Nine humanoid robots rose simultaneously from the stage floor. There was a deliberately reserved empty space in the middle. Jensen Huang, the founder and CEO of NVIDIA with gray hair, was wearing his iconic black jacket. He quickly walked from the left side towards the reserved empty space, stood still, smiled, and then spread his arms.

At that moment, it seemed to me that Jensen Huang was trying to embrace the future. But in fact, he was just fully showcasing the human figure so as to make a comparison with the humanoid robots beside him. However, he then announced, "The intersection of computer graphics, physics, and artificial intelligence occurs at this very moment."

This is just one of the "great snapshots" of humanoid robots around the time when the "ChinaVenture 2½ In-depth Closed-door Salon on Robots" was held.

A few days earlier, Figure AI, a startup that had just received $675 million in investment from giants such as Microsoft, NVIDIA, and OpenAI, released a demo. In the video, the robot, which was connected to OpenAI's visual and language large - scale model, smoothly completed a series of tasks such as visual and language recognition and understanding, dialogue, and action execution.

The development abroad is in full swing, and the situation in China is no less impressive. Recently, prominent enterprises such as Unitree Robotics, Innoway Robotics, and Galaxy Robotics have successively completed several rounds of huge financing. Looking further back, UBTECH successfully listed on the Hong Kong Stock Exchange, becoming the first publicly traded company in the field of humanoid robots.

The sudden increase in the popularity of humanoid robots has made this salon 注定 to be a communication that is both unified and full of doubts and controversies. The unity lies in the fact that all the guests are on this track towards the future, and most of them firmly believe that "humanoid robots are the most suitable physical carriers in the AI era". However, there are full of doubts, controversies and even oppositions regarding issues such as the development speed, industrial impact, application scenarios of humanoid robots, and even whether they should be in the "humanoid" form.

Regarding the discussion on the stage of "humanoid robots: technological breakthrough or capital frenzy", everyone is very concerned and has well - founded arguments, which perhaps represents a certain kind of industry - wide anxiety. There are also more specific questions, including whether humanoid robot companies should develop their own motion large - scale models, how the gap in large - scale models between China and the United States will affect the development of humanoid robots, whether humanoid robots are worth investing in now, and how to invest in them.

Many questions will not have answers for a long time. And even if the discussions are intense, when it comes to the actual course of action, people may still choose to move forward out of inertia or simply fail to solve any problems. After all, it is human nature to resist the uncertainty of the future and long for order and understanding. However, at this moment, which is bound to be full of changes, the greatest significance of all these might be that, through the descriptions of the present reality and future prospects by various industry participants, people can adjust themselves, convince themselves to let go of anxiety, and focus on moving forward.

Technological Breakthrough or Capital Carnival

In fact, when the doubts were bluntly put forward, the originally somewhat stiff atmosphere of the salon venue seemed to be ignited all of a sudden.

Shi Jiadong, the founder and CEO of Maxonic Technology, has practical and straightforward reasons for his pessimism about humanoid robots. It is still too far from large-scale application. Firstly, humanoid robots can be an excellent carrier for technological research and development, but they are not necessarily the most optimal product form in many scenarios, especially in the industrial field. Secondly, although large language models have greatly promoted the development of humanoid robots in aspects such as language, recognition, and environmental understanding, they still cannot solve the "Moravec's paradox".

In the field of humanoid robots, there is much discussion about Moravec's paradox. This paradox shows that high - order intelligent abilities unique to humans, such as logical reasoning, actually require very little computing power, while unconscious skills, movements, and intuitions require extremely high computing power.

Cheng Hao, the CEO of Acceleration Evolution, originated from the Tsinghua Robot Football Team. After leaving ByteDance, he founded Acceleration Evolution and obtained angel round financing including that from Innoangel Fund. Entering the humanoid robot field at the most feverish moment, he gave reasons from the perspectives of the market and technology. "In fact, we had repeated discussions about whether to get involved. It was only when GPT 4.0 was released in early 2023 that we were certain that the opportunity had arrived. If the entrepreneurial team doesn't get involved at the crucial moment, there will be no chance for us later on."

Although in his own eyes there is an endgame - that humanoid robots present an opportunity for a productivity revolution, Cheng Hao believes that "it's not something that one company can handle from the underlying technology to the final application all by itself". Therefore, the goal that Acceleration Evolution focuses on in the first stage is to provide developers with a hardware development platform, systems, and tools.

Over the past few years, Capstone Partners has been keeping an eye on the robotics industry. Luc described the current market as "fragmented": "There is one group of people soaring in the sky, while another group is struggling upwards in the mud, and there is no visible middle path that can effectively combine cutting-edge technologies with industrial applications." The former represents the future, and the latter is the reality. Capstone Partners has invested in Interspace Dynamics, which focuses on cutting-edge technologies, and is also a shareholder of FulltimeRobotics, which is committed to industrial applications at the scene.

Investment emphasizes allocation, and in short, each investment has its own value. What Luc is worried about is when China's intelligent robotics industry can form a sound and virtuous development ecosystem. When compared with the US market, companies like Tesla are involved in various aspects such as the market, algorithms, data, and technologies, and they are capable of integrating and making the most of these elements on their own. Moreover, star startups like Figure AI also have shareholders such as Microsoft, OpenAI, and NVIDIA, which provide all-round support in terms of hardware, algorithms, products, and orders, thus forming a virtuous ecosystem where each party can leverage its own strengths. However, the domestic humanoid robot industry is still in its initial stage, and industry players do not yet have the awareness of cooperation and win-win results.

"Those soaring in the sky think that those struggling in the mud are too practical and even shortsighted; while those who are down-to-earth think that those in the sky are blindly optimistic and even naive. This has led to the fact that it is very difficult for us to find projects that can achieve a relatively good balance between technology and application."

Mou Min, the executive general manager of Yuexiu Industrial Fund, said that the cycle from the research and development of humanoid robots to their actual large - scale application is likely to be very long. Taking autonomous driving and its related industrial chain as an example, "some scenarios have been gradually implemented, but there is still a long way to go for Level 5". As the person in charge of the artificial intelligence track of Yuexiu Industrial Fund, Mou Min and his team have looked at many top - of - the - market projects in the field of intelligent robots. After long - term contact, some projects were ultimately not invested in because the core application scenarios were not clear. Therefore, the first thing to consider for humanoid robots is "what kind of rigid - demand problems to solve and what the core application scenarios are exactly."

However, Mou Min also believes that the current state of humanoid robots is still in the initial stage of industrial development, and it is far from being a "capital carnival". "It will truly be a carnival when many mainstream PE funds start investing in it."

Jia Jing, a partner of Delian Capital, is relatively optimistic, mainly stemming from her belief in technology. Due to the rapid development of large models, the entire industry is iterating rapidly, and an inflection point is approaching. "The large model for robots is very close to complete convergence in academic terms." However, in actual investment, Delian Capital is moving steadily step by step, and it can be said to be quite prudent. "All things need to return to the essence of business."

According to the development stages of technology and the market, Jia Jing divided the robot projects invested by Delian Capital into three generations: The earliest investment was in ROSHI Robot. The three founders all came from Poten Robot. At the beginning, it provided solutions of six-axis robots in the industrial field to replace manual labor. Currently, it is also focusing on collaborative robots for human-machine collaboration and has a large quantity of shipments. Later, it invested in Mecanum Vision Technology Co., Ltd. which is "equipped with a bit of vision". The key words are 3D camera + AI + robot. It has landed in the industrial field in a more intelligent way and expanded globally. Recently, in terms of applications in specific scenarios where the operation and execution results of robots are directly related to products, it has invested in Quanyu Industry, a polishing robot company.

Ivy, the managing director of Oasis Capital, and Wang Sheng, a partner of Innoangel Fund, both stand on the side of "extreme optimism", but the roots of their optimism as investors are not exactly the same.

Ivy takes a relatively macroscopic perspective and has a strong sense of belief. She has gone all in on AI and humanoid robots, that is, embodied intelligence. Oasis Capital has also made a large number of arrangements, including MiniMax, UnlimX, Interspace Dynamics, etc.

"Oasis firmly believes that AI is a revolutionary change on the scale of an industrial revolution, which will ultimately bring about social changes. Among them, the greatest opportunity we see lies in robots. History has proven that all paradigm revolutions are changes in interaction and terminals. And now, robots are the most suitable carriers for multimodality and for bearing the interaction end. We have committed our fund, invested our money, and dedicated our time, and we are moving forward together with the entrepreneurs of this era."

Wang Sheng also has faith in AI. Regarding humanoid robots, he believes that the progress of hardware such as core components is linear, while capital pursues exponential growth. The fundamental logic behind the "big bang" of humanoid robots is AI. However, his confidence in investment stems from the strategy and ability of investing in early-stage and small projects. "Innoangel can get in touch with all early-stage projects related to intelligence, and then be the first to invest in them."

Innoangel was among the first to invest in unicorns such as MegaRobo and WeRide, and the same is true for Acceleration Evolution. Wang Sheng said, "I'm afraid there are no projects in this wave of humanoid robot projects that haven't talked to us."

Therefore, Wang Sheng doesn't quite understand why Zhu Xiaohu doesn't invest in large models. "The large model project we invested in more than a year ago has earned dozens of times the return. There are plenty of projects that are like 'flying in the sky', and projects that are like'struggling in the mud' are also making money rapidly. As long as they can make money, why not invest?"

Was it unrealistic to make PCs in the 1970s just because one firmly believed in the Internet?

If you bring your vision closer, you will find that the challenges faced by humanoid robots cannot be ignored. This is not only the reality but also a shackle. When you pull your perspective back, the ticket to the future seems so elusive. And no matter from the perspective of the industry or capital, the core of the guests' debate on the prospects of humanoid robots boils down to two words - scenarios.

Ma Lin, a partner of SEE FUND, suggests that humanoid robot companies should try more application scenarios, flexibly adjust their technologies and products, and take the path of commercialization well. At the same time, with an open mindset and thinking, they may find new applications.

UBTECH was established in 2012 and officially listed on the Hong Kong Stock Exchange at the end of 2023. In early March, its stock price soared, with a single-day increase of more than 88%, and the highest stock price was more than three times the issue price. It is truly the first publicly traded humanoid robot company in China. However, Chen Jiapeng, the head of UBTECH's Global Strategic Investment Department, frankly stated that UBTECH is "the company with the most experience in struggling in the quagmire, having tried the extended application of humanoid robot technology in many scenarios."

Therefore, Chen Jiapeng advised everyone to "be more down-to-earth." Currently, UBTECH's revenue mainly depends on the extended application of the core technology of humanoid robots in scenarios such as artificial intelligence education, smart logistics, and consumer-grade robots. In the process of technological research and development and application promotion of humanoid robots, the extension of these scenarios is, first, for the sustainability of enterprise development, and second, to form a feedback and interaction with the technological research and development and data accumulation of humanoid robots.

This coincides with the view of Yu Jun, a partner of SenseTime Guoxiang Capital. As a representative of the AI 1.0 era, the industrial investment fund under SenseTime invested in the popular Galaxy Robotics at the beginning of last year. Yu Jun believes that robots in the 1.0 era solve specific problems in vertical scenarios, while Galaxy Robotics is a representative of the 2.0 era. One of the main differences between the two is the change in the team paradigm. The 1.0 era has a more mechanical and electromechanical industrial gene, while the 2.0 era has more backgrounds in algorithms, computers, mathematics, etc.

Therefore, the strategy of SenseTime Guoxiang is to "grasp both ends and extend towards the middle." "Both ends" refer to investing in the most outstanding teams that are deeply rooted in scenarios to complete data accumulation, thus minimizing product costs. At the same time, invest in the most cutting-edge embodied large models in China like Galaxy Robotics. The "middle" refers to those scenarios and industries that currently seem to have a high degree of uncertainty. "The penetration rate will increase, and the cost will continue to decrease. There is no need to worry at all."

In contrast, entrepreneurs who have already entered certain scenarios have more detailed feelings and confusions about the technological progress and applications of humanoid robots. There is no concept of right or wrong here, only experiences and judgments under different situations, strategies, and goals.
FulltimeRobotics is Li Zike's entrepreneurial project. Currently, it has many customers not only in China but also has achieved remarkable results in going global. It has received investments from institutions such as Capstone Partners. FulltimeRobotics focuses on construction robots, specifically on applying putty and latex paint to interior walls. Li Zike said that the complex working conditions and constantly interspersed processes at the construction site pose great challenges to the automated operation of construction robots. Compared with other scenarios, the automated construction of interior painting projects has a relatively high degree of standardization and will be the first category of robots to achieve a breakthrough.

Li Zike firmly believes that technology should serve the industry. "There are very few markets as large as the construction market, which reaches 30 trillion. If it can't solve problems, it's useless." Therefore, he is generally conservative about humanoid robots, and he looks for solutions more based on scenario problems, focusing on costs. "As long as the overall cost can be slightly improved compared to before, I think it is commercially mature."

More than twenty years ago, Zhang Jianzheng, the founder and CEO of Smartmore, started researching robots at Shanghai Jiao Tong University and had many exchanges with teams such as Musk's and Boston Dynamics'. Over the years, Zhang Jianzheng's experience is that for humanoid robots composed of traditional components and structural methods, it is difficult to increase the load ratio, and the cost is too high, so there are many limitations in actual scenarios.

Now, the direction determined by Smartmore is a mobile operation composite robot with coordinated "hands, eyes, feet, and brain," and the annual shipment volume has reached several hundred units. "Whoever can combine technology, scenarios, and the market will be the king." Taking the transportation of water cooling plates for new energy vehicles and semiconductor wafers as examples, Zhang Jianzheng pointed out the difficulty of robots deeply intervening in scenarios. Various algorithms such as visual recognition, spatial identification, automatic generation of movement methods, and precise micro-operation can only be continuously iterated, learned, and adaptively adjusted. "This is not a problem that can be solved by a single intelligent large model."

Poten Robot was established in 2010 and has rich industry experience. Deng Peng, the CMO, starting from the market perspective, pointed out that humanoid robots are currently at the critical point of explosion, which depends on the marginal cost of product promotion. "It can only explode rapidly when the marginal cost approaches 0 infinitely." But from a practical perspective, Deng Peng said that humanoid robots should still have specific application scenarios to continuously reduce costs and then "blossom in all aspects."

Deng Peng roughly divided scenarios into structured and unstructured scenarios according to their complexity. Taking welding as an example, he pointed out that vehicle welding itself is a simple structured scenario, but if welding a ship outdoors, it is completely different. The light and temperature are complex unstructured scenarios. Therefore, Deng Peng believes that humanoid robots will also make progress step by step, continuously breaking through from structured scenarios to unstructured scenarios.

Many entrepreneurs present agree with Deng Peng's view. Li Yuhao, the CEO of Myan Robotics, also believes that currently, service robots are only in the early stage of commercialization, and the penetration rate of related household products is still low, and humanoid robots are even further away. In the future, on the one hand, there will be more breakthroughs in structured scenarios, and on the other hand, there will also be opportunities in unstructured scenarios such as household cleaning robots.

Li Yuhao introduced the concept of "technological water level," that is, as technological capabilities increase, many technical problems and challenges, like one "mountain peak" after another, will be "flooded," and the corresponding difficulty will also become higher and higher. Therefore, Myan Robotics focuses on the research and development of commercial cleaning robots in terms of technology to achieve a balance between functionality and robustness, in the hope of expanding the relevant market. In terms of business, the core is to resolutely go global and seek profits overseas, starting from the fact that the cost of the domestic supply chain is still low.

Jinding Capital serves listed companies and is deeply rooted in the industry. Currently, it cooperates with about 50 listed companies, covering multiple fields. Naturally, it also pays special attention to the scenario implementation of humanoid robots, what pain points can be solved, and what kind of impetus the explosion of humanoid robots can bring to the upstream industry. Zhang Shouhe, a partner, believes that "at present, humanoid robots are still divorced from the physical world."

Therefore, Zhang Shouhe is still seeking the combination of technology and scenarios. Although industrial investment cannot be "groundless," he is quite confident about the development of humanoid robots. This confidence not only comes from his background in science and engineering, which gives him the belief in technological optimism but also, more importantly, from a historical perspective. Even though it is still a long way for humanoid robots to be widely applied on a large scale, the technological breakthroughs brought about during the research and development process can still have a positive impact on the upstream industry and society.

Among the investors and entrepreneurs present, Cheng Hao believes that excessive focus on scenarios is still the thinking mode of the era of specialized robots. "In the 1970s, if I had to figure out a scenario clearly, think of the Internet and Yahoo before making a PC? It's too difficult. This is predicting the future, and even if the prediction is correct, no one will believe it."

From the perspective of an onlooker, I understand and highly appreciate the diversity of these views and experiences. To put it more simply, whether it is looking up at the stars or being down-to-earth, they all truly exist and deserve respect. More importantly, express oneself fully based on these. People naturally admire the spirit of working hard and saving up for a big move, and they should also recognize the resources, rights, and long-term influence brought by words. As Xiang Biao said in "Return to the Subject: The Anthropological Imagination of Xiang Biao," the hype may cool down, but specific stories and views will remain in people's minds, gradually changing everyone's perception of daily life and giving rise to a new sense of life.

You're talking to me about the long term, but my guys here need to make a living.

Huang Renxun said that humanoid robots are "the intersection of computer graphics, physics, and artificial intelligence." Perhaps it may also include other basic disciplines, but the large artificial intelligence model is undoubtedly the foundation for the explosion of this wave of humanoid robots. The investors and entrepreneurs at the salon basically recognize the value of the large language model. However, when it comes to whether to enter the industry to work on large models or embodied intelligence, everyone has their own considerations.

At the beginning, Chen Jiapeng from UBTECH issued an invitation for discussion. He said that as a startup, Figure AI is supported by its shareholder OpenAI. It has raised $700 million in financing in less than two years since its establishment, and Tesla's R&D investment is even higher. The R&D investment of competitors is extremely large. Domestic enterprises need much larger investment and stronger industrial cooperation to achieve breakthroughs in embodied intelligence.

Mou Min from Yuexiu Industrial Fund also put forward relatively conservative suggestions. He believes that it is difficult for early-stage enterprises to balance cost reduction, efficiency improvement, and large-scale R&D investment. Therefore, in addition to having a clear positioning, the most important thing for early-stage companies is to figure out where the money comes from to maintain their viability and advantages. The current capital market has changed a lot. To achieve these, first, it is necessary to control the company's operating costs and save more money than before; second, it is necessary to grasp the financing rhythm and the growth rate of valuation. A high valuation may mean that it will be very difficult to raise money in the future.

"We still need to have confidence. In the past, in the field of artificial intelligence, companies like SenseTime, CloudWalk, and Cambricon are still making losses, but many institutions have still made money."

Deepblue Technology, an underwater robot company, has entered its 12th year of entrepreneurship. Wei Jiancang, the founder and CEO of the company, has a profound understanding of the survival and development issues of the enterprise. Currently, Deepblue Technology has two major business divisions: Deep Sea Hi-Tech and Underwater Intelligence. According to Wei Jiancang, Deepblue Technology has actually endured through difficult times. It was only when it acquired the ability to "solve major national needs" that it truly ushered in explosive growth. The company's revenue reached 300 million yuan last year.

Therefore, Deepblue Technology is currently in the stage of product iteration and cost reduction and efficiency improvement. Since the deep-sea environment cannot be directly reached by humans, has obvious unstructured characteristics, and communication is relatively difficult, it inherently requires AI to make autonomous decisions. Wei Jiancang said, "Just like humanoid robots, the industry has already invested a large amount of money in research. In the future, we hope to achieve a breakthrough as soon as possible, so that the AI capabilities of underwater robots can reach the level of the submarines in the movies."

Shi Jiadong from Maxonic Technology clearly stated that he is optimistic about the application of large models in robots. However, as a startup, there are not enough resources, and it is impossible to invest in basic large models. He hopes to be a "transporter" at the intermediate layer, "Move the open-source large models over, find suitable application scenarios, and implement them. I also hope that investment institutions will invest more in open-source large models."

Li Zike from FulltimeRobotics emphasizes responsibility and has a chivalrous spirit. "You're talking to me about the long term, but my guys here need to make a living. We can't bet our lives on the next round of financing, right? We have to make money." But sometimes survival and growth are a paradox. If you want to expand your business in more complex scenarios, it won't work if the technology research and development can't keep up. Therefore, it's not easy to find an outlet.

In response, Yu Jun from SenseTime Guoxiang gave a very direct suggestion - that is, to increase the ability to target VC. "SenseTime has many 'hammers' but doesn't know where the 'nails' are. If you already have the 'nails', according to the generalization ability of today's large models, many complex scenarios are no longer a problem."

Deng Peng from Poten Robot still put forward the idea of putting more pressure on VC from the perspective of cost. "Many big shots mention that there is no supply chain and no solution. In fact, many of them exist, but the cost is so high that many customers simply can't afford it." The explosion of the humanoid robot industry still depends on when the intersection curve of machine cost and labor cost arrives. In the early stage, it still depends on the investment of investment institutions.

Several other investors and entrepreneurs also shared their experiences and expressed their opinions on this issue. All of them started from practical tactics and also had a strategic height. No wonder Li Zike exclaimed that he "had a free strategic and financing course." For me, the professional content and views I gained from the whole salon are secondary. The key is that I felt a very strong upward spirit, which is more precious than gold at present.

This has nothing to do with being optimistic or pessimistic. Let's put it this way, my favorite science fiction novelist, Ted Chiang, has a short story about the Babylonian miners who risked their lives to build the Tower of Babel. There is a sentence that goes like this. When the miners reached the top of the tower, "Everything that could be understood by humans was in sight... The priests led everyone to pray to Jehovah, thanking him for allowing them to see so much; then they begged for his forgiveness because they still wanted to see more."

Who wouldn't want to see more scenery?