Tesla may have just leaked Optimus Gen 3 inside its own app, but one of the top robotics experts covering humanoids thinks it probably isn’t real. And the reasons why tell us exactly what to look for when the real Gen 3 finally appears.
00:00 Optimus Gen 3 Leak 02:35 APK Assets Breakdown 05:10 Comparing Gen 3 Silhouette 08:21 Production Ramp Reports 18:13 China Supplier Audits 21:06 Giga Berlin Data Capture 25:32 Deep Dive Article Findings 29:34 Armor Covering Strategy 32:08 Shoulders And Neck DOF 37:12 Hip Overhaul Predictions 44:25 Kidney Style Waist 48:13 Hard Cover Or Fabric 50:19 Hand Complexity Tradeoffs 54:09 Reveal Timeline And Wrap
This video analyzes leaked images and technical reports regarding Tesla's Optimus humanoid robot, debating whether a \"Gen 3\" model exists or if recent visuals are mere placeholders. Experts discuss the mechanical challenges of scaling production, the transition to a Robot-as-a-Service (RaaS) business model, and the complex engineering trade-offs required for industrial durability.
Can you believe this? Tesla just either accidentally or not so accidentally leaked what looks like
the Optimus Gen 3, their next version of its humanoid robot, right inside the Tesla app.
But is this the real deal? Our guest today, one of the top robotics experts covering humanoids,
took it apart piece by piece, and his answer might surprise you. Plus, we'll also give an update on
where Tesla stands in the production of these bots, with some reporting now saying Tesla is at
And others like the information reporting Tesla is facing challenges manufacturing Optimus hands
at scale. And we have hints now on who and how Tesla will sell to its first customers.
So here's how this leak happened. An image found inside the Tesla Android app is making the rounds
that claims to be our first look at Optimus 3. And of course, the community let up. Then Dr.
Scott Walter, one of the most trusted robotics voices covering humanoids, took a closer look.
And his verdict is probably not the real thing. Here's why he's skeptical. The Optimus 2.5 shown
next to it doesn't match Tesla's official photos. The design keeps hard outer shells while every
competitor has moved to soft fabric coverings. And overall, it reads like a facelift, not a new
generation. But here's where it gets very interesting. Even a fake tells you what to
watch for. He says the real changes should come in the hips and the waist, a redesign
that would finally let Optimus bend forward, something it has never been able to do. And he
believes the hands with 22 independent movements is the piece holding up the reveal. We've still
never seen it working. So is this the real Optimus 3? Probably not. But the real one is coming. And
now you'll know exactly how to judge it. Joining us today is Scott Walter. Scott is an aerospace
engineer with a PhD in mechanical engineering and has co-founded two robotics companies. Scott is
head of research at RoboStrategy. We may talk about companies today that the fund may hold or
trade. And as always, keep in mind, nothing in this video is a recommendation to buy or sell
any security. Welcome, Scott. Hi, Herberto. I think it's going to be a really interesting breakdown
today. I'm excited. I'm excited. Obviously, a lot of us Tesla investors have been waiting for so
long for Tesla to finally show what Gen 3, Optimus Gen 3 is going to be like. Elon has hyped it up.
He was going to reveal it. And then he pulled back saying, we don't want to share it until
after we've started production. And people are saying it's going to look as much like a human
as possible. So is this a leak? Was it intentional, not intentional? I don't think it was intentional,
but we'll go through it in detail. But here's what happened. In Tesla's Android app, there is a
version where inside of that app, you can actually see these images, these asset packages contains
three files, RobotCloseupGen3.png. It's an image. RobotHomeFullBodyGen3. RobotHomeInactiveGen3. All
of them say Gen 3. This confirms that Tesla internally labels the asset Gen 3, though it
may not represent the final production design. Here is the image that they shared. So it looks
different than Gen 2.5. And this is kind of like a comparison here between the two. This is just
another person saying that the assets were discovered in the APK. And so you can see here's
Gen 2.5 here. You can see the hips were still, you know, very much like robot hips. You can see the
kind of like these, what do you call these actuators at the shins. Here's what it looks
like here in 3.3. We'll go do that. Now you wanted to share this photo. What's this about?
Well, I guess the first thing, of course, I was hoping like everyone else, that we would get a
sneak peek at Tesla at some point, or Gen 3 at some point. And it turns out it's like one of the
one of the first times it was actually a benefit to have the Android version of the Tesla app.
Everything is on the Apple. So something is in there. And I just suspect that it's a placeholder.
The reason why I suspect it's a placeholder is just kind of look at this, is that basically,
if you take the concept optimists, and you take the Gen 2 optimists, and if they had a baby,
that's what it would kind of look like. So someone might just said, let's just go ahead
and kind of blend these things together, because we need to have some sort of asset for this app
as a placeholder for the development. So I don't put a whole lot of stock in that necessarily being
the right thing. However, remember, Elon keeps on talking about people doing frame by frame analysis.
Right now, I'm going to take one frame, and I've written like a super long article that I think is
about 2000 words long, or something like that. So this is what he's afraid of, like, even just a
little image like that. So if it was, even though we can't see anything that's behind it, we can
intuit quite a bit of what the possibilities are. And even if it's not the real thing, we, we can
still kind of speculate what they are probably going to do based on the old design and what we
are hoping to see in the first place based on because the way everyone else is moving the
direction of, let's say, the whole industry. Yeah, I know, I can't wait, because I your
analysis is, is gold. You explain why some things make sense, and some things may not make sense.
So that's why it's very important to understand this. So here's, here's a visual of what Gen three
versus Gen 2.5 looks like. Here's the video of that. That Yeah, that's gonna be the Gen three
right there. Yeah, so you can kind of see the contrast between the two. And you kind of get the
the feeling that he put on a few pounds. And I made a little, little remark about that. It doesn't
mean necessarily, it's just that when you kind of look at it does look a little bit heftier. Maybe
he's, you know, gone to the gym and everything else. And of course, when you start putting
coverings to take some of the exposed areas that when they're you, you would not be surprised.
But again, this could be easily just something that someone generated with AI just said, Hey,
grok, take the two. And what would it look like? Literally, of course, could have been.
So here's humanoid daily who, who overlaid the two together. And then we're going to take a look
at that. But he's saying the Gen three head is 12% wider, the neck is 11% thicker, the shins are 15%
wider, the shoulders sit lower hips and calves, fully covered. Let's watch. You want to talk us
through this image this video? Yeah, that's pretty good. Because you know, instinctively,
you look at the two next to it, one of them, the arms look a little bit thinner, the other bit
thicker. But it's interesting, he was able to kind of come up with some numbers that indicate
that, you know, if that is indeed the case, it looks like it's added mass. But then again, it's
just covers. And we don't know what's behind the covers can be a lot of air behind there as well,
that keeps a little bit lighter, or using some sort of different material, it's also a lot lighter,
you know, they could be going to magnesium or aluminum or something like that, if they wanted
to reduce it, because we have some other reporting that we'll get to later, that indicates that Gen
three should be lighter than than the previous version. So it doesn't mean it couldn't look a
little bit bigger. But still, it couldn't be a little bit lighter. Okay, so that's what I was
going to say, because this does look bigger, the heads 12% bigger neck is 11%. The shoulders are
lower, right? And one of the things that I like there is the profile is that something I pointed
out in my article is that if you look at Gen two, you know, the shoulders seem like this. And now
the shoulders seem to be dropped down, and they seem to be a little bit more human life. But again,
we're reading is something that may not exist. But if that is the case of the direction they're
heading in, you can see Oh, yeah, the shoulder profile definitely looks a little bit better.
The forearms look a little bit thicker, we can swipe with the leg and everything else.
So that's just a silhouette, we could probably get into some of the actual details of what we think
what would be going on with us. But and also, you know, why, there's many reasons to be sort
of skeptical that this would be a way that they'd be leaking that, you know, that easily. And of
course, not as like I say, it just looks more like a refresh as opposed to a major overhaul.
Yeah, think about it. And you know, you've been working in a large corporation. So by
if this is such a secret project, think Apple, they hide everything that triple badge to get in.
Nobody's gonna go, okay, we're gonna get the Android app. And we're gonna put you know,
here's images, go ahead and put it there ahead of time. But don't share it yet. And somebody
screwed up. That's not likely going to happen only when they've decided to go for it. Will they
then share all the assets around including the images and so forth. But let's go closer into
the analysis of what the bot should look like. You'll go through your article. But before that,
there was an article written by the information and they are saying some interesting things.
They first say that the optimist hits snags in the hands, suppliers as scale up. So both
they're saying two things at the same time, right? One, they're saying there's 1000 per week
optimist being built now, but two that they're having difficulties with the hands. I guess both
could be true. Optimist production increased tenfold. But Tesla faces manufacturing snags.
Robots intricate hands pose major manufacturing durability challenges. Optimist AI needs
extensive training with 500,000 hours of data collectibles. So does every other bot does too.
So they go this must much still has to be done to produce a robot Tesla can sell.
Yes, every robot, the robots coming off its Fremont factory line. Optimus v3 are not the
exact version the company plans to sell to customers. We already knew that version four
will be the version for you. The company is still fine tuning that version which will have to meet
stricter durability and reliability thresholds before Tesla offers a commercial that will be
next year. The v3 is lighter, equipped with more cameras and designed to look more human
than earlier versions of the robot.
You're familiar with the design set. So again, that sort of indicates what I was just saying,
the expectations, it would be lighter by this. And of course, we expect it to look more human,
because it's been said that before that it that it would look like a person in a suit.
And you know, more cameras, I'm not surprised about that. I'm not exactly sure how visible
they would be. But that's why I don't believe that what we see in there is the actual design
of the v3. And I said it might be just a canard to get everyone to start talking about it.
Now that overall article, it does not surprise me that that article has been written, I had
absolutely nothing to do with article was not a source of article in any way at all. However,
it matches up with a lot of other things I've heard in my recent, my recent trips out to the
valley and everything else. And that there have been a lot of snags in the manufacturing,
definitely of the hands. I've heard it more from like the durability of the hands after they've
been built, not so much with the building. But that does kind of make sense. And, you know,
some of the reporting there seems it could be the sensors are a problem. And that they're thinking
of actually going to a glove to cover. And that would kind of make sense. As I always thought,
it's like, don't don't build the sensors into the hand, because there's all kinds of issues.
Just go ahead and get a glove that you can put on. And that makes more sense,
because then that's a glove a person could also wear. And then you have the ultimate
kind of interface right there is you have a five fingered hand, which is getting tactical
information that you'd be able to do that. So it sounds like that's what's going to do.
And we've always heard that, you know, the V4 would be the one that would be
not quite sold, but let's say made available commercially, because we suspect it's still
going to be a RAS model. And that internally, they will accept the fact that maybe it has
durability issues, because that's the growing pains you're going to go through, they're going
to put them into production, they're going to be testing them out, finding what have to do,
refine it. And then maybe by version four, they would have something that would be available,
which I think Elon had mentioned something like, they would be available to customers by sometime
in 2027. Maybe the end of the year, so maybe it's lining up towards there. You know, I suspect
because these timelines have been slipping is probably more like early 2028, something like
that happens. Okay, let's go through a few more points that the article said last month,
Tesla produced 700 robots a week. Okay, don't know how true this is. But that's amazing. 700
robots a week, up from a few dozen robots per week during the second quarter when the company
was doing small batch testing for production. According to two people familiar with the work,
managers have told staffers, the company wants to set up a continuous automated production line
capable of producing more than 1000 robots a week by the end of the year, people said that is still
far short of its eventual plan to produce around 20,000 a week, but still 1000 a week sounds pretty
good. Yeah, that's that's about 50,000 a year. And of course, the goal here is to get a million out
that plant, you know, they're building another one to do 10 times that. So if that's the goal,
it sounds like maybe they're planning to have 20 lines. Because if they're able to get 1000
out of one line, if you do 20 times that you'd be able to do it, or maybe they plan on getting it
up. It's a bit interesting. Now, there's three things that are said in the article. I know that
like we showed a little bit of the summary that there was like a tenfold increase, the actual
wording used in the article was roughly tenfold. Okay, so roughly tenfold to me usually means not
quite 10. But let's just say it's 10. Several dozen exactly what the several dozen mean. Well,
let's just say that some people, it might mean two dozen, somewhere between two and three dozen.
And if you do a tenfold on that, that means you're somewhere around two or 300, something like that.
The term several hundred also can be it's usually like 300 or more. So they might be right around.
So if they are in that ballpark, it seems like well, whatever is kind of holding them back right
now, which seems like, you know, manual assembly, the normal kind of production ramping, you know,
in the next 90 days, to be able to do a 3x of that, you should be able to hit 1000 potentially.
So it doesn't seem like it's a big deal yet. It seems like this, there's this command that comes
down. And let's go back exactly to the wording of that, because I think that's very, very important
to look at. There's two ways of reading what was just said. One is that they plan on hitting
1000 units a week by the end of the year. Or they say they're getting ready to build an automated
production line capable of 1000 units per week by the end of the year. And it doesn't mean they're
going to be producing 1000. It just means they finally got this line built with basically the
nameplate production numbers on there. It's like this is capable of doing up to 1000 a week right
now that has been built does not necessarily mean it will be producing 1000 a week. Because
there's a lot of other reasons that that may not happen. And one is a supply chain is not there yet.
Okay, so that so they're having issues that they may not be getting enough components or,
or the components that they are getting are not of the quality that they need. So like small
batches are fine. But once they start ramping the batches, then they start having quality control
issues with that. That's the reason for, you know, that that audit that was well reported by
the Chinese Tesla paparazzi for the past two weeks, following the Tesla executives as they
went around the Chinese supply chain, talking about them looking at auditing these different
things, because it seems like they are having probably issues with being able to get the
ramping to production. So it's one thing again, to have the nameplate on there to say, Yep,
you can go ahead and produce that number. It's an absolute it's another thing to be able to feed
it. Yeah. First of all, this is what Ilan is saying. It's excruciatingly slow at the very
beginning part. There's going to be the hands that we've never built any of this before,
it's all gonna break, we're gonna need to fix it and figure it out. But you have to charge forward,
have these problems, find a solution for them and catch up. If they can get to 1000 per week by the
year, I'm perfectly happy. Like you said, the production automated production line 1000 per
week, and it doesn't mean that they've already produced it now. And this is the other thing
that's interesting is like, right now, if they're producing a couple hundred, is it not automated
yet? Or it's not fully automated? It seems like there may be still a lot of manual work that's
going on partially automated. And it might be look, we want to get the whole automated lineup,
which means it's only like marginally faster than the manual assumption, but does production but
it also seems like the automation equipment they're having is a bit cantankerous right now,
all the pieces aren't quite going the way that they wanted.
Well, yeah. And then by the way, if Tesla is having this issue, I'm sure the others will have
to face it to Jeff says that these pre production units can be in the hundreds per week kind of
thing. It's doesn't mean that it's, you know, it's the you know, if you think manual, you're
thinking small numbers, but no, they can have semi automated models. Before we move forward
to the the Chinese supplier topic, you know, this is where it says roughly tenfold in recent months.
So that's great. All right. But they're struggling to manufacture the machines reliability at scale.
It's production lines are grappling with problems involving the robots, intricate hands,
automated equipment, supplier constraints, just like everything you just said. The other thing
that they revealed in this article, right, was who and how they're going to sell it to customers
in the future. Gen four. Yes, it sounds like they're kind of selecting the customers that
would be ideal customers for them, those that have sort of similar logistics and manufacturing
issues as them. So they're going through and it's going to be a RAS model. They're going to lease
it because they want to be in control of the bots. They want to own the data that the bots are
collecting. So that's part of like you're going to use it, the data that gets collected, used for
training purposes. So that itself is also a big contract discussion, but probably going to be a
requirement. And that means they'll be able to take the bots back and then get all the performance
information on there, find out the wear and tear to make sure the next generation is better. And
that makes sense for a variety of reasons that once you sell it, that means people can start
doing these tear downs. You never know what's kind of going on. The best thing is like, no,
we're going to be in complete control. Basically, we're going to be selling you robotic labor.
And again, it's probably going to be a lot of their supply chain that the tier suppliers are
working with, no one that's outside. People that, let's say they don't want to cross Tesla because
they want to be selling components. And so they'll stick very closely to what the contract details
are and everything. And so, and they're already probably used to dealing with all sorts of NDAs
and all. So if you have familiarity, that makes sense. And that's always what I felt is where the
traction is going to be going that way. You learn a lot. And then maybe by like the fifth or sixth
generation, you start getting those that can start going closer to home-like environments.
I don't, doesn't mean in the home, but it means maybe you're getting out of these like
manufacturing kind of things that you can now go into softer settings that could be
commercial settings that are storefront, hospitality, restaurants, things like that.
Yeah. So the Chinese suppliers, here's the rumor that you were talking about.
Tesla has reportedly completed its first round of factory audits for Optimus,
upgrading three Chinese suppliers from tentative to certified mass production partners.
The three suppliers are Tupow, Joyson, and Zhejiang, Shanhua. Now this is, was rumors.
And then it became like written up by the Chinese media. And then it became like each one of them
kind of like two of them denied it outright. The other one said no comment. So the likelihood is
that this is actually true. At first, about a week ago, I said, they're just rumors. They're
just rumors. But at this point, first of all, it lines up that there's going to be these three
suppliers that the Tupow group supplies core linear and rotary joint actuators,
dexterous hand drive modules, lightweight structural chassis components. And then
Zhejiang, Shanhua, intelligent controls, rotary and linear actuator modules. So you can have,
you know, multi supplier at micro motors for dexterous hands,
Joyce and force and torque sensors for tactile feedback arrays. So yeah. Oh, and then it says
here, apparently, according to the certifications, Tesla plays initial order batch totaling 5,000
that was 5,000 units. Okay, that that sounds right. That that so that would be like for
5000 of optimus or just 5000 units from those suppliers, right? Right. Yeah. So okay. So
so if it's just 5000 units, that means the number of optimize coming from a lot less,
but if it's about 5000 optima, that would be a little bit closer because we were hearing
numbers like 50,000. And I'm like, Oh, something was lost in translation there. They are not
ordering ordering that much, even though you're thinking that they want to have a line that
will be up to about 1000 by the end of that could be, but right now they're not. The other
thing is, is Tesla is not sole sourcing any of the components that they have, they want to have
multiple suppliers for the and the other thing is they want to have multiple suppliers across
all their components, because the problem is, if like one supplier is building the majority
of your components, that means that supplier knows everything that's going on and how optimus
is being built. And they they like do everything to make sure that you know absolutely nothing
the final structure is. So there's this game of cat and mouse to be able to work with the
suppliers at the same time to make sure that your IP is protected. Because a lot of times
you will go into places and I was talking to CEO yesterday was talking about Yep.
You know, my stuff and someone else's stuff was being built on the same line, right next to each
other. So it's like things like that kind of happened with some of the suppliers. So yeah,
they're trying to do a lot with that. And that that's a lot closer than the original reports
of which were were numbers that were much higher than because I've been hearing again
from back channels that like there's no way any orders of that size had been produced.
Yeah, very good information. Thank you, Scott. Appreciate that. You know a lot because you have
your back channels. Like you said, the other thing that came out this week was that and again,
I don't know how true this is. But in Giga Berlin, there are photos, there's an area where you can
see it looks like gen two hidden behind this kind of frosted glass. It says work in progress.
And this is a real photo. This is really there in Giga Berlin. And this guy says,
this is new in Giga Berlin, the prototype of an optimist robot is currently located behind frosted
glass. The factory management confirms upon inquiry, starting mid August, which was last
month, selected teams from Giga Berlin Brandenburg will be part of the global training for optimist
global training for optimist. This means that selected employees will wear special backpacks
and helmets with cameras during their work, so they can record their movements as a basis for
optimists. Makes sense that they started doing this. It'd be odd if they hadn't done this, right?
Yeah, we had heard reports about that. And again, the information article, I recommend everyone go
ahead and find it. It's a very good read. They talk a little bit about that. There's some other
locations that are going on. And also the possibility that not all the workers are on
board with that idea. So I'm kind of surprised that it's I actually think it was like the US
might have been the problem. So I'm surprised that they're actually okay with it in Berlin.
It's interesting that they're throwing the backpacks on there. I know we had seen that early
in the early days with like, you know, with the original gen one and how we're doing that they
kind of had a backpack on there. I was expecting to be something a little bit lighter, you know,
just a little more camera. But you know, it could be that they're getting some gloves,
and there's some other reason that they have the backpacks on there. I don't think an exoskeleton,
but maybe that maybe there's something else. So that's, that's, that's unusual,
because everyone else is just doing egocentric data with basically like a GoPro just the camera
down just a camera, maybe this tracking every part of your might be some some other tracking
that they're going on. There's some reason they want to have it for like the full body control
to get nuance and a lot of other information out of that, especially if they're doing it in the
factory, they want to get it right, as opposed to, we're distributing it to like, just about
every household in the world. So everyone around the world, you know, all day long can be tracking
everything they're doing, that kind of limits the quality of data. And it's a little bit of
a sidetracking here. But there were a couple articles out there this week, kind of talking
about data capture and egocentric data, and how the business model is completely broken. And it's
very tough. And part of it is that everyone is like, we don't even know the quality of the data
and whether we're grabbing the right stuff. And most people is just like, yeah, we're getting a
little bit on there. But it just seems like we've got to get a little bit more. And it's possible
that Tesla by doing that is going to find that, yes, there's this bitter lesson that we probably
have to gather a lot of data. But it's also the quality of the data. If the data is junk,
that's not going to work. And the better quality you have, it might mean that you you're able to
scale a lot faster, because there's enough modality that now everything starts to make sense
as opposed to like, there's just kind of some loose information about stuff that kind of helps
your models improves a little bit, but doesn't get you to that 99.9%. You know, it gets you in the
high 80s, you're all excited. And then you're wondering why you can't make that breakthrough
to 90%. And it could be because you're missing some important modalities. So that backpack,
I think is an intriguing piece of information. Oh, that's very good. I hadn't. I've shared that
info with a lot of people. No one pointed that out. Yeah, very cool, Scott. That was early day
stuff. It's not something that we've heard recently. Yeah, this is, well, this is recent.
So okay, so all of that sets us up, right? You have the leak, you have the supplier information
saying, maybe it's 1000, they're having troubles with their hands. You've got the Giga Berlin
saying, hey, we're already tracking things. So things are moving. Some people are probably right,
some are wrong. Let's go back to the leak. If you, what you're expecting is Gen three,
not Gen four, Gen three will still, I mean, remember how Elon was going and saying,
can't wait till you see Gen three. It's going to shock you. There were some employees in a video
that said, it's going to, it looks so human, like they said. So this does not look like,
no, no. And, but anyway, from from this point, though, you thought, you know what, there's still
a few things that we can learn from this. So you wrote an article, you want to walk us through this?
Yeah, okay. So I put the scare quotes around Optimus three, because you know, I was skeptical
that it actually is. And so I didn't really believe it was real. But I think there are a lot
of, you know, at this point, we can say a lot of things that we would expect to see an Optimus three
without even seeing a picture of it based on a lot of what has been said. And so, you know,
the first thing we can kind of see here is that there's some inconsistencies between this image,
or not this image, but if you actually look at the Gen two image, so if you're going to put stock in
what's going on here, at least a Gen two image reflect Gen two. Now, Gen two, or actually saying
2.5. The history of that was, there was a Gen two bot. And then we saw a production line about a
year ago, sorry, a pilot line about a year ago, where we could see there was some slight cosmetic
changes of what they're trying to do for what's gonna be the original ramp up because last year,
they were hoping to hit 10,000 units until they abruptly kind of slowed that that whole program
down. And then a little bit later on, we saw Mark Benioff suddenly post this video of a golden robot
at Tesla HQ, walking around and everyone was like, well, what is that? That must be Gen three.
And it was pretty clear. It's like, well, what they did is they took the pilot line robots,
and they had gone ahead and then just decided to make it the cyber cab kind of gold color.
So they had changed it like that. And also, we had noticed that the forearms are a little bit
suspicious. So the forearm we're looking at everyone thought that must be the new hand was
like, it wasn't a new hand actually was a mannequin hand because these were walking around without
hands because you don't need hands just be walking around, especially if you fall down, you might
break the hand. So there were a lot of them that were coming off the line without any hands because
Now with that, the original pilot line bot that was white had this like this bib for the torso.
The original one, you did not have what looked like a bib. They had that bib that was kind of
going up yet for the Benioff one where you saw that the first golden one, they kind of spray
painted that and you'll see there's some images that you can see the bib line there, but you
notice they want it to be all black to be kind of consistent with that. So it's like, well, wait a
minute, if that's like the official color coding for that, then that should be the same thing in
the assets that you're showing in there. Why aren't you doing that? It's like, okay, something's
a little bit inconsistent. Okay, the bib line is really there. It's a cosmetic thing. But that's
kind of why, hey, I don't trust it. If you can't even get a detail like that, right? There's going
to be something wrong. And unfortunately, that goes back to my days as a kid where I was like,
always finding little differences between things. Yeah, that bothered me to this day.
Now this thing here is that we had this other which, okay, the the genesis of this is coming
from a Brock video that was released in late January, around when the first wave of all these
AI slop videos was hitting the waves, especially with robots. Okay, this came out and everyone
thought, okay, this is Optimus three, and it was very clear that it was showing that this is what
Brock was capable of doing. But what it was capable of doing is like all these other AI videos is that
if you give it a good starting image and tell it what to do, it could suddenly produce something
out of it. So there's, and I think there was like another image, but I haven't been able to find it,
I tried tracking it down. But I think there was like another image out there of this,
of what they call like the engineering box. So it wasn't actually Optimus, but it's probably
what they were designing around. This is this is the form, the silhouette that we want to have.
And it would have been standing there, and it probably would have been pretty close to what
they wanted to where they would have been showing like a lot of external fabric exactly what the
line should be on that. And I suspect that there was actually a picture that Elon took with it and
then fed that to Brock. So it's kind of similar there where the indication is it looks like it's
going to be the soft powder coating. And again, it looks a lot like a man in a suit. Of course,
the thing that's wrong there is the hand is like a real person's hand. But that's like,
that's your normal AI kind of slop getting everything wrong at that point. So again,
there was like another image of it that looked kind of close to that told us that
he probably fed it into Grok because he was still pitching Grok at the time. Yes. And but if
anything, this would be what I would expect the next demo, which would look like it should look
like this, they would have it fully in a leotard or whatever you want to call this thing, some sort
of fabric thing. Because number one, they don't they don't want you to see it. They don't want
you Scott Walter to see the kind of hips and knees. Yeah, yeah. And with one x Neo is I had
to guess at what the hip architecture was, because they told me yes, it is an FAR architecture. But
I couldn't really tell and say that because it was so well covered. Now, if you also look at
XPeng with their latest version, what I call the Iron Maiden that has like a complete suit on that
you also can't see that. But fortunately, fortunately, they actually showed us without
the cover. So we could tell exactly what the kinematics are like, not something I expect
Tesla to be able to do. So I'm expecting the same thing. And of course, figures done a very good job
of covering up the shoulders. I haven't done the waist yet. But you can see like the soft outer
coverings, it's possible. So if their competitors one x figure in an XPeng are able to do something
like that, I'm expecting them to. So this is this is like a cosmetic change. We're using what I call
this chain mail and armor approach, where wherever you have flexibility, you've got these coverings
that are on there. And then you've got this hard outer shell, which is on there. It's like that's
the optimist version two approach. That's not going to be the optimist version three approach.
That's why I like super skeptical and saying it looks like basically if concept optimist and
optimist 2.5 had a baby, that's what you would get. This is optimist 2.6. Yeah, yeah, exactly. I
would call it something like that. Optimist 2.6. Now, the you know, what could we intuit from that
is I think we did see like, we may have seen a later version, I'm not quite sure of optimist 2.5
that might have had another version of the 22 DoF hand that they've been working on. And we did see
the 22 DoF hand at the shareholder meeting that was kind of out in the open, you know, it was not
actually on Tesla, but it was there for display. And the form factor of it was like a little bit
thicker versus what I call like the Benoha version, which was just a mannequin hand. And it would make
sense that I bet the forearms are going to be a little bit thicker. They're not gonna be Popeye
size, but they will have a little bit more to be able to kind of fit everything in there. So I
would not be surprised if that silhouette is right. The other thing about the elbow is I don't think
the elbow is going to change from the way they've done it before, because they're probably going to
do the linear actuation that when you could put an actuator down there, you end up making your
elbow joint kind of big, you know, looks like it's swollen or something like that. In your bicep,
you've got room for a nice linear actuator, they've been dealing with linear actuators, I don't think
they're going to just drop them everywhere. And it kind of makes sense to put it there. So I think
that will be the same. And the shoulder, I don't think there's really much room for innovation in
the shoulder short of actually coming up with a spherical joint. So they're probably going to
still have that, you know, the same far kind of shoulders, just figuring out how to do the
covering. And I don't think they're going to add more degrees of freedom, like expanding ended up
putting in the hugs and shrugs, they just added degrees of freedom, unnecessarily, just to make it
look super human, which comes with all the drawbacks of like cost, probably low payload, you
know, run runtime of the battery. But if you want something that looks convincingly human, you would
do that. But I don't think they're going to do that, they're going to have the same thing. But
they're going to figure out how to get a nice covering for that. When we get to the neck,
everyone agrees that you probably have to have at least two degrees of freedom for your cameras to
be able to go around. Now, everyone thinks, well, just doing your head with cameras like a like a
fly, and you don't need to move your head. And it turns out, that's not a good idea, because you
need to then do all the fusion of all the cameras that takes a lot of compute. And the other thing
is that you want attention, the idea that the robot really needs to be looking down at what its
work is, humans are kind of giving the hint by moving their head and saying, oh, that's what I'm
focusing on. So you want the robot to do itself, you need at least two degrees of freedom, the
question is whether they decide to do what's the maybe so you get like, you know, the yes, and no.
And the question is like, whether you have that movement. That's kind of there for like emotional
response. If you have like a real face, as opposed to a faceplate, maybe you do that. But
without it, I think it's not you don't quite need it. I don't see the other reason. It's like an
extra group. So it's free. Yeah, yeah. And it'll probably the same actuation would be two rotary
actuators, unless they decide to do the thing like with a Terminator where you get these two linear
actuators that kind of go up like that, maybe, but I think right now, it's just cheaper to do
in simpler, just go ahead and use rotary actions. Okay, so that's the upper torso.
Want to move on to the next? Yeah. All right. I would not be surprised. Of course,
of course. Yeah, I mean, right now, there, there's no reason why they couldn't be covered in the
first place, except they just didn't want to put the, you know, some coverings around there. It
could be for thermal issues, because they just wanted to be able to cool down. They're probably
going to stay with what's called an SPU architecture. This is a spherical prismatic
universal joint. This is something we've talked about before it does parallel actuation with two
actuators. Other robots do something similar, like the unitary robot, you will see it there,
agility bot and others. They use an architecture called RSU, which means rotary spherical
universal joint, where they're using a rotary actuator instead of a linear actuator to be able
to do basically the same thing. Again, the packaging is much better because you get so
much room in the shin, whereas everyone else is trying to figure out how to get these small or
what they're trying to make the big rotary actuators to fit somewhere in your, in your
shin area that don't fit quite right. This is in some ways, I feel like a more elegant, much nicer
way of being, you know, it just fits in really good. And you get the torque sharing from those
things and those things. That's not going to change. It's just a question of what kind of
actuation strategy they use for the, for the linear axes. You can either use inverted planetary
roller screws, or you can use a ball screw. And there are other companies that have used linear
actuators like, like Aptronic and Neura, of course, GPeng for their, their walking bots.
And some of them have actually used a ball screw versus inverted planetary roller screw,
because they're cheaper. Their, their backdrop ability is a bit better, but they have also,
you know, some of their issues. So the question is which one Tesla is going to use down there.
I suspect that they're going to stick with what they've had there before,
which is inverted planetary roller screw, which again, has been like a headache as far as
cost production, stuff like that, and being able to get the supply chain to be able to produce
it with a position and everything. Part of the reason is because that's where like, you know,
the rubber meets the road in the sense, that's where the impact and shock is coming from.
In many cases, that's the one that's going to be able to absorb that much better before it kind of
goes through the system. Other places where they, if they choose to use linear axes and other places,
they might go to a ball screw. The hand for sure. We know they're doing linear axes in there from
what we've seen and everything else. You cannot get an inverted planetary roller screw micro,
you know, made down to that small. At that point, you end up having to use something like a ball
screw drive. So that's probably where we definitely know they're going to have some.
I suspect they will also do it in the, for the elbow joint. They'll probably go there just for
cost reasons. And also has a certain amount of black back drivability, which like for safety
reasons will kind of help. But you do have enough accuracy. I mean, there's still pretty good
mechanisms taking that approach. So I think they're going to be mixing it up in different
places and not just have one type of linear type of drive. Makes sense. Wow, you know,
which means which means it's a headache in the supply chain that you're gonna deal with more
multiple suppliers. Yeah, they're trying to make use of one linear and across the board,
but now they probably need to have a few hips changed a lot. If this is the case.
Yes. And it's time for an overhaul. Because this is like the biggest design flaw that Tesla has
had for a couple reasons. Now, first, hips and waist have to be designed together. You know,
they may look like they're separate things. But you're talking about really the core of your body
and everything, you have to kind of see how they work together. And they have kind of been done
that way. Because the whole idea was like, oh, well, I could just use the hip flexion,
instead of having flexion at the waist. In other words, you know, being able to lean forward,
I don't need to have that. So already, that means people are thinking of like, how my
what I am able to do with my hip how it's going to affect my waist. But the initial approach,
and this is the early days, everyone was thinking the exact same thing is that you want to get what
your reflected mass or basically, you want to have all your mass more proximal, I mean, towards the
beginning, as opposed to distal instead of towards the end. So if you have one actuator,
which has to carry this very heavy kinematic chain, it's going to have to do more work.
So we'd like to do is get rid of a lot of the mass and some of that mass can be these actuators
that are causing things that look like to be moving different directions. So the whole idea
is take the most important joint, which is the flexion joint that allows you to walk and make
it like third and the kinematic chain and make the others to one and two, so you don't have to carry
that mass. And so they, there are different ways that you can arrange that, to have it go down
there, but it's a little bit tricky. But when you decide that you're going to use a linear actuation
strategy, especially for the flexion, it comes with a cost and that is the flexion axis is very,
very low. So everyone was using what we call these ARF or RAF kind of architectures, rotation,
abduction, flexion, or turning it around, abduction, rotation, inflection, reflection came last.
In there. And so Tesla was ARF, but then the next generation, when everyone started to think about
it said, ah, you know, well, it's not that important. What we want to do is come up with
this better packaging, going to a rotary actuator. And actually the flexion is at the top of the
kinematic chain, which shocked everyone. I mean, all the dynamists is like, why would you want to
do that? Now we have to carry this mass, but by getting the other actuators close to the center
of rotation, they don't have quite as much of an impact as if they were down on a lever arm that
you had to move them around. So Unitary kind of was the first one to really unveil it at scale
and show that it was functionally a good way to approach. And everyone has followed suit.
So they have the flexion first as a rotary actuator, then abduction second, and then built
right in the leg, you have the rotation coming on down there. And it makes sense to kind of do that
way. So I suspect they're going to move into that direction as well, because if you go back to that,
the first slide that I showed there kind of going up is that we can see if we want to intuit
something from the image, that based on those covers, if those are hard covers, that means the
flexion has to be very high. Otherwise, those covers are going to expose themselves a lot,
they have to be in some place that they're able to kind of move around. And that means it's up
higher where it should be, versus the current optimist, you see the current optimist, its
flexion axis is very low. I mean, that's much lower than any of us. I mean, you'd have to break
your leg to be able to flex at that point. And by getting it up much higher, it does give them
the opportunity to not have to really have flexion in the waist, but I think they will.
Anyway, even though a lot of designs are sort of tempted to say, when we get it up that high,
then the waist, we do not need to have that vertical or what you call flexion, forward
flexion in that. So what's the possible solution that they could have? Is it going to be far? So
I'm going to say that there's three possible ones. G-Peng is probably the closest, it's not a Tesla
clone. It's a unique design on their own. So I would not say this is a copy of Tesla anyway,
but they are going through and also innovating in the design. Their hip is a bit kind of complicated
and similar to what Tesla had. Their first one was a RAF architecture. The second one,
they changed it up to that ARF architecture, where you can see that green actuator there.
It's right there, like where the buttocks would be. And then they have the blue actuator, which
is the rotation, which is rotating the whole thing. And then they have the flexion actually
coming from a linear actuator. But they changed it. With Tesla, the linear actuator is actually
down in your hamstring. And then they have a second actuator also where your quadriceps are
for the knee. And that's the way G-Peng did it in their first generation. The second generation,
they moved it around. And they realized that, hey, if we move that linear actuator actually
to where the quadriceps would be, we can actually get that flexion axis up a lot higher where we
want it, need it to be. So that was a clever thing. The other thing they found is like,
so that means the actuator that's doing the knee is kind of behind. And the knee has this weird
Hurkins mechanism. It's a four bar mechanism, which was always kind of protruding out.
They have it kind of behind now. So that mechanism is almost not exposed at the front of the knee,
but it's hidden in behind. So that was like another little clever solution. So Tesla might
ironically copy them from looking at their covering. But I suspect they may not. But I'm
pretty sure, pretty sure that they're still going to have a linear knee. I'm not sure
about flexion. I'm going to say it's 50-50 on whether they decide to go with the flexion.
And if they are going to go with the flexion, then it's probably going to look a lot more
like the Iron Maiden. Otherwise, it will be a far architecture where the knee joint won't be
a rotary knee joint, but it will also be a linear axis. And that one is like, this is where I'm
going out on a limb. And I'm saying that one's going to be a ball screw. It's not going on a
limb. Right, right. So, so that's see, they could go between one or the other. But at that point,
I think they're isolated enough from the ground shock that they could probably go ahead and use
that. And it would be more of a cost savings thing to go with that. And I think the other
reason they want to do it is that there is, I think with a far architecture, they'll be able
to walk a little bit better that the linear driven bots are not walking very fast. For instance,
Iron Maiden walks really well, but she does the catwalk slow, and the catwalk is so slow.
So, so you're, you're so enchanted by her walking, you don't realize how slow she's going. So,
so we've never really seen those go fast. I mean, if you look at the Chinese humanoid games,
they were, they were all rotary. You know, everyone who was breaking Hussein's ball,
those are all rotary. There was like nothing linear on down there. So, so that that's kind
of a lesson there. If you want to get a little bit of speed. Now we have seen like a version
of Optimus that's been able to do a decent clip. But I think, but it was like only once
where you haven't really seen it consistently. But what we have seen is that they'll probably
do that. But the reason to have that is it could be for like, if they want to do a deep squat,
if you want it to be able to get down and lift something heavy, you want to have something
that's pretty strong. And that's the advantage of the linear actuators is that sometimes the
strength, when you really need that, that torque, it will be able to get it for you.
That might be a little bit harder, especially a holding torque. So I suspect there's squats
like that. And again, anyone that is really designing something for the factory,
they're really not trying to compete in the decathlon or set world records.
They're thinking about what do we need in the factory? I don't need to do cartwheels. I don't
need to do an eight second hundred, but I do need to do maybe a 300 pound squat. Okay. So, so that's
why you would go ahead and do something like that. So that's where I think the change is going to be.
Now, finally, let's get up to the waist and why the waist. Now, if we really do see covers like
this, this is telling us a lot, because if you look at both figure three and at Optimus,
you will notice that there's almost no waist. I mean, most people are jealous of like the,
of the size of the Optimus waist, right? It's like, my goodness, I wish I could
lose that much around my waist. That's how you can tell it's not a person in suit because
no Halloween costume would be able to get someone that thin around there where that is,
is your waist size. Okay. The unfortunate thing is, is if you look around there, they're like
all this vibe, there's all of all this, what I call wasted space, pun intended there,
of where they're like, there's nothing there. There's, there's just air. And why not use that
volume for something, but you can't hang batteries down there or electronics or anything else,
because that's kind of where it's moving. So why don't you think about putting something else there.
And this is this idea of, of what we call the kidney style waist. And that's the next image
that we see from G-Peng. Unitry does it as well, but, but it's the next one up. So this is from
it's because it's one of those, the best exposed images you'll find out there. The unitary one is
so well hidden. You can't really tell the older Atlas did this newer Atlas does not the original
figure one had this kind of approach. It's basically that RSU architecture I talked about
that everyone was using for the ankle moved up to here where you have these two rotary axes
and these linkages down there in a, and basically a universal joint in the middle.
And what that does is it gives you the forward lean and it gives you the lateral flexion back
and forth. Okay. So forward flexion, lateral flexion, it doesn't give you rotation. So what
you do is you stick an actuator underneath that whole unit there, that platform that
rotates the whole thing. And you get three degrees of freedom there. Part of the reason for the far
hip. And also this antedral angle in the far hip is it opens up enough space for you to be able to
put that actuator in there for the hip. So it fits in there really nice. But the idea of the kidney
style waist is that those actuators are now kind of a plot occupying that free space down there,
they're literally where your kidneys would be. And you get torque sharing. And that's nice,
because if you're leaning forward, sometimes, you know, you want to lift something up. It's like,
that's why you get back pain, because it's a lot of strength. Now I've got two actuators
that I can work together to get that up. And meanwhile, also with the differential trick,
I can go ahead and do that. That now means by doing that, and then having the deflection axis
much higher, it'll be a lot easier for optimist to be able to get into the cybertruck by itself.
Whereas, you know, right now, I think someone has to do it, because it's just so stiff, it's just
going all the way down there, there's absolutely no movement from the hip, I mean, even below the
hip, all the way on up. So it makes it very, very difficult the way it sits. So So now it will be
more natural, it will be more human like, so when you put it into any sort of leotard or outfit,
it's going to look right. Yeah, I was thinking, well, first of all, it's very complicated.
Your knowledge is through the roof. A lot of that went over my head. But you know, it's a lot of
trade offs, isn't it? Like, you have to decide where you want to do, where do you want to go?
And each robot company is making these decisions. But, you know, my stupid idea would be, well,
if you don't need kidneys, maybe you can put the, you know, make it as human as possible. What do
humans have their kidneys, but why not put, you know, cooling, some sort of temperature control?
Oh, yeah, yeah. Or something, you could, you could probably fit something in there. But yeah.
Yeah, yeah, yeah. Absolutely. Absolutely. Absolutely. So the. So yeah, I would say that
that's kind of what we get into it. And say, I'm trying to think, say anything else we got along
there. So we basically gone from head to toe, looking at everything's right. Yeah, we're pretty
well done. So I mean, that's, you know, that's very interesting. So what's weird is, at first,
you said it doesn't look like it is, because that wouldn't be gen three. But then you said,
look, I actually think that the hips should have changed anyways. Exactly that. And then they
covered it. They had to raise the legs up anyways. So it's like my point being that if this was just
2.5, and then they just put it through and says, you know, put some coverings around it. Well,
that doesn't apply right here. If you put that would be a major change. That would absolutely
be a major. So again, it gets closer to the original concept that they wanted to kind of
look like that, where I think it is, is that, well, wait a minute. Now you could make that
fabric. There was one creator out there who made kind of that look like a suit that you put on
there that you could have the same kind of what champagne and black that they're talking about
coloring. So it's possible those aren't hard coverings. But if you kind of look at that image,
it's really suggested that those are hard outer coverings. It could be that it's off that is like,
no, we're actually going to knit a suit that looks just like that. And you're going to think
it's a hard to your cup. But I don't think so. I think it is more like that. And the thing we'll
tell is, if the waist looks like it's got bulk in it, then we know, okay, they've gone through
different way structure. Now the last thing that I kind of talked about is the possibility that
rather use the rotary actuators there, they might go to linear. But there's only been one case of a
with Iron Maiden. And so you have a choice. And I think part of the reason they did is they also
made their waist, it's either like four or five doffs. I think it's actually a five doff waist,
which because they talked about it being a spine with a lot of movement in there. So they get some
redundancy in there along with a hip redundancy. That's why I can do the catwalk. That's why I can
do a lot of other things. And it's very convincing as a person. But every degree of freedom that you
have is a point of failure. I mean, it's like if you're lifting something heavy, it's either
something could overheat or the bearing itself that there is now carrying something it can break.
So if you want to be deploying these in factories, you only want to have the degrees of freedom that
you need. You don't want to add more. And that way, the ones that you do have, you can put more
reinforcement there. So let me then ask you that question. The article we read about the information
said that the hands were giving them a lot of challenge, which makes a lot of sense. That's
these guys are creating not only 22 degrees of freedom, like very intricate hands. Do you think
that Elon made a mistake by going too far with what he wanted with the hands? Do you really need
all of that? You know, but now you're slowing down your you know, too many too much over designed
for this stage of the development? Yes, yes, I would have said that should be like a long term
research project to go ahead and have a hand team that's kind of working on it. But intermediate,
I would come up with something that's more robust and kind of reduce some of the degrees of freedom.
Because you don't need to play the piano to be able to do things on the factory floor. There's
a lot of operations that you can do that, you know, you might think you need that kind of
dexterity, but you can probably do a lot without it and think of how to bulk it up for the different
operations on there. So again, I would sort of think like, are we, we should be attacking the
first 80%, not the last 20%. It seems like the focus has been kind of on the last 20%. And there's
a lot of learnings that everyone has to find out about, you know, what kind of actuators you need,
can we get the suppliers, the actuators are the actuators that are failing? Is it the tens they're
failing? Is it the wear and tear going there? We're hearing for sure it's a tactile sensors,
that that's one part of it, but could be a variety. And it's also really hard to put together.
Evidently, that's the part that they cannot automate. They're trying to, but all that is
actually ironically being done manually. In other words, the hands are being assembled by hands.
But I guess I are on the other side, you know, hit it now, get, learn it now, get going on it,
because if you can do the, the intricate hand movement that others can't, that opens up your
TAM of what it can do. It can sew. It can, it can, it can do things that, you know, it can,
maybe it can cut lemons better than if you didn't have those ability to really hold the lemons
because lemons all different sizes, whatever it is, you know what I'm saying? Like just so many,
I could do it, but if you can't do that or anything else, you have no TAM. Okay. So,
so that's the whole thing is you get the TAM and the Sam. Okay. So what's your
servable addressable market go after that first. And don't worry about this large market
by getting kind of the learnings, because there's so many people who don't care about you being
able to peel lemons. They'd like you to see, to be able to do some of these other things.
So there's no reason why they couldn't have different hands and one that they would kind
of scale up for later use, but they, they have their own ideas that might, a lot of might have
to do with just the embodiment data capture. We don't want to have to force people to move
their fingers like this. If they're naturally doing something like that, that could be the
argument. And then again, we saw Unitary just came out with their, their new hand for the G1.
And the interesting thing about it, it's a much smaller hand. And, and I would say the hand is
the appropriate hand because it's designed for the G1. It's designed for the payload of the G1,
the size factor and everything else. But they put like this degree of freedom,
the fifth metacarpal there, it's 22 doffs. So they have a really nice CMC joint.
They have it there. And I'm kind of questioning, do they really need that? I mean, they were kind
of showing off because the hand is so small. You really don't need it to come around. And
especially if you have three doffs down there, to me, it was like, that's an extra one. It's
kind of cool to have it. Maybe it's for data capture reasons or something like that. But
if I was building an industrial hand, I would think twice about, should I put that thing in
there? Because I'm just asking for, for more headaches. And that's why you see Wuji is like
20. Everyone's like, everyone's between 20, 21, 22, something like that. And you can sort of decide
where do I cut it? And again, they're not putting more degrees of freedom in the, in the shoulder
or in the waist or up here. Then if, you know, if you're going to put everything in the hands,
why not put it everywhere else in the body? Because it comes with a payoff or it comes
with consequences and a cost. What a great show today, Scott. Amazing. Your knowledge is just
through the roof. Unbelievable. Thank you for sharing that. Great analysis. We'll see what
happens. When do you expect Gen 3 to be actually revealed? Well, I get, I think we'd safely say
not in the third quarter because we're only a few days away from it. Yeah. I'm hopeful by the end of
year, but my suspicion is we probably won't see it until early next year. Yeah. Thank you so much,
Scott. Appreciate it. Thanks everybody. Bye-bye. Yep. Keep building everyone.
RoboStrategy is a closed-end fund focused on robotics, physical AI, and emerging technologies.
This discussion is provided for information and educational purposes only, and does not
constitute investment advice, a recommendation, or an offer to buy or sell any security. Any
opinions expressed are those of the speaker, as of the recording date, and are subject to change.
Forward-looking statements and opinions are based on current expectations, estimates, projections,
and assumptions, and are subject to change without notice. Actual outcomes and results may differ
materially from those expressed or implied. Any references to prior investment experience,
portfolio companies, or investment outcomes relate to activities conducted outside of RoboStrategy
and are provided solely for background and informational purposes.
Any reference gains, returns, or investment outcomes may be unrealized and are not indicative
of future results. Investing involves risk, including possible loss of principle. References
to companies, technologies, or investments are illustrative only and should not be interpreted
as investment recommendations.