Since its inception, FSD has always warned the driver that it needs to be supervised by the operator. The earliest versions required frequent and near constant input by the driver to remind them that it is not fully autonomous.
The resistance to a name change from FSD to supervised FSD has nothing to do with safety measures put in place to make it clear to the operator the limitations. It’s a reflection of the liability risk and the need to mitigate it.
In California, over the counter sleeping pills carry a warning “may cause drowsiness”.
When streaming some series lately I saw an ad for Ebglyss (for eczema, Lilly). It included the warning “Do not use Ebglyss if you are allergic to Ebglyss”.
FDA requires over-the-counter (OTC) sleep aids and medications containing sedating antihistamines (like diphenhydramine or doxylamine) to carry specific warnings about drowsiness.
Next you’ll be blaming California for the humidity in Mississippi.
And that was in the context of “California authorities stepped in and required a name change, which Tesla grudgingly made”
Then you made up a regulation attributed to California… which has nothing to do with California. Does California over-regulate, especially with vehicles, we agree.
Don’t make stuff up, I won’t have to comment on it.
I think the examples provided are a sign of the times of the absurdity of product labeling rather than an admission by the product manufacturer of fault. California’s requirement of a name change is not in and of itself evidence that Tesla has been deceptive with the FSD product. More to the original point on the fatality data, there is no evidence that FSD is more likely to result in a catastrophic injury or a fatality.
I do think that there could be ultimately be a link between driver’s becoming less attentive and more heavily reliant of FSD. The Peltzman effect talks about this idea with other safety measures. Which probably means, FSD has to reach a near perfect standard to go full hands off, simply from a litigation risk standpoint. Which is a shame, there is probably a threshold where it improves safety over human drivers but before reaching that ultimate standard. I remain a long term patient shareholder, hoping they reach that end.
That’s the Tesla bull narrative and there may be some merit to that line of thinking. I don’t think it is nearly that cut and dried though. One reason is they have different business models. Waymo has said they have no intention of manufacturing vehicles. Instead they will license their tech, and leave the low margin manufacturing to others. And they could potentially license their tech to any manufacturer, not just EV makers. Easy to scale and the TAM is huge.
Another reason is that ride hailing permitting is a patchwork of state and local regulations and autonomous vehicles add another layer of complexity on top of that. It took Uber and Lyft years to work through the permitting process and Tesla (and Waymo) are looking at similar long permitting timelines. The difference is Waymo is permitted and operational in 11 cities, with probably an additional nine more this year.
I don’t think Waymo faces any different scaling problems than Tesla does. At least, not under both companies’ current plans.
Back when it looked like the “Tesla Network” model was their business plan, that might have been true. That was the model where actual autonomy was going to one day be able to be activated on every car that Tesla sold, so there would be an existing “fleet” of millions of cars that could be pressed into taxi service with just software changes.
But that’s not where Telsa’s going anymore - at least not in timeframe relevant to scaling. They’re now doing exactly what Waymo and others are doing - building bespoke purpose-built robotaxi cars that they themselves will own. These cars have such markedly different hardware than Model 3/Model Y vehicles that it is exceptionally unlikely that any of those existing cars will be able to be used for robotaxis.
So they will face all the constraints on scaling that Waymo and other fleet-based robotaxis will have. It takes time and capital to build the fleet, you have to buy land/get leases for places for them to park and charge and get maintained, you have to build a workforce in each city for remote “call for help and rescue” services and returning lost items and doing all that fleet services, etc. To say nothing of marketing and advertising to build consumer awareness that this is a service that exists - since you don’t have the individual drivers/owners driving awareness, you’ll have to do it yourself.
True enough. But Tesla doesn’t have cruise control. It has “Auto-pilot” and “Full Self Driving.”
Those names imply far more than they are capable of doing.
I wonder if this lady’s family would have a case against Tesla for misleading their car owners into thinking these systems can do more than they actually do.
Ripped from today’s headlines: Waymo is blocked in New York due to political opposition.
Waymo is increasingly facing political roadblocks as it tries to roll out its self-driving taxis powered by artificial intelligence nationwide. After early successes winning over politicians in California —its home state — and elsewhere, Waymo has stumbled in unlocking some of the biggest markets in the country.
My thesis of the scaling advantage doesn’t have anything to do with converting or borrowed use of existing owned vehicles. Tesla just simply is a world class manufacturer that is obsessive about vertical integration. Everything about their approach from camera only to the initial two seater use has cost and efficiency in mind. Google doesn’t operate in that manner, which has been a luxury, and has limited manufacturing experience at scale. In addition, I don’t think even under the most optimistic scenarios the manufacturing costs are likely to be close between the two. Experts are all over the place on costs but some think Waymo gets to $50k-$80k total cost per vehicle with its 6th generation vehicle that might include a 42% reduction in sensors. Tesla is targeting a sub $30k production cost.
I have been a long time holder of shares in both companies and think they are great at their core activities. But, Tesla thinks like a start-up and has no hesitation about pivoting when something doesn’t work. Culturally, it’s a huge advantage for a large company that thinks in this manner.
It is a bottleneck to full scale production but I don’t think it disproportionately impacts one vs the other. If Waymo had an advantage in this area, I don’t know that it offsets the differential in the cost of production. To the extent WAYMO and TESLA simply provide the fleets and a third party leases them or owns them with a fee share agreement, it becomes neither manufacturers problem.
Sure, but that’s while they’re still doing their artisanal, handcrafted conversions. At some point, they’ll reach sufficient scale where it makes sense for them - like Tesla - to just build robotaxis that have everything in them at once.
All Tesla’s done is decide to build that robotaxi plant a few years before they actually need it. Their production cost for the Cybercab is sky high right now, because when you spread the capital cost of the facility over just a few hundred cars you get insanely high numbers. By the time Tesla gets to the point where they need enough volume that their production cost will be sub $30K, Waymo will also be at scale with a much lower cost for their vehicles. Perhaps a bit higher than Tesla’s, since a car that seats two passengers is cheaper than a five passenger car, but Tesla will have to eat some of that in the cost of their artisanal Model Y conversions and operating costs. But not enough to matter, when all is factored in.
I think that’s right. Which is why I’m a little skeptical about their robotaxi business. With Musk, his secret isn’t that everything he does works. It’s that he does a lot of things and is comfortable taking risks on things that have massive payouts but might not work. Which means he does a lot of things that don’t hit home runs (Solar Roofs, Dojo, Cybertruck, Boing Company), but it means that something he does may be a home run.
The differential in the cost of production is probably going to be insignificant compared to operating costs. These cars will end up going about 200K-250K miles. If there’s as much as a $10K difference in the production cost of the cars (and LIDAR isn’t that expensive anymore), that’s barely a few pennies per mile. Unlikely to be enough to matter materially.
Tesla’s main advantage over Waymo was the Tesla Network concept - being able to do the same trick as Uber and have a zero marginal cost fleet, made up of cars that people already owned. Now we know they can’t do that, at least for the intermediate term (and perhaps never). Which means Tesla doesn’t really have any major advantages in this competition, and quite a few institutional disadvantages (like, they’re quick on innovation but have no idea how to manage a consumer-facing brand).
No, that’s the estimate for the full massed produced vehicle. But, like I said, estimates vary by expert and it may be difficult to know for sure. What is for certain, Tesla already has the production facilities in place.
They are high but not as high as production costs for WAYMO. The production runs parallel to the model Y production in Texas. Your fixed costs are then spread across the production of several lines.
You could say the exact same thing about Google. They have thrown gobs of money at things that have not worked out. There are no guarantees and I understand no winner has been crowned or that robotaxi will be a huge financial success. However, it could be the autonomous capability provides a licensing solution for other manufacturers fo the actual vehicle itself. There is a world of possibilities.
Again, I think you are minimizing just how difficult it is to manufacturer at scale in lower margin businesses. I never bought into the concept of the tesla network. It’s a liability and logistical nightmare. There is going to be a cost to build, own, and manage the fleet and Tesla is just better at this kind of problem solving. We haven’t even touched on what advantages Starlink may provide to robotaxi if not available to competitors or with more limited features.
Yep. They’re bearing that capital cost years before they need the cars. It doesn’t make sense to have a dedicated production line to mass-produce a car that you only need a few hundred of.
They’re still sky high. Even in an existing factory, the cost of just the production line for mass-market automobile production is several hundred million dollars (search results say between $300M and $1B for a line in an existing auto plant). That’s before you spread the fixed cost of the factory (land, building) across all the lines. If you’re looking at the capital cost of opening that production line a few years too early over a few thousand cars (if even that many over the next year or two), you’re looking at $40-50K additional per car for just that. Again, Musk is always insanely optimistic about how far away actual deployment of autonomy is - so this doesn’t look like 3D chess. It looks like a costly error that he’s got his Cybercab production line open when his deployed fleet is 135 vehicles nationwide.
Not minimizing it - but lots of companies do it. There’s scores of major automakers around the world. I don’t think Google’s going to make these cars. I think that Geely and Hyundai will make these cars. Right now, they make non-robotaxi base cars and Waymo installs the robotaxi eyes and guts. That’s expensive. But it makes sense, because Waymo only has a fleet of 1,500 cars, and doesn’t need enough new cars added to the fleet to justify a dedicated production line to just their vehicles. Once that changes and Waymo has scaled to the point where they need 10K (or whatever) new cars per year, then Hyundai and/or Geely will have a production line that makes the full robotaxi (including eyes and guts) for Waymo.
Does Tesla have a small advantage in manufacturing efficiency over Hyundai or Geely? Maybe - I have my doubts. With Tesla’s massive automotive margins having come back down to earth with some actual real competition, I think the jury is still very much out on whether they’re much better at being a manufacturer at scale in a low margin business than other carmakers, rather than having just run up big margins at a time of vehicle scarcity. But even if they have a bit of an advantage, it’s exceedingly unlike to have any material effect on the cost structure of the service - let alone enough of one sufficient to compensate for Tesla’s lack of skill at managing consumer-facing brands or their lack of experience in managing a ride-hailing service.
This is irrelevant for autonomy and has been for a decade.
Expansion of L4 autonomy is a city-by-city, neighborhood-by-neighborhood, driving scenario-by-driving scenario grind - much more of a bottleneck than manufacturing known hardware - by far.
For both Waymo and Tesla and everyone else.
Waymo is scaling faster - they have known autonomous hardware and software that works and are iteratively optimizing both (eg, reducing sensors).
Tesla hasn’t even reached its first generation of working hardware and software for scalable L4 autonomy - why in the world are we discussing how they can scale?
Tesla has nothing yet to scale.
Let’s see Tesla accrue a few million miles of true autonomy with rigorous, public safety analysis.
I don’t think so. In fact, if the long term idea is both solve autonomy and plan to own the vehicles or sell the platform to others, the cost of manufacturing becomes incredibly important, where the low margin provider wins.
It is if your system is highly reliant on mapping. There will be bottlenecks for both in terms of storage and charging but those are temporary problem.
If Tesla solves FSD, it easily overtakes Waymo because it can then scale much faster. It’s like the iPhone in the early days with its interface with Microsoft products. It kept blackberry in the game with the business community, until that problem was solved.
You must have glossed over my earlier comments. IF Tesla solves autonomy, the marginal cost to scale and manufacture becomes the more important differentiator. That’s why it matters. It will be irrelevant if Tesla doesn’t solve it.
In fact, the best evidence is to the contrary: even with working autonomy, Waymo’s spread across regions/cities is a grind.
If you tripled Waymo’s fleet today, they wouldn’t scale much faster, if at all faster.
Tesla’s spread of supervised vehicles is even slower. Why would unsupervised vehicles, with the vastly increased risk and complexity, be faster?
Tesla does the equivalent for each city: they drive around thoroughly taking lidar and other measurements - no different operationally than mapping. There is also evidence that Tesla did all kinds of city-specific training for the Austin effort - hardly a scalable process. All the evidence shows that Tesla has a painstaking process for each city, slower than Waymo.
And this may not even be only - or primarily - a driving software grind. It’s going to take time to build out your infrastructure.
You have to figure out how to support the fleet. If you enter a new market, and don’t have access to data from a partner like Uber or Lyft, you don’t know the demand patterns yet. Sure, you have a general idea of where people come from and go to - rush hour to the downtown, weekend trips to the nightspots, etc. But nothing granular.
So you don’t know in advance where you want to put your parking spots for overnight. Where you put your server dispatch locations. Where you put your midday refuel/emergency clean/lost items depot. You start with supporting your initial small fleet - some short-term leases (or licenses) at places like gas stations near the urban core. You have to scout all those locations, enter into lease and contract negotiations, hire a workforce, etc. You have to get licenses and permits. You might have to install some chargers - though hopefully you’re finding people willing to lease you spaces overnight at commercial uses that have the chargers. Etc.
But you keep that small. These might not be the right spots. You might need to pivot. So these are all short-term arrangements, picking them up where you can, and they might have to be abandoned if they don’t fit the evolving network. So it can take time to find spots like that, because you’re trying to find little unused chunks of parcels mostly being used for something else, rather than buying a bit vacant lot and building on it.
As you expand you have to keep doing that. As your service area and fleets grow, you need more physical locations - maybe building buildings, maybe purchasing locations. You need more staff, in more areas. All of which takes time.
Completely different scaling problems than an Uber/Lyft model - and (mostly) unrelated to how well the software/hardware works in the actual car.
Waymo’s grown to the point where they’ve just outsourced some of that to a professional fleet-management firm: