Tesla Optimus Sub-Prime Robot

[Knightshade]

Yeah, like I said. They will do millions and when they do millions they will need all that those job listings say. But that is not a proof that they will not do meaningful scale in between

If you had any understanding of scale manufacturing, at all, it clearly would be though.

I don't think there's any more times I can explain it that will help at this point however.

You don't make a large number of the same something before you have those 9 steps all completed.

Tesla did all those things, first, for every product they have ever made before moving past hand-made/prototyping stages. So have all other companies mass producing in volume. I've already given you a number of reasons (and it was hardly an exhaustive list) you just keep choosing to ignore them.

 

[Knightshade]

This is gen 2 that was supposedly due to be ramped starting November 23:

According to whom? Ramped in what way and to what degree?

The video Tesla released said nothing about ramping, just about the fact Gen2 existed and had improvements.

They might be building 1k to 10k of these this year for exclusive use in Tesla factories. Probably all handmade.

What do you believe they could learn from 10,000 units they can't learn from 100 units at vastly lower cost and without needing to rework 10,000 units when they find an issue or need to test something new?

A much more likely approach right now, since they're still making major improvements, and still figuring out the fundamental design requirements, is you have a thing in mind, you make a tiny # of prototypes based on it- then you test it. You iterate your prototypes based on the testing.

You want to make the lowest # of test units that provide useful data, because each unit has (quite high compared to eventual production) cost- and it's a LOT easier and cheaper to make 5 to 10, test, then make 5-10 improved ones. Then repeat that 5-10 times ending with 50 total units... than it is to 50 of one version, find an improvement, and then have either 50 useless ones, or 50 you need to go back and revise.

Gen 3 will likely ramp in 2025 and be partially built by Gen 3 robots. New hires will mostly be working on gen 3. Are you both referring to gen 3?

I'm referring to whatever their first model that will be mass produced is, could be V69.420 for all I care about version number. You don't do that, at all, before you've done all the stuff in that job listing.

Apart from the myriad OTHER reasons you can't mass produce without those things- you can't even validate the things you're making are correct without those steps.... (and indeed Tesla ALSO has jobs posted for people to design, then build, the systems and machines to DO that testing after the specs against which they'll test are created).

So the idea such mass production beginning THIS year- the ORIGINAL thing Alex was quoted as claiming and Heltok seemed to be arguing for- is pure fantasy.

 

[Buckminster]

According to whom? Ramped in what way and to what degree?

The video Tesla released said nothing about ramping, just about the fact Gen2 existed and had improvements.




What do you believe they could learn from 10,000 units they can't learn from 100 units at vastly lower cost and without needing to rework 10,000 units when they find an issue or need to test something new?

A much more likely approach right now, since they're still making major improvements, and still figuring out the fundamental design requirements, is you have a thing in mind, you make a tiny # of prototypes based on it- then you test it. You iterate your prototypes based on the testing.

You want to make the lowest # of test units that provide useful data, because each unit has (quite high compared to eventual production) cost- and it's a LOT easier and cheaper to make 5 to 10, test, then make 5-10 improved ones. Then repeat that 5-10 times ending with 50 total units... than it is to 50 of one version, find an improvement, and then have either 50 useless ones, or 50 you need to go back and revise.






I'm referring to whatever their first model that will be mass produced is, could be V69.420 for all I care about version number. You don't do that, at all, before you've done all the stuff in that job listing.

Apart from the myriad OTHER reasons you can't mass produce without those things- you can't even validate the things you're making are correct without those steps.... (and indeed Tesla ALSO has jobs posted for people to design, then build, the systems and machines to DO that testing after the specs against which they'll test are created).

So the idea such mass production beginning THIS year- the ORIGINAL thing Alex was quoted as claiming and Heltok seemed to be arguing for- is pure fantasy.

Musk said:

“So, we’ve actually had to design our own actuators that integrate the motor or the power electronics, the controller, the sensors. And really, every one of them is custom designed. And then, of course, we’ll be using the same inference hardware as the car. But we are, in designing these actuators, designing them for volume production. So, they’re not just lighter, tighter, and more capable than any other actuators wherever that exists in the world, but it’s also actually manufacturable. So, we should be able to make them in volume.”

The first “sort of production candidate actuators integrated and walking should be around November-ish,” Musk added.

So as I said, pick V2 or V3 and then you can have a better debate.

V2 is clearly good enough to be put to work. Building production capability is the biggest issue so Elon won't mind building thousands of V2s.
I would guess V3 ramp starts end this year.

 

[Knightshade]

So as I said, pick V2 or V3 and then you can have a better debate.

Why does version # mattter?

"The first one they mass produce" is what the original thing from Alex is talking about. It's the same thing I'm talking about- and I've been crystal clear on that.

 

[Tiger]

(only partly joking, the need for robot mechanics is going to rise exponentially)

Unless robots fix themselves, or each other. There won't be jobs left where humans are needed.

 

[Mengy]

So the idea such mass production beginning THIS year- the ORIGINAL thing Alex was quoted as claiming and Heltok seemed to be arguing for- is pure fantasy.

I do think it's likely Tesla will build a decent amount of bots in 2024 to deploy into their own factories for testing / training, say a couple hundred or so by the end of 2024. I would not call it "mass production", although I do think it likely Tesla could set up a slow little Optimus line to make them in very low volumes. Plus having such an early production line could give insights into how to better design the eventual high volume production line, much like what Tesla did with the current Semi low volume line.

Just because Tesla isn't ready to build the high volume production line does not mean they aren't going to build a temporary slow line. The two are not mutually exclusive.

 

[Buckminster]

Why does version # mattter?

"The first one they mass produce" is what the original thing from Alex is talking about. It's the same thing I'm talking about- and I've been crystal clear on that.

Because you are talking cross purposes. Mass produce is a relative and woolly description. I might say 10k hand made V2s is mass production in 2024. Others might say that 50k V3s made by V3 robots in 2025 is still not mass production.

Much of your discussion has been about design of pcbs. V2 design is complete and they are ramping production. V3 design has perhaps just begun.

One alternative is that v2.1 is mass produced. Design again mostly complete.

 

[Knightshade]

I do think it's likely Tesla will build a decent amount of bots in 2024 to deploy into their own factories for testing / training, say a couple hundred or so by the end of 2024. I would not call it "mass production", although I do think it likely Tesla could set up a slow little Optimus line to make them in very low volumes. Plus having such an early production line could give insights into how to better design the eventual high volume production line, much like what Tesla did with the current Semi low volume line.

Just because Tesla isn't ready to build the high volume production line does not mean they aren't going to build a temporary slow line. The two are not mutually exclusive.

Didn't say they were.

And we know for a fact Tesla will do a prototype line--- in fact they're hiring for a dude to eventually build it.

https://www.tesla.com/careers/search/job/staff-systems-engineer-tesla-bot-216268

What You’ll Do

You will work closely with hardware and software engineering teams to take robot design in initial concept through prototype development and into full production
Provide manufacturing insights during product definition to streamline design for manufacturing
Drive the development of manufacturing line from initial concept to full-scale production
Spearhead the in-house development of innovative and intricate prototype manufacturing equipment in low volume line
Formulate assembly sequences and define critical quality parameters for efficient production
Partner closely with the technician team to swiftly implement design modifications on the production line and expedite the robot’s assembly
Lead internal/external equipment suppliers to design production equipment in high volume line
Identify and troubleshoot a diverse range of challenging mechanical, electrical, software, control, data, and algorithmic issues
Calculate equipment capacity and line balancing

They're clearly calling that out a low volume prototype line.... then they figure out all the things they need to further change and improve, iterate on that line until it meets their goals, and a couple steps further down the job listing you find "Lead internal/external equipment suppliers to design production equipment in high volume line" based on what they learned from the prototype low volume one.

And again, they're just now hiring the dude whose job, several steps in the future, is to build the LOW volume line first then do the other stuff after that.

Alex seems to think they're WAY ahead of where they obviously, actually, are based on the hires they're still in the process of making.

Because you are talking cross purposes. Mass produce is a relative and woolly description. I might say 10k hand made V2s is mass production in 2024. Others might say that 50k V3s made by V3 robots in 2025 is still not mass production.

Much of your discussion has been about design of pcbs. V2 design is complete and they are ramping production. V3 design has perhaps just begun.

One alternative is that v2.1 is mass produced. Design again mostly complete.

Your fixation on version numbers is kinda weird?

Model 3, from launch up until highland, was "V1" of the car--- but it had a TON of iterative changes over those 6 years of production... (though all the fundamental design specs, schematics, etc were done ahead of initial mass production- and took years to complete through prototype iterations in tiny volumes). We see the same with cybertruck.

These early pre-production, pre-even-prototype MFG equipment, is where you're constantly iterating in small (or sometimes major) ways, but only making a very few of each version.... I've already explained why it makes no sense to make large #s of one specific prototype iteration.


Also still waiting on your specific citation for "ramping production" of v2.

 

[Buckminster]

Didn't say they were.

And we know for a fact Tesla will do a prototype line--- in fact they're hiring for a dude to eventually build it.

https://www.tesla.com/careers/search/job/staff-systems-engineer-tesla-bot-216268

They're clearly calling that out a low volume prototype line.... then they figure out all the things they need to further change and improve, iterate on that line until it meets their goals, and a couple steps further down the job listing you find "Lead internal/external equipment suppliers to design production equipment in high volume line" based on what they learned from the prototype low volume one.

And again, they're just now hiring the dude whose job, several steps in the future, is to build the LOW volume line first then do the other stuff after that.

Alex seems to think they're WAY ahead of where they obviously, actually, are based on the hires they're still in the process of making.






Your fixation on version numbers is kinda weird?

Model 3, from launch up until highland, was "V1" of the car--- but it had a TON of iterative changes over those 6 years of production... (though all the fundamental design specs, schematics, etc were done ahead of initial mass production- and took years to complete through prototype iterations in tiny volumes). We see the same with cybertruck.

These early pre-production, pre-even-prototype MFG equipment, is where you're constantly iterating in small (or sometimes major) ways, but only making a very few of each version.... I've already explained why it makes no sense to make large #s of one specific prototype iteration.


Also still waiting on your specific citation for "ramping production" of v2.

The first “sort of production candidate actuators integrated and walking should be around November-ish,” Musk added.

A production ramp of Optimus should occur shortly afterward, he said.

I agree that you don’t need to be interested in v5.2 and v5.3.
But at this stage it is very important. Design is still very much in flux and you can’t swap out one bit at a time.

V3 or v4 will get hw5 which will be a big change.

 

[Knightshade]

"shortly afterward"

Ah... so 2025-2026 in Elon Time

Seriously- they're only now interviewing candidates for some of the jobs that have to be filled, and left to get some not-done-quickly work done, before there's any sort of serious ramp.


But I mean there were only like 5 bots in July 2023... so I suppose if they get to 50 total by July 2024 that's a 10x "ramp"

That's still miles from mass production.

 

[Buckminster]

"shortly afterward"

Ah... so 2025-2026 in Elon Time

Seriously- they're only now interviewing candidates for some of the jobs that have to be filled, and left to get some not-done-quickly work done, before there's any sort of serious ramp.


But I mean there were only like 5 bots in July 2023... so I suppose if they get to 50 total by July 2024 that's a 10x "ramp"

That's still miles from mass production.

Production ramp probably means at least hundreds per annum or 50 per month. Why would they wait? They could be at 5 per month now and 10 per month in Feb.

I didn't mention mass production - that's where V3 etc. comes in.

 

[Knightshade]

Production ramp probably means at least hundreds per annum or 50 per month. Why would they wait?

Because it's really dumb to build a LOT of the same prototype? Tesla didn't build 50 of the same cybertruck prototypes. Why would they with the robot?

Why would they NOT wait is a better question.

Again- what does 50 of a test unit get you that 5 does not?

They could be at 5 per month now and 10 per month in Feb.

Per month? Sure. 50 not so much. I mean, they COULD be...You just still haven't given any reason they WOULD be.

You build the absolute minimum of a version you KNOW is not production ready, figure out what worked and what did not on that iteration, then you repeat.

Absolute minimum to be defined by things like "How many different people or teams need to test this iteration" or "How many different environments does this iteration need testing in"-- things like that.


Now, if you're saying they might collectively end up with 100 -total- for the entire year, by iterating in small batches-- sure.

But spitting out 50-100 of a specific prototype per batch doesn't make much sense until MUCH MUCH later in the process... when you've gone from prototype to production candidate.

And the specific hires they're JUST NOW making tells us they remain a LONG way from that.

I didn't mention mass production - that's where V3 etc. comes in.

No- the actual post I originally replied to did.

You seem to want to have some other conversation though.

 

[Cosmacelf]

I think y'all are crazy. No way Optimus gets more than 100 of the latest gen units made until late 2025. They have a LOT of hardware and advanced, never before done, AI to figure out before mass manufacturing. Not to mention that it takes about a year, once the hardware is fully designed, before mass manufacturing starts. Other gating factors include battery supply, inference chip power usage and capability and data center capacity for training and even some amount of each bot compute.

We may still see Optimus in some factory job in the next two years, but it'll be engineering testing and validating and proof of concept stuff.

 

[heltok]

I guess the main discussion is around what we can inference from this sentence:
Spearhead the in-house development of innovative and intricate prototype manufacturing equipment in low volume line

My inference is that Tesla often have a pilot line that is separate from production lines. Think Kato Road vs Austin for 4680. Sometimes the pilot line will actually go close to "mass production", but it's mostly meant to be a learning experience which they later will scale in other production lines.

So I think the language of the listing is to indicate that the worker will not be working at the final production lines, but at the development production line.

I think the inference that Tesla will not do meaningful production in 2024 is incorrect.

Btw the number I quoted(2000) was hypothetical to have a discussion around the semantics, not my guess. I think it's more in the hundreds for 2024, but I don't discard the thousands possibility yet. I think they could do thousands if they wanted, but it's not the best use of their resources as they need to focus on software and data and I don't think they, at this point, can gather meaningfully better data with thousands than they can with hundreds and more robots bring costs in form of more support and service.

 

[Knightshade]

I guess the main discussion is around what we can inference from this sentence:
Spearhead the in-house development of innovative and intricate prototype manufacturing equipment in low volume line

...So I think the language of the listing is to indicate that the worker will not be working at the final production lines, but at the development production line.

So... you didn't bother to read the actual job description I gave the link to then- since it directly and explicitly debunks your thinking here.

Why do you keep ignoring like 90% of the content to which you then try and reply?

This isn't a job for a "worker" for the line. This is a job for the guy to design and develop the line itself

And it mentions him doing that for BOTH the does not exist yet and needs to be developed prototype low volume line
and
the actual production line that comes even later than that.

I think the inference that Tesla will not do meaningful production in 2024 is incorrect.

Ah- so now we've moved the goalposts from Alexs original "mass production" to the even MORE nebulous "meaningful"

as they need to focus on software and data and I don't think they, at this point, can gather meaningfully better data with thousands than they can with hundreds and more robots bring costs in form of more support and service.

Funny story. They're also hiring HUMANS to do the data gathering.

Apparently that's cheaper than building hundreds of prototype robots they're not remotely ready to actually build yet.

https://www.tesla.com/careers/search/job/data-collection-operator-tesla-bot-213088

First 3 listed tasks are:
Walk a pre-determined test route daily for data collection
Wear a motion capture suit and perform designated movements and actions based on project requirements
Start/stop recording devices and perform minor equipment and software debugging

 

[heltok]

So... you didn't bother to read the actual job description I gave the link to then- since it directly and explicitly debunks your thinking here.

Why do you keep ignoring like 90% of the content to which you then try and reply?

This isn't a job for a "worker" for the line. This is a job for the guy to design and develop the line itself

And it mentions him doing that for BOTH the does not exist yet and needs to be developed prototype low volume line
and
the actual production line that comes even later than that.





Ah- so now we've moved the goalposts from Alexs original "mass production" to the even MORE nebulous "meaningful"





Funny story. They're also hiring HUMANS to do the data gathering.

Apparently that's cheaper than building hundreds of prototype robots they're not remotely ready to actually build yet.

https://www.tesla.com/careers/search/job/data-collection-operator-tesla-bot-213088

First 3 listed tasks are:
Walk a pre-determined test route daily for data collection
Wear a motion capture suit and perform designated movements and actions based on project requirements
Start/stop recording devices and perform minor equipment and software debugging

Semantics.

Anyway, I don't think we are getting anywhere. I stated what I believe. You state what you believe, I appreciate that. We disagree. Neither of us is convincing the other with our arguments.

 

[Knightshade]

Semantics.

You claimed the job listing told you he would only work on the prototype line not the production line.

I pointed out the job listing literally says he will work on both. And not merely "work" as you wrote, but design both lines.

That is not semantics. At all.

I stated what I believe. You state what you believe

No.

You stated what you believed.

I pointed out Teslas own words literally disprove your belief (and not for the first time).

Then you hand wave it away as semantics.

I agree we are not getting anywhere- but it's primarily because you continue to refuse to actually make any attempt to read or understand the content to which you are replying. You instead ignore it, post things directly contradicting what Tesla themselves tells you, then dismiss any attempt to point that out.

 

[heltok]

I pointed out Teslas own words literally disprove your belief (and not for the first time).

I disagree on what the words prove/disprove.

Anyway. I will end my part of this discussion now.

 

[MC3OZ]

Tesla could absolutely see an excess of 5 Billion in profits from the Bot per year. You are right that as competition comes, there will be price and margin compression. This is simply 1 example of many, many tasks that could be automated. But each year that goes by, the capabilities of the Bot will grow non-linearly and can be sold into new markets.

If the AI scales in that way, the market will truly be supply limited for a long while. Given their hardware / software competency, I fully expect a Tesla Bot factory in whichever market to be COGs competitive. Gross margins will certainly not stay at 50%, but even downward pressure alongside massive growth in revenues can ultimately lead to tens of billions in annual profts while only reaching 10-20% bot market share.

I've decided to reply in this thread...

Humanoid Robot Competition - China

Tesla Optimus Sub-Prime Robot

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...

teslamotorsclub.com

This Chinese robot looks like a copy of Optimus Gen1, because it is the video is "proof of copy" rather than "proof of concept" because the aim was to replicate everything a Gen1 Optimus could do..

One video estimates $30,000 to build the Chinese robot and $20,000 to build Optimus... both numbers seem about right to me.

One the hardware side this is how I see the Chinese progressing, always one generation and 1-2 years behind Tesla.

For robots that can work in car factories and many other locations in China, the Chinese robots will be more than good enough...

Outside of China Optimus has one major advantage - It isn't built in China

Most countries will not make it easy or legal for large number of Chinese Robots to work in factories in their countries - Optimus will have a higher level of trust,.

Humanoid Robot Competition - Startups
Tesla has a big advantage here in relation to resources, experienced staff, factories, engineering and design, data and training hardware.
The data advantage will come form a large fleet of robots...
In some ways the Chinese robot competition is similar to Chinese EVs . except they will find it hard to be cheaper than Optimus..
US and EU robot startups similar to US and EU EV startups, most startups will not make it, but will be a value source of staff and resources for Tesla and perhaps the few that do make it.
it is hard for a startup to compete with a much better resourced startup, and that is the bottom line..

Dumb Robot Competition
These are definitely cheaper, but are one dimensional doing one task reasonably well, but generally making humans more productive, not replacing a human.. Cost to make one of these more like $1000 so definitely an order of magnitude less that Optimus..
Perhaps may allow a restaurant to operate with 3 waiters rather than 4, but when one waiter calls in sick, adding another dumb dish carrier does nothing.
My hunch is around a 50/50 split between people and robots, so 2 human waiters and 2 Optimus, the restaurant might still have the dumb dish carrier, because the dishes can go straight to the Optimus that loads the dish washer...
Now if a waiter calls in sick, the restaurant may be able to borrow an Optimus from the kitchen, hire one at short notice, and borrow one from somewhere.
Operating with one human waiter and 3 Optimus waiters is far from ideal, but most of the time an extra robot will be easier to get than an extra human.,
With the 50/50-split, I still see most able bodied humans who want a job being able to get one...

Resources
Tesla has obviously pivoted and the Optimus project is well resource, staff, equipment, labs, training hardware..
Building the fleet relatively fast should yield a data advantage... similar to cars, Optimus should gather training data..

Secret Sauce
Does Elon think Tesla has some type of advantage or lead in AGI that will take time to replicate?
Pay attention to comments Elon makes like - "we are solving general artificial intelligence"
My impression is Elon might think that, and we might not understand what he is really talking about...
I can't judge whether or not Elon is right or wrong.. but he seems bullish, and I don't know all that he knows.
if it exists, the secret sauce might be one more reason why the competition isn't a short term threat.

Project Status
Seems to me like they have built some "Gen2 Release candidates" and they are testing that the hardware can do everything necessary before starting trial production. They are testing object manipulation, walking, dexterity, etc rather than detailed training on specific tasks.
The next step would be a small scale pilot line perhaps 10 per week eventually ramping to 30-50 per week.

Deployment
I think they can phase in small numbers of Optimus robots on easy well defined task at Tesla factories at anytime, that will double as fleet and longevity testing.. The size of the general training lab might not increase for a while, that lab could be trying to refine the product and increase the range of tasks Optimus can do. A separate lab at a factory might be slowly trying to increase the number of factory tasks an Optimus can do.
The constraint is probably training compute and trainers, hence no need to a rapid ramp up in robot volumes.
After a while Tesla might spin up a program to start early training at other worksites, both within Tesla and external to Tesla.

End State at Tesla
My hunch would be a 50/50 split, some places might have a lower percentage of robots, but a lower percentage of human workers might be problematic. Tesla will eventually produce more Optimus than they can use in house, so these can be sold/leased..
1 Million Optimus sold/leased per year by 2028-2030 seems like a reasonable target, after that the question is TAM. I think many work places could get to a 50% share of robots, or at least a 30% share, Offices and educational facilities might be the exception, any job with a high intellectual content seems problematic. Customer enquires less so, because the robot can follow a predefined script and search for answers.. Sales probably needs a human. 1-2 robots per household is a reasonable long term projection.

 

[Buckminster]

* Dave Lee did a youtube video here on Tesla's Optimus folding laundry video. He basically sees the video as demonstrating the dexterity of Optimus's hands. He feels that balance is coming (other robots do it) and by getting the hardware right like this Optimus could see surprisingly quick growth of software that enables a plethora of useful tasks.