StartseiteArtikel

Unitree is just one step away from going public: WANG Xingxing and his "listing exam"

AI大模型工场2025-11-25 17:54
How did Wang Xingxing build Unitree, valued at 12 billion yuan, in nine years?

From struggling alone in 2016 to having a team of thousands today, Wang Xingxing spent nine years honing his skills and achieved a magnificent transformation from a technology follower to an industry leader. Finally, on November 15, 2025, he completed the IPO counseling.

Wang Xingxing once said, "We are pushing forward with some standard IPO procedures. In a sense, we've always regarded the company's listing as a process of learning and growth. Just like facing the high - school entrance exam or the college entrance exam, I think this is a stage in the enterprise's journey towards more mature management and more mature operations. It's also an account of our efforts over the past nine years and more, and an explanation to our shareholders."

The journey here was full of hardships, but now it's a scene of prosperity. In this exam that lasted only 132 days, Unitree clearly passed with full marks. However, we still need to study and analyze their years of efforts frame by frame.

01 With Courage Alone, from Startup to Leadership

In 2013, Wang Xingxing, a master's student majoring in mechanical engineering at Shanghai University, was inspired by Boston Dynamics' BigDog robot and thus embarked on his career in embodied intelligence.

At that time, instead of following the mainstream hydraulic drive solution in the industry, he pioneered the use of the external - rotor brushless motor direct - drive technology route and independently completed the research and development of the full - degree - of - freedom quadruped robot XDog.

Surprisingly, this technology route effectively simplified the mechanical structure, reduced the R & D and manufacturing costs, and became the foundation of Unitree Technology's core technology route of "low cost and high performance".

Time passed to 2015. Wang Xingxing's XDog participated in relevant competitions in the Shanghai area. It won the second prize and an 80,000 - yuan bonus. This fund became Wang Xingxing's "first pot of gold" for starting a business and also strengthened his idea of industrializing the technology.

As the industry's attention gradually increased, the test video of XDog spread on overseas social media, attracting the attention of potential buyers and investors.

Thus, Wang Xingxing received 2 million yuan in angel investment, laying a financial foundation for the startup project, and Unitree Technology began to take shape.

On August 26, 2016, Wang Xingxing finally registered and established "Hangzhou Unitree Technology Co., Ltd." in a 50 - square - meter office in Binjiang District, Hangzhou.

Although the initial team only had three or four people, fortunately, each person covered a wide range of fields and worked hard.

Wang Xingxing served as both the CEO and CTO, fully responsible for the company's strategy and core technology R & D.

The name "Unitree" also has a special meaning - "The tree of the universe, the foundation of technological innovation", which clearly defined the company's development orientation centered on technology.

In the following years, Unitree Technology gradually completed the leap from technology verification to large - scale development.

In 2018, the company received Series A financing from investors including Sequoia China Seed Fund and Shunwei Capital, and its business model was officially verified by the market.

Three years later, Matrix Partners China, Sequoia China and other institutions jointly participated in the Series B financing. The funds obtained were used for the mass production of the consumer - grade product Go1.

Until the first half of this year, the company finally completed the Series C financing, led by companies such as Tencent, Alibaba, China Mobile, and Geely. After the investment, the company's valuation reached 12 billion yuan.

With business expansion and capital injection, the company's governance structure continued to improve.

In July this year, with the help of the counseling institution CITIC Securities, Unitree Technology launched IPO counseling, clarified the plan for the use of raised funds, and improved the company's governance structure (appointing three independent directors and establishing four special committees of the board of directors).

The counseling record report shows that Wang Xingxing is the controlling shareholder and actual controller of Unitree Technology. He directly holds 23.8216% of the company's equity and controls 10.9414% of the company's equity through Shanghai Yuyi Enterprise Management Consulting Partnership (Limited Partnership), controlling a total of 34.7630% of the company's equity.

Recently, Unitree Technology successfully completed the IPO counseling in just 132 days, which is not only an affirmation of the accumulation over the past nine years but also a stage - by - stage account to shareholders and the team.

Over the past nine years, Wang Xingxing has led Unitree Technology to steadily promote technology iteration and the commercialization process.

02 From 0 to 1, the Gradual Implementation of Humanoid Robots

Although Unitree Technology didn't start to involve in humanoid robots until after 2022, in just three years, it has become an industry benchmark.

Wang Xingxing said, "Our goal is to make robots as popular as mobile phones and enter every household."

Remember the H1 that performed the yangge dance at the Spring Festival Gala last year. At that time, it was often ridiculed by netizens, and Unitree became well - known because of its robots' dancing and running.

Of course, Unitree also faced many doubts. People wondered what's the use of just being able to dance?

In response, Wang Xingxing repeatedly mentioned that Unitree should "focus on movement first, then perform tasks".

Just as humans learn to walk and run first from infancy to adulthood, robots must first have stable and efficient movement abilities before they can undertake more complex tasks.

While undergoing IPO counseling, Unitree also had a major event - the release of Unitree's first wheeled robot, G1 - D.

It not only has a price advantage but also high performance value.

The design of G1 - D is very ingenious. It combines wheeled movement with lifting, and Unitree's innovation here gives G1 - D a larger working space. The vertical working space can reach two meters, the waist joint movement space is Z ± 155°, Y - 2.5°~+135°. Isn't it great?

In addition, as Unitree's first wheeled robot, the whole - machine degree of freedom without the end - effector is 19. Each arm has 7 degrees of freedom, the waist has 2 degrees of freedom, the column has 1 degree of freedom, and the chassis has 2 degrees of freedom.

The robot's head is equipped with a high - definition binocular camera, and the wrist is also equipped with a high - definition camera. In particular, a mobile chassis can be optionally configured. The speed of the flagship version of the mobile chassis is 1.5m/s. It can rotate 360° in place without any problem, and the battery life is up to 6 hours, only two hours less than my one - day work. Moreover, it doesn't slack off and doesn't need to take breaks for meals.

Previously, Unitree focused more on bipedal movement. The debut of G1 - D was an exception. Although it simplified the humanoid robot, the wheeled base greatly improved the robot's stability and was more friendly to physical scenarios.

On November 6 this year, Unitree also released a full - body teleoperation platform. That is to say, under human teleoperation, G1 can smoothly perform daily household chores such as throwing garbage, wiping tables, sweeping the floor, washing dishes, and organizing sundries.

At the 2025 Artificial Intelligence + Conference, Wang Xingxing proposed that "in the next decade, AI technology will endow robots with the ability to truly 'understand the world'."

Seeing this, one can't help but sigh that the day when humanoid robots enter families to do work may be just around the corner.

03 How Does Wang Xingxing Layout the "Large Model", the Brain of Robots?

Wang Xingxing emphasized that the most critical challenge facing intelligent agent robots now and in the future is the robot large model, which is also the biggest obstacle restricting the large - scale application of humanoid robots.

Unitree started the research and development of large models in 2023, mainly focusing on the UnifoLM (Unitree Robot Unified Large Model) series. At that time, it had already combined AI technology with the robot's hardware.

This year is a crucial year for the explosion of Unitree's large - model technology and the promotion of its ecosystem. The UnifoLM brain model released in September can deeply understand the interaction rules between robots and the environment. It can also serve as an interactive simulation engine, generating realistic environment and feedback videos based on the robot's future actions.

The UnifoLM - WMA - 0 model also integrates the general framework of "world model + action decision - making", specifically designed for general robot learning and suitable for a variety of robot bodies.

It has two core modes: decision - making and simulation. It can predict interaction information to assist in action generation and simulate environmental feedback. It has been deployed and verified on the Z1 robotic arm and the G1 humanoid robot, and can perform tasks such as stacking blocks and assembling cameras.

In this way, with a "brain", robots can do more than just repeat instructions. It is a crucial step towards diversity and a crucial step towards becoming "human - like".

In addition, while independently researching and developing large models, Unitree also strengthened the robot's model application ability through cooperation with leading technology companies.

On the one hand, at the 2025 World Artificial Intelligence Conference, its humanoid robot demonstrated the ability to be equipped with Tencent's Titanium Screw Integrated Intelligent Platform. With the help of basic model services such as multi - modal perception models and planning action models provided by Tencent, it can achieve autonomous navigation and instruction execution.

On the other hand, Unitree also uses Alibaba's Qianwen model. The open - source basic model of this model has excellent adaptability to dexterous hands, which can well meet the needs of Unitree robots' dexterous hand - related operations, complementing the self - developed UnifoLM series.

Rejecting "building a car behind closed doors" and adopting an open, inclusive and win - win attitude, Unitree's strength and vision in all aspects are evident from its position as a leader in the field of embodied intelligence.

Wang Xingxing also said that "the critical point of the robot large model will come in 1 - 2 years at the fastest and 3 - 5 years at the slowest."

Conclusion

From the brave exploration in a 50 - square - meter office to industry leadership with a valuation of 12 billion yuan; from the initial appearance of XDog to the practical application of G1 - D, and then to the possibility of UnifoLM large model endowing robots with the ability to "understand the world", Wang Xingxing has written a growth history of a Chinese science and technology innovation enterprise in nine years.

This 132 - day IPO counseling "exam", seemingly a final sprint, is actually a confirmation of the perseverance in the technology route, prudence in commercialization, and an open - minded attitude towards the ecosystem over more than nine thousand days and nights.

Listing is not the end but the starting point of a broader stage.

Wang Xingxing once said that the next decade is the decade when robots will become "life partners".

Now that embodied intelligence is gradually entering our lives and the critical point of large models is approaching, we have reason to expect that the dream of "making robots as popular as mobile phones" is slowly becoming a reality with Unitree's progress.

This article is from the WeChat public account "AI Large Model Workshop", author: Yuwan, editor: Xingnai. It is published by 36Kr with authorization.