Tesla Optimus Sub-Prime Robot

[Rarity]

V2 does look like a treat. That said, it appears that Tesla and other Musk, Inc. companies have an acute shortage of compute and Optimus is a bit down the line of priorities. You can see this with the quote from Larry Ellison about xAI's appetite for compute and Tesla's subdued free cash flow lately. Maybe it's better to roll another version of Optimus hardware before going to scale in order to put more of the supporting compute in place.

‘We got enough Nvidia GPUs for Elon Musk's company, xAI, to bring up the first version -- the first available version of their large language model called Grok. They got that up and running. But boy did they want a lot more, boy did they want a lot more GPUs than we gave them. We gave them quite a few, but they wanted more, and we're in the process of getting them more. So, the demand, we got that up pretty quickly. They were able to use it, but they want dramatically more as there's this gold rush toward building the world's greatest large language model. And we are doing our best to keep -- give our customers what we can this quarter and then dramatically increase our ability to give them more and more capacity each succeeding quarter.’

https://twitter.com/x/status/1734760528927437154

 

[heltok]

Some nice pics here:

https://twitter.com/x/status/1737196832584057208

 

[UkNorthampton]

Some nice pics here:

https://twitter.com/x/status/1737196832584057208

"Just imagine the possible LORAs that TeslaBot will train offline while it sleeps. This is how TeslaBot will automatically learn new skills without a software updat"

 

[Buckminster]

https://twitter.com/x/status/1738433162819088759

 

[heltok]

Has anyone found any decent analysis of the Tesla bot from outside of the Tesla community? I find the silence strange. Reddit is all over the place with Elon derangement syndrome and newbies claiming to be experts talking about Honda Asimov and Boston Dynamics Atlas.

The only reasonable source who analyzed the previous bots I could find was:

https://twitter.com/x/status/1734960212786696407

Where he basically says nothing and then ask Tesla to do a fun challenge that has nothing to do with getting it to scale and doing useful stuff, clearly not understanding Tesla's strategy of getting it to market.

Any other good analysis? Or have the enthusiasts, the professionals, the roboticists, the futurists etc decided to just ignore it? Why?

 

[UkNorthampton]

Has anyone found any decent analysis of the Tesla bot from outside of the Tesla community? I find the silence strange. Reddit is all over the place with Elon derangement syndrome and newbies claiming to be experts talking about Honda Asimov and Boston Dynamics Atlas.

The only reasonable source who analyzed the previous bots I could find was:

https://twitter.com/x/status/1734960212786696407

Where he basically says nothing and then ask Tesla to do a fun challenge that has nothing to do with getting it to scale and doing useful stuff, clearly not understanding Tesla's strategy of getting it to market.

Any other good analysis? Or have the enthusiasts, the professionals, the roboticists, the futurists etc decided to just ignore it? Why?

Same reason many environmentalists have never heard of Tesla Masterplan Part 3: ignorance

Whether wilful or not I can't tell Green influences and media silent.

With regard to both subjects, non Tesla people don't understand the seriousness of the intent to mass produce megapacks and bots. So used to talk, despair, prototypes.

Tesla will surprise, hopefully bring genuine joy when people understand.

 

[Cosmacelf]

Has anyone found any decent analysis of the Tesla bot from outside of the Tesla community? I find the silence strange. Reddit is all over the place with Elon derangement syndrome and newbies claiming to be experts talking about Honda Asimov and Boston Dynamics Atlas.

The only reasonable source who analyzed the previous bots I could find was:

https://twitter.com/x/status/1734960212786696407

Where he basically says nothing and then ask Tesla to do a fun challenge that has nothing to do with getting it to scale and doing useful stuff, clearly not understanding Tesla's strategy of getting it to market.

Any other good analysis? Or have the enthusiasts, the professionals, the roboticists, the futurists etc decided to just ignore it? Why?

Because it isn't a market yet. There are zero AI based robots for sale. So even though a large visible company like Tesla is working on it, it could very well be like FSD, always late and not quite ready.

I mean, look at the Tesla semi. It doesn't even appear to have any gating technology, Tesla just has to gather field testing data, tweak the design a bit, and then start mass manufacturing in 2024 (latest info from Lars, pretty good source). The semi will upend the transport business starting in a year. OK 2-3 years for the volume to be felt (first year will be lots of fleet pilots like the 40 or so at Pepsico). So, yes, obvious disruption coming our way, yet no one talks about it.

And Teslabot has a lot of gating technology. Elon even said it'll take another year for them to figure out the hands to the level they want.

OK, so that's why the general press isn't talking about it. What about the technical press or other tech companies? Look, there are literally only like five, maybe ten startups (including Optimus) working in this space. The technical community is tiny. Everyone who knows anything about it has got their head down working for one of these companies, and until someone has a product near completion, there's nothing to talk about. We have heard about Digit's trial at, I think it is Amazon?, but that's the only trial I know of.

New technology is like this. It takes a while for it to become well known. And this technology is very much unproven. Heck there are lots of people that still thing EVs are a flash in the pan and won't amount to anything. Humanoid AI robots might not get going for another 10 years for all we know. I think it'll be sooner than that, but it isn't out of the realm of possible that it will take a lot longer.

 

[Mengy]

New technology is like this. It takes a while for it to become well known. And this technology is very much unproven. Heck there are lots of people that still thing EVs are a flash in the pan and won't amount to anything. Humanoid AI robots might not get going for another 10 years for all we know. I think it'll be sooner than that, but it isn't out of the realm of possible that it will take a lot longer.

The thing about humanoids is that they will get going in modest capacities first, meaning their abilities will be limited at the onset. Afterwards their abilities and capabilities will grow via software improvements, and very quickly people will realize the true potential of what they are.

This is especially true for Optimus because Tesla is so incredibly good at both the manufacturing AND the software via OTA updates. My hunch is production will start within a few years but the bots capabilities will be simplistic at first, like moving boxes or very light assembly work on manufacturing lines. Then we'll see OTA's which add capabilities on existing hardware, expanding their utility greatly with ease.

 

[jhm]

For me personally Optimus has been a sleeper issue. I've also been distracted with my seminary studies. But now with the O2, I am waking up to the potential. So pardon my delay in catching up with this thread. I'd like to start sharing some of my emerging perspectives.

There is a gathering convergence of technologies now making Inteigent humanoid robots (Bots) inevitable. So we need to be attentive to all these emerging technologies and their respective needs to innovate, scale and productionize. The timing of all this is critically important.

I now see 02 as the most consequential endeavor of Tesla todate. Bots is a force multiplier that will hasten the transition to sustainable energy and transport.

Obviously, the first deployment of O2 should be within Tesla itself, both in manufacturing and maintenance.

Next, I would submit that Tesla should become the leading Bot parts supplier.

For example, both Tesla and Figure AI have found that they needed to develop their own actuators. Nothing on the market was quite satisfying for humanoid robotics. This implies immediate market demand for Tesla to supply O2 actuators. Tesla obvious wants to drive down the cost of hardware through a scale up of manufacturing. They can scale up and market these O2 actuators immediately, regardless how production ready O2 as a whole product might be at the moment.

We can do a similar analysis on all the components of O2. Note also that software, specifically NN components are also critical and marketable "parts" too. As massive capital flows into the Bot space, thousands of startups and even larger corporations will want to get their hands on the latest, best and cheapest parts.

Money will not be made in final assembly. Rather, Bots themselves will do final assembly, along with maintenance and upgrades. What the Bots will demand are the parts!

Consider the O2 hand. Tesla can market that hand along with NN components as software drivers. There is both hardware to manufacture and sell and NN to continuously improve and offer on subscription. Additionally, Tesla can market wrists and whole arms. The value of the O2 Hand comes down to how intelligent/dexterous it is, how physically capable and durable it is, how easy to integrate it is (plug-and-play), and how affordable it is. Tesla could focus optimizing every attribute that a Bot startup would want in a Bot hand.

When Bot startup are trying to optimize their own Bots for particular markets, I think they will want to mix and match parts. Some niches will be willing to pay $5000 or more on a highly dexterous hand, while others just need a cheap stump for $50. Many others may want the best hand available under $500. Fostering a robust ecosystem of swappable parts will enable the Bot market to realizes economies of scale rapidly.

So I see selling fully assembled O2 as inconsequential. Selling a box of parts from which an O2 can be built will suffice for many years. And this means Tesla has some new products to scale and market in 2024.

 

[Tiger]

Next, I would submit that Tesla should become the leading Bot parts supplier.

Tesla will here probably become supply restricted like in all its other businesses. It will not have spare to sell to third parties, as it will sell out directly to customers and license the software features on SaaS fashion to print money. Yes, not all bots will need all parts, but like cars, you can tweak them to your needs in addition to available factory options.

 

[jhm]

Tesla will here probably become supply restricted like in all its other businesses. It will not have spare to sell to third parties, as it will sell out directly to customers and license the software features on SaaS fashion to print money. Yes, not all bots will need all parts, but like cars, you can tweak them to your needs in addition to available factory options.

I am specifically challenging this business model.

On problem with ramping up a car is that one missing component can halt production. Selling parts a la carte, you can sell all you can make without being slowed by the least available part.

 

[heltok]

I have listened to most of the Tesla Bot videos on youtube etc. Tesla-youtubers seem to think that Tesla might make 100k robots next year or 2024. Imo they fail to understand what will be the bottlenecks of a project like this.

I believe that Tesla could produce 100k robots next year. It's a lot easier to assemble bots than cars and it requires fewer parts and less material than a car. Software the bot can already do simple object manipulation which is enough and it can walk and balance good enough. So why will they not just make 100k? Because
1. They are still making rapid improvements in both hardware and software. Whenever you make a product you will never be done. There are always improvements you can make. But at some point you have to say ok, this is version 1.0 and go into production. But as long as you are making large rapid improvements you keep pushing development.
2. Getting to scale would improve data gathering, so that's a reason to go to scale. But if you have issues with failures at customers, suddenly you have a messy complex service/support situation and developers will have to deal with this rather than develop. So you don't go to scale before you are ready to.

Tesla and Elon know how to do disruptive product launches now. Production hell, Starlink, Autopilot etc. Imo youtubers should stop assuming that they know how to run a project like this better than Elon.

 

[Cosmacelf]

I am specifically challenging this business model.

On problem with ramping up a car is that one missing component can halt production. Selling parts a la carte, you can sell all you can make without being slowed by the least available part.

There's no market for humanoid robot parts. Everyone is a startup with no factories. And even a perfectly build hand is useless without the software to manipulate it. Tesla isn't going to spend the time to onboard customer engineers on the required software to run the hands, for example.

 

[jhm]

There's no market for humanoid robot parts. Everyone is a startup with no factories. And even a perfectly build hand is useless without the software to manipulate it. Tesla isn't going to spend the time to onboard customer engineers on the required software to run the hands, for example.

You don't see Tesla selling Dojo compute?

 

[Cosmacelf]

You don't see Tesla selling Dojo compute?

Not in the near term (3 years), no. They will use all the dojo compute they can internally. And it is much harder to build a software ecosystem around cloud than people think.

 

[jhm]

Not in the near term (3 years), no. They will use all the dojo compute they can internally. And it is much harder to build a software ecosystem around cloud than people think.

Ok, if you think Tesla will be compute constrained for the next three years that definitely would slow a lot of progress in the Bot space. It would actually worry me that Tesla could fall behind and miss out. From my perspective there is a massive ecosystem that Tesla wants to be a part of.

 

[Knightshade]

Ok, if you think Tesla will be compute constrained for the next three years that definitely would slow a lot of progress in the Bot space. It would actually worry me that Tesla could fall behind and miss out. From my perspective there is a massive ecosystem that Tesla wants to be a part of.

Fall behind whom? everyone is constrained on compute and expected to remain so for years.

 

[Tiger]

I am specifically challenging this business model.

On problem with ramping up a car is that one missing component can halt production. Selling parts a la carte, you can sell all you can make without being slowed by the least available part.

Parts is a low margin business, chinese companies will soon 3D print all your stuff on aliexpress for 1/10 the price. What you want is a moat which you can only achieve with vertical integration from hardware to software to payments at high margins like Apple.

High margins eat low-profit volumes and production halts for breakfast.

There will be more Tesla Bots in the world than there are iPhones. A person only needs one working iPhone but several Tesla Bots can work for them.

 

[Buckminster]

You don't see Tesla selling Dojo compute?

Not in the near term (3 years), no. They will use all the dojo compute they can internally. And it is much harder to build a software ecosystem around cloud than people think.

Ok, if you think Tesla will be compute constrained for the next three years that definitely would slow a lot of progress in the Bot space. It would actually worry me that Tesla could fall behind and miss out. From my perspective there is a massive ecosystem that Tesla wants to be a part of.

Thread on this subject:
Project Dojo - the SaaS Product?

 

[jhm]

https://twitter.com/x/status/1739719880499425506

CEO of Figure AI sees AI capabilities coming faster than hardware. 2024 will be a big year for humanoid robots. Figure and others will be eager to aquire hardware to match the AI.