XPENG stellt seinen neuesten humanoiden Roboter IRON vor – mit bionischen Muskeln, Festkörperbatterie und beeindruckender Rechenleistung. In diesem Video zeigen wir alle Details zu Technik, Design und dem ambitionierten Zeitplan für die Serienfertigung ab 2026.
😇 Dein Abo hilft uns: https://tublo.eu/abonnieren
✅ Source: XPeng
➡️ Mehr Infos: https://www.tuningblog.eu/tipps_tuev-dekra-u-co/xpeng-iron-743027/
Der IRON ist mehr als ein humanoider Roboter – er ist XPENGs Vision der verschmelzenden Zukunft von Mobilität, Robotik und KI. Präsentiert auf dem AI Day 2025 in Guangzhou, basiert IRON auf einem hochmodernen Vision-Language-Action (VLA)-Modell, betrieben von XPENGs eigenen Turing-KI-Chips. Mit 2.250 TOPS Rechenleistung, 22 Bewegungsfreiheitsgraden (DOF) in den Händen, einem sensorisch aktiven Soft-Skin-Körper und einer vollintegrierten Festkörperbatterie zielt XPENG auf eine menschennahe, natürliche Bewegungs- und Interaktionsfähigkeit ab.
Der Fokus liegt klar auf kommerzieller Nutzung – etwa in Serviceumgebungen – bevor private Haushalte angesteuert werden. Ein zentrales Merkmal ist das modulare Design: Körperform und Funktion lassen sich anpassen, um verschiedenen Einsatzbereichen gerecht zu werden. Die Kombination aus bionischer Wirbelsäule, muskulösem Aktuatorensystem und lernfähiger KI zeigt deutlich: IRON ist nicht nur ein technisches Konzept, sondern ein nächster Schritt auf dem Weg zur embodied intelligence – KI in physischer Form.
Im globalen Wettbewerb humanoider Roboter will XPENG mit IRON einen neuen Standard setzen – sowohl in Technologie als auch in Integration mit anderen XPENG-Systemen wie EVs und Fluggeräten.
#XPENG #Robotik #HumanoidRobot #KI #XPENGIRON #AI2025 #FutureOfMobility #EmbodiedAI #tuningblog - das Magazin für Auto-Tuning und Mobilität!
😇 Dein Abo hilft uns: https://tublo.eu/abonnieren
✅ Source: XPeng
➡️ Mehr Infos: https://www.tuningblog.eu/tipps_tuev-dekra-u-co/xpeng-iron-743027/
Der IRON ist mehr als ein humanoider Roboter – er ist XPENGs Vision der verschmelzenden Zukunft von Mobilität, Robotik und KI. Präsentiert auf dem AI Day 2025 in Guangzhou, basiert IRON auf einem hochmodernen Vision-Language-Action (VLA)-Modell, betrieben von XPENGs eigenen Turing-KI-Chips. Mit 2.250 TOPS Rechenleistung, 22 Bewegungsfreiheitsgraden (DOF) in den Händen, einem sensorisch aktiven Soft-Skin-Körper und einer vollintegrierten Festkörperbatterie zielt XPENG auf eine menschennahe, natürliche Bewegungs- und Interaktionsfähigkeit ab.
Der Fokus liegt klar auf kommerzieller Nutzung – etwa in Serviceumgebungen – bevor private Haushalte angesteuert werden. Ein zentrales Merkmal ist das modulare Design: Körperform und Funktion lassen sich anpassen, um verschiedenen Einsatzbereichen gerecht zu werden. Die Kombination aus bionischer Wirbelsäule, muskulösem Aktuatorensystem und lernfähiger KI zeigt deutlich: IRON ist nicht nur ein technisches Konzept, sondern ein nächster Schritt auf dem Weg zur embodied intelligence – KI in physischer Form.
Im globalen Wettbewerb humanoider Roboter will XPENG mit IRON einen neuen Standard setzen – sowohl in Technologie als auch in Integration mit anderen XPENG-Systemen wie EVs und Fluggeräten.
#XPENG #Robotik #HumanoidRobot #KI #XPENGIRON #AI2025 #FutureOfMobility #EmbodiedAI #tuningblog - das Magazin für Auto-Tuning und Mobilität!
Kategorie
🚗
MotorTranskript
00:00Musik
00:30Musik
00:34Actually, I have mixed feelings now, because over the past seven years, in order to make the walking very lightly and very elegantly, we did a lot of work, and we try to mass-produce Aaron as well. Thank you, Aaron. You can go to have a rest.
00:52This is a very difficult process, because today, Aaron is still under the R&D stage, but we plan next April, we are going to enter the mass production preparation stage with hardware and software.
01:20Today, the iron you see very gently and also working like the model.
01:34A lot of robots in the industry, you can hear the noise of working very loud, and also the impact on the floor is very big.
01:45In the past, the robots will work like this, but we are actually always want to preparing for the mass production.
01:54During the rehearsal, some of the colleagues were saying that, is there a real human inside the iron? No, actually not.
02:02Next, I want to, first of all, tell you guys that we are already starting the mass production preparation of our humanoid robots.
02:23We hope it can be lower than 1.7 meters, also can be lighter, can be more good looking and safer for our customers.
02:31We will also enhance the reliability of our mass production robots. This is what we are doing.
02:36Next, I want to use a video.
03:06This is a video.
03:16This is all muted.
03:18Just this isnal box.
03:23i found it on the advisory board Au Patientlaw
03:30Ich habe ihn gesagt, dass er eine intelligenter Mensch ist.
04:00That's why it's more like intelligent human has the intelligence to create a better life with human beings.
04:09That's why the next-gen iron has very flexible bones and also very solid muscle, bionic, and also very soft skin.
04:19We hope it can also have a similar height as human beings, inside out, human-like.
04:25That's why we designed the human-like spine for the next-gen iron.
04:30It will be able to actually go down, bent, like human beings.
04:36I cannot touch the floor, so I will not show you, but iron can.
04:40And also around the lumbar area, we can also have flexible movements.
04:45No matter if it's lying, standing, getting up, doing some easy gestures and movements, all of them can achieve on the next-gen iron.
04:58In addition, we also have bionic muscles on iron.
05:03It can be more like human with different body shapes.
05:07You can choose a little bit fatter iron, or like me, a slimmer iron.
05:13Or you can customize your iron based on your preferences.
05:20We also offer you the full-coverage soft skin, so the robot is warmer and also more intimate.
05:31In the future, it can go to the households.
05:33This skin touches very softly, and on this new-generation iron, we also create some touch sensors.
05:44Different areas, like the hands, they can have interaction using the detectors.
05:49With humanoid bones, muscles, and skins, we can say that iron can have different body shapes and sexes.
06:01As you can see on the slide, we have two male robots and two female robots.
06:09I suspect that just like you buy the car, you can choose different colors, exterior, interior.
06:19In the future, when you buy the robot, you can choose sex, and you can choose the hair longer or not, or the clothes, what kind of purpose.
06:29So do you want to achieve this?
06:30If you want, then let's see some hypothesis.
06:40If iron wears different clothing, it will become actually people from different industries.
06:48I believe in the future, we have different robots.
06:51We have human for sure, and also we have iron, this kind of intelligent human.
06:57What's more, for our next-gen iron, it also has the head-mounted 3D curved display.
07:06A lot of sensors inside, including the visual sensors.
07:10We have the cameras, we also have the millimeter wave radar.
07:17We also have a lot of very sophisticated components for the head, for the face.
07:24You can have different expressions.
07:26With the screen, the display here, you can see the emotions of iron.
07:31We are also developing based on other kind of features, like ear, and also the mouth, and so on.
07:40So it's actually the integration of different capabilities.
07:44We also developed the bionic dynamic shoulders for our next-gen iron.
07:51So our humanoid robot can stretch their shoulder like human beings, and they can achieve the same level of flexibility.
08:00And next point is the most difficult one, dexterous hands.
08:05This actually is very difficult to mass-produce, because for the hands, not too big, very small.
08:14But we need to be more like human hands.
08:19For human hands, actually, it's the most sophisticated organ.
08:24For iron, we developed 22 joint flexibilities for a single hand.
08:34And also, we developed the smallest frequency regarding the joints.
08:41So single hand can achieve 22 different degrees of flexibility.
08:47And for iron, the fingers can actually support a very small item, as you can see on the slide.
08:56In addition, the working of iron, no matter the gesture or the dancing or the sitting gesture, very gentle and also very soft.
09:10We add the flexibility at the tipping point of the feet, just like the models working on the stage.
09:23We also actually adopted the first all-solid-state battery in the industry on iron.
09:30A lot of people are asking, are you talking about the all-solid battery or semi-solid-state battery?
09:41And why you don't use it on the cars?
09:44Because for robots, it's only less than 2 kilowatt-hour battery.
09:48So I think humanoid robots will be the opportunity for us to mass-produce the solid-state battery,
09:56because for humanoid robots, the safety requirements will be more stringent.
10:00For cars, we are facing the roads, the all-doors, but for our robots, we use it in office, in shopping malls, in your households.
10:12So we need to achieve the top safety standards.
10:15In addition, we use three Turing AI chips on the next-gen iron with the maximum computing power of 2,250 tops.
10:25I believe this is the most powerful robot in the industry, and very possibly next year will be the smartest robot,
10:36because using the computing power that I mentioned, we actually combine three large models, VLT plus VLA plus VLM.
10:50So as we all know, VLA actually comes from the cars, VLM is also from cars.
10:58But we add the new VLT large model.
11:02This large model is the concept and also is the model first created by Xpong.
11:09It's also today the first time for us to announce VLT large model will be the core engine for robot autonomous action.
11:18It's actually like the brain of our humanoid robots for their decision-making, especially in the real world.
11:27For instance, how the body is going to move, how the hands are going to adjust, and how they are going to take different actions.
11:36This VLT large model is extremely valuable.
11:40Also, the VLA big model will also be applied on our Xpong next-gen iron, but we need to say that it will be much more difficult,
11:54because on the car we have one engine going forward, going backward, going left, and going right.
12:00But for a robot, we have 82 joints.
12:09So you can think about actually tons of combinations, from hands to feet to shoulder to lumbar to head.
12:19A lot and a lot of gestures will be a challenge for us, but we firmly believe that this VLA large model will be extremely useful for the body control and have unified intelligence.
12:36Also, Xpong, last year we have established the first embodied intelligence data factory in Guangzhou.
12:45We believe no matter your computing power is high or not, data will be the key.
12:51How can we receive massive data over time?
12:56I think this is very important for our robots to be more generalized.
13:03Here, let me allow to make it still confidential, because in the future I will disclose more and more to you guys how Xpong can use the massive data to support the mass production of our robots.
13:18Regarding our iron, we also want to mention about the three laws of robotics, because now we extend these three laws of robotics.
13:29As we all know, the first law is that robots cannot harm humans, cannot let human beings get harmed because of no action.
13:41The second law is that robots need to obey the instructions of humans unless this instruction is in conflict with the first law.
13:49The third law is that robots need to protect its self-existence unless it is in conflict with the first and the second law.
13:57Now, for our company, we are thinking the fourth law that robots cannot disclose any human privacy because they have ears, they have eyes to hear and also to see everything around you.
14:12This kind of active safety protection will be actually very important for our company to develop for our customers.
14:21We all know that in cars, we have AEB feature. Through rules and algorithms, we can protect the occupants.
14:33But for robots, can we have similar feature? Maybe not every one of you knows the complicated process of robot development.
14:42Sometimes it will go crazy dancing. Suddenly, you can think about 70 kilograms robot. If he kicks you, then what you can do?
14:52That's why we need to have this kind of active safety protection without harming people. This is crucial.
14:59We can see that very frequently, including in our company, we have shoot this kind of video.
15:06We can kick the robot and the robot can still be balanced. Is this safe? No. Because during the movement, if you kick him, if the floor is not flat or your pet is on the floor, then the robot will actually step on the pet or step on the pit on the road.
15:27That's why we need to think about how can we protect actually the pets, the people, the conditions around the robot. All of this are belonging to our working scope regarding the iron development.
15:40Next is about commercialization. Many may concern that this is your eighth generation of robot and when can you actually make it a commercial product and mass produce it?
15:55It is something that many would ask. Actually, two years ago, we've been contemplating about it and we would like to share some of our insights and observations.
16:05First, should we introduce our robot into factory to tighten the screw?
16:10It is actually the easiest settings we believe robots can perform. But after one year of trial, we found that it's not suitable. Why the case? Because first, the most difficult part of the robot, we believe, is hand.
16:28And when you perform the work, like tightening the screw, usually it would just take one month for the robot to have its hand wear. Not to mention it has two hands and the cost for it that is very costly.
16:44Not to mention the manufacturing skills between Chinese and American workers are different. Therefore, it's very cost effective to hire Chinese workers in our factory.
16:55That's why we believe it is not the best way to introduce the robot into factory from both commercialization, from both the technology and usability.
17:06Therefore, having the robot to tighten the screws is not our priority. It may come later, but not for now. So that's why we cancel it.
17:17Then how about the household chores? We often run into videos, and I'm also excited about the vision because I'm kind of a lazy person. For all those household chores, I would rarely perform myself.
17:33Can I have a robot to actually do that for me? But still, we believe that it is still not viable. First, it's because of safety.
17:43As I've said, for a family, you usually have about 100 square meters area, not to mention all of these obstacles, the items that you stacked in your house. It's very difficult for your robot to not get tripped over or just run into your furniture, the pads.
18:08I believe no single companies and robotics can actually ensure safety in the household. That's why I think safety would be the greatest challenge to hinder that development.
18:20The second is generalization. Because you have different household settings. How can you actually ensure the robot can learn your settings and perform accordingly? I think it's very difficult based on our current technology.
18:35That's why we decided to pause it for now, even though we may choose to deploy it in our household in the long term. That's why, after all of these considerations, we decided that we want to prioritize our commercial settings.
18:55The first, we expect that it would be deployed in three kinds of scenarios. When I was in the mobile internet industry, we often say about world protections, guarantee, and I think it's similar in robots. We often followed the same kind of rules.
19:19Like for robots, we believe that it can perform work like tour guide, shopping guide, and even reception in the company. So these three kinds of capabilities is very likely to be deployed in the real world.
19:33For our next-gen iron. For our next-gen iron, we hope that it will not be remotely controlled. Instead, it can be instructed via natural language. But of course, we need to undergo some basic training, instructing it how to better guide the people.
19:50You need to have those materials. You need to have those materials that input into the system so that this robot can actually introduce themselves.
20:02And next year, our robots are very likely to show up in our expo stores to introduce our product. You will probably see our robots introducing our product. But sorry, they can only introduce the car.
20:17They cannot actually test drive with you. Not to mention, it will probably be to lead your guide in our manufacturing base. In the coming future, this may generate a new kind of occupation.
20:34Occupation that we will require people to teach, instruct, and manage the robot. For example, one manager in charge of a group of seven robots,
20:46that can perform the work that dozens of people can do. This would be a dedicated occupation probably in the future.
20:56Next is about mass production. Because we really take it seriously.
21:03And after our development of our next-gen iron, many of our labors would be actually concentrated into our mass production,
21:17helping that we can achieve the kind of capability, making sure that our robot can see, listen, understand. Even though it may perform slowly, but it can ensure that it's operable.
21:29And we often say that it's more difficult to mass-produce the robot than to build a robot ourselves. And we actually took many detours.
21:39And we would like to share some of our experiences with you on this AI day.
21:44First, in the past, when I actually got into the auto industry 11 years ago, we believed that car is actually hardware-defined.
21:58And to think about the software, how to better be compatible with the hardware.
22:03Therefore, the software providers can actually assist the car OEMs. However, it's totally different in robotics, because it is a software-defined hardware.
22:19You can see that our iron, they actually learn how to walk through the reinforcement learning.
22:25Take some simple steps, because at that time, it can only do so based on that hardware and the software.
22:34But if you want to power a robot using the new model, due to the different architecture and different software, the hardware may probably need to adjust it accordingly.
22:47And in this field, well, to be bluntly, many of the car OEMs, the suppliers or the partners, they may have found that a lot of the software providers in the car manufacturing industry can hardly perform very well in the robotic industry.
23:12Second, people would ask about whether the robotics, what is the standard for the robotic quality.
23:22And from our perspective, we believe that the quality of robots will likely go beyond that of the stringent automotive grade.
23:31Because we also work on the flying cars, we believe that these two would be, well, very rigid, more rigid than the car.
23:40For example, our battery designed, well, is more stringent than that of the car.
23:46We worry about catching fire in the household or buildings, not to mention the robot has 82 joints.
23:56So, imagine a car, if the engine fails, it cannot drive.
24:01But there are 82 joints on the robot. If one fails, well, the hole may collapse.
24:07Oftentimes, we will have situations like a joint fell and the electricity will break down, the robot will just fall over.
24:16That's why the robot would have very high requirement for safety, much higher than that of the car.
24:24And for perception design and domain controller designed, we actually benchmark against the car and even higher than that.
24:34The third lessons we learned is that we find that most of the car, the car makers, they would like integrations and innovation to do this at both the same time.
24:47But for robotic manufacturers, it's more about fusion, fusion and innovation.
24:53How can it better feed the software and integrate to make it coordinate?
24:59Because the eyes, the hands, the legs of the robot need to be coordinated.
25:04These are powered by two different systems. It's very difficult.
25:09That's why we required full fusion.
25:13Fusion and innovation is actually the foundation for robotic development, and this is a great challenge.
25:22I believe that for the future robotic manufacturer, many of them would prefer the full-stack self-development.
25:30And for those who choose for integration, well, it may be very difficult for them to venture into this field.
25:38That's why we believe that mass-producing the advanced humanoids is as difficult as building a robot taxi with an immature hardware and a software supply chain, just like back in 2015.
25:53That's why I would take the lead to build a robotic team.
25:57And so far we have 10 R&D teams working on the robotic development.
26:03So apart from the robotic centers, we have other departments that would also work on the R&D.
26:12Not to mention we have more than 20 partnering departments, over 1,000 people that will support our overall development.
26:20And again, thank you for listening to our technology projects.
26:25The flying car is one that must bring people higher rise by telling women our customers that are interested.
26:36This is something to learn who lives in an organization andジャ examination in the room be prepared for their own 선택er,
26:41Untertitelung des ZDF, 2020
27:11Für mehr Videos einfach abonnieren!
Kommentare