The Year 2025: The First Year of Humanoid Robots - "Carnival" in the Living Room, "Wealth" in the Workshop
In the past six months, capital, factories, the media, and the public have all set their sights on humanoid robots.
In China, Unitree Technology became an overnight sensation after its humanoid robots appeared on the Spring Festival Gala of the Year of the Snake. Wearing red cotton-padded jackets and holding red handkerchiefs, the humanoid robots "danced the yangge" on the big screens in the living rooms of national audiences. Since then, humanoid robots have witnessed an unprecedented "carnival."
Looking deeper into the industry, humanoid robots have had more and more opportunities to "make an appearance" in recent years. However, at present, the real "wealth-creating" scenarios for the entire industry are not in the living room but in the workshop. Now, the humanoid robots of leading domestic and foreign enterprises such as Ubtech, Tesla, and Figure have successively entered various workshops for practical training, and the commercialization seems to have reached the critical point of large-scale development.
The Two Sides of Humanoid Robots in 2025: "Carnival" in the Living Room, "Wealth" in the Workshop
In April this year, the world's first half-marathon for humanoid robots was held in Yizhuang, Beijing, quickly attracting the attention of global media and audiences. Throughout the race, there were constant controversies, hotspots, and topics. Finally, "Tianguang Ultra" developed by the Beijing Innovation Center for Humanoid Robots, "Little Naughty N2" by Songyan Power, and "Walker II" by Shanghai Zhuoyide won the first, second, and third prizes respectively.
What did such a race bring to humanoid robots? The answer is a traffic carnival. After the race, the products of Songyan Power, the runner-up enterprise, received high market attention. Among them, the same model of the running robot, "Little Naughty N2," was viewed by nearly 40,000 people on the JD auction platform and finally sold at a 42% premium.
Similar situations have been repeated this year. Many humanoid robot startups have appeared and become popular on social media. Users can see different brands of humanoid robots showing off their skills, such as dancing, running, crawling, and fighting, while swiping their phones at home, creating one carnival after another.
Meanwhile, leading humanoid robot enterprises are focusing on practical training in various workshops. Elon Musk is vigorously promoting the mass production plan of Tesla's Optimus robot while putting the Optimus Gen3 into Tesla's factories to perform tasks such as battery sorting and equipment debugging, constantly verifying the practical ability of humanoid robots in industrial scenarios.
At BMW's Spartanburg plant, Figure's 02 robot has also started high-intensity logistics sorting operations for 20 hours a day, as well as tasks such as body handling and quality inspection. In China, Ubtech, the leading humanoid robot enterprise, has carried out practical training in more than 10 automobile factories such as Dongfeng Liuzhou Motor and Geely Automobile, completing processes such as seat belt inspection and body quality inspection.
This year, Ubtech launched Practical Training 2.0. Dozens of humanoid robots carried out collaborative practical training in Geely's ZEEKR 5G Smart Factory. For the first time, group intelligence technology was introduced into humanoid robots, breaking through the task scope of a single humanoid robot and expanding it to the flexible production line-level requirements that can be completed by multiple robots working together.
Dozens of Ubtech industrial humanoid robots Walker S1 collaborate to carry large and heavy material boxes.
Why do leading humanoid robot manufacturers such as Ubtech, Tesla, and Figure attach so much importance to workshop practical training? In the view of "Intelligent Relativity," the industrial scenario is precisely the key for humanoid robots to "dig for gold" next.
Figure refreshed the industry record with a valuation of $39.5 billion, a 15-fold increase compared to its 2024 valuation. Behind this soaring valuation is Figure's layout and large-scale practical training in the industrial scenario, as well as the support of its mass production plan. In March this year, Figure officially announced that its BotQ mass production factory had been completed. The first production line has an annual production capacity of 12,000 units, and it will be expanded to 100,000 units per year in the next four years.
In China, Galaxy Universal set a new industry record with a single financing of 1.1 billion yuan. Its cumulative financing exceeded 2.4 billion yuan, and its valuation exceeded the "unicorn" threshold. The reason for reaching such a valuation also lies in Galaxy Universal's focus on actual workshop operations in factory scenarios such as industrial assembly and quality inspection.
Ubtech has taken the lead in realizing the application of humanoid robots and has become the world's only humanoid robot company to announce cooperation with multiple automobile enterprises. Its Walker S series is the industrial humanoid robot that has entered the most automobile factories for practical training globally. Unitree Technology has successively won multiple industrial-level orders, such as the inspection project of Shenzhen Hydropower Station and the humanoid robot training platform of Tongji University. The highest single order amount exceeded 8 million yuan, accelerating its penetration into the industrial scenario.
All these signals indicate that the "wealth" of humanoid robots, including financing and large orders, comes from the "workshop." Only by constantly entering the workshop, carrying out industrial practical training, and solving various business problems in the industrial scenario can humanoid robots at the current stage make money.
Humanoid Robots are Evolving in the "Workshop"
As more and more humanoid robots enter the industrial field, the "evolution" targeting the actual business needs of the industrial scenario is also taking place simultaneously, guiding humanoid robots to perform more complex operations in the workshop.
Facing the complex and production line-level task requirements in factories, Ubtech relies on the key core technology systems of humanoid robot group intelligence, such as cross-field fusion perception, intelligent hybrid decision-making, and multi-machine collaborative control, to build a group brain network technology architecture system generated by the connection of the "super brain" and the "intelligent cerebellum." Based on the Internet of Humanoids (IoH), this architecture realizes the collaborative scheduling of different system modules through a cross-level organic interconnection mechanism, expanding the autonomous task execution scope of a single humanoid robot to the flexible manufacturing requirements of the production line level and forming a hybrid intelligent system with multiple robots working together.
Thus, Ubtech achieved the integration of group intelligence in multi-dimensional scenarios, supporting the collaboration of complex tasks in fields such as industrial manufacturing and service scenarios, and building an intelligent agent network ecosystem with automatic organization and automatic optimization capabilities. This progress directly promoted the participation of humanoid robots in production lines and tasks, allowing more and more humanoid robots to enter factories in groups and cooperate with each other to handle complex tasks.
Overseas, Figure also proposed its own technical architecture for industrial scenario practical training. In February this year, Figure released its new-generation Helix large model. One of the cores of this model is the dual-system collaborative architecture. By innovatively adopting the "System 1 (S1) - System 2 (S2)" hierarchical architecture, it achieves the coordinated balance between semantic generalization ability and high-frequency real-time control.
Specifically, S2 is responsible for "slow thinking" about high-level goals, while S1 is responsible for "fast thinking" and real-time execution and adjustment of actions. At the same time, S1 can quickly adapt to the constantly changing actions of the humanoid robot while maintaining the semantic goals of S2. Simply put, S2 thinks at a low frequency and is responsible for overall planning, while S1 moves at a high frequency and is responsible for real-time operations, thus achieving a division of labor and cooperation mode similar to Ubtech's "brain and cerebellum."
Ubtech industrial humanoid robot Walker S1 collaborates with Wali unmanned forklift F1200S to complete the warehousing of automobile parts.
The Helix large model enables Figure's humanoid robots to perform tasks more like humans in terms of reaction and logic. For example, in a factory, when a robot needs to carry a box, the "super brain" can disassemble the task, confirm the target, and plan the route, while the "intelligent cerebellum" is responsible for real-time execution, such as carrying operations and dynamic obstacle avoidance.
In addition, based on the Helix large model, Figure also enables humanoid robots to achieve multi-machine collaboration. By allowing two robots to share the same AI model, they can achieve real division of labor and cooperation. According to the demonstration video released by Figure, two Figure 02 robots can already cooperate in a division of labor. One is responsible for holding the container, and the other is responsible for picking up things to carry out item sorting.
From Figure to Ubtech, the progress of these technologies and capabilities continuously promotes the evolution of humanoid robots from "machines" to "humans," enabling them to work, think, and cooperate like humans. This will also continue to deepen the "gold-digging" ability of humanoid robots in the industrial scenario.
Industrial Capital Can't Wait to Set Its Sights on Humanoid Robots in the Workshop
These series of signals are favorable for the development of humanoid robots. Facing such a hot field and its explosive potential in the industrial end, industrial capital has been pouring into the humanoid robot field on a large scale and at an accelerating pace. In fact, since its listing on the Hong Kong Stock Exchange in December 2023, as the world's "first humanoid robot stock," Ubtech has directly driven industrial capital to pay high attention to humanoid robots, benefiting latercomers to some extent.
In the first half of this year, the influx of capital into the humanoid robot field was even more intense, attracting about 30 billion yuan in crazy fundraising. Among them, the financing proportions of domestic and foreign markets were each half. In China, Galaxy Universal, Unitree Technology, and Independent Variable Robotics led the way, attracting more than 60% of the financing amount. Abroad, Figure and Apptronik led the way, and their valuations soared.
The "group intelligence" of Ubtech's humanoid robots conducts collaborative practical training in Geely's ZEEKR 5G Smart Factory.
Behind the capital frenzy, there are almost the figures of industrial giants both at home and abroad. For example, the investors of Unitree Technology include China Mobile, Tencent, Alibaba, Ant Group, and Geely, while the investors of Figure include Microsoft, NVIDIA, and Amazon. These industrial capitals not only bring money to humanoid robots but also bring all kinds of industrial-level resources that the industry development urgently needs, such as technologies, data, supply chain management, and practical scenarios targeting the industrial scenario.
For example, through a strategic investment of 1.1 billion yuan, CATL deeply implanted its 4680 battery technology and BMS management system into Galaxy Universal's humanoid robots, increasing the battery life from the traditional 4 hours to 8 hours. At the same time, the heat management technology solved the heat dissipation pain point of robots during high-intensity operations.
Microsoft led Figure's Series B+ financing with $500 million. It also opened up its Azure AI infrastructure and multi-modal large models (such as Claude 3.5), helping Figure-02 robots achieve an end-to-end closed-loop of "natural language instructions - action planning - execution control." The semantic understanding accuracy rate increased to 92%, enabling them to better execute industrial instructions.
Similar capital-industry collaborations are constantly emerging in the humanoid robot field. For example, Tencent AI Lab provided motion planning algorithms for Unitree's H1 robot, and Google DeepMind opened up its reinforcement learning platform D4RL to Apptronik.
While humanoid robots are highly concerned by the whole society, it means that the industry development has officially crossed the days of working alone and entered the stage of industrial collaboration. Whether it is the services provided by humanoid robots or the industry where humanoid robots are located, collaboration will be the keyword for future development.
Learning to cooperate in groups is also a characteristic of human civilization evolution. Humanoid robots may not be an exception.
This article is from the WeChat official account "Intelligent Relativity," author: Chen Bocheng. It is reprinted by 36Kr with authorization.