Nvidia-Manager treten selten gemeinsam mit Wang Xingxing und Wang He auf und besprechen eingehend die "Schwächen" und die "Geschäftsmöglichkeiten" von Humanoiden Robotern.

According to a report by Zhidx on August 12th, during the 2025 World Robot Conference, Rev Lebaredian, the vice president of NVIDIA Omniverse and simulation technology, Wang Xingxing, the founder, CEO, and CTO of Unitree Technology, and Wang He, an assistant professor at Peking University, the founder and CTO of Galaxy Universal, and a scholar at BAAI, jointly provided detailed answers to a series of key topics at a small - scale press conference. These topics included the difficulties in large - scale commercialization of humanoid robots, key challenges, technological bottlenecks and trends, the necessity of humanoid and bipedal designs, the advantages of Chinese enterprises in physical AI and robotics, and the application scenarios expected to be popularized first.

NVIDIA is currently the most comprehensive provider of robot infrastructure. Its chips and platforms are widely used in robot enterprises. It is also collaborating with Alibaba Cloud, Beijing Humanoid Robot Innovation Center, Fourier Intelligence, Accelerate Evolution, Ubtech Robotics, Galaxy Universal, Unitree Technology, and Zhipu Robotics to promote China's development in humanoid robots and AI solutions.

Unitree Technology is one of the most well - known robot unicorns in China. It launched its IPO on July 18th this year and has deployed NVIDIA's full - stack robot technology for its new humanoid robot R1. The R1 is priced at 39,900 yuan and is available in stock. Mass production is expected to be completed before the end of the year. Unitree Technology has also self - developed a dexterous hand and hopes to achieve natural interaction in the next one to two years. The technical goal of its combat robot is to generate any action in real - time.

As a representative unicorn in the field of general - purpose robots with embodied large models, Galaxy Universal has completed financing of over 2 billion yuan in the two years since its establishment. The Galaxy Universal G1 Premium is one of the first humanoid robots equipped with NVIDIA Jetson Thor, using a wheeled chassis. Its next - generation humanoid robot project will adopt a pure bipedal design.

Zhidx has sorted out 15 informative Q&A sessions and shared the key layouts and progress of NVIDIA, Unitree Technology, and Galaxy Universal in the robot industry later in the article.

01. The Difficulty in Commercializing Humanoid Robots Isn't Cost; Overcoming Speed Challenges is Key for Factory Deployment

1. What are the current difficulties in promoting the large - scale commercialization of humanoid robots?

Wang Xingxing: For the commercialization of humanoid robots, cost and hardware are not the key issues. In a sense, as long as a machine can be used, even if it costs 100,000 or 1 million yuan, there are still many scenarios where it can be put to use. In the past one or two years, the hardware has been sufficient. Of course, improvements are needed in terms of reliability and cost, but it is not a limiting factor.

The most pressing issue currently is that the embodied intelligence models lack generality, and their practicality needs to be greatly improved.

2. What are the key technological bottlenecks restricting the large - scale deployment of humanoid robots?

Wang He: It's simple - the robots' work capabilities are not strong enough, and the types of tasks they can complete are limited. However, if a very general level can be achieved within these limited skills, many scenarios can be empowered at once.

The main breakthroughs of Galaxy Universal at present are in grasping and moving. As long as the robot can grasp any object, move its lower limbs and extend its upper limbs in the scene, and finally place the object accurately, many application scenarios can be realized.

Behind this, a truly accurate target recognition and positioning system is required. Currently, we are promoting this technology through synthetic data.

Of course, even if this key problem is solved, there are still many tasks that robots cannot complete for the time being. But once the target recognition and positioning problem is overcome, the humanoid robot market will be at least in the scale of hundreds of billions, and visible results can be achieved within five years.

After solving this key technological bottleneck, with such a huge market investment, robots will surely unlock more skills and move towards a trillion - dollar market.

3. What key challenges need to be solved for humanoid robots to be truly put into work in factory workshops?

Wang He: This year, many humanoid robots have started training in automobile factories. We have noticed that when most companies promote humanoid robots in factories, they mainly focus on two aspects: one is handling, and the other is sorting.

In terms of handling, in the recent robot video shown by Galaxy Universal, the handling speed has approached the human level. Calculated, the number of items handled per hour is comparable to that of humans. At this stage, it is very close to actual factory deployment. I expect that by the end of this year, dozens of Galaxy Universal robots may be put into actual use in factory workshops.

However, handling is just the first step. In addition to handling, the closed - loop ability of palletizing also needs to be achieved. Only when both handling and palletizing form a closed - loop can the robot truly handle the entire work process. Otherwise, doing only half of the tasks will not yield ideal results.

Sorting is an even greater challenge and has not yet reached the level of human workers. It still requires a certain period of technological iteration and breakthrough.

Whether it is taking items from a conveyor belt or from a shelf, the current biggest difficulty is speed. Skilled workers can pick up items very quickly. Robots currently struggle to achieve this efficiency in terms of both models and hardware.

When we developed retail robots, taking items from shelves or desktops is technically similar to industrial sorting. However, the requirements for rhythm in retail are lower, and the consequences of picking up the wrong item are less severe. In an industrial scenario, such as an automobile manufacturing plant, a one - minute shutdown of a production line may mean a loss of tens of thousands of yuan. Therefore, the requirements for the accuracy and speed of sorting are extremely high.

4. How to address the challenges of energy consumption, thermal management, and volume limitations in physical AI (especially in the field of robotics)?

Rev Lebaredian: Each generation of NVIDIA products significantly improves performance per watt and performance per dollar. Relying solely on Moore's Law is no longer sufficient to solve many of the problems we face. We foresee that the effectiveness of Moore's Law in CPUs and general - purpose computers will gradually come to an end. Therefore, we are committed to building specialized computers for specific algorithms.

This kind of specialized computer requires not only optimization at the chip level but also overall optimization at the algorithm, software, and application levels to achieve maximum performance. This is not achieved by a single factor, such as making the chip smaller or faster, but through full - stack optimization.

This is a very difficult engineering task and is also the core competitiveness of NVIDIA.

We initially applied this approach to computer graphics rendering (especially in the game industry) and then extended it to other fields. After the launch of CUDA, we began to apply it to physical simulation. Later, with the rise of deep learning and AI on GPUs, we continuously specialized our processors. Each generation of products has achieved significant performance leaps under the same power consumption and cost, and this will continue in the future because our innovation journey is far from over.

02. Is it Necessary to Design Robots as "Humanoid" and "Bipedal"? The Scale of Future Humanoid Robots will Exceed the Total Output Value of Industrial Manipulators

5. Is a "humanoid" robot necessary?

Wang He: In the long run, humanoid robots must be able to integrate into human life. From an end - game perspective, if a robot can work, reach up to 1 - 2 meters high with its hands, and move around in our environment, there is no other form except the human form.

In the next few years, humanoid robots will gradually evolve from mobile composite robots towards the end - game state.

If it is a fixed - point robot, the tasks it can perform are limited to what is in front of it, which is highly restrictive. Therefore, mobility is inevitable. A mobile cart can only carry goods and cannot perform any operations. The robots we are building today are equipped with a mobile platform, which can be raised, lowered, and folded, and are equipped with two mechanical arms and hands to carry boxes.

I believe that in the next ten years, the proportion of humanoid robots will not be small, but it depends on what it is compared with. The global total output value of large industrial manipulators is only about 10 billion yuan, which is not high. A leading automobile manufacturer can sell cars worth 10 billion yuan in a year.

I expect that the output value of humanoid robots will multiply by 10 every three years. Currently, the leading companies sell 1,000 units. Three years later, it will be 10,000 units, and another three years later, it will be 100,000 units. If 100,000 units are sold at a price of several hundred thousand yuan each, the output value of working robots will reach 10 billion yuan, exceeding that of a single company and even exceeding the total output value of the entire industrial manipulator industry.

In the next ten years, we will witness a robot market that surpasses the current scale of all industrial robots. In the following ten years, it may become a trillion - dollar market that surpasses the scale of the automobile and mobile phone markets. Therefore, we should not underestimate it, but it won't reach the scale of the automobile market overnight.

Wang Xingxing: In a sense, I regard humanoid robots as an important carrier of general - purpose robots. When real general - purpose AI performs tasks, it inevitably relies on robots, especially general - purpose robots.

Currently, the humanoid robot is the most ideal form of professional robots. Although it seems complex, in essence, its structure is composed of several joint motors connected in series, so the overall design is relatively simple. In contrast, tracked carts or other forms of robots are more complex.

I have always believed that when general - purpose AI matures on a large scale, everyone will be able to easily build a humanoid robot, just as people can buy computer components to assemble a computer today. In the future, if AI is powerful enough, the requirements for hardware will become lower and lower.

6. Why design robots with two legs?

Wang Xingxing: The real question is why not design them with two legs. Because having two legs is relatively convenient, and most importantly, it provides more general - purpose capabilities. In a sense, locomotion ability is a relatively weak form of AI ability. For example, small animals, even ants and insects, can walk very well, but their AI ability is weak. So I have always thought that for a truly general - purpose and work - capable embodied AI model, the mobility or leg ability is actually secondary.

If a robot can work, then the control of its legs will naturally not be poor. If it can't even control its legs well, it means it hasn't reached the stage of a very general - purpose AI model as people imagine.

Additionally, since two - legged design is relatively simple and our company specializes in leg technology, it is a logical and interesting direction for us, and it is also widely favored by people. Moreover, if everyone uses wheeled chassis, it will lead to homogeneous competition, which is unnecessary.

I think there are differences between wheeled and legged designs. At different time points, the lower - body solutions for robots will vary.

Our company focuses on leg technology and hopes to improve the overall movement and work capabilities of robots. I also work on wheeled chassis. Currently, wheeled chassis are very stable and energy - efficient in industrial open spaces and when moving between shelves, but they may not be able to pass through complex environments. If the chassis is made smaller, stability will be lost.

I firmly believe that legs are the future because they can cover all reachable spaces of the upper body and can flexibly adjust the flexibility of the waist.

However, at different stages, there will be the most suitable form for actual applications, and we will not be limited to a single solution. We are simultaneously conducting research on lower - body and even full - body control using wheeled chassis and Unitree's humanoid robots.

03. China has Unique Advantages in Physical AI and Robotics

7. What are your views on the demand and challenges in practice for China's physical AI and robotics?

Rev Lebaredian: China is both an important market and a production base for AI technologies and products. In the field of physical AI and robotics, China has unique scale advantages, combined with talent advantages, forming a unique ecosystem.

Robots are the bridge that brings computing and artificial intelligence into the real world. China is the best place to achieve this leap because it has unique conditions:

Firstly, top - notch AI talent: China has leading AI universities and a large number of intelligent, well - educated, and enthusiastic AI researchers and developers. Nearly half of the world's top AI talents are concentrated here.

Secondly, electronic and computing technology capabilities: China not only has strong R & D capabilities but also has an unparalleled global electronics manufacturing industry. China has profound expertise in manufacturing electronic hardware and key components required for robots. Such an ecosystem and manufacturing scale are difficult for other countries to match.

Thirdly, a large - scale manufacturing foundation: There are real scenarios for large - scale deployment and testing of robots, where data can be quickly collected and algorithms can be iterated to enable continuous robot evolution.

This enables companies like Galaxy Universal and Unitree Technology to mass - produce robots, learn, and iterate rapidly. China's unique comprehensive conditions provide a solid foundation for the rapid development of the physical AI and robotics industries.

04. The Biggest Problem with Household Robots is not Technology; the Number of Robots is not Directly Related to the Fertility Rate

8. In which scenarios will robots be widely popularized in the future?

Wang Xingxing: In the future, robots will definitely become more practical. Whether in the industrial, service, or household sectors, the overall time cycle will be longer, especially in the household sector. The biggest problem with household robots is not technology but the extremely high requirements in terms of ethics and safety, which significantly raise the threshold for their popularization.

The maturity of each industry requires a long cycle. More than a decade ago, people also thought that new - energy vehicles would develop rapidly, but it took a long time for them to mature.

The new technologies we are using now are completely different from those ten or twenty years ago. Both hardware and software have undergone tremendous changes. However, when many people talk about robots, they like to refer to the situation ten or twenty years ago and think that the industry has been developing for a long time.

Actually, the technology of the new generation of humanoid or general - purpose robots has only been developing for two or three years and needs more time. However, judging from the current development speed, I personally feel that the industry is still growing rapidly, and it is very likely that the number of personnel and the shipment volume will double every year in the next few years.

Based on this, if more powerful and general - purpose large AI models emerge in the future, these models will enable robots to perform better in more general scenarios such as factories and households. The more general - purpose they are, the easier it will be to popularize them. On the contrary, if they are not general - purpose, promotion will be more difficult.

9. Since robots can replace many jobs, will the fertility rate decrease?

Rev Lebaredian: I don't think people's decision to have children is directly related to the number of robots. However, one thing is certain: whenever the population of a society grows, the GDP and productivity will increase. The output capacity of a society is directly related to its population size.

Currently, the population decline trend faced by most countries will lead to non - growth or even shrinkage of the economy. If no measures are taken, our economy will decline.

Therefore, it is an urgent task to build robots to create an "artificial population" to assist in various tasks and improve productivity, not only to maintain the current productivity level but also to promote the further development of society.

05. Challenges in Robot Computing Power, Models, and Data: Improving Energy Efficiency, Unifying and Aligning Architectures, and Supplementing Multimodal Data

10. What are the differences between NVIDIA Jetson Thor and the previous Jets