0:00 Real world and then it's gonna go in there and then suddenly it's gonna notice the oh wait a minute, I know when I'm done because you can feel and literally the voltages will slightly change.
0:11 And so these are the subtle little cues that are just so hard to tease in a simulation.
0:15 They're so hard to get from egocentric data and why they want to do more and more of the
0:19 data capture.
0:19 And then of course the other things they can do is they may have like a proxy.
0:22 We're not sure before they have all the alphabetic.
0:26 It seems that they have alphabetic outfits.
0:28 That means that's almost like a string of nine.
0:36 This is not just the hands.
0:38 almost have the whole thing as which is basically how Tesla did full self-driving is that people were not driving the car remotely.
0:46 They were sitting in the seat, right?
0:48 Literally driving the car the way the car is supposed to be driven.
0:51 We'll we'll get to that.
0:52 it's a very good comment that you made.
0:54 Everyone else puts on they get the p humans to put on glasses and they put on gloves.
0:59 And you're saying actually what Tesla you think is they're doing is they're actually putting on exactly what the robot has, like a battery pack in the back, the actual headgear.
1:09 as if they are the robot and that's interesting we'll we'll cover that a little bit more but tell me about what will it take for Optimus 3 the new robot we haven't seen
1:19 To be able to do the things Elan says we would do.
1:21 You put together a deck here, let's let's walk as good as
1:26 so basically there's the three things.
1:27 You have the Swiss wash, the
1:28 hundred miles per hour fastball and chainsaws and motorcycle.
1:31 We're gonna break them down one at a time to get kind of an idea of of what's actually meant.
1:36 So yeah.
1:37 So you know the the first thing
1:39 is you know what would it be like to assemble a Swiss watch?
1:42 You know, it's with one hand, with the head looking down, right, at the at the work, and then suddenly has to catch a hundred mile per hour fast.
1:50 It's not like it's going to drop something to be able to get it.
1:53 You know, thrown from basically an angle, 'cause he he said, you know, thro obliquely I think was the word.
1:58 So
1:59 And everyone's wondering what that means.
2:00 And what I think that means is not straight on.
2:03 Okay?
2:03 It's gonna be coming from the side somehow.
2:06 So and
2:08 And there's actually probably some advantage to doing it that way anyways.
2:12 You know, one of them is like if it misses you know, if it doesn't get the catch, it just goes by, it doesn't like hit it right in the head.
2:17 It's also gonna be a lot easier.
2:19 And then finally, what kind what's required for like the torque requirements and everything else and responsiveness in the actuators to be able to do something like that
2:29 juggling chainsaws on fire on a unicycle and everything else.
2:33 And it seems like individually a lot of these things could be done.
2:37 The question is you can eat
2:38 you put that all into one machine.
2:41 So let's see.
2:43 Let's go ahead and start breaking it down.
2:46 All right, so.
2:48 It has to be a fair fight.
2:50 In other words, the robot's not allowed to use tools that the human doesn't have, but at the same time we're not going to take tools away from the robot that we need as humans.
2:59 a Swiss watch, we are not doing it barehanded.
3:01 we have a lot of things that are helping us, you know, a loop or a magnifying glass, you know, tweezers.
3:07 In many cases there's also something
3:08 to rest your your wrist on to keep you from shaking and everything else.
3:11 So that's fair game just so long as it's not specifically modified to be robot friendly.
3:17 Okay, it has to use the same stuff.
3:18 And if it doesn't need it, fine.
3:20 But I would say it's there.
3:22 the other thing is that when we talk about it's not really blind.
3:27 It does have peripheral
3:29 view but it should not be relying on any external cameras or anything else.
3:33 It literally has to be whatever it sort of has on there and they have it set something else up.
3:38 So
3:38 For instance, this kind of stuff of tracking balls, there's lots of robots who can do it.
3:42 We've already been seeing that there are bots that can play tennis and ping pong and everything else.
3:46 Industrial arms have been doing it for over ten years.
3:48 be able to do that tracking, but usually it meant they mounted other kind of camera systems around there to be able to do the response.
3:54 even some of the tennis playing bots that we've seen, I believe they were not doing
4:09 this idea because there are people who are able to juggle chainsaws and I think maybe even on fire.
4:14 I'm not sure if on a unicycle.
4:16 I know there are jugglers on unicycle.
4:18 I don't know if anyone tried that.
4:18 with a chainsaw.
4:19 But those are props.
4:21 Those are not the chainsaw that's down in your garage.
4:24 They are much lighter.
4:25 They're balanced different.
4:26 The handles are different to be easier to grab.
4:28 They're dulled and everything else, the the blades a little bit shorter.
4:31 You know, a bunch of things like that.
4:33 So it's like, okay, it can use the same ones that humans would be using to do that particular trick.
4:38 But
4:38 it has to use the same unicycle that a human would put.
4:42 So we may see some some bots in unicycles.
4:44 As a matter of fact, I think there already was an example back in the spring
4:48 of a G one on a unicycle.
4:50 What wasn't clear is whether it was actually a pedaled unicycle or whether it was an electric unicycle that was able to do self balancing on its own.
4:57 That's cheating, right?
4:58 I mean
4:59 Using another mechanism to do the balancing.
5:01 And the other thing is like that's a G one that's not human scale.
5:04 So you wanna have something that that's fully human scale.
5:09 things are just going to be a back of the envelope calculation to kind of see is it feasible or not and could we tease out the direction that opt this might be going if they really believe can have.
5:18 So the first thing is you have to make some assumptions.
5:22 So the only thing we know is you mentioned it's a baseball.
5:26 A hundred miles per hour.
5:27 Now when I think of a hundred miles
5:28 an hour baseball think of a pitcher delivering it from a pitching mound which is 60 feet six inches away and of course the World Series is right now and there's a pitcher for the blue for the the brewers that can
5:39 throw like 104 miles per hour.
5:40 So I think that's why that kinda comes up.
5:42 So let's assume you can't throw it from ten feet away.
5:45 It's it's gotta be a reasonable distance.
5:46 So let's assume it's being shot
5:49 out you know some sort of pitching arm or something like that, that's gonna cover that distance in under half a second.
5:55 So that means in less than half a second the robot has to
5:59 acquire it, figure out where it's going, react, and put its hand in the right place.
6:03 So we can kind of break down what's gonna happen.
6:05 Humans are well, yeah.
6:07 Humans can, not all.
6:10 Those have been trained, who have good reflexes, are are able to do that.
6:15 And this is kind of what goes on.
6:16 And again, this puts our world model.
6:19 we are very good at being able to track incoming ballistic things.
6:24 Now we we've we certainly heard in Warzone that they do tracking of of incoming warheads.
6:28 And the reason why you can do that is that they're on a known trajectory, and once you get enough of it, you can figure out where it's going to land and you can try to intercept it, unless they do evasive maneuvers.
6:39 can only do the evasive maneuvers by putting spin on it to make it curve.
6:42 But basically we as human beings have seen enough things fly through the air that with just a few images in our eyes, we can almost predict.
6:48 where it's gonna land.
6:49 That's why we can we can grab, you know, ru know where to run to grab a a catch or something like that.
6:54 So that's what the same thing that's going on.
6:56 You have a a short window
6:58 in which you've got to figure out where the ball's gonna go so you're gonna react.
7:02 And a batter has to do the same thing.
7:04 So a batter has to very quickly kind of guess and decide whether they wanna swing the pitch.
7:08 The same thing with getting your hand out there.
7:10 So there's a little bit of a breakdown here we have of kind of looking what's going on.
7:14 First, you know, looking down isn't the same as being blind, because we we will see something.
7:19 And the only question is whether it's gonna look up, but I think he was saying it's still gonna be looking down almost as if it's bored and it goes and catches the ball Optimus.
7:27 where are the cameras on the Optimus?
7:29 Don't they have extra extra cameras?
7:30 Yeah, they have extra cameras.
7:31 Yes, yes.
7:32 And and they may even be cameras from behind.
7:34 we don't know if there's gonna be risk cameras.
7:36 We we have have no idea.
7:39 someone in the also in in the torso.
7:41 So it's impossible there there are other cameras there.
7:43 Right now I think there's stereo cameras.
7:46 It may be like a chin camera or something like that.
7:48 But we can assume there's
7:50 Yeah.
7:50 It's like it's like the car.
7:51 The car has eight.
8:00 But there's another reason is that it's a lot easier to track a throw coming from the side to figure out where it's going to go to
8:09 to one that's coming right at you.
8:10 If it's coming right at you, you don't have enough information.
8:12 So you do want to actually have a little bit of obliqueness.
8:15 Okay, so which is which is fair.
8:17 Okay.
8:17 If he said oblique
8:18 That means it's like it doesn't have to be straight ahead.
8:21 He's being f very straightforward and saying it's gonna be to the side, so we're giving a little bit.
8:25 So you have really as as you see when
8:28 when you throw that, only about a hundred milliseconds to make a decision on where it's going.
8:33 So you you get enough and how many data points you have in there is gonna depend upon the frame rate of your camera.
8:38 That's kind of very important.
8:40 So you get that, then you have to make a decision to commit where the ball is going to go.
8:44 And at that point you have, you know, about a quarter second or
8:49 something like that to move to wherever the hand is up to where it needs to go with a with a peak velocity of around three meters per second.
8:55 And three meters a second is obviously something that people can do and these robot arms should all
8:58 also be able to something like that that at peak and th there should be enough torque and everything got be able to get over to where it needs to be.
9:04 So so you know this has implied that there's gonna be some some latency in
9:16 Now how far the ball
9:18 moves between frames.
9:20 So most cameras have maybe 30 hertz.
9:23 Okay, so that that's what a lot of them are running at.
9:25 And if you're going at that, that means it's moving well
9:29 over a meter between frames.
9:31 Okay.
9:32 So that that that means at a hundred miles per hour.
9:37 So that means that
9:38 That first look you're gonna get in the first hundred milliseconds you're only gonna see like three frames, which is barely enough to do the fit.
9:44 Now you can do the curve fit because you can assume it's a parabola, okay?
9:48 It's it's not
9:48 Like I'm seeing three data points, I have no idea what curve to fit to that.
9:51 It's like we know what curve to fit to it, so that's kinda sufficient.
9:54 But the more data you have, the better it's gonna be because you've got to worry about blur and a lot of these
9:59 other things so it it could be that that Tesla is already running maybe a little bit higher 36 frames at least in the Tesla cars we don't know
10:08 what the cameras are going to be doing.
10:10 So pretty much to pull this trick off, you would want to have cameras which are greater than 200 hertz.
10:17 And that means you know not
10:18 just that you're capturing images that you now need to have a central processor that can do that.
10:23 I don't know if AI four can do that.
10:25 Maybe AI five can.
10:27 So you you know
10:28 You're getting a lot of data now what are you gonna be able to do with it.
10:31 So you gotta worry about motion blur as well.
10:34 And that, you know, those kind of exposures to see how much smear you have in the ball.
10:39 You know, these are all things you gotta kinda tease out.
10:41 And then if you have stereo depth, that helps as well.
10:44 So if you just have a single camera, it's gonna be very hard to figure out.
10:47 I mean you can do
10:49 stereo through basically time.
10:51 The f the fact that a little bit of time is moved or you move your head back and forth.
10:55 But if you're kind of fixed and things are quick, it's really good to have two cameras there.
10:59 stereo to be able to kind of figure out where that depth is and and get a lot quicker.
11:03 So it's possible to do the tracking with with something a little bit higher and it's probably possible to get the
11:08 hand in that position.
11:10 But the real question is going to be, what about the catch?
11:16 This is where the concern is.
11:19 Miles per hour is pretty darn fast.
11:20 Have you ever tried to catch baseball barehanded?
11:26 I mean you do see people that actually catch the baseball barehanded.
11:28 handed when they they snag a home run, but it's not coming in at 100 miles per hour.
11:32 So it's it may have left the bat at well over it, but because of wind resistance, everything else, when it comes on down there, it's
11:38 It's a little bit slower.
11:39 It could still sting a little bit, but but basically you can't.
11:43 But there's a huge difference between a fifty mile per hour changeup and a hundred
11:49 You know what that difference is?
11:53 It's really the energy.
11:55 Even though 100 miles an hour is twice as fast as the energy.
11:58 as fifty, it has four times more kinetic energy.
12:02 Mm-hmm.
12:03 Which is why it hurts a lot more when it hits you.
12:06 It doesn't hurt twice as much as a fifty.
12:08 It hurts
12:09 four times more.
12:10 And so there's that energy has to be dissipated and has to be dissipated very rapidly and very quickly.
12:15 And so it's a question of reflex.
12:18 So
12:19 You're talking about assembling a Swiss arm a Swiss watch.
12:23 Is it going to be a one handed assembly or kind of two are down there?
12:26 One's holding tweezers, the other one is keeping
12:29 something steady and suddenly its hand just goes whoop up and catches the ball because there's not time to put on a glove, that's for sure.
12:36 Or it's gonna have to already have some kind of padding.
12:39 because the the hand will shatter attempting to capture that thing for a variety of reasons.
12:45 It doesn't have the compliance to be able to to do.
12:48 that so it could stretch and break the tendons.
12:50 the tendons could stretch so quickly that it could cause damage to the gearbox and the forearms very easily.
12:57 Fingers could potentially still be able to do that.
12:59 snap off the and it it's a question is probably not enough reaction.
13:02 So the way you get around it is that let me go ahead and and get my little prop here is that either the
13:08 The ball is just gonna come in and smack me like that, right?
13:11 And then the hand everything in the hand's gonna come back.
13:14 Well first thing, I have to get my hand up like this, right?
13:16 And I want it to come in.
13:17 Do I want it to hit my fingers?
13:18 I want to hit it right here on the palms because when it hits the palm usually hurts more.
13:22 Usually you want to try to get it there so your fingers will comply a little bit.
13:26 Your fingers are not going to be down like this.
13:27 They've got to be open.
13:29 open because they have to then come for the grasp a little bit later.
13:32 So that means they're gonna be at c almost full extension.
13:35 When it comes in here, these things are already at their hard limit, which means you could go back
13:38 and snap it.
13:39 There's not gonna be anything to absorb it.
13:41 The only way to do it is if it's tracking it well enough that it's smart enough to move the hand maybe back at a very high rate.
13:58 relative velocity is fifty and you capture some of the energy that way and get the absorption.
14:03 The thing is whether it could respond enough.
14:05 So you've got to do that move at the right time.
14:07 If you do it too early
14:09 you know, you may be getting your hand out of the way and you may not be able to react fast enough that as soon as you hit, the damage is done.
14:16 You've got to start moving beforehand.
14:18 So you you
14:18 maybe feeling that before it goes.
14:21 So the real thing is that hand is not going to be able to handle a hundred mile power of passport.
14:25 So so that's like the first place we say, that's incredulous unless
14:29 you come up with some other ways to get around this very, very difficult to do that.
14:32 A glob.
14:32 And the assumption is that after it does it, it goes back to assembling the Swiss Army watching.
14:36 Hopefully it's not with a broken hand.
14:42 Okay, now the juggle.
14:44 This is a different kind of catch that that we're doing on here.
14:46 So if you're juggling and putting
14:48 something up now the first thing is you said juggling.
14:52 So when you say juggle, how many objects are you referring to?
14:56 Three.
14:57 I I think everyone I mean
14:58 In order to juggle something means you've got more items and you have hands.
15:02 So it'd be a minimum of three.
15:03 It's not like, oh, we're just gonna do this.
15:05 This is not juggling, okay?
15:06 So I've gotta get them all up here.
15:08 handed two.
15:18 One catch every second.
15:20 And you've got to throw it up there long enough to give you that one second while you're manipulating the other one to move around.
15:25 That's the whole part of juggling.
15:26 So when you're taking that, it's it's coming up, it's coming.
15:28 coming down, you've got to catch it, you've got to absorb the energy of that and you've got to put that energy back in again.
15:33 So again, we might be looking at more theatrical chainsaws, which might be as low as like
15:38 three kilograms of weight.
15:40 Most chainsaws are maybe around five to seven, something like that.
15:43 so it'd be a lighter one with a lot of things kind of taken out of it to make sure it doesn't hurt anyone.
15:49 So there might be a chain on it, but there won't be any teeth.
15:52 and you need to be able to throw that up.
15:54 It's gonna have a certain launch velocity that you need to be able to do.
15:58 amount of energy.
15:59 So you see we're looking around thirty-five joules there, which is a lot less than the 145 that we're kind of deal with.
16:04 And when we we capture it, we already are kind of know we're gonna catch it and go down.
16:09 to absorb the energy as opposed to having this thing like just smash right on into us.
16:13 So that means we get an idea of what's that going to be like.
16:19 And it's gonna be somewhere around like having a ten or twenty kilograms in your hand.
16:24 And we know that's around the payload capacity.
16:26 So if if that's possible and
16:28 And if we kind of look at what the peak torques are going to be out there for all the motors and everything, it's within the possibility.
16:34 So based on what Tesla has announced before going back to AIDS.
16:39 Day two, where we get an idea of what the the torques are going to be in the various motors, it's within that.
16:45 And I'm assuming those actuators might actually be a little bit stronger.
16:48 think they've made them weaker.
16:49 I assuming they're they're probably even a little bit stronger than theirs.
16:52 So that means it would have the strength to be able to pull this off.
16:56 And it should be able to have the cord
16:58 coordination.
16:59 I mean it it seems to me that this is something you could coordinate.
17:01 Again, I I'm pretty sure we've probably seen some humanoids.
17:04 There there have been robots designed for juggling, you know, so almost everything we're talking about, someone
17:09 has built something like this already.
17:10 Someone has already built a simple little bot that has been able to ride a unicycle.
17:14 It just wasn't a humanoid.
17:16 Someone for juggling, someone for doing a catch, all those
17:18 different things.
17:19 Now we're trying to fold them in.
17:20 So we know physically it's possible.
17:22 it's just a question whether a humanoid would have it and whether the specs meet what we expect to be
17:28 in Optimus or what we've seen being Optimus.
17:31 So it's kind of within the range.
17:33 The the biggest trick of course is that when you are juggling these things, assuming you have the other strength,
17:38 You are on a literally a wobbly base, which is a unicycle.
17:42 And balancing that will be a bit tricky, but for these control systems they d they'll just be able to react fast enough.
17:48 So
17:48 Yeah, every I'm pretty sure they get the balance.
17:51 Every limb is that's the beauty.
17:53 That's the they got multiple cameras, but they have every multiple that they can manage.
17:57 Yes, two years ago this was
17:58 was not really possible for a lot of companies because they didn't have whole body control.
18:02 They they really had kind of this locomotion policy and they had the policy up there and while the hands could move around and you know
18:08 the system was correcting for the bounds.
18:10 Now it's gonna be much better.
18:12 It it really is knowing what the hands are doing and the sort of the expectation of what the reaction is about to be and it can prepare us.
18:18 Certainly
18:20 And this is something that's actually fairly easy to simulate in a physics simulator because it's really coming down to rigid body physics and dynamics.
18:29 dynamics and doesn't require super high fidelity that you would for other things.
18:33 So this is something that they could probably nail in their physics simulator and get it trained up.
18:39 So now the the whole thing is about whether they do this live and you know running
18:49 engines with fires and stuff like that.
18:51 maybe something that if you do it live, you're gonna really have to make sure it's kicked off.
18:56 You know, it's like just about
18:58 And anything could go wrong with that.
19:00 But you know, for the most part, everything here seems like it's plausible and could be possible from the standpoint of the unicycle.
19:08 But what we should really look is at what of all these things that we've we've talked about, we haven't got to the the the Swiss clock yet or the Swiss watch, which do you think is probably the most challenging?
19:21 I still think the switch watches.
19:23 The tiny little things.
19:24 I am not I'm not concerned about the unicycle or the juggling or the catching.
19:28 I think those are all possible Yeah, yeah.
19:31 And again it goes back to John Carmack.
19:33 He nailed it.
19:33 He just said like it's it's these fine little fiddly things that are very, very difficult.
19:37 So the bot could go out and
19:38 bring a jack-o' lantern into you for you to do the repairs, but you know, the you know, getting that is is the real challenge.
19:45 So so the the point is that the bot is able to kind
20:06 And the dexterity is a very good thing.
20:08 Bots, yeah, they can show you they're walking and running, but that doesn't take a lot of you know, AI chip power that they don't have.
20:15 Memory.
20:15 Yep.
20:16 All that.
20:16 Yep.
20:17 And now when
20:19 So again, this is a Swiss watch.
20:20 This is not a Swiss clock.
20:22 It's a watch.
20:23 So we can assume that is things that are you know, would go on your wrist, which means it's very small.
20:28 to be made up of these small movements with lots of little gears and you know very, very, very, very small parts.
20:34 And again, this is not something that any ordinary person can do.
20:39 assembling a s a Swiss watch.
20:41 It takes a lot a lot of training and it just needs the right tools.
20:44 So that's the difference between a jeweler and you and me is that they own the right tools.
20:48 And they know how to use those tools.
20:50 and the thing is we as humans do not have anywhere near that kind of accuracy that's demanded to put these things together.
20:59 We use a lot of props and tricks and tools to give us the precision for that.
21:03 And a lot of it is patience.
21:05 The other is like magnifying glasses and special little tools.
21:09 So we're able to correct for the fact that we are very sloppy meat bags that go around with absolutely no precision because we are looking at what's going on
21:18 and we'll make little adjustments and movements until we get it right and it falls in.
21:22 So in some ways it's difficult and challenging.
21:24 The other time it's just a processing power thing.
21:26 I I I think that's more of
21:29 a mind over matter kind of issue as opposed to that you need to have something that's super accurate.
21:33 So again, we've seen these robots that have been able to thread a needle and everyone's impressed.
21:38 Oh look at the
21:39 That's not a big deal because they were already given the needle, they're already given the thread.
21:43 The real challenge is picking those two items up before you go ahead and do that.
21:46 But even then, if you if if there's
21:48 They're going, you know, threading the needle like that, that's impressive.
21:51 But they're going like what we would until you make the adjustment and you see it's right and you move on in.
21:57 Same thing with the Swiss watch.
21:59 is that when you're dropping that little gear in there, you're not just getting it right the first time.
22:03 You're kind of getting it over very, very close and you keep making these micro adjustments until eventually it goes in.
22:08 Because you're using supports.
22:10 You're making sure there's something there.
22:11 If I was just like holding this up in my hand, I would probably have difficulty doing that.
22:15 But so long as I have like a nice fixture to put it on and everything.
22:19 else and let's say the the bots allowed to cheat just the same way that we're allowed to cheat.
22:24 Okay.
22:25 Just can't do anything extra than that.
22:26 So that will be a challenge.
22:28 but I can believe that it it could multitask and like do this grab the ball at the same time.
22:36 That's it so the question is.
22:39 Who's gonna do that first?
22:40 I I have a feeling a couple of these, some Chinese robot companies will probably do it before Elon gets a chance to show it up.
22:48 Which is what I I told him it's like don't don't tell them what you're gonna do because within two weeks one of 'em is gonna do part of your demo.
22:56 I'm not worried about that.
22:57 I mean the whole point is it's a
22:59 demo versus a real thing.
23:00 Can it really do it?
23:02 Okay, so what is your yeah, your your kind of thing here?
23:05 You're basically saying yeah, you can do these
23:09 things, you've got it.
23:10 Seems like it's possible, right?
23:12 Based on the implied specs.
23:14 It's it's possible assuming those are the specs that Optimus is going to have.
23:18 So
23:19 if if Elon really believes and I you know, if it's not something that he just kind of pulled out of his hat to say, All right, this is what I want to do but he had been discussing it and
23:28 thinking about that that that's kind of the capability and they all agree that it it has the has that.
23:33 It tells us something about the potential specs, you know, of of what Gen three may potentially have.
23:39 Gen four by the way.
23:40 He never did get a date.
23:41 I mean he says, you know, it could be gen four.
23:43 You know, that could be the future.
23:44 So here we go.
23:45 Catch 100 mile per hour with a mit.
23:47 You said it's possible.
23:48 Yes, it's ha could be happening.
23:52 Juggle balls, very possible.
23:54 I've seen we've seen that.
23:55 Juggle props, very possible.
23:56 Not just plausible, possible, very likely.
23:59 likely.
23:59 Right.
23:59 Not on stage.
24:01 watch likely.
24:03 What's unlikely is the full movement.
24:04 So maybe the case in hand, but certainly not the full movement.
24:07 And all at once.
24:08 Yeah.
24:08 not not gonna happen.
24:10 Yeah, yeah, yeah.
24:11 So i again it kinda comes down to this like I'm sure it could do a watch battery pretty easily, just like I could do a watch battery, but th but a lot of the other stuff I would have a hard time.
24:19 Unless, you know, I was really given the right tools, then I I might surprise myself on how far I could get along in it.
24:25 It's just that I could probably take it apart, but then it wouldn't go back together.
24:29 Right.
24:29 Gotcha.
24:30 Okay.
24:31 So that was, you know, very cool.
24:33 So these are metrics though.
24:35 You do think that this is the like it's i he I think it's
24:39 You know what he's doing is he's pushing the goalposts in terms of like, yeah, it's one thing for you to pick and place.
24:44 That's not what we're going for.
24:45 We're going for a human a a humanoid that is super hum
24:49 human.
24:49 It can do these things.
24:50 That would be impressive if it can do things that human that's what Unit tried to show you.
24:55 This thing can lift a car.
24:57 If it's true, it'd be cool.
24:59 Of those those three things, which is the most practical in your mind?
25:02 The switch watch.
25:03 I mean obviously the others are just jokes and fun things, but if you can actually you know it's it's
25:08 When John Carmack said, right?
25:10 It's like, yeah, I don't want you just bring in the thing, I want you to put in the batteries.
25:14 Can you really put in the batteries?
25:15 Because you know how hard it is for us to like you know.
25:18 No, the Costco battery thing.
25:22 How do you how do you you try to cut it, the plastic thing?
25:25 It's hard.
25:26 It's hard or like trying to open the back
25:41 And then you can put the batteries in the right order, right?
25:43 If you get them if you get the positive and negative
25:48 At the end of the day, Elon believes, no no no, this is possible.
25:52 It whatever human can do, the bot can do.
25:54 And that is why he's designing the hands to be very he talked about it's not.
25:58 Not just 22 degrees freedom like everybody else's hands, all the other kind, but no no, he wants skin.
26:05 He wants you know the the the ability to do this he pointed out right
26:08 Right.
26:09 So I just wanted to also show that this these guys are really like th they have the academy.
26:17 So you you do a great
26:23 Yeah, and I think the last thing to remind everyone is that we already saw two years ago Optimist catch
26:29 A tennis ball.
26:29 So we know the jug that it can already but yeah, I think that was where the telephone is those who are involved in the development and maintenance of open source software.
26:39 According to the FBI and U.
26:41 S.
26:41 Justice Department, face dealer hackers use fake or hijack social media accounts.
26:49 to invite potential victims into video calls hosted over imposter versions of Zoom, Microsoft Teams, and Google Meet.
27:01 Shop.
27:03 RPM now.
27:05 You just don't get it, man.
27:07 You say, you know.
27:09 bringing dating partners so stupid.
27:12 I mean maybe if you're like really trying to win, right?
27:17 And then beyond
27:19 that I mean chicken had like Beanks as like his one of his partners or whatever.
27:23 She gets eliminated, he goes on to win the fifty K.
27:26 So it's like
27:31 It ultimately didn't hurt him.
27:59 And that's a slow pitch.
27:59 That was a lava stuff like that.
28:01 So we know all these things are are coming and are possible.
28:05 And again, the Swiss Army are possible.
28:17 If you have twenty
28:18 20,000 optimists and they're playing around.
28:22 That's how you train them to be that much more precise.
28:26 And remember when one bot figures it out, the other bot's
28:28 get it right away.
28:29 So they're all helping to get better, smarter, smarter, smarter.
28:33 this is where they're going to build those 10 to 20,000.
28:36 We know we have the free mod factory in California.
28:39 That's supposed to do a million robots per year, but like you said, not till 2028 or 2029.
28:45 But they'll have 10 or 20,000 coming out of the Fremont factory.
28:49 But this is actually the Giga Texas.
28:52 In near Giga Texas, this is where the op the Optimus factory is.
28:56 It is massive.
28:57 And you can see where it's
28:59 progression is it's already to the point where Joe Techmar believes it will open in about six months where you can actually put things in it.
29:06 So as soon as the
29:09 you know it probably won't start building robots for another year or two but it will be housing the academy that's where the 5000
29:19 robots could go.
29:20 10,000 robots can go and they can start training.
29:23 Yes.
29:23 And and that's probably a reasonable assumption because again, we we've talked about what do you do with 5,000 bots.
29:29 And it requires a l a lot of room and a lot of space.
29:31 So either Tesla is renting a lot of warehouse space around Southern California that we're not aware of, for the bots that they're gonna be producing, or
29:38 you know they're gonna have to build something internally and I don't think they have that much room in in Palo Alto to be able to do it.
29:45 They are being manufactured in Fremont, so I assume they're kind of being housed there.
29:49 or very closely for now.
29:51 At this point, you know, we don't know how many they have.
29:54 I I would say they probably have in the order of a thousand, maybe approaching to
29:58 two thousand this point, based on some of the reports of what's coming out.
30:01 We don't know if if those are full bodies or not, whether they're just the bodies and they they don't have all the hands yet because they were having subtles of hands.
30:09 But based on on the reporting at 250 well, it was several hundred, okay.
30:14 And so I'm assuming that several hundred might mean around two fifty a week.
30:18 early September.
30:20 So they probably are north of a thousand by now, trying to hit five thousand by either the end of the year or the beginning of next year.
30:27 That would be
30:29 just enough to really start the Alpins Academy that they're going up.
30:32 However, it sounds like they're not waiting for them to come around to start doing the data collection.
30:37 And that's something you can do
30:39 well in advance of them being there to start that flywheel so that when you have the optimize to go in and close at Cinderel Gap, they have actually something to work on.
30:55 Yeah you you talked about this so number one we already know that in game
30:58 Giga Germany, Kiga Berlin, those staff were interviewed and they said yes, they're actually being trained.
31:05 They have to put on, you know, stuff in t equipment.
31:09 to show optimists what to do.
31:10 So we know that they're doing it in factories.
31:12 Then a couple shows ago we showed that there's huge hiring.
31:16 Tesla is hiring a lot of people all over the place.
31:19 and in US and you said, you know, I think that that's gonna be the service centers.
31:23 And Lars Morvey was just interviewed by Munro Live and he confirmed that A, that Optimus
31:28 human or robot will be helping to put the chargers into the cyber cabs and the robotaxis to charge them and will help in a service center.
31:37 So you're you're hit
31:38 you would hit that one out of the park that they're you're gonna use the service centers as ways to train the bots.
31:44 Then we know that they they they they're hiring people to put on
31:49 equipment.
31:49 And what you figured out, I just pointed out earlier, is that they're not just putting on glasses and calling it a day.
31:55 No, they're they're like actually being outfitted as if they
31:58 are Optimus.
31:59 They got the backpack battery, they got the thing so that they're almost like a cosplay as an optimist.
32:05 Like every every part of the Optimus is probably around them.
32:08 And then
32:09 there's a leak here from Robert Scoble.
32:10 So Robert Scoble is a well-known tech insider.
32:14 He's been covering as a journalist technology since the beginning
32:19 in Silicon Valley.
32:24 He said this.
32:25 Tesla, this is just yesterday.
32:26 He said Tesla optimist training regime.
32:29 It's a leak.
32:30 Workers and its Palo Alto R and D facility have been holding robot university for many months already.
32:36 Number one, he's saying this is happening already a long time ago
32:42 Workers are wearing twenty four pound suits, this is what you've been saying, with up to forty cameras and grids of three D sensors to keep it.
32:48 capture every detailed video in 3D fields of whatever the worker does.
32:52 There are multiple configurations.
32:54 They are given different tasks to do.
32:56 They don't know what the tasks come from.
32:57 Multiple squads
33:08 cameras.
33:09 This leads to a data flywheel that is orders of magnitude more data than the car captured.
33:13 Separate teams are capturing home data like making a sandwich at home, like in a house
33:19 Like making a sandwich while other teams capture jobs and factory floor like installing HVAC systems in cars.
33:25 Elon isn't playing around.
33:27 I can't tell you how I learned
33:31 So do we trust him because he hears a lot of things?
33:38 The information already kinda reported something like this back in September and they they were talking about some of the Alpha suits and that they were wearing backpacks.
33:45 They didn't have specifics of how much it weighed.
33:48 I heard that.
33:49 I said, Oh, okay, this is deja vu and if you go back four years ago we actually saw everyone an outfit exactly like that.
33:56 So this may not be exactly what they're wearing because this
34:01 But we kind of looked at it and said, wait a minute, they just took the cameras out of the Optimus head and just mounted it onto the Optimus.
34:07 Operator and then they probably took all the compute and everything in the back and stuck that on there and said what we want to do is we wanna capture it as if it's the robot.
34:16 So there's
No line has that.
Comments
Discuss on OpenVibe.Community