First half goes through the slide deck:
And it turns out David Moss’s intervention-free FSD driving has ended at 12,961 miles due to Snow + weather + rural Wisconsin roads.
PS: Also some sensor fusion discussion towards the middle.
Whole lotta’ hype in that presentation. I don’t believe that TSLA Cybercabs are going to be any cheaper to operate than Waymos. They’ll still need human fleet monitors just like Waymos.
Plus, I don’t see anyone paying $60K+ for a TSLA and then renting it out as a robotaxi. Too risky.
Finally, Optimus is a solution in search of a problem. Specialized robots rule the day and no one needs a generalized robot.
1 Like
See the slide at 6:36 in the video. Tesla will have the same supervision, fleet management, insurance, and other platform costs as Waymo, but potentially benefits from lower vehicle depreciation and maintenance costs.
Your post on Optimus won’t age well, I predict. Just like in 1985 some thought we’d never all have cell phones because “who needs a phone in their pocket all the time?”
1 Like
Everyone used phones in 1985. No one today uses generalized humanoid robots. There is no use case for them.
There is a use case for making a production line sufficiently automated that one can use extremely low cost humans to do the work that isn’t automated … so if a general purpose robot can be trained to the job and is even lower cost, especially considering the hours it can work, why no use case?
1 Like
I think it’s just a variant of the discussions we’ve had here from time to time. There’s a use case for robots, but not a use case for general purpose humanoid robots, because the nature of work in an auto factory (and many other applications) fits specialized robots better than general purpose ones.
Among other things, a specialized robot requires a well-defined, automatable, stable over time use case and a sufficient market to justify development costs.
Whereas, a generalized robot may be quickly trainable and can just be put to work … especially if provided with optional specialized attachments.
1 Like
Curious. I think it’s just the opposite. A single use robot is easy to train, it just has to do one thing and do it well. A generalized robot has to do a thousand things and do them all well enough to replace a human, which is not easy given that humans are capable of solving problems on the spot.
Developing a Roomba to vacuum floors, or an IED de-fuser may take more than one thing or two, but imagine trying to develop one that also takes the dirty clothes, puts them in the washer, dispenses the correct amount of detergent, transfers them to the dryer, takes them out and folds them and puts them away. (And that’s just one simple task for a human. It’s a dozen for a robot, and there are literally tens of thousands or even hundreds of thousands it might have to “learn.”)
1 Like
When you train one humanoid robot you have trained ALL his brothers. Knowledge is just one download away.
The Captain
2 Likes
I’m posting to say I agree.
See, it’s indeed possible.
Edit/
Oops, haha, I thought this was the captain.
Lol, maybe it’s not possible.
Meanwhile, no-one else wants to talk about robotaxi economics?
Typical for this crowd.
Sure, I will.
I think it’s vastly overstated, at least in the near term.
It’s clear that current on-the-road Tesla’s will not be able to make the demands of a Robotaxi service at scale , at least not yet, perhaps not ever. Even Tesla is running their training wheel services with remote monitors in case of trouble, which means they can’t shift “flip a switch” and have it immediately come live.
Uber and Lyft were able to achieve scale quickly because there was no hardware component to deal with. Drivers supplied the cars, drivers supplied the phones to send and receive instructions and payments, all Uber and Lyft had to do was coordinate a back-end which was truly “one size fits for any market, all at once.”
Tesla is going market by market, entering with Lidar mapping (strange) and other precautions, not just “turning it on” as has been hyped for many years. More to the point, there are now regulations in place which any provider must abide and fulfill, making it slower yet.
I’m not saying it won’t be a business, heck taxi-cabs are a business. I am just doubtful that people will give up their cars in favor of hoping there’s a Robotaxi ready, I’m doubtful people are going to buy a new car and then invite strangers to take it over in great numbers, and I wonder about the economics of a self-standing business (not using customer cars) where cleaning, depreciation, and dead-head costs are involved.
I am even more doubtful that significant numbers will give up the “second car” in favor of rental, although I happily admit there will be some in economic situations or dense urban situations where the costs may prove advantageous. I just doubt it will be enough to move the needle for car manufacturers in a seriously detrimental way.
All of my speculations are based on a short to medium term and from an investing standpoint. I’m 78; more than 10 years out I don’t care about much. I might have a different perspective if the “flip a switch” hypothesis came true, but I really don’t expect that nor even see how it’s possible with the current fleet. That argues for a more gradual build of such a business. That certifiably could happen, but not on the time scales that the acolytes seem to think supports the kinds of inflated stock prices we see today.
4 Likes
Here’s a slide snapshot from the video:
Tesla has the same regulations that Waymo has, and benefits from Waymo paving the way in 5 cities. Waymo hasn’t scaled, mostly because it doesn’t have enough vehicles - only 3500 iPaces. It’s working on Zeekrs (tariffs hurt that effort), and then Hyundais, but it still seems unlikely that either will approach the CyberCab cost, and hence depreciation advantage.
This is going to be a scale thing. Waymo and others are partnering with Uber precisely because they don’t have enough automated vehicles to service customers in a wide service area in a timely fashion. If Tesla is also produce not single digit thousands, but hundreds of thousands of CyberCabs a year, then the wait times will be small and as such more people will use the service.
I completely agree. This customer car thing is only valuable for your own family’s use - have your car pick up your kid at school and bring them home, for instance.
I do believe a widespread, convenient, and cheaper than having your own car robotaxi does disrupt personal car ownership, though, and in fairly short order. I believe it will, at first, greatly reduce second car ownership as I believe economically that will work out.
As usual, these things will happen more slowly than some pundits (eg Tony Seba) predict, but also faster than many doubters predict. The combination of viable robotaxis in multiple markets combined with personally owned Teslas doing their own not just L2++ but L4 driving should have a big impact. But, yeah, I don’t have data for that hunch.
1 Like
BTW, in case you missed it, there’s at least one other company working on an L4 capable vehicle for personal ownership, Tensor:
It’s also built on the Nvidia Drive stack, with 4 Thor computers.
Supposed to ship in production later this year. We’ll see.
1 Like
Agreed. On the other hand, a single use robot has to be individually engineered and manufactured for that specific purpose. The market for that purpose will be strictly limited, which means that up front cost is spread over a limited number of units. There are clear use cases for both types, and the market will decide where the tipping point is. (Make that where the tipping point moves to, as it will be dynamic.)
In the early days of Tesla’s manufacturing they learned that some of the tasks they tried to have done by specialized robots just weren’t suitable. The example they gave was handling something floppy, like a wiring harness. That might be an example of a task that an anthropomorphic robot could succeed at.
1 Like
Developing a Roomba is likely to be simpler in isolation, but that is unlikely to be an important measure. For starters, the humanoid robot can probably do some jobs even if thus far trained only on 5 of the 1000 skills imagined for it. Job changes and it is missing a skill … train for that and, as the Captain reminds us, now they all have that skill. Have 10 factories, each training on a different new skill, and soon one is doubling and tripling the skills available everywhere.
1 Like
FWIW, four observations on the chart that Smorgasbord excerpted above:
- There’s no good reason to assume that Waymo and Tesla will have the same operating costs. They’ll be independently setting up their ridership algorithm, operator support networks (remote drivers for trouble spots and the like), maintenance and charging locations, advertising and the like. No reason to believe that their costs will be identical, any more than any other two companies that are in the same industry have the same identical costs.
- Related to that point, this simply assumes away the fundamental difference between the two companies. Namely, it assumes that Tesla’s vision-only cars will operate just as well as Waymo’s more-sensor cars will. The main difference between those two approaches is Waymo’s determination that more sensors are necessary for the cars to function well. If that’s true, then Tesla’s operating costs will be higher (they’ll need more remote operators, more remote dispatch units, etc.) for the same equivalent service, because Tesla’s cars will need more “help” than Waymo’s. Right now, that manifests in Tesla’s cars needing an actual person in the car - which of course puts the operating costs through the roof.
- The vehicle cost comparison is certainly wrong - too high for Waymo, too low for Tesla. The $170K cost for Waymo cars might have been true several years ago, but component costs for LIDAR have dropped precipitously (both as the technology improves and as their volume of the sensor suite has increased) so even their hand-crafted vehicles today are much less expensive. And Tesla’s current cabs are more expensive than $30K, because they’re not stock Model Y’s - they’re also bespoke vehicles outfitting with different camera suites and upgraded comms systems. If this is intended to not actually be “off the bat,” but reflect Tesla’s expected Cybercab costs (as suggested by Tesla getting lower costs for a “simpler, purpose-built vehicle) then Waymo’s costs should be much lower - because by the time Tesla gets up to volume on the Cybercab, Waymo should be up to volume production at their facility in Mesa, Arizona.
- Finally, and most importantly - both companies are super expensive, which undercuts the Tesla value proposition. Seriously - more than $2 per mile? That’s completely uncompetitive with driving your own car, which is at around $0.70 per mile even for new cars, and closer to $0.30-0.50 for what most people experience. A service coming in at that price point is not going to be displacing much driving other than existing taxi/ride-share service. If there’s no material price difference to Uber (ie they’re not super-cheap), how do they drive adoption with their…questionable brand position and lack of advertising? And if they’re coming in at that price, how much could this service actually be worth? Uber is “only” a $175 billion company - and that’s based on world-wide rideshare and all their delivery and other services. Their North American ride-share alone component is at most a third of that value.
I don’t know. I look at that chart and see a business model that simply assumes the most contested point about Tesla (that they can make vision-only work as well as multi-sensor cars), and even if they can they’ll have a product that costs just about as much as Uber does but lacking the decade-long head-start that Uber has in getting their app on people’s phones and becoming a synonym for hailing a ride. How do you scale against that? A sizable but not enormous TAM (the existing hailed ride market, not much of the self-driven market) with an enormous entrenched competitor against which you have no material cost advantage?
The Tesla vision for rapidly taking over the market was based on getting to Level 5 (avoiding the $2.00 per mile operating costs in this chart) and using the existing fleet (millions of existing cars that are already substantially depreciated). This chart shows really rough economics for the Tesla business model “off the bat,” even after they get the Cybercab through the production ramp into high volume.
4 Likes
But plenty of people use generalized humans to do stuff for them.
An engine is quite different than a horse, but we still all switched over.
1 Like
It’ll take 50 years at least before we have robots that are intelligent as humans. Maybe more like 100 and there is no guarantee that Tesla is going to be the winner in the space.
Betting on generalized humanoid robots today does not justify TSLA’s 300 PE.
2 Likes
The ironic thing about this board is that it encouraged at least two participants to short TSLA before it hit ATHs last year.
As a Roadster owner and TSLA investor since 2011, I’ve lived through doubters on EVs, battery gigafactories, reusable rockets, brain implants, etc.
I despise what Elon has become politically and socially over the years, but he’s made me a ton of money and his companies continue to thrive pushing the boundaries of new technologies.
PE is a backwards looking metric. Robots don’t have to be as “intelligent as humans,” (whatever that actually means) to be a profitable business. Humanoid robots will be a big business, and Tesla is likely to be a big winner in that business, if only because the world we live in has been designed for humans, so mechanization of that world has vastly less friction with robots who have a similar form. Add some intelligence and the change will come, in far less than half a century.
3 Likes