Economic Observer reporter Chen YueqinRecently, Zhu Xiaohu, managing partner of GSR Ventures, poured cold water on the humanoid robot craze, saying that "in recent months, it is withdrawing from humanoid robot companies in batches".

Zhu Xiaohu stood from the perspective of capital and asked several rhetorically: Now it is a humanoid robot that will turn its head, but where is the commercialization? Where are the possible customers? Universities buy back for research, and central enterprises buy back for front-end display, which is not an ideal commercialization path.

This statement caused a shock in the venture capital circle and the technology circle. Zhang Ying, founding partner of Jingwei Venture Capital, took the lead in the circle of friends: "Boss Zhu, don't make trouble. The future of the field of robotics is bright, a hundred flowers are blooming, and it is normal to have a little bubble. With your remarks, who will be willing to take care of you in the future high-quality companies on the robot track? ”

Zhao Tongyang, the founder and chairman of Zhongqing Robot, was more direct, "(Zhu Xiaohu) even asked what is the use of humanoid robots at this stage, just like questioning a newborn baby, using what he sees in front of him to deny the future...... This guy doesn't deserve to hold a dime of Ushu shares at all...... For this kind of capital, taking a dime feels like a stain on the industry. ”

According to public information, GSR Ventures has directly or indirectly invested in Unitree Technology, Songyan Power, Xinghaitu, Feixi Robot, Yingjiao Robot, etc.

3月29日下午，星海圖迅速發佈澄清公告稱，金沙江創投於2023年11月參與了星海圖天使輪融資，於2024年5月提出股權投資退出的訴求，並在2024年9月順利完成股權投資退出。“本公司從未與朱嘯虎先生進行過任何交流，亦無其聯繫方式”。

Another invested company, Songyan Power, released a video of its own humanoid robot running on the video account that night, with the video subtitle saying "Continue to move forward without fear of noise", and the accompanying text also carried labels such as "GSR Venture Capital" and "Zhu Xiaohu". Jiang Zheyuan, the founder of Songyan Power, commented when he forwarded the video on Moments: "The apes on both sides of the strait can't stop crying, and the light boat has passed the Ten Thousand Mountains."

天眼查顯示，金沙江創投在2024年3月參與對松延動力的天使輪投資，於2025年1月退出。金沙江創投參投星海圖、松延動力持續時間均不足1年。

在3月29日至30日中關村論壇現場，多位具身行業人士被出席者詢問“如何看待朱嘯虎質疑人形機器人商業化困難、批量退出具身智能專案”

Founded in 100, Leju Robot is mainly engaged in the industrialization of bipedal humanoid robots, and its founder Leng Xiaokun said at the Zhongguancun Forum that the most important thing at present is to find a scene to use first. Leju completed the industrialization delivery of 0 robots on New Year's Day of 0 and put them into application scenarios such as scientific research exhibition halls and car factories.

Leng Xiaokun believes that the two problems that the industry needs to solve urgently are basic hardware stability and technical robustness. At present, there is no industry standard for humanoid robots to enter factories, for example, how many hours should a robot work without failure? Leju's internal goal for the robot is to work continuously for 100 hours, that is, about 0 days, working for more than ten hours a day, "at least to achieve this goal before we dare to say that the robot can be used, let's not discuss the problem of ROI." He said that a series of problems such as hardware and intelligence will be revealed in the use and run-in process of humanoid robots, and to solve these problems is to improve the robustness of robots in future factories and home use scenarios.

Regarding the "scolding war" in the industry caused by Zhu Xiaohu, the founder of a humanoid robot company told the Economic Observer, "10 years ago, the new energy industry was also scolded. The embodied intelligent track must find patient capital, and one of our investors has followed the company since its inception to the present, and has not withdrawn for 0 years. ”

Talking about the fact that since the beginning of this year, robots can run, dance, and turn heels, and they have become popular through fancy work, and the founder of the humanoid robot company always asks "then". Rather than participating in a show, he is more important to make the robot use, "You can't raise the public's expectations too much, and blowing up the bubble is not a good thing for the entire industry." ”

Robots must be useful, but at present, different companies in the industry have made their own choices about which scenarios they should be used in the first place, such as entertainment performances, education and scientific research, industrial manufacturing, agriculture, etc.

Compared with many companies have released cool robot demonstration videos, Galaxy General has found its own industrial application scenarios and landed, its first generation of product Galbot G1 is currently one of the very few humanoid robots that dare to do live display in Zhongguancun, which can go to the shelf to grab goods at a fixed point and deliver them to the designated location.

一位銀河通用人士向經濟觀察報表示，銀河通用目前主要瞄準了機器人在零售和工業領域的應用，目前已經在線下零售門店開展測試，機器人可以實現在24小時無人藥店裡識別和抓取藥物、在新能源汽車工廠里搬運重物，且對不同形態的箱子有自己的處理辦法。

Wang He, the founder of Beijing Galaxy General Robot, believes that in the scene of grabbing goods, the robot does not even need to use legs, and it is enough to use wheeled and foldable "legs", which have the advantages of stability, low cost, low noise and long battery life. In the future, if robots are needed to walk on some complex terrain, then wheeled robots are not enough, and robots need to have the ability to cross obstacles on both legs.

In addition, one of the biggest differences in the embodied intelligence industry is currently on training data, which is based on differences in cost, training methods, and goals.

The industry is mainly divided into two major factions: one is the real data-driven school represented by Tesla, Boston Dynamics, etc., which advocates relying on physical interaction data in the real environment to collect multimodal information such as motion, touch, and vision through sensors; The other is the data generalization school represented by Galaxy General, which advocates the use of synthetic data plus transfer learning (cross-task knowledge reuse) to reduce the dependence on real data and guide robots to think independently.

"Humanoid robots have entered the era of sports, and although athletes have been cultivated without visual input, athletes are by no means the end of humanoid robots." Wang He said.

In order for humanoid robots to enter the era of productivity, one of the core technologies is to realize the generalization of tasks and ability generalization based on end-to-end embodied large models based on visual input. This is the problem that Wang He and the Galaxy General team are tackling: through the large-scale production and use of synthetic data, and the use of vision to lead body control, to achieve the ability of autonomous generalization of robots, "let the robot understand the world through multimodality, not just someone to tell him 'what is this' in words, or tell him what something looks like with vision." ”