if it’s part of the training data, but it would have a decay function as roads get constructed.

you should get out more, tesla is so so far ahead.

What's your basis for saying this? Waymo actually serves rides and has a safety record. If you work in robotics you know that eval is everything -- what's Tesla's L4 driving record? It doesn't even have one, so how can you possibly compare?

L2 driving record vs L4 driving record are insanely different. You can't fall asleep in an L2 car without risking death. You can fall asleep in an L4 car.

I see, that's exactly the kind of "proof" parent comment is looking for

the proof is out there on the road

https://www.youtube.com/watch?v=iYlQjINzO_o&pp=ygUJdGVzbGEgZ...

There is a driver behind steering wheel all the time. It is not a self driving vehicle.

I agree completely with everything you've said, but it's worth keeping in mind that if legal/compliance requirements were to be loosened, a company could expand their offerings with a less robust safety record. And someone close to the incoming presidential administration would be in a position to influence what those requirements are.

Why would this help Tesla more than Waymo?

Even supposing slanted regulatory rule making, Tesla’s autonomous capabilities are vastly inferior to Waymo. The rules can’t be that complicated, either you can drive a car autonomously on a given road or not.

FWIW I’m a Tesla owner who rides Waymos whenever they are available for a cab ride, and won’t turn on FSD despite a number of free trials

>either you can drive a car autonomously on a given road or not

This doesn't seem obvious to me. My impression is that the rate of improvement on autonomous systems is at least somewhat gradual. You've seen this in Waymo's slow roll out, moving from offering the service in initially less complicated settings (suburbs of Phoenix) and progressing to more challenging terrain (San Francisco). If that's the case, then a competitor could provide a worse product but, due to the fact that accidents are somewhat rare in general, have this go undetected for a while. In the intervening time, the worse performing company would benefit from collecting additional training data that they couldn't have been able to collect under a more strict regulatory regime.

Definitely possible that Waymo would benefit even more though! They definitely seem well prepared (except for the unfortunate hiccup of working with a Chinese OEM right before a bunch of EV tariffs got put in place)

My impression is that we’re regulations to soften, Waymo would quickly expand to freeways - the lack of their ability to do so now constrains their usefulness

If you won't turn on FSD how do you know Waymo is vastly superior?

What I mean is that I’ve used it and don’t like how it drives / had bad experiences, so I won’t use it now even when it’s free

Tesla has data from 5 million vehicles in real world conditions.

What does Waymo have, a few hundred cars in San Francisco and Phoenix? In a field where data matters more than anything else, Tesla has an insane advantage.

And yet waymo appears to be ahead commercially?

Because Tesla is yet to start commercial operations in this area.

Waymo (and every other AV company) mostly use simulated data. The real world stuff is very repetitive and not that important; they also don't collect nearly as much as you think they do.

And it doesn't help with the main issue, which is Tesla uses cameras and cameras don't work reliably at night, in glare, when they're dirty, etc.

> a company could expand their offerings with a less robust safety record

But would people take it?

I mean, if one service is 25% cheaper but there are a bunch of horror stories of their vehicles crashing into highway barriers (like this past October), I'm not sure people are going to get in.

Driving is 100% cheaper, and there are a bunch of horror stories of people crashing their vehicles into highway barriers. We’re not starting at zero here.

Is driving really 100% cheaper? Owning a car has pretty high fixed costs.

Also, it's different when you crash a car yourself vs a robot doing it. If you are driving, there is clearer accountability than a robot, so the bar must be much higher for the robot.

Self driving cars don't have to be on par with human drivers, they have to be WAY WAY safer else public backlash is likely to ban them.

It's not like Waymo couldn't be rolling out faster, but it's obvious they're doing everything they can to maintain the public's trust in them because once that's lost it'll be incredibly hard to get it back again.

And ultimately the law is one thing, but if people legitimately see Cybercabs as dangerous to them personally they'll take things into their own hands. Someone assassinated a CEO in New York a few days ago, you think pissed off people won't molotov a cybertaxi?

What is this based on? Some made up target numbers? Tesla isn’t going any of this stuff yet so they have no idea what it really costs.

Er, CyberCab is MAGA. I'd rather walk uphill both ways.

Oh ye of little faith, why wait for cybercab when the model 3 was driving NY to LA driveway to driveway in 2018. Waymo died years ago

possible context... summoning from NY to LA around January 2018 was a real promise by Musk.

https://x.com/elonmusk/status/686279251293777920

edit: removed talk of downvotes, just providing the tweet.

thanks for the context, shitposting is hard to detect!

It was circa the same period that humans were supposed to be back on the moon (now delayed to end of 2027) :’(

Very possible, but first you've got to ship.

Am I living in a bubble or is much of the excitement behind autonomous cabs just hype? Like, sure, it will be popular, but the way some folks(and not just Tesla zealots) talk about it, it will be 25% of global GDP.

Beyond, idk, saving my relatives the task of driving me to the airport, I don't see my family using an autonomous taxi at all. I think I'm like many Americans in that I like my personal vehicle because it's mine. I'm comfy in it, it's filled with only my stank, and I have tailored it to my liking.

if you understand why Uber, Lyft and taxis are such a big industry than you understand why driverless rides will be such a big industry. It's the same industry. But driverless requires a fraction of the labor cost, hence the potential for cheaper rides that will only grow the industry even larger. And I don't have to tip a robot.

I think it will be Tesla. Regulatory pressures for them just lifted, and they move faster than all the other driverless companies, courtesy of Elon's fairly hardcore mode of operation.

It may not be a perfect driving solution (probably 1:1 teleoperator to car in the beginning), but their go to market speed is undeniably better than Waymo et al. Waymo, Cruise, etc have to map out an entire region and have long rollout timelines. In theory Tesla could say they will deploy in X city, talk with the local government to get necessary approvals, and get it going after a few test drives. Thats the tail, anyway. With that kind of go to market speed you can start acquiring customers pretty easily and start scaling up the business. Its basically the typical SaaS setup - cut some corners for market share and become the defacto player in it after some time.

I wouldn't write them out purely out of spite. Their solution is the most conducive to scaling laws, on the ML, operations, and business side.

Maybe I'm missing something but are there any current regulations that make it difficult to operate?

My understanding was that in many states, it's mostly just paper requirements and not any strict testing, etc.

Most states and communities aren't ok with unproven AVs on their streets causing accidents and injuries, even if they are superficially politically aligned with the AV manufacturer's CEO.

That's a good thing, we shouldn't be lowering the bar for AVs.

There’s a chicken and egg issue with this in that to become proven the AV needs to drive somewhere for a while.

> to become proven the AV needs to drive somewhere for a while.

There's a straightforward solution to this that doesn't compromise public safety:

Employ paid training drivers who monitor the AV while it is collecting training data in environments where it does not yet meet the safety bar. These drivers are paid to be observant and intervene to avoid accidents. Furthermore, in their interventions, these training drivers provide critical training data used to improve the system.

Uber did exactly that and got excoriated when the paid driver was on her phone and the car killed someone on the freeway. I think many people raising are concern trolling this because they either don’t want private cars in favor of public transportation or because they are luddites towards automation of driving jobs.

> Uber did exactly that and got excoriated when the paid driver was on her phone and the car killed someone on the freeway.

This is what Waymo does also.

Perhaps Uber should have been more careful whom they hire, or put stronger controls on what they are allowed to do in the car.

AVs already exceed the quite low bar for humans. The real crux is liability when they screw up.

They exceed the minimum viable human driver. They're some combination of a teenager and grandma. They generally get where they're going safely but anyone who's not a minimum viable driver is cringing pretty often.

Nobody wants to let a minimum viable driver chauffeur them around every day. Accepting the reduced quality of performance and increased risk a short duration cab ride is another story.

The only federal requirements are FMVSS, which have nothing to say on the subject of AVs. Some states have their own testing programs with paperwork requirements.

Nov 18 (Reuters) - Tesla (TSLA.O), opens new tab shares rose more than 5% on Monday after Bloomberg News reported that President-elect Donald Trump's transition team was planning to set up federal regulations for autonomous vehicles.

The report comes days after Trump named the automaker's CEO, Elon Musk, as a co-head of the incoming administration's government efficiency department. Trump's team is looking for policy leaders for the transport department to develop a federal regulatory framework, the report said, citing people familiar with the matter.

Last month, Musk criticized the state-by-state approval process, required for self-driving vehicles, as "incredibly painful", weeks after unveiling a two-seat "Cybercab" robotaxi without a steering wheel and foot pedals, set to go into production in 2026.

Interesting... so Musk may be angling to get a federal law passed that would force all states to accept autonomous vehicles with less testing or data sharing than say, California currently requires.

If each car has someone piloting it because the tech can't do it, maybe the tech is the problem? I don't see how cameras can succeed. Sunlight and fog are constant and not practical to work around with visual data. I'm glad there's regulation keeping those lithium incendiaries piloted by people. People are bad enough at driving.

Evs catch on fire less than ICE. Sunlight is not a problem. High dynamic range cams

Let's pretend EV fires aren't exceptional. Sunlight certainly is a problem and you can see that in Tesla's own data.

Can you link said data please

I'm not sure what you want. The NHTSA investigated using Tesla's data and found 956 crashes caused by these types of system failures: https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf

If you simply search for things like "tesla direct sunlight", "tesla camera condensation", "tesla camera bad weather", you can see thousands of instances of vehicle cameras not working.

zero instances of sunlight in that pdf or “tesla’s data”

The entire investigation was based on Tesla's data if you're unfamiliar. I don't care what you believe. Seems like you're one who consumes the marketing rather than data.

Tesla’s data talking about sunlight? No please do link a single data point

Tesla is a veeeeery distant second. Elon is full gung ho on vision only. They even removed ultrasonic sensors for parking. Tesla even removed radar. Vision only is Tesla’s bet.

Meanwhile Waymo bets on jamming the car full of sensors to ensure it has superhuman 360 perception in all conditions with multiple sensor redundancy. Waymo takes safety insanely seriously. Tesla doesn’t care if 100s die in their car while on FSD. Waymo doesn’t want to make cars or manage fleets. They want to purely focus on automating the driver end to end.

Unless Tesla has a magical perception algorithm breakthrough, Waymo may expand all US states that allow it, while Tesla won’t even get an operational robotaxi with 100+ rides per year.

Elon has tremendous success in SpaceX with falcon & starlink, that’s a fact. But Tesla FSD is so much hype compared to substance.

Tesla is a distant first.

https://www.youtube.com/watch?v=iYlQjINzO_o

https://www.youtube.com/watch?v=hoZhER_-nhg

https://www.youtube.com/watch?v=NeNKYRWGgBc

We’re talking about robotaxis.

All those are channels that claimed that FSD was great 3+ years ago.

Not exactly reliable sources of information.

Omar is a fanboy yes, but AIDRIVR and DirtyTesla have been openly criticizing FSD for years and show all the problems openly, now waxing lyric over the latest version. Omar, while sometimes cherrypicking videos, has reportedly had 1 safety critical intervention in over 1000 miles of driving in SF and LA.

Having been in a Tesla in FSD mode recently, it felt quite a lot like Waymo's car did about a decade ago. That's a purely subjective opinion taken off the cuff.

Karpathy said in a recent podcast that they pre-train on LIDAR data so no, Tesla is not gung ho on no LIDAR. They "just" do transfer learning with inference on vision-only, which might be a great idea in the end.

I can’t wait to see how badly this comment ages

Tesla will likely kill lots of people, but the regulatory environment will be so loose no one will care.

As a Tesla customer I’m inclined to agree. I paid in full for FSD six years ago, but never received anything even remotely usable in my region. They just took my money and kept lying. The feature is actively dangerous.

There’s no evidence that this is even going to be a viable business model, commercially or technically. There’s no “moat” around it or obvious first mover advantage. It seems very early to worry about competition.

I mean Waymo already has competitors, Uber, Lyft, and regular taxis. Right now Uber and Lyft are typically strategically priced to be slightly cheaper than taxis are but I think Uber has enough margin to cut costs to compete with Waymo as they enter more markets, and Waymo needs to get people on board to make robotaxis a real thing people use and to offset their dev costs.

First mover advantages:

a) Early technology adopters are more forgiving of minor problems and are your free word-of-mouth marketers. The company benefits from lowered GTM costs.

b) Fast, reliable service depends on density; there needs to be enough cars on the road that there's always one available within a few minutes when you open the app; users will develop a habit of choosing whichever provider is reliably faster and that habit will become ingrained, making market entry for later movers more costly. First movers also have the advantage of their early adopters helping pay for their fleet as it's built up; later movers have to flood new markets with subsidized capacity to get to equivalent density. That works (Uber did it for rideshare) but it's expensive. So again: lower GTM costs.

Regarding whether this will be a viable market - Lyft/Uber have already demonstrated there's tons of demand for on-demand transportation services, and Waymo in SF is already very popular. One can back-of-the-envelope whether the economics work but generally speaking any time you replace labor with automation in a large market, it does.

In the long run though you're right that there isn't a moat that can't be solved with massive capital investment. But you could say the same about many markets; turns out one can have a successful company within a competitive market the old fashioned way.. by just being really good. It's harder, but it works.

Waymo is also in competition with non AV ride sharing companies. They'll need to make it more cost effective and/or (continue to) deliver higher quality rides

They're in competition with the individual drivers, not with the ridesharing companies. The expensive part is owning the cars.

Didn't Waymo's CEO admit self driving cars were a pipe dream a few years ago ?

Can't edit my comment but seeing the number of downvotes I might as well enlighten a few people: https://www.cnet.com/roadshow/news/alphabet-google-waymo-ceo...

That was 6 years ago. I find it odd for a CEO to talk so cautiously (odd but perhaps refreshing), but I'm certainly not holding anyone accountable for what they said in an interview 6 years ago when it relates to a bleeding edge industry.

Well yes of course accountability clearly isn't a thing at play here, Musk said self driving cars were coming in two years back when I was still studying, that was ~10 years ago...

We're still not even close to anything remotely capable of being autonomous, even on sunny straight californian roads... Bring these to switzerland snowy mountain passes or serbian dirt roads lmao

> a bleeding edge industry.

The only thing that this industry is bleeding is money so far

Waymo has fully autonomous commercial operations in 5 cities now. Have you been in one? They are amazing.

Most taxi/rideshare rides occur in cities, not the snowy mountains of Switzerland. Maybe they never make it to those snowy mountains. Who cares?

Five cities in CA/AZ/TX, aka 90 degrees angles grid, 0 inclines, sunshine 300+ days a year, no snow, no rain, perfect visibility, &c. Musk was 100% talking about self driving cars for the average joe, not robotaxis in gentrified city centers. I also have a fully autonomous vehicles, my vacuum cleaner navigates my living room flawlessly...

And again that's exactly what waymo's ceo said, it won't work everywhere / in all conditions, and as it turns out: outside of deserts and the west coasts these conditions are far from edge cases, they're the norm in most of the world

Without even talking about edge cases like mountains, put one of these in Paris or Rome and I give it 15 minutes before it gets stuck/crashes, people underestimate how hard the last 20% will be. 100 000 cars go through that place every single day: https://youtu.be/JgWhagB4d_g?t=79

Zero inclines? No rain? Waymo operates in San Francisco

You have it wrong buddy. Tesla is the one that needs competition, they’re going to have a robotaxi monopoly soon.

Tesla hasn't even launched a pilot program. Cruise has actual commercial road hours logged and Waymo has a fully operational commercial service that keeps expanding. Tesla has a LOT of catching up to do. Cities won't just let them launch a service on a whim. It's going to take years to catch up.

We shall see…

Have you ridden in a waymo? I've found it to be about 95% as good as FSD, and FSD goes anywhere.

I've been very puzzled at people treating FSD as non-existent when it's very reliably good.

Yes, I have, it's been than anything else out there and it's not even close. There was never a moment in time that Waymo was not a distant leader despite how competitors wanted to appear.

I’ve taken a few dozen Waymo rides, but it doesn’t matger because this is not something you can really test as a person. The entire issue is whether you have, say, 7 or 9 nines of reliability (e.g., 10x or 0.1x times more dangerous than a human), and you can’t extrapolate that from how smooth it feels or whatever. FSD doesn’t have that reliability, which is why they don’t let you sleep.