Unitree G1 Actuator Teardown: How This Humanoid Robot Moves

Hello everyone. Uh, today it's Paul and Armen from Monroan Associates and we're giving you a deep dive into the actuators in the Unit G1. Right. So to

ground everybody, we've been tearing down or have torn down a Unitry G1 here.

So please maybe find our other videos.

But uh what we're going to do today is look at what makes how do they actually actuate um their their robot. Right? So here we

have a left leg and arm that we kept together as reference material. We tore

down the other side of uh the robot and we're keeping these kind of you know assembled as it's just mirrored. Um but

maybe Paul, how about you start us out with talking about the general strategy and architecture of of the actuators in the unitary.

Right away you can see that there's that for the leg you have different size larger actuator than you have in the arm. The arms have

similar but smaller size actuators.

And let me just talk about a little bit what an actuator is. So this is the thing that moves the the robot limbs

and then for each degree of freedom for the limb you have to have a separate actuator. What an actuator is is an

actuator. What an actuator is is an electric motor. Um

electric motor. Um that and the electric motor has a controller and usually the electric motor has some kind of gearbox because you want to

magnify the torque delivered by the electric motor to be able to move the limb. And so you have for each actuator

limb. And so you have for each actuator position you've got different requirements. So for the hip and the

requirements. So for the hip and the knee um you have very extreme stringent requirements for a lot

of torque in a small package and similarly with the arm elbow and shoulder you've got a smaller space but

uh lesser amount of torque you need and so you can put a smaller actuator in.

Yeah.

So, the package space is sort of driving most of the decisions and also like all of the other things as we've mentioned on some of the other videos

um unitry motivation is to try to get to market quickly with something that does the job

well because developing the software to operate the robot is at least as big a challenge as yes the hardware. So get hardware out that

the hardware. So get hardware out that works quickly and and then iterate from there.

And so we see that kind of thinking also in the actuator itself in these robots. This particular one has something like two dozen of these

actuators. And when you look at the bill

actuators. And when you look at the bill of materials and the costs for everything that's in a robot and actually you could buy our

report and get uh a detailed breakdown of the bill of materials. You'll notice

that the actuators make up a disproportionate size. With 24 of these little actuators,

size. With 24 of these little actuators, you end up with lots of money in the in the actuator. And so designing these

the actuator. And so designing these efficiently, getting the the lowest cost solution is absolutely critical to the whole business case um surrounding the whole robot.

And I'll add to that, this model in particular had 23 or 24 actuators.

Um this is a relatively simple model, right? So even even the G1 goes up to 30

right? So even even the G1 goes up to 30 or so, right? So it increased the cost even more. And in the future, most

even more. And in the future, most traditional humanoids that have two legs are going to be in the 30 to 40 kind of number for actuators, right? So there's

a lot of cost there. Um, but maybe Paul, and correct me if I'm wrong here, uh, looking overall, you seem to have two frame sizes of motors, right? You have

these bigger ones that you kind of see here, which is on the table, right? And then the smaller actuators in

right? And then the smaller actuators in the arm and these are also used in to actuate the foot. So in the calf

but then maybe in your analysis I think you've also noted that the stack height changes right between the different overall sizes. Is that right?

overall sizes. Is that right?

Yes. You can see that uh the smaller motor they increase the the height of the motor just a little bit. You slap

them together just a little bit. a

couple millimeters to give it a little bit more torque in this tight package.

So, yeah, that's the that's the thing. You you have to design each one of these for the requirements that it's it's needed. There isn't really a

needed. There isn't really a one-sizefits-all when it comes to actuators. So when you go to actuator

actuators. So when you go to actuator design and let's look at what's inside the actuator at it. At the core there's an electric motor. This is the stator

and this is the rotor of the electric motor. This particular rotor has 16

motor. This particular rotor has 16 magnets. So that's 16 poles.

magnets. So that's 16 poles.

And the stator has 18 teeth or 18 slots.

We always have the same number teeth and slots.

this is a good choice for this motor.

Um, and so, uh, electric motor designer has a lot of options available to them on ways to make an electric motor. So,

in this case, the rotor is on the inside. It spins inside here. The stator

inside. It spins inside here. The stator

is on the outside and the the windings are wrapped tightly around each tooth.

And that keeps the axial length of the motor an absolute minimum. So the length of the copper sticking out past the end

of the steel is a minimum when you can wrap right around one tooth. This is a little bit of a compromise though to get the minimum package size

with 18 teeth.

trying to create a 16 pole magnetic field is not perfect. You end up with a magnetic field that has a lot of

harmonics in it. It's not a beautiful sine wave.

By contrast, when we do a traction electric traction motor, the designers will put many many

teeth in. And so

teeth in. And so the windings have to go from one tooth to another and they have to kind of weave their way past the other windings

in order to get from one slot to another. But when you do this, the

another. But when you do this, the benefit is that you create can create a a beautiful sinosoidal

magnetic field um that produces torque with very low loss. So this is an an extremely efficient way to do a motor, but you get this extra length out beyond

the steel in order to get the wires from one slot to the other. Is is that just purely driven by packaging that you have so many teeth and so many windings that you just end up needing the room to make

that turn?

It's yes the um and in this case efficiency was valued more highly than the package space. And in this case

fitting a maximum amount of torque into a minimum size package was the driving force. And so they traded off a little

force. And so they traded off a little bit of efficiency in order to get the minimum package size.

This is also a motor that does that same thing where it's the windings are wound around the teeth, but in this case the

rotor is on the outside and it spins.

So it's the magnets are the part that's spinning on the outside. And this

method, this is for a automotive radiator fan. This type of motor is a

radiator fan. This type of motor is a kind of an inside out motor.

The beauty of this is that you get you move the air gap as far out as possible.

You get lots of torque. But the problem with this is now the windings are in the inside and there's this thing spinning on all around them. It's really

difficult to get good cooling for the windings. And the windings are where all

windings. And the windings are where all the heat or most of the heat are is produced. And so this type of motor is

produced. And so this type of motor is is good when you have like air flow like a like a fan that can get through inside the motor. But when everything is

the motor. But when everything is enclosed like the actuator has to be then you want to get your windings out near the the outside.

And so here the windings attached directly to the outside frame which is aluminum and it uh anodized aluminum and also they've even made these little

ridges in the aluminum to increase the surface area and they clamp they clamp this additional aluminum around this

tightly so you get intimate contact between the motor and the windings and the the aluminum of the heat sink uh

which is the rest breast of the leg. And

so we can get the heat out of the motor by by putting the windings on the outside and having the rotor on the inside. This type of motor also is nice

inside. This type of motor also is nice for an actuator because it's kind of a donut type shape. It's hollow in the

middle. Thin w thin stator thin rotor

middle. Thin w thin stator thin rotor open space in the middle and we take advantage of the open space. The

designer of this took advantage of the open space to package the first stage of the gear reduction on

the inside of the of the motor.

So this pinion gear which is the sun gear of a planetary gear set is attached to the rotor on the inside. This one the

rotor is so this would be like this and this is inside like that.

So, that's how this is set up. Uh,

you've got the two two-stage planetary gear set. The

first stage is packaged inside the motor and the second stage is packaged here right next to the motor. Second stage is larger and

of course it has to handle higher torque. So the torque from the motor is

torque. So the torque from the motor is magnified first through the first stage planetary gear set, then magnified again

by the second stage gear set to give us a total of 7.77 to1 with this particular gear ratio. With a

two-stage planetary, we've seen cases up to about 20 to1. Mhm. Then Paul maybe uh talk a little bit about what are your

advantages or your tradeoffs between having a two-stage and a single stage here, you know? So the single stage could theoretically get to roughly that

ratio, right? So why do you think they

ratio, right? So why do you think they chose to do a two-stage given that added complexity? Yeah, that's often that's a

complexity? Yeah, that's often that's a an important point that a motor an actuator designer is going to have to look at the gear reduction needed to do the job and see whether or not they can

do that with a single stage gear reduction. And 7.77 is right on the line

reduction. And 7.77 is right on the line where you could conceivably do that with a single stage gear reduction. But when

you do then where the teeth meet you have this contact angle. And when you have a high

contact angle. And when you have a high gear ratio like that the contact angle is not very perpendicular to the um axis

of rotation. And so you end up with the

of rotation. And so you end up with the teeth rubbing against each other and that produces extra loss. And so

and wear right and wear and reduces life. um and it reduces noise. And so, yes, it can be

reduces noise. And so, yes, it can be done, but it turns out that a two-stage um gear reduction for this ratio is a

little more efficient and because of the contact angles you can get with the lower um gear reduction.

Yeah.

So, those are some of the tradeoffs that are involved. And in this case also,

are involved. And in this case also, they're using planetary gear reduction.

And in some cases when you want to go to higher ratio then we see strain wave type gear reduction and cyclloidal gear reduction

to get very high you know gear ratios from 20 to 50 and then even as high as 100 or more.

The beauty of this type of ratio this is the same kind of uh gear ratios also that um traction motors use. highly

efficient and easily back drivable. In a

vehicle, you need regen braking to help improve efficiency. You actually do

improve efficiency. You actually do regen braking um with this type of a a setup as well. So, every time uh this

thing uh the robot jumps and lands, it absorbs that impact and uses the motor as a generator to replenish the battery and improve

your efficiency. It's interesting

your efficiency. It's interesting because it physically has to, right? Any

any robots when if you heard somebody, you know, throw out marketing terms and they tell you that their robot is generating and recharging its batteries, it's kind

of that would be just marketing, right?

You have to from physical principles of essentially making the magnet spin.

Yeah. Whenever you try to produce negative torque uh or you know resist uh motion then the the motors are technically generating

this. So this also one of the things

this. So this also one of the things that I love about this motor and is this control board this single control board

um relatively small compact control board processes quite a bit of power over 100 amps at peak load. So you've

got the six switches like like other inverters we've talked about you know one two three four five six switches three of them one two three are attached

to the three phases and the three phases are you know go to straight to the motor and then the other three are attached to

the DC bus and so this little thing turns the DC from the battery into the AC power that

the motor needs to spin. And on this side, it has the the co the controller.

So down through the these little wires, you get a command. It says, I need X amount of torque from the from this

actuator. This turns that torque command

actuator. This turns that torque command into the appropriate current command and then sends the voltages needed to these

voltage commands to the switches and that sends the current into the motor to produce the torque. So this little board

does an awful lot in addition to that.

So this board is sensing the position of the rotor. So it knows exactly how to

the rotor. So it knows exactly how to turn these switches on and off to control the motor. The outside the output of this motor has there's a

little shaft goes through here. It comes

out here and that shaft is connected to a gear, a little plastic gear that's just under this board. And that plastic

gear is a a position sensor for the output. And so

output. And so the controller knows the precise position of the rotor.

But after you go through this gear reduction and each of these gear sets has just a little bit of slop, a little

bit of play in the gears. And so you don't know for sure exactly where the output position is. And so this actuator

senses the output position and sends it back to the controller so that it can give the overall controller precise position and it makes the whole thing

much easier to control even though there's a little bit of play uh in these gears. A couple of the features that I

gears. A couple of the features that I wanted to mention as we talk about the gearbox. The last bearing, this outside

gearbox. The last bearing, this outside bearing, um, this bearing is a relatively large bearing on the outside diameter of the

last planetary carrier, which is the output of the of the whole actuator.

This is of course necessary to be able to allow the thing to rotate, but it also has to be able to handle off-axis

loads. So when someone is holding a a a

loads. So when someone is holding a a a heavy load then that torqus puts a big off-axis torque on this bearing and so

this bearing has to be robust enough to handle that type of offaxis load and uh yeah we're working in a real place.

[snorts] So this off-axis load and the bearings that that can handle it make this not so cheap a bearing.

Yeah. But it's much better to put a a robust bearing inside the actuator than the other alternative, which is to put a separate bearing on the outside of the

actuator. Um, and and a whole another

actuator. Um, and and a whole another additional structure outside of the actuator to carry that extra load. So

what we're talking about here if you imagine you had you held up a mass on the hand and uh like this right your

your load uh the torque that's imposed by the lever arm is uh unaxis now of the of the actuator axis right so but if you

held something like this right it would be an offaxis load and now this bearing becomes very critical right furthermore How you design this joint uh also

affects the the local offaxis loads, right? So you see here unitry went for

right? So you see here unitry went for um a one-sided connection there. There's

other designs where you would hold it on both sides, right?

Which which they do right here, which exactly which they do up here.

Right. So obviously this lever armor is even longer. So it's more critical to

even longer. So it's more critical to better support your bearings, right? So,

they decided, okay, let's um put a yolk on here versus here's a shoulder lever arm. They get away with just connecting

arm. They get away with just connecting to one side of the actuator.

Okay, last feature. How do you get electricity to a thing that is rotating?

So, if you have 24 actuators, one option is to run power directly from the battery with an individual wire for each actuator. That turns into a giant wiring

actuator. That turns into a giant wiring harness that is going to cost about as much as your rest of your robot combined. And so, there's got to be a

combined. And so, there's got to be a better way. And Unitry does have a

better way. And Unitry does have a better way.

They run the power down the center power and signal down the center of each actuator through this bundle of wire.

And so each of these motors are daisy chained together. So the power of one

chained together. So the power of one come exits and goes out and the next motor is attached. They're hooked up in

a daisy chain manner. The only trouble with that is that now this wiring harness has to be able to twist.

So at through it's not doesn't go more than 360 degrees around. There's hard

stops that prevent it from going more than 360 degrees, but it goes about 360 degrees one way than the other. And so

over many many cycles that's going to fatigue the copper and eventually cause

wires to break. And so I mean this is DC power. And so in principle you could use

power. And so in principle you could use a couple of big solid conductors to go through but then the big solid conductors are would not handle this

kind of twisting and would fatigue and break after only two or three cycles.

So instead they use very small stranded wires and and even that they they go to small bundles and in individually

insulate each bundle.

And so that this gives the the power cables the ability to to be robust to a certain amount of twisting.

And this type of arrangement can can survive through say a couple hundred thousand cycles.

If you need a robust connector or need something that can handle millions of cycles, um then you might have to go to a different solution like for example what

cars do with the steering wheel. So

steering wheel can go round and round and we have to there's a whole bunch of stuff, you know, buttons and things on the steering wheel and all the wires that go to those buttons on your

steering wheel in a car, they're fed through something called a clock spring.

And so it's a spring that um has a a ribbon of wire ri wire ribbon connected to the spring and it allows it to wrap

and then unwrap um as you twist the steering wheel. And that kind of a setup

steering wheel. And that kind of a setup can do millions of cycles as they do in automotive steering wheels every day.

But that's a trade-off. It adds cost. So

to get that kind of robustness, you have to add a part.

Yeah. And just for context here, if you if you think about the use cases or industrial use cases of humanoids, uh a couple hundred thousand cycles simply isn't enough, right? So

it it depends, you know, if if you're in a joint that um isn't flexed a lot um so it it's not

it's it doesn't do the full range of motion often. It it does the full range of

often. It it does the full range of motion infrequently. Yeah,

motion infrequently. Yeah, then maybe this solution works very well. These are things that um the

well. These are things that um the designers have to trade off and look at.

You know, I think we just scratched the surface here um with how you how actuators are how complicated these things are. Yeah,

things are. Yeah, these things are currently made in the thousands and usually by um suppliers

that are doing custom applications for each customer.

And so that drives the cost of these up.

So if every time you're doing a a slightly different actuator customized for the customer applications, it's going to be at one level. If you if

you're going to be making millions of these, um, then you're going to want to go to kind of a you can have any color as long as

it's black approach and you're going to have to simplify the system down so that it assembles a little bit more easily.

And so that these are things that it it excites us here at Monroe about, you know, what are the kinds of tradeoffs that you would have to do to scale this type

of actuator up from where it's currently at to the millions that are coming in the near future.

Certainly a lot of um uh evolution in humanoids and in particular in actuators because they are such a cost driver.

we're going to see uh many different actuaries come out that are going to employ or rely more heavily on automotive methods right as as uh the

sector grows. Um but with that Paul I

sector grows. Um but with that Paul I feel like I I learned a lot. So thank

you for joining here today. Um, if you are involved in advanced robotics and development, uh, or even other industries, automotive, um, and you'd

like to talk to me and Paul about your specific needs from a cost perspective, benchmarking, right? Uh, we'd love to

benchmarking, right? Uh, we'd love to talk to you and the best way to do that is to either uh, reach out via LinkedIn or just um, go to our website,

muncommercial.com, and you'll find a contact us section there. So with that, thank you very

there. So with that, thank you very much.

Yeah, thanks for listening.

See you next time. Bye-bye.