Name: Embodied AI, Oh My!
Uploaded: 2025-12-02
Duration: 17 min 27 s
Description: Traditional robots are usually pre-programmed with specific, repetitive tasks in controlled environments. The new wave of robots are embodied with AI to perceive, learn, and adapt autonomously through physical interaction with the unpredictable, real world. A chat with the company that is innovatively mass-producing these new humanoids, one of which just topped Fortune China’s Best Design List.

Traditional robots are usually pre-programmed with specific, repetitive tasks in controlled environments. The new wave of robots are embodied with AI to perceive, learn, and adapt autonomously through physical interaction with the unpredictable, real world. A chat with the  company that is innovatively mass-producing these new humanoids, one of which just topped Fortune China’s Best Design List.

Embodied AI, Oh My!

Conferences

So both Autoflight and Everheart are on this year's China Fortune China best design list. Uh, a list we've been, um, publishing for 4 consecutive years. Uh, also on this year's list was a humanoid robot. Uh, from the our next company AGIO. This new wave of robots features machines no longer just pre-programmed with repetitive tasks. They are embodied with AI to learn and adapt in the real unpredictable world. Here to tell us more is AGIbot's chief marketing officer, Heng Qiu, in conversation with Clay Chandler. Please welcome them. Uh, thank you so much for joining us and, and welcome. Uh, this is a real treat. We've heard from companies large and small today. Yours is a relatively new startup. Tell us about AGIBott and when it was founded and where it's based, and how long it's been created. OK, so, uh, AGI bought a new company built in 2023, so less than 3 years, but we develop very, very fast. We think now we are already one of the leading companies in this area. Just as our company name, AGI bought means our target is to realize artificial general intelligence. Right, so in other words, these are robots that don't, you don't have to sit behind them with a remote control and steer them and show them where to go, but can, uh, navigate themselves, or do something like work in a warehouse, take a box off the shelf, bring it back to a truck, something like that, right? So you've got a bunch of different robot products. Can you tell us about the robot product line and how many different types of robots you're, you're manufacturing? OK, so in fact we have 3 series of humanoid robots. Uh, the first one we call it a series. This is a full-size humanoid robot. The, the height is 1. 3 m and the second series is called X series. It's a half size robot around 1.3 m. So it's a child height and the third one is a wheeled robot with a wheel. So move easily in the factory. So we have three series of humanoid robot and also we have dog robot do yeah, yeah, uh, really interesting, so, um. So the, the one that's kind of almost full size, that's the the A series, right? And the most sophisticated, most advanced one is the A20, that is, um, we have A2 X2 and the G2. These three series for yeah, for, for, for different scenario, yeah, and we were, we were chatting a little bit backstage about capabilities, but um. And you were saying that that the technology is really advancing very swiftly, that it changes every quarter, you'll have another series of robots out next year. What, what's changing? Tell, tell us what's progressing. Is it AI where the real innovation is, or is it, uh, is it something else? Uh, I think AI contribute a lot to the progress of humanoid robot. So for, for us, for AGI port, um. Um, strategy is one human, uh, one robot body plus 3 intelligence. You you understand for the hardware, we need the robot body and for the intelligence we need three types of intelligence. locomotion means for the movement, interaction, for talking, for gesture, and so on. And thirdly, the manipulation intelligence. So combine all these three kinds of intelligence, robot can really work, can provide. Productivity. So we combine all these together. Maybe that's something um special for us because not every robot company can do all this, can do all those things. Um, I mean the location challenge is an interesting one because it requires sensors responding to the visual signal, interpreting that, creating a map in its algorithmic brain, whatever that is, and, and then. And then responding to that, right, navigating the space that it it senses. Um, that seems really hard to me. OK, but I don't think that's the hardest part for a robot. Really, what's the hardest part? Um, just now we talk about like a locomotion interaction and manipulation. So for locomotion, maybe now you can see many company can do that dance, jump, and so on. You can see all kinds of videos on the internet, right? But for interaction. I think the level may be less than the locomotion level check the whole industry. For example, if only you make a conversation, maybe it's not so difficult, but you don't want to only talk with the radio. You want it to have a gesture, has a, OK, just like a human, so that makes things more difficult. Then talk about manipulation. You need to work as good as a human or even better than human. That's even more challenging. So I think manipulation intelligence is the hardest one in action, the second locomotion, maybe right now it's the easiest part. So I read a thing um in the news the other day. And, and maybe you can talk a little bit about this. It was about um Your robots, um, uh, I was an A2 robot that took a 3 day journey from Suzhou to the bound in Shanghai, yeah, more than 100 kilometers continuous working. Yeah, tell us, tell us about that. Was that hard? Um, yes, in fact, that's not easy, um. I want to talk about several parts. So first, uh, for the stability, you know, even for a human work more than 100 kg, you will feel very tired and maybe in the beginning, no problem, but at the end, maybe you feel uneasy or even. For, for, for now, maybe, but for a robots our robot, no problem. And also on the way you need to judge where to Avo is an obstacle, right, because you know so long distance, there are many, many different kinds of environment and difficult situation he needed to handle. And also, uh, since when you walk such a long distance, you need to work at a daytime and at nighttime for different scenario for outdoor scenario, it can handle. So in fact it's not easy. Yeah, it's not easy at all and the and the terrain, this was on public streets, right? It was on roadsides and sidewalks and had to navigate intersections and It sounds very, uh, very complicated. How about the battery for that longer trip? Uh, we change the battery, um, in a hot swap way because when you, it has inner battery, right, when you change the the the big one, it can still stand there. So you need them to do some uh complex operation, just a stand change and then work again. I say it's a kind of battery battery swapping, um, I've, I've seen some really interesting videos of the A2 and it, um, you know, you talk about manipulation, the, uh, and motion, um. Apparently these things can do even very, I mean many things that we think nothing of that we think is very simple tasks as humans are very difficult for humanoid robots like for example, I've seen videos of it picking up an egg and knowing exactly the lightness of grip necessary to keep from crushing that uh egg, but how challenging is something like that. Um, OK, so during this kind of process, I think you need several key technology behind. One of course is the intelligence. It can grasp the, the egg and also we need the dexterous hand can have the sensor to feel how heavy. And what is the material of this egg because if it's very heavy, he needed to grasp tightly. If it's not so heavy can Not so tightly, so don't crush the egg. So both the sensing the the judgment and then manipulation needed to needed to be combined together. So in fact, of course, if only for egg, it can be optimized. But since our target is AGI, we are not for egg, right? We are not for Right, uh, bottle for everything we need to handle all these issues based on a lot of data based on big model, a model to handle all this. Yeah. And how will you train these robots to, to move, to do tasks? they train by watching humans perform tasks, or how will it, how will the training happen? Yeah, in fact, we have a robot school for each data collection center. And what is special for us is we combine three kinds of learning technology. The first one we call it imitation learning means there's a teacher hand by hand, remote control the robot to do something. So in this way, the robot imitate how people do this kind of a task. So this is for the imitation. Imitation learning accumulate high quality data for the robot to do something. Then with this data we can train the big model. But of course, the data volume is not enough because if you want a lot of data, you need a very long time and high cost. So we have the second one. The world model class seem too real. The the the methodology. So it means we build the world model to do the simulation. So with the just now imitation model learning, we get a lot of high quality data. With this data, we build a model in the world model simulation. And then we do millions of times the simulation to get enough data to trade the big model. Then the big model can work. Then we have the 3rd learning scenario. We have the Real world Reinforcement learning means the robot with a bigger model do. Try, make a mistake, adjust this scenario. Then during this process, he just like an internship student, as this work side do make some mistake, then change, adjust, then better, then adjust better. Then During not a long time, it can work perfect. So combine these three types of learning methodology, our robot can evolve very fast. Yeah. Fascinating. So you've got this, uh, robot university in Shanghai, you're training the robots, um, maybe only school, university, maybe to that. Um, when they, when they graduate, what kind of, uh, what kind of jobs do you envision them taking? What kind of sectors do you think humanoid robots will be most, most valuable? Yeah, in fact, now we have 8 scenarios that our robot can work and can get a salary. OK, means that he feels proud that this job is well done. I can get a salary, OK. 8 scenarios. The first one is for the uh reception. Like uh for the 48 uh automobile car shop, they should do the reception, do the guidance, and so on. That's the first task. And second is for commercial recreation, like a dance, sing and so on. And the third one is for intelligence uh uh manufacturing for some like um. Uh, consumer electronic production line. And the fourth one is for the logistic sorting. The package come, they turn the package to show the barcode to the scan camera to show to Beijing to Shanghai and so on. And the 5th is the security patrol like the dog. Uh, to see whether there's a fee for something like that and for internal, uh, for indoor scenario, we have the robot to check the data center, whether this battery is overheat or something like that. And the 6th 1 is for commercial cleaning. We have a cleaning machine for the big mall and something like that do the cleaning. And the 7th 1 is from the 7th and the 8th 1 for university. The first is the 7th 1 is the data connection for the big model training and the 8th 1 is for the education and research. They we use our robot for how to learn how to build a robot, how to train the robot, something like that. Pretty sizable set of markets. So can we meet an A2? Can we bring one out on stage? Yeah, yeah, sure, of course. OK. We have one waiting to join us in the wings here. How do we invite him out to the stage? There he is. He's about, you can see there's a heart on its screen. A heart, yes, there is. Does he have a name? Uh, yeah, his name is A2. Hello everyone. My name is AGI. I'm really delighted to be here in Macau. This is my first time attending a fortune event and I'm so happy to meet all of you. I hope you all have a wonderful day. Now, I'd like to present a short performance for you. Please enjoy this Tai Chi demonstration. There he is. Well, not quite Bruce Lee. Thanks. Have a nice day. Uh, I can see the audience like a robot better than us because he come they take photos, but they don't take photos for us, yeah. All right, yes, there he's he's leaving us. Navigates pretty well. Good sense of balance. So this is the same one for the more than 100 kg to Shanghai walk, yes. Wow, uh, that's fantastic. Um, really, uh, impressive, and we look forward to, uh, seeing future iterations, uh, of your, of your human robots. So, thank you very much for joining us. Yeah, thank you very much. The proudest thing I want to emphasize is that our robot is really working now, is getting cellular now, not like many, maybe other cases they are just. Shooting the videos. It's actually up and running. Yeah, great. Hangzhou, thank you so much. Thank you very much. Thank you.