Unitree G1 Humanoid Robot Teardown

[00:00:16][music] Hello everyone, it's Armen Fchenovski with Moon Associate.

[00:00:30]Today we're talking humanoid robotics with uh our friends from Sheffller. Al Mackey I think humanoid head of humanoids at Sheffller. Correct.

[00:00:40]>> In North America.

[00:00:41]>> In North America. Scott Walter old friend of the channel. Uh second time on uh you are your job description is R2-D2 at >> correct >> at Robo Strategy Strategy. And the R2-D2 comes from robotics research diligence director.

[00:00:58]>> How's that? That's That's not a bad job title.

[00:01:01]>> And I brought my socks today just to prove [laughter] it.

[00:01:04]>> There we little closeup of the socks.

[00:01:07]>> Yes.

[00:01:07]>> There we go.

[00:01:07]>> The R2-D2 socks. There we go. A first on Monroe Live, huh? [laughter] >> And Professor Paul, um, our actuator, >> who's the expert, >> not a professor.

[00:01:19]>> We'll call him that.

[00:01:20]>> So, um, yeah, let's, uh, let's begin.

[00:01:23]Maybe Al, uh, you brought another friend here.

[00:01:25]>> Sure.

[00:01:26]>> Sam.

[00:01:27]>> Sam. Everyone's dubbing him Sam.

[00:01:29]>> What is going on here?

[00:01:30]>> Uh so uh Sheffller is a I think is a known industry and automotive supplier.

[00:01:36]Uh a couple years ago we've decided that um with humanoids we see the market need because of labor shortage and many other reasons and we can use these in our uh manufacturing facilities. So uh our vertical integration including things we do in controllers, motors, gearboxes, bearings, sensors, batteries, you name it. Uh there's a lot of value in humanoids that we can capture. And here we have a couple of examples on display.

[00:02:04]It shows some rotary actuators that have strain wave actuators at the as the key transmission. We've got quasi direct drives with planetary gears. Uh some linear uh gearboxes and actuators including force sensors from ball screws to planetary roller screws and then even uh even small uh small ball screws. So very uh wide portfolio covers over 50% of of the content in a in a humanoid but we're actively working on optimizing these for the humanoid market and making sure they scale in a similar way that automotive technologies have in the past in the past.

[00:02:41]>> Yeah. Very interesting. Right. So you're optimizing for cost for function capability torque density. Right.

[00:02:48]Exactly. take all your expertise, right?

[00:02:50]All your manufacturing scale and try to >> optimize >> to make these robots more efficient.

[00:03:00]Right. So, >> exactly. And one of the key things is there's a lot of iterative development that's happening in humanoids right now.

[00:03:06]So, flexibility is needed, but one thing is we're making sure we're optimizing for performance, but we have our eye on scalability and cost and minimal capex uh for the future. Yeah. Yeah. Great setup.

[00:03:22]>> Some of these actuators that we have are based on strain wave gearing. Others are based on uh planetary gears. But this one specifically, I mean, you can tell lightweight is an absolute big problem.

[00:03:33]Thermal uh uh thermal efficiency as well. And you can see in terms of what Sheffller does internally here, anywhere from controllers, so PCBAs, gearboxes, in this case it's a hat style strain wave gear optimized for back drivability. Uh for the humanoid market, we have main bearings that are crossroller bearings.

[00:03:58]>> Mhm.

[00:03:58]>> As well as an input bearing for uh for the strain wave gear. And you can see that this actually has an ellipse to drive the to generate the wave for the strain wave. And then last but not least, a torque sensor. And typically high gear ratio actuators that to get that torque density we're talking about, you need torque sensor for the feedback.

[00:04:18]>> So for the viewers that might not be familiar with a string wave gear, it's actually probably best if you pause the video and just >> look it up.

[00:04:26]>> Yeah.

[00:04:27]>> Watch [laughter] a YouTube video. So you have this internal ellipse.

[00:04:30]>> Yes. that's deforming.

[00:04:33]Uh maybe I should be able to to show this.

[00:04:39]>> There are two main contact points that it's oblong. So the the inner driver is oblong and then you should be able to see uh the uh the gear part here is uh also taking the same shape and the number of teeth on the input is off by one compared to the circular spline here. And that's what generates the high high gear ratio.

[00:05:03]>> So you can get really high gear issues with that like 100 to one and higher.

[00:05:07]Right.

[00:05:08]>> With very little mass. Yeah.

[00:05:10]>> And then almost zero backlash.

[00:05:12]>> Almost zero backlash. And one more thing to and this is where optimizing for the humanoid industry is very different than optimizing for traditional robotics.

[00:05:21]Traditional robotics you typically have position control. Uh so stiffness is really important a lot more important than back driving or efficiency for humanoids. It's slightly a different problem to solve. So we have to adjust and do some fancy things with the gear geometry and the bearings to reduce noise because you don't want that to be audible to improve back driving and starting torque and improve efficiency also.

[00:05:46]>> And then uh some of your actuators also use gear boxes, right? We have some here.

[00:05:52]This is a much much bigger gear, but you'll see this here. It's a very typical architecture.

[00:05:58]>> You have uh typically a sungeear and the sungeear drives the planets and then the output of the first so this is a two-stage planetary gear. The output of the uh first stage is through the planet carrier and then the planet carrier is connected to the second sun.

[00:06:16]>> Then the second sun drives the second carrier and that's this the output. So you have two independent ratios, >> correct?

[00:06:23]>> They multiply, >> right? And uh so this one in particular is about 15 to1, right? So depending on what kind of ratio you need for your certain actuator, you have different options to go with.

[00:06:35]>> And uh so >> correct.

[00:06:37]>> What we're showing here specifically is from our unit tree tearown.

[00:06:42]>> Mhm. And and typ and typically the reason why you see this architecture that I so sun to planets to sun to to carrier is is to really minimize input inertia into the system and that gives you a lot more that's what allows one of the things that allows for agile control too.

[00:07:00]>> Yeah.

[00:07:00]>> Is the low low inertia.

[00:07:02]>> Okay. So, Eric, I don't know if you have panned across the table yet, but uh we have lots of parts, right? These are from a Unitry G1 that we've recently dissected here at Monroe. And uh what we've been doing earlier, Sheffller and Scott and and the Monroe guys went over and kind of brainstormed around and analyzed how this robot functions, right? So, maybe let's start with the actuator here since we're on the topic.

[00:07:33]Uh we'll take off the shuffler parts.

[00:07:35]>> And what I think it would be rather interesting to kind of compare the two here is that yeah the idea of the strain wave is you can get a really high gear ratio usually like a 100 to one very easily and you'll see that's very light and also very compact. But one of the things about it is that it's not back drivable because the gear ratio is so high. So we can put the input in and get an output but if we try to reverse it and turn the output it's very very difficult to go in the opposite direction. So for for joints like the knee or something that could take a high impact that could cause the flex plane here to break for example under impact >> because it's flexible >> because yeah and and it's not back drivable. So >> yeah it's not back drivable. Now you'll see that if you actually in the hand this feels a lot heavier than this and you notice we're going through two stages and we're only getting a 15 to1 where here's we have 100 to one. So that really helps with torque density but this thing is easily back drivable. So, wherever the output is, I can do that or I can do the input. That's one of the huge advantages. This is typically what you would see what they put in what they call a QDD, which is a quasi direct drive. Um, it's basically just a way of saying it's like it's a low enough gear ratio that you do not you can measure the output torque from basically the input current. You don't actually need to have a torque cell out here to know what it is. And that torque transparency makes it a lot easier to control. not so easy with the strain wave without actually having to put a torque cell on there to get an idea what the output is.

[00:09:00]So those are the pros and cons of the two is like >> cost how light it is volutric >> space.

[00:09:08]>> Yeah. However, in this particular actuator, right, they are actually measuring >> uh not so much the torque of the output, >> the position to the output because this unfortunately has the disadvantage of having a little bit extra slop the uh >> backlash of the of the gears.

[00:09:28]>> Now, usually the strain waves or the harmonic drives are in industrial robots because they don't have that backlash which means we can control a position really really well whereas here we can't. And you got two stages. Each stage has its own. So that's only going to add up, >> right? There are strategies to get rid of backlash and planetary gears. You can split the gear, add a torsion spring, you can taper one of the one of the gears as well. There's multiple strategies you can you can implement.

[00:09:53]But then you add preload and then that starting torque really becomes tricky for for controls. So usually avoided.

[00:10:00]>> Yeah.

[00:10:01]>> And so that unitry decided to to go this direction.

[00:10:06]uh for their actuators and and they have the same actuator in different sizes at at each joint of the the robot. Um so they kind of scale this same technology for the hip, the knee, the elbows, the shoulders, all the the major joints are using this same kind of uh technology to planetary gear set. Um this is a style of motor that is um relatively low cost but not as torque dense as some of the uh permanent magnet synchronous motors that you see in industry. So um there's potential for some improvement there.

[00:10:55]Just you can see the amount of space there is in there uh where there could be copper. So the the winding resistance could end could be a little lower if they used a different winding technology.

[00:11:08]And then heat is a big problem. And so if you can reduce the winding resistance, you reduce the heat. The best way to handle heat is to not make it in the first place. It wouldn't be a a Monroe live video if we didn't complain about the number of fasteners.

[00:11:25]>> Yeah.

[00:11:25]>> They have. And this here again we see that for example the the rotor is attached with little screws >> and uh >> so this goes together as this.

[00:11:39]>> They worked really hard to reduce the weight of this thing by machining out parts and then they go and add little fasteners to to bolt it together >> which makes the assembly um quick and easy. Well, the assembly process is not as easy as it could be, but in terms of development time, it's it's quicker to to design a fastener than it is to design a snap fit or or some other type of method of connecting it. Um, and this is replaceable. It's it's repairable.

[00:12:14]So, there's some advantages. And clearly what we see all over this this robot was time to market was one of the the big driving forces in the engineering decisions that they made because you know you don't take something like this and and hog it out of a block of aluminum unless your plan is get get to market fast. You can kind of see the >> machining marks on it.

[00:12:40]>> The machining marks where things that you would normally expect to be a cast part instead are hogged out of solid blocks of aluminum um using a CNC method. Now, it's CNC is less expensive in China than it is elsewhere because of the capacity they have. But even so, if you do this anywhere in the world, a at at higher volumes, a casting would be much more cost-effective.

[00:13:13]But this allowed them to get the parts to get them the thing to market >> um much much faster because you don't have to wait for do all the development on the tooling, >> right, >> to get it in.

[00:13:26]>> Yeah. And and then also from a development perspective too, right?

[00:13:30]They're quickly iterating, making new parts, right? So, you're using CNC's.

[00:13:34]You finalize your design and you already know how to make it. All right. If it's doesn't break the bank or the business case completely, let's just keep making them like this, right? And and offer them to labs and consumer markets, right? Get them out in market space. Be first mover. And uh yeah, that's how you end up with basically all these silver components are CNC billet parts, right?

[00:13:58]That's quite striking to see.

[00:14:00]>> If I had to guess, it's like you said, time to market and also probably maintaining flexibility in the early stage of the development because if you might if you buy a a bunch of dyes to cast all this and you only intend to make 5,000 robots of this generation, >> might not be might be a necessary evil to start with uh with machine parts. And then you have a a complete production robot that you can start working on the software for and you can get the software development more mature and software is in these probably the the biggest lead time item.

[00:14:44]>> Yeah. While we're kind of on structure development and we talked about the actuators, should we spend a minute on the cooling strategies? I thought that was >> Yeah, they have several that we've already talked a little bit about. I'm just anodizing the outside of the >> the housing >> housing >> um as a part of it. But, you know, they're they are clamping every one of these motors in around the outside of the outside diameter of the motor. And that helps transmit heat into the aluminum parts and gives them the, you know, a little bit more thermal mass for a time constant. and uh conductive heat transfer. That's one of them. In the CA in the one the parts that have really high duty cycle like the knees um that we see that they put this in um this is a heat pipe. So it's a it's inside the copper is hollow. There is a a type of refrigerant that when the motor heats up, it boils and turns into a gas phase and moves up here. So it's it's clamped up against this and then it it transmits the heat into the aluminum but when it it condenses so it vaporizes here and condenses here and so you get heat transfer with no delta T. So with no um change in temperature.

[00:16:16]>> Y >> and so you can very efficiently move heat from the hot thing to the cool thing.

[00:16:24]Yeah, it's very interesting. The energy actually goes into the phase change, right? And then you go from the liquid to the gas. So that's where your energy, your heat goes from the actuator and then the gas travels up, condensates, goes back one phase and that releases the heat here, right? Right. So that heats up locally and then that just sinks into the aluminum which is relatively uh to what you um need to transport in terms of thermal energy massive and can then easily rad it away into the environment. Right. So very effective way to transfer heat.

[00:17:02]>> That's the I mean it's the same technology we use for cooling CPUs >> laptops. Y >> yeah. And so I love seeing adoption of technology from one one industry into another. You I I don't know of another case where the um the stator of a motor is being cooled using a heat pipe.

[00:17:27]>> Yeah, exactly. You only see that on CPUs, not on motors, >> right? Yeah. What you also see in laptops and CPUs, right, is air cooling.

[00:17:34]Everybody's familiar with that, right?

[00:17:36]So for uh the hip actuators here and these two specifically that uh move the leg forward and backwards, they have two fans in the back here right here. Uh centrifugal fans and uh they will blow air out of this hole into this again billet CNC machine parts. So from that hole to that hole and [snorts] then these heat fin looking structures that are machined in they they're air guides airflow guides and they guide the airflow 180° along one side of the actuator cooling essentially from here to there and exits through this hole. There's also some diversion of the air that blows into the electronics in the front here to keep some of that cool. And uh so that's a relatively more common cooling strategy, just air cooling. This is what you hear when you are close up to the robot, right? The the the blower fan sound.

[00:18:40]>> It's coming in around like this, right?

[00:18:41]Because we can see the vent holes over here. So it's then coming up the torso.

[00:18:44]So most of it is coming out here but some is getting redirected to go through the body essentially and coming out these holes >> coming out there which means it will then kind of come up to here as well impinge a little bit.

[00:18:55]>> Um yeah somewhat but essentially it's pretty open to the atmosphere around this area.

[00:19:02]>> And the other thing is that uh this version is just a single axis on the tor. So they have another one with these what they call these kidney actuators where they have two up there along with this.

[00:19:10]>> It would be here.

[00:19:11]>> They're there. Uh yeah it would be there but they're not there. They're not there. Right. So that's the option you uh between the different versions of the G1.

[00:19:18]>> Yeah. Then Scott, maybe you can talk about this. Is this you tell me would this function similar to how they're actuating the ankle?

[00:19:26]>> Yes.

[00:19:27]>> Maybe you can talk about that for a moment.

[00:19:28]>> Yes. So um pretty much I mean the first thing to talk about the robots is they have joints and joints have to be actuated somehow. And what they've done is they actually have made the actuators and the joints the same thing. So you can do a remote actuation strategy where you have a hinge joint there and then the motor somewhere else and then a tie rod. They go ahead and move it or in this case the actuator is forming the knee joint or it's forming the different hip joints. We see it up here in the elbow and the shoulder and everywhere else. The problem with that is that the actuator has to be a hinge as well which is why you have those cross roller bearings as well. It has to handle a lot of that load and trying to figure that out. It also gets very bulky and at some point you want to make things as compact as possible and ideally you don't want to have what they call distal mass. Any mass that's at the end of your structure, it makes it harder to carry.

[00:20:18]The more you want to have as much of it as proximal as possible to try to pull it up.

[00:20:22]>> Now, when you come down to an ankle, if we think of our ankles or wrist, we think of these things as being very small. They don't take up a whole lot of space. So, how do we make an ankle that doesn't have the actuator actually built in here like we do there? Otherwise, you're going to have a big swollen ankle. Um, there are some robots like the Boston Dynamics Atlas is now trying to form them actually out of actuators.

[00:20:43]And you will see it's really big and bulky. So, the trick is they are using an actuation strategy called RSU. That stands for rotary spherical universal.

[00:20:52]Explain exactly what that is. It's also a similar strategy we see in the Optimus robot, but in the Optimus, they call it an SPU, spherical prismatic universal, where prismatic basically means you're using a linear actuator rather than a rotary. So, let's break this down. The first thing we form the ankle joint with this universal joint down there or sometimes called a cardon joint. And it has basically two pins that are concentric that go through each other that allow us to do the ankle pitching this way and then also the roll that way. And you notice we can get more movement in the pitch than we do actually in that roll. And you also notice that I this degree of freedom is locked out. I can't do it. So if you look at our own feet, we'll notice we can pitch our feet quite a bit that way.

[00:21:38]And we do a little bit of this. And the reason we do that is for balance. So that if we're we're out like this, we want to keep our feet flat to the floor and not angle that way. The other thing you'll notice is that while we can point her toes in any direction, that movement does not happen below the knee.

[00:21:53]>> No.

[00:21:53]>> Basically, your knee always points the same way. Yeah, we have a little bit of compliance that way, but for the most part, it's only two degrees of freedom.

[00:22:00]This locks it down. You only have two degrees of freedom. Now, how do you actuate that? Well, you do it remotely with this tie rod. So, this tie rod goes up to here and it's connected to one of these motors. So, you'll see you have a little lever arm here that when you move on that will allow it to go back and forth. But you'll notice there's a second tie rod there as well. And they both have to move at the same time. And they're each connected to a different motor. Now, that means it gives you torque sharing. So, that means you have two motors that can give you a lot more power in that direction, which means, hey, I can jump now because I have two motors doing that. But there's a little bit of movement this way. In order to accomplish that movement, you actually do differential movement of these tie rods. So you'll notice when I did this, one goes in the positive direction, one goes in the negative direction, and that gets kind of split that way. Now, as soon as you do this, the tie rod is going completely out of the plane. So if you were normally just articulating that way, everything stays in a plane, and all my joints can be pins. The problem is when I do that, I can't. And now you have to put in what's known as a spherical or him joint right in there for that connection. So that's not a pin joint. That's actually a little sphere.

[00:23:10]Probably from that direction you can see it a little bit better. And not only do you need a spherical uh joint right there, you need a spherical joint up here and another one up there. So hence the rotary spherical universal. That's where that comes from.

[00:23:24]>> There we go.

[00:23:25]>> Yeah, >> that makes sense.

[00:23:26]>> Yes. Now the danger of this is because it's a parallel mechanism, the basically the torque transparency is a little bit tricky. So normally we want to be able to read the motor torque so we can figure out from our RL policy how to walk. And if we really don't know what's happening down here, we're at bit of a loss. We don't want to put sensors down there. And because it's being shared, it's really hard to kind of split the two um signals out on what's actually coming from pitch resistance and what's coming from rolling resistance. So now you're seeing some designs that are trying to change that up. However, this RSU strategy you will also see in the Optimus wrist except it's using it with this the um uh linear. So that's it's an SPU and they also took the same thing and then built it right into here.

[00:24:12]>> Y >> now you have other strategies. You'll notice what's interesting. They articulate it from the front of the foot. You will see some robots where they say, "Well, rather than have them come down here, why not have the control rods come to the back?" I guess that's just a choice and where you want it to be. Um, you have another option is that you could take these motors and make them go 90 degrees this way because you'll notice both tie rods, if we take this thing off, we will see where they connect up there, they can get in each other's way depending upon where you are in the workspace. So, you have to pick and choose where you um attach them onto your motors so that you have the most movement. And the other problem, of course, you have is that your lever arm keeps changing and you can get down to a point of kind of a deadlock. So, pros and cons of being able to do it, but the main thing that motor mass goes from here up to there, >> move it up. And the other thing that scares me a little bit with this, this is I mean know the use case of this, is not used in production. You know, it's maybe lab work and that type of stuff.

[00:25:12]But you see this open cardon uh joint and the grease. If this worked for a long time in your living room sweeping dust and vacuuming or in a production setting, that's going to dry out, get sand in it, and uh >> the joint won't like that, right? So, >> um definitely not ready for prime time working in a production. The way it's I guess automotive, you would see, you know, the boot over this kind of like you see on a CV joint.

[00:25:41]>> Yes.

[00:25:42]>> Right. That that type of >> definitely do that. Yeah. Now, this is rather interesting. Yeah, with this uh we see this both here and up higher on the joint is a circuit board with a bunch of capacitors on it. And so this is connected to the DC line uh we call a DC bus that feeds all of the motors and all of the motors are kind of fed as a daisy chain back up to the battery. And since every motor can be both a motor and a generator, you end up with the the energy coming from the battery.

[00:26:26]You'd like it to be a nice constant voltage, but every time this thing generates a little bit, >> see the green light, right? It's generating >> Yeah. energy >> just by moving the the motor generates a little bit voltage. So the the DC voltage with all these motors in the string and motor each one doing its own thing sometimes motoring and sometimes generating your DC voltage is all over the place and you need to be able to control these things and your the torque you get from the motor depends on the voltage and so if the voltage is going all over the place it's really difficult to control precisely uh what each motor is doing and so they put in these capacitors along uh on the line to store some of the energy and help flatten out some smooth out the ripples. Um, these are performing a filtering capacity uh smooth out the ripples in the voltage waveform on the DC bus and makes all of the motors much more controllable.

[00:27:31]The way that they did it is they sort of packaged it in the locations where they fit.

[00:27:38]And so they're running extra cables and connectors to these things.

[00:27:45]It'll be interesting to see how Unitry does things on their next generation and if they are packaging these things in more optimal locations. Again, time to market, get it in in in production, start running it, run into the problems, and fix them >> and then feed that back into your next generation. That seems to be the the philosophy behind the design.

[00:28:10]>> And this and >> that fits right in. So >> it does. Yeah.

[00:28:13]>> Yeah. It's interesting how they package this, right?

[00:28:15]>> Yeah.

[00:28:16]>> So the the capacitors they go into the lightweing features, right? So you kind of have to push it back in.

[00:28:22]>> The capacitors get warm.

[00:28:24]>> They Yes. They they are both uh they're storing energy and they have a little bit of energy dissipation. So there's a little bit of heat that's that's produced and keeping them cool. The capacitors have a very narrow operating range that they like to be in and so keeping them cool is necessary.

[00:28:41]>> So that means it'll be conducted away.

[00:28:44]Yeah. They can wick it away from >> and and of course you know you you are trusting the fact that uh you're you're going to have some rest periods and some heavy work periods. And so the temp you're not going to have continuous high level of operation. uh so the capacitors have time to dissipate the heat and so again this kind of thing means that you you need to kind of build a virtual model in the computer of your whole system. So you need to work with a company that is able to um deliver create before they deliver the heart the hardware they will deliver to you the kind of virtual hardware that you can put into your model and then run the model and experience the model will show this this DC bus instability and how much then you'll be able to design how much capacitance you need to absorb it.

[00:29:42]Yeah, 100%. May quick comment. I mean the whole SIM to real gap is a big topic and delivering like we have the responsibility to deliver good digital assets of the gear train of the controller etc etc to enable these OEMs to do the simulations that you're saying and really do worthwhile or meaningful uh optimization before they even have their hands with the with hardware basically. Yeah, that's that's the key that otherwise um you're forced into >> hardware [clears throat] iteration which can take months instead of days.

[00:30:19]>> It's interesting with this unitry that you only see rotary actuators, >> right? So because a lot of companies are using as you know right you provide both rotary and linear actuators but um there's advantages for each >> um actually maybe with that uh can you talk compare and contrast some of the advantages and disadvantages of the different actuation strategies? Yeah, absolutely. I mean, um, yeah, this is a very interesting topic. Even, okay, even within rotary, you see there's strain wave gears and you see there's planetary gears. Even within linear, you see there's ball screw and planetary roller screws.

[00:30:58]>> Mhm.

[00:30:58]>> Even further within planetary roller screws, there are inverted and non-inverted planetary roller screws, right? So, uh, I give you a quick example. Um, if packaging is really really tight and the stroke that you have to actuate uh isn't excessive, you can use something uh like this uh with an inverted style either ball screw or planetary roller screw. Uh what inverted means is it means you uh rotate the nut, you have an anti-rotation feature and the spindle translates. Uh the benefit here is packaging because you uh glue the magnets directly on the on the nut of the inverted screw and then Yeah, exactly. That's that's the actuator configuration. We can we can pull that as well.

[00:31:43]>> Um but then then there's the other aspect is that normally >> Yeah.

[00:31:47]>> Oh no, that's ball screw. That's the ball. We have a roller screw down down there.

[00:31:52]>> Yeah, that one roller screw right here.

[00:31:54]What's interesting the roller screw is that you're talking about inverted is now the thread's on the inside rather than the outside. So if I'm putting a thread on the outside, that's really easy >> to manufacture.

[00:32:05]>> How do you put the thread on the inside, right?

[00:32:06]>> Well, you can for like the first couple millimeters and then as you start getting deeper and deeper, it gets really hard.

[00:32:11]>> Yeah, you have to you have to come in with a grinding uh wheel and do this all all the way.

[00:32:18]>> And how long does it take just to make that?

[00:32:20]>> Uh that's a great question. It depends on how good you are with design and grinding and how deep the the the depth is. So anywhere between from what I've seen anywhere between a few minutes to 15 minutes per part to 20 minutes.

[00:32:36]>> It's a lot of processing time. That's cost.

[00:32:38]>> And then you got this piece here which is where we actually have the planetary roller screws.

[00:32:42]>> Think of ball screws and automotive for EPS for steering. Everything started as ball screws are these ground precision industry parts. Then automotive was interested in EPS. So now you see ways to roll the spindle, you optimize the grinding for the nut. Those exact same approaches have to be applied here as well to get to get the cost down. So that's >> yeah, I'm excited about seeing that happen because we're looking at, you know, thousand actuators right now >> and you have 24 of them in something like this. uh when we start applying that kind of automotive technology, then that likely is going to go from $1,000 per actuator to $100 per actuator on that scale as and because that's what we saw happen in in the automotive industry.

[00:33:33]>> And so I'm excited to see that kind of thing happen here. And it might not be applying automotive technologies necessarily, but the automotive approach to scale. Um, it might be different products, different level of precision that we eventually might need, but then approaching it with that same, you know, mentality for scale is is really important. Yes.

[00:33:55]>> Mhm.

[00:33:55]>> Absolutely. But back to the quickly to the different. So roller screws are really good for getting you a big uh uh packing a lot of punch and as compact of a pos of a package as possible. Uh ball screws give you some more efficiency. So it depends on the application which which way you go. And then typical uh non-inverted roller screws. This is where you have a very big stroke and this is where you spin the spindle and the nut translates. So it's the other way around. If you have a long stroke uh that you need to actuate, typically you end up going with a a traditional roller screw.

[00:34:31]>> Yeah. And so one thing that we discussed earlier that caught my eye is how you approach >> force sensing in these linear actuators, right? So >> often times you'd see >> strain gauge or you know PZO based approaches.

[00:34:47]>> You've looked to uh other technologies.

[00:34:50]What do you have there?

[00:34:50]>> Yeah. And I mean this is where our background in automotive helps. We uh we have these MEMS elements for pressure sensing and in diesel applications today uh lines that are optimized for millions of pieces a year. Uh but it's a pressure sensor but the concept is the same.

[00:35:09]Instead of dealing with pressure, you're dealing with a diaphragm and a force uh implemented the exact same way as you do for high volume production. And then that brings the cost significantly down from a typical load cell in an industry application to a commoditized uh pressure or force or torque sensor even.

[00:35:28]>> Yeah. So this might be multiple times less costly than other solutions.

[00:35:33]>> Exactly. But this is exactly the approach to take on every last component in the humanoid. So >> okay.

[00:35:40]>> Yeah. And Armen, I think maybe just to go ahead and wrap up, why did you choose rotary versus linear? Yeah.

[00:35:46]>> Now, if you look at it at the human body, everything we have is basically a rotary joint. Unless you're ET, you don't have anything that's translational at all. Okay? So, when you So, when you want to build it, you're saying, "Well, wait a minute. Since everything's rotary, why not just go ahead and put a rotary motor there?" Um, because if you are going to do linear, you're basically taking linear motion, trying to figure out how to turn it into rotational motion. And it gets even sillier because think about these linear actuators start out with something that rotates to try to create something which is a translation to now do it. Why would you then go ahead and do it? And a lot of it comes down to sort of the the strength to weight ratio you want to have. There are some things like maybe down in your knee where you need a lot of force on that load or in your elbow. The other thing is it just packages really well. I mean it's like I can fit a linear actuator in my forearm much better than I can put a really big rotary motor right here. So, a lot of it comes down to actuation strategies, packaging strategies, the amount of strength you want to have, and the cost you're willing to bear. The others that we've talked about is that if you put the actuator right here, it has to carry the load. So, maybe down at your knee, you want to put a hinge joint there. So, it's taking all of that load, the cross load, and then all your actuator has to do is apply a forward load and doesn't have to worry about anything that's off axis.

[00:37:00]>> So, these these are engineering decisions.

[00:37:02]>> Yeah. Um maybe to wrap up there's few engineering exercises as complex and multid-disciplinary as humanoids right so it's a very interesting space to watch that uh is really taken off now and uh after going through this uh it's very inspiring very interesting and uh it certainly seems like we're in for a lot of entrance uh into the space right so it'll be interesting to watch >> absolutely >> and then I think we see a really good example here of thinking about that. See the elbow joint there? That's a cantal lever, which means you have to design this to be able to take that loading without failing. But up here on the shoulder where they say, "Hey, you're taking even more of a load." You notice they've got kind of a Yolk there to be able to do it by pinning it on both sides.

[00:37:51]>> They didn't put it on this side here because probably adds mass. But again, you have to decide, is it going to be able to handle that? Are my cross bear cross bearings going to be able to handle that? Yeah, I know you have side forces on it, right?

[00:38:03]>> Oh, a lot. Absolutely. I mean, that's basically, you know, your ACL tear is because your knee decided to do something it's not supposed to do.

[00:38:11]>> Okay.

[00:38:13]Anything else? Any last comments?

[00:38:15]>> This is great.

[00:38:16]>> No. Okay.

[00:38:17]>> All right. Uh, with that, I think we'll wrap up. We'll call this a video and, uh, please tune back in. We'll have more of these. Uh, there's actually Dexter's hand torn down behind us. Uh, but we'll keep that for another video. Um, all right. Al, thank you.

[00:38:33]>> Scott, thanks. And Paul, right, >> so take care, guys.