How we actually test a robot hand | RLWRLD

Come up with the idea to test with what kind of grasp or what kind of item it can grasp.

Overall, yeah, this was actually for heat management. These two hands are the best

management. These two hands are the best hands that can accomplish most of the tasks that we have set. That's why I brought this hand. We are trying to have

a rough feeling of which one is better at which criteria and which task. If I

had to choose one between Sharpa or Ubi 2, um Hello guys, we're back. Did you guys like the previous video? Today we are

back with actually working hands. So

today we can show you guys like how how it's performance is like comparing the different uh hand like Sharpa.

The way we tested on the performance ability of our hands is we really have like 20 plus tasks and see how easy or hard it is to actually

complete the task. Some of them are like grasping some weird looking mouse.

Defining what robot hands is good robot hand. That kind of stuff was really hard

hand. That kind of stuff was really hard because there was no testing metrics for dexterous hand. So we had to define what

dexterous hand. So we had to define what is dexterous hand.

A hand like this is pretty difficult to pick up. So we we tried to divide the

pick up. So we we tried to divide the performance cost into as many objects as possible to play with cuz in real life every

object has different uh shape or weight distribution or that's why we have like 20 plus objects to test on.

It It won't do it.

Yeah.

Yeah.

If they really want to make it like numerically valuable, we should test at least 30 times. Yeah. And actually, this move is really hard move for the robot hand. It's very hard. It's really

hand. It's very hard. It's really

impressive that over time can do this stuff.

Because usually the thumb ROM is really weird. It doesn't match the human. So,

weird. It doesn't match the human. So,

it's really hard to press something like keypad or pen. But since Uchi 2 has really good thumb joint arrangement, it's it's easy.

This kind of action is pretty hard. Uh

one of the task that I found most impressive was that it can do a lot of uh human-like uh thumb action.

Oh, yeah, flipping the phone. Yeah.

Like Joe said, ROM back-drivability, like special silicone part, every everything makes this uh all

this kind of movement possible. Let me

Let me uh try it. Yeah, it's

try it. Yeah, it's it's it's hard for Uchi 2 as well cuz the thumb is actually pretty short compared to other robot hands. So, it's

not only just ROM or the strength or um the back-drivability, but Oh, there we go. I guess Uchi 2 is

specialized in tasks that involves very, very delicate control of the thumb, like pressing numbers on your smartphone or uh doing flippings and all that.

We're grasping some non-uniform object with Uchi. And what's really good about Uchi

Uchi. And what's really good about Uchi is that you don't have to think about it. You can just close your palm, that

it. You can just close your palm, that it will grasp the like like plastic bag or box, anything, because it's back drivable. It just adapts to the object.

drivable. It just adapts to the object.

So, that really makes the total operation a lot more easier.

It It seems easy, right?

Yeah. It's actually very hard to do it with a robot hand.

For all it's due to the lack of friction.

Uh friction, yeah. I mean, they say it's easy to like this kind of glove and like friction like that. But, the glove that has a lot of friction on it.

But, none of them actually did except for Uzi Uzi.

Right?

Yeah. And the squishy part, the the the part that we mentioned before on the previous one.

Do Do you think it's actually normal hand?

I'm not sure about that. Yeah, yeah,

yeah.

The heat management is way better than Uzi one, but it's still pretty warm.

Sometimes, though, we even up to like 56 degrees Celsius.

And one more thing, one more thing. This

silicon part in hand, this this I think this shape really helps the flipping the phone.

Right?

Yeah. And they did very cleverly so that there's no gap at all. So, it's a lot easier to grasp when you're like like that holding it like this way.

Yeah, like this kind of thing.

Yeah, yeah, yeah.

Thin object which can hold it.

Most of human uh robot hands cannot hold it like this.

But, Uzi can because they have squishy parts.

Yes, yeah.

We actually call this a minimum grasping diameter. And it's it's it is also a

diameter. And it's it's it is also a part of our testing Yes.

parameters.

Yeah.

Yeah.

We added one more extra criteria for testing robot hands. One of

them this was actually for heat management to see like how high the temperature goes up to.

It's like 45 degrees, right?

Mhm.

Not that hot.

Mhm.

And And what I found out is that middle finger is the hottest one.

Yeah?

Always, yeah. Some other robot hands goes all the way up to 60 degrees.

Yeah. If you If you touch the 60° wall, it's really uncomfortable.

Okay, so now we're going to doing test with the Sharp hand. Sharp hand is like twice more expensive.

No, four times.

Four times, so it's really really really expensive.

It's like $50,000 and Uchi 2 is $13,000.

per hand Yeah, 50 per pair, well, 30 per hand.

I I chose to compare Uchi 2 with Sharp hand because these two hands are the best hands that can accomplish most of the tasks that we have set. This

hand is too big. The the one thing about the human scale comparability is if it's too long and too big and the the composition is different, it's harder

for retargeting.

Which means it's hard to teleoperate and hard to collect data and hard to use at actual work sites for robots.

Compared to Uchi 2, this task is a lot easier for Sharp hand cuz the thumb is long longer.

Sharp Sharp hand definitely has more power in it and Uchi 2.

But the thumb retargeting is bit awkward compared to that. And the back drivability, I mean it it really depends on how you control it, but I feel like

it's a lot more stiff, so I am bit concerned about its durability because of the lack of the back drivability. Let's say this is more

drivability. Let's say this is more superior in every aspect compared to Uchi 2, but if it only lasts like two three days what what can you really do with it,

right? But overall feeling is just that

right? But overall feeling is just that I think Sharp hand is more like well well made.

Yeah, it's well made.

I think the quality control is really good good on Sharp hand way because if you actually break it down and see under the cover, I think it's more mechanically reasonable.

Like the gear teeth module and the size is more reasonable in Sherpa Wave. I

think like like the safety factor in the oof Uji 2 is like bare barely over one.

Oh, yeah.

I'm not going to do Can you borrow me here?

No, I'm kidding.

I think Uji maybe violated some some some kind of safety measures in terms of mechanical strength. But

mechanical strength. But if you take consider the vector ability, they will be approximately the same because this is less vector able, so it's more

vulnerable to impact.

Sherpa has its own advantage in terms of its grasping force compared to Uji 2.

Sherpa is better at picking stuff from the ground. Like one example would be like thin stuff. Like this kind of nut, M6 nut.

Cuz it's so strong it can easily grab them.

But Uji 2, not so much.

Yeah.

But like why would you need to pick stuff from flat surface?

What stuff a place that the robots will be deployed?

It's not for picking stuff from flat surface but like using tools.

Yeah, yeah. Or like using tool. Yeah,

yeah.

Okay, so first one is price.

Yeah.

Is Sherpa worth its 50k price?

What do you think?

I'm sorry, Sherpa.

Uh I think that's no.

Probably not.

So, why?

Huh?

It's just too expensive. If if you were to place a robot at a work to replace a human being, deployed It says too expensive.

too expensive.

Yeah. If we buy both hand, that's all already 100k. And for Uji 2,

already 100k. And for Uji 2, Yeah.

it is worth it?

Worth it. It's just really remarkable to pull this kind of uh, back-drivability force hit management with that price 10.

If we were to make it here, you can throw so much money at it, but still you can't make it. Machining all the warm gears and spur gear, winding the motors

in the coil, doing the cable management, like even doing this kind of silicon fixture it's all really just incredible engineering.

I think 13k is not not that bad at all.

But, like one thing that makes me concerned is it's durability.

I durability We have tested its back-drivability and its grasping force and everything, but we haven't tested its durability.

Yeah, I know.

Yeah, so we can't technically say say, yeah.

Yeah, yet.

Yet. But,

we're still testing the durability. So, we actually finished the testing and we put this Uji hand to work. So, we will see about the true durability.

So, back-drivability and and ROM ROM Uji 2 win.

Yeah. And the the thumb Uh, the thumb. Yeah. The thumb

configuration It's just too hard to control this thing without my hand.

Mhm.

Yeah.

Uji 2 is way way way better. Like

way better. Like Can you do this?

[laughter] ROM is also, yeah.

I mean, sharp If you If you push it to the bare limit, you can mimic it, but like you can't really compare it's back-drivability.

So, back-drivability and and ROM.

ROM, which is two wins.

Heat management. Oh.

Oh, yeah.

It's a Sharp pal wins.

Sharp pal wins, for sure.

Yeah, because it have fan fan in it.

Yeah, the fan in it. I think the fan really helps it everything cool down.

Yeah, yeah. I don't know why Uzi doesn't put a fan in it.

Maybe it's too thin.

Too thin?

There's no no space to put the fan in it.

I mean, you can just put it here and make it like two or three centimeters longer. They just didn't they didn't

longer. They just didn't they didn't want to do it.

Maybe they thought it wasn't necessary.

Heat management, Sharp pal wins.

Uh surprisingly, you'd think that uh if Uzi 2 has better vector ability, it would have uh worse repeatability compared to Sharp

pal, but no.

Like, both of them are very very precise. Uh I was very surprised cuz

precise. Uh I was very surprised cuz uh when I tested it, the three standard deviation uh like that covers 99.97%

of the whole try was only like 43 micrometers, which is like 0.043 millimeters. This is pretty good.

millimeters. This is pretty good.

Yeah. So, in terms of repeatability or the precision you say, um both of them are at very best.

So, 43 micrometers for the Uzi and 25 micrometer for the Sharp pal.

They are still Sharp pal is a little bit better, but that's also yeah yeah yeah negligible.

Yeah, yeah, yeah.

And also, the palm silicone pad.

Oh, yeah.

Yeah, yeah. So,

as you saw, because Sharp pal is really flat and hard, the phone just slipped.

Yeah. Yeah. And it's

you just don't want to put your valuable stuff on cuz you feel like it will break it. But for Uchi, it seems more safe.

it. But for Uchi, it seems more safe.

It feels like just putting a phone on a flat surface or just Uchi 2 because it shaped the silicone pads really well. It

like just naturally holds it there.

Yeah, yeah. No.

For teleoperating, the thumb ROM is a lot more similar to ours than to Sharp's. So, I felt a lot easier when

to Sharp's. So, I felt a lot easier when controlling or testing Uchi 2.

I feel that all the pilots will say the same thing.

Yeah.

That that's pretty much it for our standards or testing metrics.

The reason why we are testing all so many different robot hands is to make our foundation model to support various dextrous robot hands. Many other

companies are doing it with grippers, but we are choosing to stick with dextrous hands so we can solve the most difficult problem very at the very last

mile. Like completely duplicating what

mile. Like completely duplicating what humans do. And don't forget our uh

humans do. And don't forget our uh all hands up page. In that page, we are we uploaded almost like 20 robot hands URDF and their pros and cons and why we

are focusing on dextrous hands.

And you can play with the URDF files on it as well.

Yeah. So, check

all hands up page. So, stay tuned.

Yes, stay tuned.

Mhm.