Tesla Optimus Sub-Prime Robot

[Knightshade]

First it would be humans building it by hand, like it is today.

Those wouldn't be final production units. Those would be the "we are still figuring out what actual parts to use" units where they're changing HW to meet design goals as testing finds issues.

Once you finalize -then- you have to go back to suppliers (internal and external) and pricing and supply chain stuff to insure an orderly flow of final parts into the process.... as well as final design/tooling on what the actual line needs to make the final product.

Again read the first 9 tasks on the job listing. All of that stuff has to happen, multiple iterations of many of those steps typically, before you build your first mass production unit.


They're not getting all that from "posting a job listing to find the guy to run that" to "we can mass produce thousands of final units" all in this single year.

 

[heltok]

Those wouldn't be final production units. Those would be the "we are still figuring out what actual parts to use" units where they're changing HW to meet design goals as testing finds issues.

Once you finalize -then- you have to go back to suppliers (internal and external) and pricing and supply chain stuff to insure an orderly flow of final parts into the process.... as well as final design/tooling on what the actual line needs to make the final product.

Again read the first 9 tasks on the job listing. All of that stuff has to happen, multiple iterations of many of those steps typically, before you build your first mass production unit.


They're not getting all that from "posting a job listing to find the guy to run that" to "we can mass produce thousands of final units" all in this single year.

We will see how much they can get done in a single year or not. It's not that complicated parts they use like gigacastings etc. Which component of the robot do you see will be that hard to make 2000 of in a year? Also they did not start this year, likely they had an idea about what parts would be needed and which had long lead times a while ago.

 

[Knightshade]

We will see how much they can get done in a single year or not. It's not that complicated parts they use like gigacastings etc.

You seem to keep skipping over most of those first 9 steps.

They not only don't know what parts they will build yet, they don't even have the final specs for what they need them to do.

That's WHY they're hiring that position- to determine all that, write it up, THEN move on to actually designing and sourcing what they need for those parts... THEN and ONLY THEN figuring out the process to assemble all of them.

Which component of the robot do you see will be that hard to make 2000 of in a year?

Once they're at step 10? None of them. It's just not a fast process to get to step 10.

When a PC company plans to make a new laptop- and I mean a pretty standard one similar the ones they've already been making for years and years just maybe with a few updated chips and small changes to things like the screen or shell-- those first 9 steps STILL take more than a year.

This is far more from-scratch work than any of that stuff.

Also they did not start this year, likely they had an idea about what parts would be needed and which had long lead times a while ago.

Obviously not, since LITERALLY the first few tasks for this job posting are that.

What part we need is not 'we need hands"

it's "we need hands that meet all the following performance specs, and hardness specs, and cost specs, and power specs, etc"

If they already HAD all that they wouldn't be hiring this position.

 

[heltok]

You seem to keep skipping over most of those first 9 steps.

They not only don't know what parts they will build yet, they don't even have the final specs for what they need them to do.

That's WHY they're hiring that position- to determine all that, write it up, THEN move on to actually designing and sourcing what they need for those parts... THEN and ONLY THEN figuring out the process to assemble all of them.




Once they're at step 10? None of them. It's just not a fast process to get to step 10.

When a PC company plans to make a new laptop- and I mean a pretty standard one similar the ones they've already been making for years and years just maybe with a few updated chips and small changes to things like the screen or shell-- those first 9 steps STILL take more than a year.

This is far more from-scratch work than any of that stuff.



Obviously not, since LITERALLY the first few tasks for this job posting are that.

What part we need is not 'we need hands"

it's "we need hands that meet all the following performance specs, and hardness specs, and cost specs, and power specs, etc"

If they already HAD all that they wouldn't be hiring this position.

I think you draw too much conclusions from job posting wordings. Imo they are hiring for making millions of bots. The wording might be thousands, but that's just wordings. Does this mean that they cannot do thousands of bots this year? No. I am not saying that they will, I am just not ready to discard the possibility that they might. We will see.

 

[Knightshade]

I think you draw too much conclusions from job posting wordings.

What do you think the first 9 tasks on the job listing mean?

Imo they are hiring for making millions of bots. The wording might be thousands, but that's just wordings.

The wording is fine for either number.

But you don't even start building the first production unit out of EITHER thousands OR millions until step 10 on that list of job responsibilities.

That's the part you keep skipping over.

You don't mass produce anything before you have your product requirements and constraints, before you've completed design of your sensors, before you're finalized and documented the actual system architecture and created design documents, before you've created schematic designs for your electronics, before you've done PCB layout and design, before you've created and defined your BOM, before you've prototyped all of that, and before you've then validated and qualified the results of all those steps.

Then, and ONLY then, do you actually start production.

They have NONE of those things today. We know this because they literally posted a job looking for someone to DO those things.

And BTW this job posting is JUST for the sensors. Which have to go through all those steps listed before production of JUST THE SENSORS.

They'll need to hire other people for other aspects of the bot to do all those same or similar tasks for their own components.

Meaning task 10 on this guys list isn't even "mass produce robots"

It's "mass produce sensors"

You still need all the OTHER parts that go into the robot to ALSO be at stage 10... and have a master plan to integrate those parts with each other (which is why other teams are mentioned during a number of those first 9 steps.... you don't want to end up with a sensor expecting 5v, but the whole time the battery team was designing a 12v system for example).

This isn't Junkyard Wars where you just bolt a bunch of random parts together-- modern manufacturing has a lot of steps you have to do, and get right, before production starts.

 

[heltok]

But you don't even start building the first production unit out of EITHER thousands OR millions until step 10 on that list of job responsibilities.

If a unit is production or not is semanatics. And I don't think there are any physical laws that would be violated by building a unit before step 10.

I don't think the Bot project has all the same properties as previous technologies. It will be able to assemble itself, what other thing could do that? And Tesla are not following the same rule books making rapid iterations. AI is rapidly improving making it even stranger.

I would not be surprised if there never is a "mass production" unit on the way from 0 to 1M/year, maybe they will rapidly iterate the whole way there. Maybe the thousands of bots assembling millions of bots will be different than the bots they are building.

Look at this picture:

I would guess the #1 was n=1, the #3 n≈30(we saw like 6 i the last video and they are in showrooms). What do you think n will be for #3? And when do you think we will get #4?

The nice thing with AI, simulation etc is that you can easily train a model to learn any differences. Just add noise to every parameter in the simulation and the bot will learn to figure out what to do. They showed self calibration. As long as the input is mostly the same or can be generated with a GAN it's no major problem if they use 6 or 8 cameras etc and the human form factor will remain human.

 

[Knightshade]

If a unit is production or not is semanatics.

it's REALLY not.

And I don't think there are any physical laws that would be violated by building a unit before step 10.

A unit? Not at all.

You are likely to build a number of units before step 10.

Just none of them will be production- they'll all have different design/spec/etc elements from each other-- because you are iteratively prototyping until you get to the version you want to make a LOT of the same one of. Which is literally what mass production means.

There's a half dozen steps in that list before you even get to your FIRST prototype though- and you keep ignoring them as if they don't exist and take significant time.

Then the prototyping also takes time.

Then you have to do the BOM and supply chain stuff based on the FINAL specs working with your suppliers, tooling for the factory, etc.

THEN you can make LOTS of the SAME something- which is the whole point.

I don't think the Bot project has all the same properties as previous technologies. It will be able to assemble itself, what other thing could do that?

Not sure how you think this matters in any way? Once it's ALREADY built it could do that.... in literally the same way a human already can.

And Tesla are not following the same rule books making rapid iterations.

The job listing says otherwise- those steps are the same rule book as everyone else designing things for mass production.

Tesla might have fewer iterations when they get to prototypes- but that's most of the way through those 9 steps already.

I would not be surprised if there never is a "mass production" unit on the way from 0 to 1M/year, maybe they will rapidly iterate the whole way there.

This is utterly nonsensical to anyone who has ever had any experience in this field.

Tesla is known for iterating on the production line in small ways. All the major design decisions were done over the course of YEARS before that. That's the stage the robot is at- they're still at the start of major design decisions. You don't start building a lot of something before then unless you love wasting money-- which Tesla does not.

Look at this picture:
View attachment 1009512

I would guess the #1 was n=1, the #3 n≈30(we saw like 6 i the last video and they are in showrooms). What do you think n will be for #3? And when do you think we will get #4?

You're seeing stuff that's basically hand built to test out ideas. Cheaper CNC machines and 3D printers make doing that a lot easier and faster than it used to be, but once you figure out what's possible you still need to go back and do all your design work, schematics, board designs, BOM creation, supply chain work, all the stuff in those 9 steps.

Otherwise when you get to step 10 and say "Ok, go make 1000 of these" or whatever large # you want, you discover you can't for a slew of reasons.


Again, they're still hiring folks to figure out the design goals. Then the other 8 steps before you actually make a production version of anything.

And again it's not like it's JUST for that job either-


Another example of jobs currently still hiring for the bot:

https://www.tesla.com/careers/search/job/mechanical-design-engineer-tesla-bot-93683

This guy will need to design joints and related structures-- not "make" but DESIGN in the first place... then insure they integrate with all the other systems- and insure they're easy to assemble and service-- and fabricate prototypes for evaluation, including designing the testing equipment, and then eventually develop final specifications for volume production.

There's also at least half a dozen software jobs that make clear Tesla is FAR from production level on that side as well.... one asking you to develop "end-to-end robotic learning with either reinforcement or imitation learning" - a thing some folks seem to mistakenly think they already have....

 

[heltok]

Otherwise when you get to step 10 and say "Ok, go make 1000 of these" or whatever large # you want, you discover you can't for a slew of reasons.

Can you list these reasons?

I don't see any major problems for parts. HW4 already there. Batteries etc there and Tesla are experts at quickly fixing supply chain issues here. Motors not that hard. The plastic/metal/fabric not that hard.
Assembly not that hard either.
Software is the main bottle neck, but most of the software 1.0 is the same as the car and the software 2.0 seems to be coming along nicely.

So what's left?

The main problem I see is dealing with customers, support etc. But if they are their own customer that's less of a problem.

 

[Knightshade]

Can you list these reasons?

If you have no documented specs how do your suppliers make parts for you? How do they or you even qualify them as doing what you need?

If you have no PCB layouts how are your boards manufactured at a plant?

If you have no schematics how does anyone know how to wire anything up correctly or consistently?

if you have no BOM you have no idea how much it'll cost you to build anything- nor any way to insure what you built this shift is the same as what you built last shift (and thus no way to figure out WHY one or the other has a specific problem). You also have no way to insure you're not idling your line waiting on any specific part rather than knowing you're good to build. You have no way to plan and control supply chain.

The list goes on and on and on...

All of this stuff is the difference between hand-building a tiny amount of non-identical test units with either off the shelf or one-off parts and mass producing production models of anything which inherently need to (at least batch to batch) be identical and meet specific criteria.

They're fundamentally different things

I don't see any major problems for parts. HW4 already there. Batteries etc there

Yes those off the shelf parts are what are going into the hand built prototypes because the actual custom versions do not yet exist

That's why they are hiring these people

To go from "cobbled together hand made stuff, and parts that were originally designed to work in totally different places" to "a complete package designed to work together in a robot and be manufactured in bulk"

Even for parts that WILL get reused- the 2170s are the same, but the pack is not going to be. Neither will the overall cooling or BMS systems. These all have to be designed, engineered, and qualified- then a supply chain spun up for them.

Same with the sensors which again are not going to be identical to the car in the production model (hence the job posting for a guy to ground-up spec all that out)

and Tesla are experts at quickly fixing supply chain issues here.

You can't "fix" a supply chain for parts that not only don't exist yet, but you haven't even written a design document for yet.

Motors not that hard. The plastic/metal/fabric not that hard. Assembly not that hard either.

I guess you should apply for one of these jobs since you seem to have this much more figured out than Tesla themselves do.

Software is the main bottle neck, but most of the software 1.0 is the same as the car and the software 2.0 seems to be coming along nicely.

Again I just cited a job posting that says otherwise.

The main problem I see is dealing with customers, support etc. But if they are their own customer that's less of a problem.

Yes, because you keep ignoring the 9 steps in that job description that come BEFORE mass production, then when someone tries to explain why they matter you just hand-wave it away saying it's all "not that hard" without seeming to understand what those things even ARE.

 

[Buckminster]

https://twitter.com/x/status/1747403399895638395

 

[Buckminster]

https://twitter.com/x/status/1747401716990095576

 

[heltok]

If you have no PCB layouts how are your boards manufactured at a plant?

Which boards? I assume they will use identical HW4 boards and making a few thousands extra should not be hard when they already make millions.

Yes those off the shelf parts are what are going into the hand built prototypes because the actual custom versions do not yet exist

Which things need to be custom? Motors yes, but they are good at making motors and that's where they started the presentation at Autonomy day and the latest bot was to verify that they were good enough.

If you have no schematics how does anyone know how to wire anything up correctly or consistently?

At lof the wiring will be through the actuators and the amount of wiring will not be that crazy. A lot will be like Model 3/Y but a lot less length and power. I fail to see how it will be so hard to make a schematics of the wiring.

Same with the sensors which again are not going to be identical to the car in the production model (hence the job posting for a guy to ground-up spec all that out)

The haptic sensors in the hand yes. But they seem to at least have 10 working ones already, likely a lot more. Will scaling these be the impossible bottle neck? Getting volume of the cameras with the right specs should be pretty easy.

I guess you should apply for one of these jobs since you seem to have this much more figured out than Tesla themselves do.

I believe Tesla can figure this out on their own. And again I think you are reading a bit too much into the job listings taking them as proof that Tesla are years away from making 2k units/year.

 

[adiggs]

So if they produce 2000 units in 2024 you would consider that mass production?

At least for me the number produced is secondary to how they're produced. If they put together 2000 robots for engineering purposes or something akin to how the original Roadster was manufactured, then that isn't mass production.

On the other hand if they have built a factory and build 2000 of them using the beginnings of a mass production process - then that's the start.

 

[Knightshade]

Which boards? I assume they will use identical HW4 boards and making a few thousands extra should not be hard when they already make millions.

Again you appear to just be pretending like 80% of the stuff they're hiring this dude to do was already done before they hired him. PCB design is literally a task they're hiring someone to do, and you act like you don't think they have to design any.

Which things need to be custom? Motors yes, but they are good at making motors

They're good at making EVs and look how long it took to refresh the S and X last time.

This is significantly more complex/different.

and that's where they started the presentation at Autonomy day and the latest bot was to verify that they were good enough.

It was good enough for a next-iteration of testing. Not remotely good enough for mass production.

Because again, they're listing jobs to hire people to GET them to good enough for production. They're quite some distance from that today.

At lof the wiring will be through the actuators and the amount of wiring will not be that crazy.

Ok, but someone has to develop a final plan for all of it, write up schematics and wiring diagrams, etc.... Again they're JUST NOW hiring design and architecture people for that--- you're not remotely close to production at that point.

A lot will be like Model 3/Y but a lot less length and power. I fail to see how it will be so hard to make a schematics of the wiring.

How many schematics for complex electronic systems have you personally written?

Is it zero? Because it sounds like zero.

The haptic sensors in the hand yes. But they seem to at least have 10 working ones already, likely a lot more. Will scaling these be the impossible bottle neck?

Since they're not remotely close to final versions you wouldn't want to scale them.

That.

Is.

The.

Point.

Of.

The.

People.

They.

Are.

Hiring.

You do a lot of design, architecture, and planning FIRST.... then you iterate prototypes until you either hit, or evolve, your plans to something that works, is easy to make a lot of, easy to source a lot of, is easy to assemble, and is easy to maintain.... only after all that do you start mass production.

Getting volume of the cameras with the right specs should be pretty easy.

You mean the specs they don't have yet, and are just now hiring people to figure out?

Then after they figure them out they need to actually get a supplier to build them- then test and validate them.

Again this isn't "Eh, buy whatever Best Buy has on the shelf this week"

I believe Tesla can figure this out on their own. And again I think you are reading a bit too much into the job listings taking them as proof that Tesla are years away from making 2k units/year.

And I believe you're just ignoring the 80% of the job listing you keep demonstrating you don't understand any of the details of.

Guess we'll see who was right in 12 months.

 

[Knightshade]

https://twitter.com/x/status/1747401716990095576

"if Tesla bot sells for 100k"

Elon said "less than $20,000" was the goal.

Does nobody bother to read anything Tesla actually says while making up their own narratives about robots?

 

[heltok]

Again you appear to just be pretending like 80% of the stuff they're hiring this dude to do was already done before they hired him. PCB design is literally a task they're hiring someone to do, and you act like you don't think they have to design any.

So which PCB will they have to design?

 

[Knightshade]

So which PCB will they have to design?

Ones that, according to Tesla themselves, haven't even been designed yet.

For one though they're not going to be using off-the-shelf vehicle MCU/HW3/4 units in the actual production-level robot because that wouldn't make much sense.

Likewise there's other PCBs in existing Tesla products (BMS/charge control for example) likely to be different from the car to the bot-- same with motor controllers and more.

As I say, bolting on off-the-shelf stuff just to get a working test item is fine-- but nobody sane does that for something you're going to be making thousands (let alone millions) of.... instead you custom design your parts for all the things I listed previously, all of which will pull out weight, cost, manufacturing time, and maintenance time across those thousands or millions of units.

 

[Buckminster]

As much as nobody is enjoying your discussion, can I suggest you talk about the same thing?

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

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

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?

 

[heltok]

Ones that, according to Tesla themselves, haven't even been designed yet.

For one though they're not going to be using off-the-shelf vehicle MCU/HW3/4 units in the actual production-level robot because that wouldn't make much sense.

Likewise there's other PCBs in existing Tesla products (BMS/charge control for example) likely to be different from the car to the bot-- same with motor controllers and more.

As I say, bolting on off-the-shelf stuff just to get a working test item is fine-- but nobody sane does that for something you're going to be making thousands (let alone millions) of.... instead you custom design your parts for all the things I listed previously, all of which will pull out weight, cost, manufacturing time, and maintenance time across those thousands or millions of units.


View attachment 1009656

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. And what's mass production or not is semantics. Imo what matters is how many robots they build, how useful they are etc, not if they are mass produced or not.

 

[jeewee3000]

I should really retrain to become a humonoid robot repair-guy, offering cheaper and faster service than Tesla whenever your good 'ol house robot breaks down.

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