Inside Wuji's Human-Like Robot Hand

[00:00:00]So, this is the motion capture glove by Wuji [music] hand.

[00:00:05]Amazing.

[00:00:12]>> [music] >> You're not allowed to do that.

[00:00:18]>> [laughter] >> He can do it.

[00:00:25]That's a good idea.

[00:00:28]>> [music] >> Okay. Pulling out one card is not easy.

[00:00:48]Yeah.

[00:00:48]>> This our first generation hand. [music] We did a lot of bionic design for it, so it can show what whatever a human teach the model do.

[00:00:58]>> Right.

[00:00:59]>> And we are going to launch our next generation hand very soon.

[00:01:03]>> Okay.

[00:01:04]>> already see some picture in our website.

[00:01:06]>> Mhm.

[00:01:07]>> Yeah. And believe it, the next one will be more durable and the stable for many task.

[00:01:13]>> We want to >> make the human dexterity hands that to be a general contact [music] in the near future.

[00:01:20]>> I think you are going for back-drivability and compliance.

[00:01:23]>> Yes.

[00:01:23]>> So, what's like what's the philosophy behind that? Or how is it advantageous versus >> Yeah.

[00:01:29]>> other type of hand.

[00:01:30]>> Yeah.

[00:01:30]>> For human hand, our finger can back drive, right? [music] >> Mhm.

[00:01:33]>> So, in this case, when we touch something or when we falling down, we will not hurt ourselves.

[00:01:39]But for the first generation one, >> Mhm.

[00:01:42]>> it can be hurt when it touch some hard thing or the robot itself falling down.

[00:01:47]>> Right.

[00:01:48]>> So, we needed the back drive to make the it can [music] >> more robust.

[00:01:53]>> The first thing.

[00:01:54]>> Okay.

[00:01:54]>> The second thing is when it can back drive, the torque transparency will be very good.

[00:01:59]>> I see. our sensor can detect the torque the change of the torque. So, you can get a 20 more sensor in your in your hand. That will be good.

[00:02:08]>> I see.

[00:02:09]>> Yeah.

[00:02:09]>> And is it because of the low gear ratio of the motors?

[00:02:14]>> Yeah.

[00:02:14]>> Uh you get better transparency.

[00:02:16]>> Yeah. Yeah. We got a very low uh gear ratio.

[00:02:19]And then the first generation.

[00:02:21]>> I see. Uh so, in terms of uh strategy, like what kind of uh applications, what kind of use cases uh you're trying to develop?

[00:02:31]>> Uh we want to you make this hand as powerful as a human hand. Mhm. So, its application is our customer's thing.

[00:02:38]They design their model and they to apply to different application. And we just uh support them to achieve that.

[00:02:44]Yeah.

[00:02:45]>> Okay.

[00:02:46]It's It's customer's imagination.

[00:02:48]>> Yeah.

[00:02:48]>> They can do whatever with it.

[00:02:49]>> Yeah. Yeah. Sometime we got a lot of feedback when they do the different things, they meet some problem, ask us to evolving our hand. I think that's a very good loop to improve this kind of product.

[00:03:02]>> And with the hands, I hear they break a lot or they require constant maintenance. And then uh you you don't get enough maintenance from the supplier.

[00:03:14]>> Yeah.

[00:03:14]>> So, how how are you going to face that challenge?

[00:03:18]>> Yeah.

[00:03:18]>> For Wuji.

[00:03:18]>> I think we will establish a new warehouse in the North America very soon.

[00:03:24]>> Okay.

[00:03:24]>> And we also find some good distributor in the Europe. I think we are building up our service network in this year.

[00:03:33]>> Okay. Okay.

[00:03:34]>> Yeah. Because maintenance is very good, as you say. Very important. Yeah.

[00:03:38]>> Very very important, right? Okay.

[00:03:40]I see a lot of researchers that are adopting and using Wuji hand.

[00:03:44]>> Yeah.

[00:03:44]>> And you're partnering with them.

[00:03:46]>> Yeah.

[00:03:46]>> Yeah.

[00:03:49]>> Yeah, you can shake hand with this one.

[00:03:54]Yes, a little bit warm, but it's okay.

[00:03:56]>> Yeah, yeah, yeah, it's warm.

[00:03:58]>> [laughter] >> Yeah, very warm hand.

[00:04:06]>> It's fascinating how this motor here is driving this moment.

[00:04:11]>> Yeah.

[00:04:11]>> All right, so the motor is here, right?

[00:04:13]>> The motor is inside this this it's following my finger. Yeah, yeah.

[00:04:17]>> Yeah, we use a kind of warm gear to make it can yeah, move this way.

[00:04:22]>> Okay, okay.

[00:04:24]>> Uh you know, Uji team made a very sensational like way to having like the actual motor directly inside the finger. Do you know some stories behind that? Like how did that happen?

[00:04:36]>> Yeah, yeah. Actually, our our CEO, Mr. Pan, has a very fascinating dream to how to get to the embody AI. Uh but he started from the motors and we build very strong motor. [music] I think for now they are still one kind of the have the highest uh torque density in the market. Yeah, and then we switch to build the human texture hand in 2023.

[00:05:05]Yeah.

[00:05:06]>> Cuz nobody had tried like putting motor directly in the finger before, right? It's in a >> I think people want to, but there are a lot of engineering issue making that thing not that possible.

[00:05:18]>> Yeah.

[00:05:18]>> For us, we also meet those problem, very different kinds of problem, but uh for now I think we fixed maybe 50% [music] of them in our first generation hand and then in our second I think we can finish maybe 75% and we'll leave the rest to the next generation. But for this one it's already very good. You can see the back drive is the >> Wow.

[00:05:40]>> different joint have a its own back drive.

[00:05:44]So it can detect the torque very sensitive and it can also be like this.

[00:05:49]>> So it's very delicate.

[00:05:50]>> Yeah.

[00:05:51]>> Wow.

[00:05:52]>> Nothing stopping it from moving.

[00:05:54]>> Yeah.

[00:05:55]So you don't need to make a collision or it's falling down. Yeah, it can protect itself. I think it's very durable.

[00:06:03]>> Oh, it's like a human's hand.

[00:06:04]>> Yeah. Yeah.

[00:06:07]>> Yeah. You can shake hand or you can For human hand is about 400 to 500 g.

[00:06:19]>> Right.

[00:06:19]>> This this one is about 600 g.

[00:06:22]>> Yeah, and the shape is also human like like because the motors are not in the palm.

[00:06:26]>> Yeah. It's kind of Yeah.

[00:06:27]>> The distribution of the weight is similar to human hand, right?

[00:06:32]>> You can use my hand to compare the size.

[00:06:34]My hand, right?

[00:06:36]From this way I can cover this one. So the size is the same. Yeah.

[00:06:40]>> The proportions are quite similar. Yes.

[00:06:42]Yes.

[00:06:42]>> Quite similar. Uh we chose the color in black is because we also have our glove in black, right? So I believe in people use glove to collecting data like like what this gentleman's do. Uh the model would record some black hand in manipulation data, right? So we think with our real hand to do the autonomous uh task, they can also in the same shape >> Mhm. [music] >> and the same color and everything will be the same.

[00:07:09]>> I see.

[00:07:10]Does it also extension?

[00:07:12]Extension?

[00:07:13]>> Yeah.

[00:07:13]>> Okay.

[00:07:14]Which is very hard to do with tendon tendon hands, right?

[00:07:18]>> Uh yeah. I think tendon hands have some have its own advantages, but in the long run I believe the industrial customer will yeah accept our choice.

[00:07:29]>> Okay.

[00:07:30]What's the price?

[00:07:30]>> For now it's 13,000 US dollar for one hand.

[00:07:36]Yeah.

[00:07:36]>> hand, not the pair.

[00:07:38]>> Yeah, two pairs two 26,000.

[00:07:42]Yeah, but but know is the the the the this market is just start. Yeah, just start from uh last year. Believe in the end every every every robotic can afford to uh [music] dexterous hand in the future.

[00:07:56]>> Okay.

[00:07:56]>> And uh since uh the body is only have about 20 or 226 degree of freedom.

[00:08:03]>> Mhm.

[00:08:03]>> But you know, single hand we have 20. A pair hand have 40.

[00:08:07]So, I think it's very fair a pair of hand can take uh about 50% of the bomb cost of a whole robot.

[00:08:15]Yeah.

[00:08:16]>> Mhm.

[00:08:16]>> That's that's our idea.

[00:08:18]>> Yeah. Yeah.

[00:08:19]>> And uh for this one there are no more heating overheating issues.

[00:08:24]>> It will be a lot of >> This one Did you touch this one?

[00:08:26]>> Yeah, for this one >> you can feel the hand temperature is a little bit higher. Yeah, for this one they can work all day and then you you will not think it's too hot.

[00:08:36]>> Yeah.

[00:08:36]>> It has more fat.

[00:08:38]>> [laughter] >> It has more fat.

[00:08:40]It's very squishy.

[00:08:42]>> Yeah. And then we will also update our glove to make uh it suit to this this size.

[00:08:48]So, uh in the end the human hand and our uh Ubi hand two can put in the same glove to collect data.

[00:08:56]>> Okay.

[00:08:57]>> And the glove can also used as the tactile sensor for the hand.

[00:09:01]>> Mhm.

[00:09:01]>> And that will be a very that will be a complete solution. Yeah. Yeah, we will ship this hand from uh late July.

[00:09:08]>> Okay.

[00:09:09]>> Mhm.

[00:09:09]>> Yeah, this is still not a mass production version. Only the AI model can control this kind of complex system.

[00:09:17]>> Mhm.

[00:09:17]>> So, I think uh people are buying this to [music] uh empower their AI to have a physical body, a physical hand >> [music] >> to finish their yeah, more com- complex uh manipulation task.

[00:09:31]>> What is the power consumption? I mean the energy, yeah.

[00:09:33]>> Uh I think since you have different movement it's uh mhm it be goes to from the 10 W to to 30 W.

[00:09:41]>> Okay.

[00:09:42]>> But since it has back drive, so we >> [music] >> we put the limit maybe it can stand for 100 W.

[00:09:49]>> Go up to 100 W.

[00:09:50]>> It can go up to 100 because sometime when you back drive, the how say put the torque trans back to the uh electricity.

[00:09:59]So it can stand for 100 W. But normally it will work under 30 W.

[00:10:03]>> I see. So normally gripping task is 30 W. Yeah, up to 30 W.

[00:10:08]>> Okay.

[00:10:09]>> Actually our normal everyday task is not that extensive. For picking something, putting something on, it's not consumption energy.

[00:10:21]But when you tapping on the keyboard, play a piano, that's a very challenging task.

[00:10:26]>> Right, right, right.

[00:10:27]>> Even for Dexter robotic hand.

[00:10:30]>> What's the upcoming road map for Uchi if you can share a little bit to us?

[00:10:36]>> Yeah. I think firstly I want to make this hand strong as possible and very affordable in next three or two five years. And the second we want to improve our data collecting system. We want it can collect the whole up body or the whole body and then make people can work wear it very easily, can collecting all the days. I think in this way since it's all product from Uchi, right?

[00:11:04]We can minimize the gap between the data side and the real physical side to make it a zero gap. In this way I think we'll be very friendly to those model supplier.

[00:11:15]>> Yeah.

[00:11:16]>> Want to be the hardware provider.

[00:11:19]>> I think >> the most comprehensive hardware provider. Yeah.

[00:11:24]>> Makes sense.