A historic moment for China's embodied intelligence—On August 19th, Unitree Technology officially listed on the Sci-Tech Innovation Board (STAR Market), becoming the 'First Humanoid Robot Stock' on the A-share market.
Thus, a scene of China's robotics counterattack has been created: in ten years, starting from scratch, Wang Xingxing has built Unitree into the world's leader in humanoid robot shipments, beginning with a single quadruped robot.
Investment community noticed a detail: Unitree's three largest external shareholders—Meituan, Sequoia, and Matrix Partners—have also jointly invested in a very low-profile embodied intelligence company, Miaodong Technology. Such a scene is extremely rare. According to the prospectus, Meituan is Unitree's largest external shareholder; Sequoia China not only made an early bet but also became the independent private equity fund with the highest shareholding ratio in Unitree through multiple rounds of heavy investment. Matrix Partners also led an early round in Unitree and still holds over 5% of the shares.
Like Wang Xingxing, Miaodong's two co-founders, Gao Jianrong and Yang Shuo, also previously worked at DJI. Yang Shuo is also the only founder in China's embodied intelligence circle who came from the core team of Tesla's Optimus. Looking closely, Miaodong and Unitree share some commonalities, such as adhering to self-developed motors, emphasizing full-stack software and hardware capabilities, and maintaining a product-oriented and user-oriented approach. They also have some differences; for example, Unitree's product portfolio is 'quadruped robot dogs + humanoid robots,' while Miaodong's is 'bi-pedal wheeled robots + small humanoid robots.'
When the three Unitree investors appear together behind Miaodong Technology again, it inevitably sparks imaginations.
A Rare Team, Background of DJI Alumni + Tesla Alumni
DJI alumni startups have been a hot theme in the primary investment market over the past two years. Companies like Unitree, Bambu Lab, EcoFlow, and Kuma have successively become leaders in their respective fields, drawing significant attention. It is reported that Bambu Lab has set a revenue target of 24 billion yuan for 2026 and plans to initiate its IPO process in 2027. This further solidifies the belief that DJI alumni companies hold the key to success in the hardware sector, establishing DJI as a 'training ground' in the hardware field.
It is understood that Miaodong's CEO, Gao Jianrong, worked at DJI for nine years. He was one of the most capable assistants to DJI's founder, Wang Tao, early on. Later, he was assigned to specific business areas, successively responsible for the RoboMaster robotics competition, DJI's supply chain, DJI's global marketing, and an independent BU, DJI Education.
According to investors who have long observed DJI, it is extremely rare within DJI to find an executive who has spanned so many business areas and always reported directly to Wang Tao.
Miaodong's CTO, Yang Shuo, studied under Professor Li Zexiang. He joined DJI during his master's degree and served as a key algorithm engineer for the flight control module, making core contributions to the Phantom product line. In 2016, Yang Shuo and Gao Jianrong jointly managed the RoboMaster competition, marking the beginning of their collaboration. In 2018, Yang Shuo left DJI to pursue a Ph.D. at Carnegie Mellon University (CMU), focusing on the control of legged robots.
After earning his doctorate, Yang Shuo joined Tesla's Optimus team, a gathering place for the world's top technical talents in robotics, where he quickly became one of the highest-ranking Chinese engineers on the team.
"The Optimus team reports to Elon Musk every two weeks, and I attended every meeting," said Yang Shuo. "The deeper I got into it, the more I realized that robotics is inseparable from China's industrial chain system. It's better to do it domestically than in Silicon Valley—the iteration speed is several times faster."
The recent wave of embodied intelligence has been largely driven single-handedly by Elon Musk. With the first-hand knowledge of the embodied field and the hardware productization capabilities honed at DJI, it is not surprising that the Miaodong team attracted top-tier institutions to invest from the very beginning.
First Product Revealed, Breaks Global Crowdfunding Record
One month before Unitree's listing, Miaodong Technology officially launched its first product, Beni, on the crowdfunding platform Kickstarter. Beni is a bi-pedal wheeled robot and also the world's first camera robot.
Currently, over 5,000 people have already paid for this product. According to industry insiders, this product has become the number one in terms of funds raised in the robotics category on Kickstarter's history.
During the crowdfunding campaign, Marques Brownlee (MKBHD), the most influential tech YouTuber, reviewed and scored recent tech products. Among more than ten products, Beni received the only perfect score of 10/10.
Marques Brownlee has over 20 million subscribers and is widely recognized as the world's number one tech YouTuber, wielding unparalleled influence in the tech field. It is understood that MKBHD is hardly influenced by brands. In this episode of reviews, the worst score was 2/10, making the perfect score for Miaodong's product particularly significant.
The great success of Beni is a victory for Miaodong Technology's strategy of 'cutting-edge technology + top-tier product power.' The form of wheeled-legged robots has existed for some time, but the industry has never solved the problem of productization. Beni is the world's first wheeled-legged robot aimed at consumers, stemming from the team's deep insight into user needs and a profound understanding of the robotics platform.
In terms of technical approach, Miaodong firmly chooses to use reinforcement learning to train all of the robot's functions. This new direction has been applied in humanoid robots but had never been used for user functions like follow-and-shoot before. A greater challenge lies in ensuring the consistency of these neural networks across mass-produced products—making one robot jump 25 cm is easy, but mass production requires consistency across tens of thousands of robots. If the robot's jumping performance is inconsistent across different environments and loads, it would significantly reduce structural lifespan. To address this, the team conducted extensive research and experiments in algorithms, hardware, and other aspects, gradually narrowing the simulation-to-reality gap and ensuring that neural networks perform consistently across thousands of robots.
Why start with the seemingly niche scenario of 'low-angle follow-and-shoot'? Miaodong's logic is very 'product-oriented': imaging devices are naturally divided by perspective—drones handle the sky, phones handle everyday moments, action cameras handle sports—but no one has filled the gap of 'ground-level first-person perspective.' Moments like pets running on the ground, a child's low vantage point, or the low-speed impact feeling of skateboarding and tennis... these moments were previously missed. Beni doesn't replace any existing devices; it fills that blank.
As the veil is lifted, Beni offers a brand-new experience: it's the world's first portable, all-terrain follow-and-shoot robot, weighing less than 2 kg. It can jump up to 25 cm, reach speeds of 28 km/h, and can autonomously avoid obstacles, jump onto steps, and get up after a fall. With Beni, people no longer worry about missing wonderful moments; it can automatically capture those fun moments in life.
Currently, Beni already has its first batch of users, including pet owners, families with children, runners, and inline skaters, etc. Based on feedback and suggestions from beta testers, Beni is continuously iterating—this emphasis on user experience and co-polishing the product with users also distinguishes Miaodong from its peers.
It is understood that Beni is preparing for its official global launch and sales in October, following the end of the crowdfunding campaign. The Miaodong team stated that their internal goal is to sell one million units within three years. If this target is achieved, Beni will become the most popular robot product in history.
Moving Towards Humanoid Products, Adhering to Full Stack, Adhering to First Principles
Miaodong's product pacing is relatively similar to Unitree's—both companies did not first launch humanoid robots but started with non-humanoid forms to create products.
Yang Shuo further explained that the team insists on thinking about this issue using first principles.
On the hardware level, the core capabilities of a robot are ensured by its motors. Startups often lack mature motor design capabilities and production experience, so they typically need to cooperate with suppliers through ODM or OEM. While this is faster, it not only increases costs but also prevents tight synergy between software and hardware. Unitree's leading global capability in robot locomotion is precisely due to the accumulation in motor design and production gained from early batch shipments of quadruped robot dogs.
After accumulating motor capabilities through Beni, Miaodong has now designed motors with angular precision and torque density not inferior to Unitree's. Especially under similar cost solutions, they have increased motor control precision by an order of magnitude, proving the team's solid hardware design capabilities.
Regarding the 'brain' and 'cerebellum,' Miaodong insists on pursuing the whole-body intelligence route. "Many people give up on the humanoid bipedal direction because they worry about not being able to achieve whole-body intelligence," in Yang Shuo's view, "In home environments for consumers, passability is the first test for a robot. I can hardly imagine how non-humanoid robots could handle the 'House' scenarios common in European and American homes."
For this purpose, Miaodong has made significant scientific breakthroughs over the past year. Based on world models, Miaodong introduced two major technical architectures for the 'brain,' DiT4DiT and MotionWAM, gradually solving the industry challenge of coordinated upper and lower limb operations in humanoid robots. For the 'cerebellum,' Miaodong's Perceptive BFM is the first in the industry to solve the problem of human-like gait walking on complex terrain.
In terms of product, Miaodong has an even more mature methodology. "We insist on thinking about products based on scenarios and needs. Users don't care about how the technology is implemented; they only consider what this product can do and whether its value matches its price," said Gao Jianrong. Through Beni, Miaodong's product has already entered households with children, pet owners, and sports enthusiasts. Among them, core users will become early co-creators for Miaodong's first home-use embodied product. Designing embodied products by staying close to users and scenarios is perhaps one of the most anticipated aspects of Miaodong's home-use embodied products.
Beyond this, what Miaodong values more is consumer-grade product definition, user experience, and the data feedback loop, extending this capability from Beni to humanoids. Subject to compliance, Miaodong can obtain data from users and continuously iterate the product itself. With the onset of mass production, Miaodong expects the daily data scale to exceed 100,000 entries by the end of the year.
Witnessing the current fervor in the industry, the Miaodong team maintains the clarity of entrepreneurs. They believe the market overestimates the importance of operation models for robots. The truly important issues are the safety and reliability of robots, and the 'vast amount of reliability engineering between a demo and mass production.'
This is also reflected in Miaodong's pacing: they do not chase high valuations and are willing to grow steadily over 3 to 5 years, "proving ourselves one product at a time." In Yang Shuo's view, humanity's expectation for a 'mechanical companion that resembles itself' has persisted for millennia. This is a field worth dedicating a decade to, and there's no need to rush.
This article is from the WeChat public account "投资界" (ID: pedaily2012), author: Wu Qiong
