- 2 days ago
Category
🗞
NewsTranscript
00:00When you talk about this whole humanoid robot industry, and really China dominates, right,
00:04this report that came out last week that found that China's humanoid robot makers are commanding
00:09more than, look at this, this pie chart says it all, more than 97% of global shipments in the
00:14first half of 2026. So that's according to industry data from Smart Analytics Global.
00:19And according to their calculations, Shanghai-based Agibot had the top spot. Now the company
00:25had 44% of those global shipments, that is some 8,400 units in just the first half alone.
00:32One industry group also estimates that China installed 300,000 robots back in 2024, and
00:38that vastly outnumbers the around 38,000 machines that were installed in the U.S. last year.
00:45This one on your screen, taking it out right now, was being tested for its real-world abilities
00:50at last year's World Humanoid Robot Games in Beijing. I think we spoke to Jesse Liu,
00:55from LEMEX yesterday, right, about what is going to be the key theme at this conference
01:00this year in Beijing. He says it's really about moving some of these robots into the real world,
01:05into the open space environment.
01:08Physical AI, right?
01:09Yeah.
01:09It's not even close when you look at how far China's actually come on many of these fronts,
01:14and specifically, of course, when it comes to robots. And of course, really, the upgrades
01:20that we then need to see and should come to expect, I think, in terms of the capacities of robots.
01:26And I think also interesting is, I mean, people talk about this as a zero-sum game. To what extent
01:31do robots, will robots replace humans? I think that is a question anyway that has somewhat been
01:38answered, and not just this time, right? You look back at the course of how manufacturing has been
01:44automated the last 120 years. I mean, there has been this takeover of robots in an industrial
01:50side. And I think moving forward, I think in the physical and in the retail front, what will the
01:55relationship be between robots and humans? And by the way, that is not a philosophical question.
02:01That is more pragmatic and practical at this point in time.
02:04Let's bring in Kenneth Ren. He's the director of Overseas Solutions at Real Man Robotics. It's a
02:11Chinese-embodied AI firm that actually runs a humanoid robot data training center in Beijing.
02:17Kenneth, good morning, and thank you for joining us on the shows. Why don't we start with that?
02:23What is the question that you guys get from the industry and your community and your clients between,
02:30you know, the future relationship between robots and people?
02:34Let's start there.
02:37Okay. For the robot and the human workers, we think that the robot cannot replace the human
02:45workers, or maybe the human workers will not lose their jobs. Because why we develop or why we
02:51design the robot? We want the robot to really help people, help human workers do something in the
02:57real task, not replace the human workers. And for example, we think that the robot can
03:04help human workers to do what the people don't want to do or what the people cannot do, even
03:10in some dangerous scenarios. And to be honest, we think that in the near future, the robot
03:18cannot replace or cannot control the whole task independently with all the humans. So maybe we
03:26need to have human and robot collaborations. For example, the robot do some standardized task, and the
03:33human do some like the judgment and some high-risk decisions. And I think that is why we use the
03:41GLN or we use the teleoperation devices, because we think that the
03:45use teleoperation devices, we can control the robot anywhere remotely. And on the other hand, we think that, you know,
03:54when we use the teleoperation devices or when we use the robot,
03:57other roles will be created. For example, we need some human workers to help us to control the teleoperation devices,
04:05and we need someone to help us to do some data collection.
04:08So it means that the work will not be replaced. They just change the work from the, you know, the
04:16direct control, direct control to the maybe managing. So that is what we think that not replace the human workers.
04:27Yeah. Can you just walk me through what it looks like in these robot schools that you have? What does
04:36it take to train a robot for one thing? And I know it could be a pretty costly sort of
04:41operation, but I'm guessing you need a tremendous amount of training data. Is there enough of that training data out
04:48there now?
04:50Yes, you know, more and more company has launched the different type of the robot. So it means that the
04:57robot has maybe beyond from the lab or some simulation environment to the real world environment. So maybe first about
05:05the robot training platform, we think maybe next generation of the training platform will focus on for example, like factory,
05:13like warehouse, or even some service environment. And we don't want to
05:18teach the robot to do some one or two or three specific action. We want the robot maybe can identify
05:25even adapt to the different task, different environment, even some unexpected situations. And that's why we think there's maybe three
05:37steps or three stage of the robot training or robot maybe robot features. First is about to use some teleoperation
05:45devices to control the robot.
05:47For example, for example, put a bottle of water on the table or do some window cleaning. Although maybe it
05:55will use the five minutes or 10 minutes. But you know, we moved to the second stage that is where
06:01we will collect some data during this process. Or maybe we can know the details of each action. And we
06:06can continue to move the third stage that will train the mode of the robot by this newly collected data.
06:14So it means in the future, when we have more and more data. So it means in the future, when
06:18we have more and more data or when we have more and more details, the robot will be more and
06:21more quickly, maybe from the five minutes to three minutes to one minute, even 30 seconds.
06:26And we think the long-term goal is that the robot can plan on making some decisions by themselves. That
06:36is what we want.
06:39Oh, okay. So, I mean, riddle me this. So in a year's time, because you're very close to the data
06:47and what robots are capable of doing. And I think part of, I think what you guys do is upgrading
06:52that skill set.
06:53What would be the most advanced thing a robot you think would be capable of in a year's time?
07:03You mean what the robot will really do, right? In the near future.
07:08Yes. We think that, for example, do some highly standardized task or do some high repetitive task. For example, moving
07:17the box. And, you know, sometimes we have some problem. For example, like...
07:23Net shift demand. For example, like some labor shortage problem. I think that is what the robot can do in
07:31the near future, because they can really help humans to solve these problems, not to replace the human workers.
07:39Okay. Now, I want to get a sense of the scale of your business, given that this is a booming
07:46industry. It's also a policy priority in mainland China.
07:51Talk to us about your industry. It's your corner of the industry. What is competition like? How many robots go
07:57to your school? I know that sounds like a strange question.
07:59Just give viewers who are watching from the outside an idea of your world, Kenneth.
08:06Actually, in our robot school, we totally have 150 robots to do different tasks in the different scenarios. And we
08:14collect different data. We also, you know, our customers also, they want to define or they want to design the
08:22different, their own scenarios. And we also can help our customers to this task.
08:27And during this process, we can, we also use the teleoperation devices to control the robot, to collect each details
08:36and each data. And we think this is a good way. And this is maybe a quick way for now.
08:43Because, you know, the model, we need to give some time to the robot.
08:47We need to give some, maybe more money to the robot for the trainings. So that is why we choose
08:53the teleoperation devices.
08:57Kenneth, what do you think the goal is for, you know, training these robots in this school that you have,
09:03right? It is, I mean, I'm guessing you can't just keep training hundreds of thousands of robots every time. Is
09:10it, we had a guest yesterday talking about developing a robot brain, for example. Is that sort of the goal
09:17for the industry in some ways is to just, just download something that I can put on, on my robot?
09:22I don't have to keep training it.
09:24In some ways. And how far away are we from that sort of reality?
09:29Actually, you know, for now, we don't think that the robot can do the whole job, right? So we will
09:36separate each job. And, for example, for the, for the, maybe simple, or maybe for the easy task, we will
09:43teach the robot to do this.
09:45So we guess maybe in the, after one year or two years, the robot can help us to do this
09:50standardized task. And for some highly risk decisions, the human will do this job. So that is why we, we
09:57train the robot. So we need to step by step to, to help, to, to let the robot really help
10:03people to, you know, do some job in the real task.
10:10This is something that we, we always look at when we look at these robot videos, because maybe you can
10:13help us out.
10:15They're always a bit clumsy when it comes to doing these simple tasks, right? Whether it's making, you know, coffee,
10:23whatever, right?
10:26How do you work on the, the dexterity aspect of, of these robots? Just a curious question.
10:32Yeah.
10:35You mean how to, uh, let the robots do this job, right?
10:39Yeah. How do you make it more precise and more human-like? Just, yeah.
10:45Oh, so you, that is why we choose the, you know, we call the GLN, means the global link network.
10:52That is, uh, tele-operation devices with the same structure of the robot arms.
10:57And our controllers will control the robot arms for the, uh, of the devices, uh, tele-operation devices.
11:04And when the, when the people do the same actions, the robot will do the same actions.
11:09So that is why we choose this, this way, or why we use these devices to control the robot.
11:15Because we think, we, uh, you know, I always have a joke with our team that our tele-operation devices
11:23also have a mode.
11:24That is the best mode, the human, which has been developed for so many years.
11:28So, when the controllers do these two actions, the, the, the, maybe the robot will also do these actions.
11:34And we can get the, each details or maybe each data of, for example, like a joint, as a camera,
11:40and as a whole robot.
11:42So, we will get data, and, um, maybe in the future, when we get more data and to train the
11:48mode,
11:48and the, the robot can do this action by itself.
11:51and we will get data on to them as a result.
11:51And if this happens, we will see between each other one and two ones.
11:52One and two, two families will get data on to them as a result and
Comments