The Rebirth of Beijing’s “Robot Data Factory” Reveals an Embarrassing Secret in the Industry
Hello everyone, I’m your financial journalist. Today, we’re going to talk about a topic that sounds very futuristic, but is actually driven by real-world dynamics and business strategies: the humanoid robot data training centers.
In simple terms, these centers are where robots are “fed” with the information they need to become smarter. For robots to learn, they must watch a large amount of video, interact with various objects, and perform numerous actions, and all of this data serves as their “food”.
Recently, the first humanoid robot data training center in Beijing, which was once very popular in the Shijingshan Shougang Park but then suddenly shut down and had all its robots removed, has become active again. However, the players and the approach have changed this time.
This is not just the story of one center in Beijing; it reflects the struggles and transformations that more than 90 similar centers across the country are undergoing. I’ll break down the logic behind this phenomenon into five key points in plain language to help you understand what’s really going on in the “robot data business”.
---
1. The Plot Twist: From 100 Robots to an Empty Space – Who Left, and Who Came?
【Key Facts】
When the center in Shijingshan opened last year, it was quite impressive, with over 100 dual-arm robots simulating 10 different scenarios such as homes, supermarkets, and factories, creating a miniature “robot society.”
But early this year, it became deserted, and the operating company, Ruirman Intelligence, withdrew.
Now the center has reopened, but with a new approach:
- Fewer Robots: Instead of having many physical robots, it uses more “motion capture suits” (devices worn by people to mimic robot movements).
- Technological Upgrades: It focuses on “4D Gaussian light fields” and “world models,” which sound more advanced, but actually represent changes in data collection methods.
- New Owners: The leading companies are Lingyun Guang and Zhongguancun Tongli, with several other companies from the industry chain participating.
**Plain Language Explanation:
It’s like a restaurant that initially featured “live cooking performances” (robots controlled remotely by a chef, Ruirman). The chef (Ruirman) might have thought the food wasn’t good or had a disagreement with the owner (the government/investors), so they left. Now the new owners (Lingyun Guang, etc.) are focusing on a “central kitchen” model (pre-prepared data rather than using physical robots). It might not be as visually appealing, but it could be more efficient and cost-effective.
Key Point: Ruirman claimed that the previous approach was too laboratory-based and the data wasn’t realistic enough; the new direction emphasizes “physical AI infrastructure,” which sounds more practical, but it also means that the initial investment (the 100 robots) might have been a waste.
---
2. The Behind-the-Scenes: Was It Technology or Money That Caused the Split?
**【Key Facts】
Why did Ruirman leave? Both parties have different versions of the story:
- Zhongguancun Tongli (the new leader) says: Ruirman didn’t meet the standards; the data quality and accuracy were insufficient, and their remote control approach didn’t align with the center’s future plans.
- Ruirman says: The initial version was just for testing; they want to move to real-world scenarios for remote control, such as in factories and homes, not just for show in the lab.
- Industry insiders reveal: There were disagreements with the local government, and Ruirman moved its registration to Chaoyang in May.
**Plain Language Explanation:
It’s like a partnership going sour. One party thinks the other didn’t do the job well, while the other thinks the other party didn’t understand their vision.
The key issues are the change in registration location and the disagreement with the government.
- Government Perspective: They provided the land, funding, and policies, expecting Ruirman to sell 1 million data points per year. But did they sell any? How many? No one’s sure. The government wants tangible results and industrial development, not just a display of robots.
- Company Perspective: They bought the robots and set up the scenarios, but couldn’t sell the data at a profit or at a price that covered costs.
- Possible Reason: It was probably a failure in expectations. The government expected quick economic benefits, and the company hoped to rely on government subsidies. When it became clear that data sales were poor, both parties were in a difficult situation. Ruirman’s departure was both a strategic shift and a way to minimize losses.
---
3. The Industry’s Dark Secrets: 90 Centers Sprouting Up – Who’s “Buying and Selling from Within the Same Circle”?
【Key Facts】
There are at least 90 humanoid robot data training centers nationwide.
A secret has been revealed: much of the data is just circulating within the same companies.
- The Pattern: The government funds the centers, local robot companies (like Leju and Zhiyuan) sell robots to the centers, which use the data for training, and then the data is sold back to these companies (or their affiliates) for model development.
- Example: Leju Intelligence won the second phase of the Shijingshan project, and its prospectus shows that the local industry company was its largest customer, accounting for 12.94% of its revenue in 2025.
- Doubts: Some founders admit that these companies rely on these centers to generate unsustainable income.
**Plain Language Explanation:
This is a classic case of “buying and selling from within the same group.”
- Government: They provide funds and resources, creating the illusion of a thriving industry and boosting GDP and employment.
- Robot Companies: They sell robots and see increased revenue and better financial reports, while also receiving government subsidies.
- Data Circulation: The data collected is often sold back to the same companies.
Problems:
1. Data Quality: Data from closed labs may not reflect real-world situations.
2. Lack of Market Impact: The data doesn’t reach real third-party users (e.g., AI model companies) but remains within the industry chain.
3. Sustainability: Government subsidies are limited; without them, the cycle breaks down.
In Summary: Many centers are just selling a story and dealing with excess inventory.
---
4. Technological Shift: From Remote Control of Physical Robots to Data Collection Without Physical Robots?
【Key Facts】
Previously, the mainstream was using remote control of physical robots: people wore gloves to operate them and record data.
The trend is changing:
- Fewer Dual-Legged Robots, More Wheeled Robots: Dual-legged robots are difficult, expensive, and unstable.
- More Data Collection Without Physical Robots: Devices like UMI (handheld data collectors) and Ego (first-person video data collectors) are becoming popular. People use these to collect data in real scenarios and then apply it to robots.
- Reasons: Remote control of physical robots is costly, inefficient, and the data lacks generalizability.
Plain Language Explanation:
It was like having robots do the work manually, which was exhausting, expensive, and risky. Now, it’s more efficient to have humans collect data in real scenarios using devices like UMI/Ego.
- Advantages:
1. Cost-effective: Instead of buying 100 robots, you only need a few UMI devices.
2. Real-World Data: The data comes from real-life situations, making it more accurate.
3. Flexibility: Humans can use multiple devices simultaneously or switch between different scenarios.
- Beijing’s Center’s Upgrade: It’s moving from using many robots to using 4D light fields and world models, reducing hardware dependence and improving data quality.
---
5. Future Outlook: Will Data Collection Centers Disappear? No, but They’ll Change and Improve?
**【Key Facts】
- National Development and Reform Commission: They’re preventing blind imitation and promoting healthy, orderly development.
- Industry Consensus: Data centers are valuable in the early stages, but they can’t rely on a self-sustaining cycle.
- New Directions:
1. Building High-Quality Data Systems: Focus on collecting accurate and useful data.
2. Serving Real-World Industries: Data should help robots work in factories, logistics, and homes.
3. Diversified Revenue: Beyond data sales, provide solutions and technical services.
Plain Language Explanation:
Data centers won’t disappear, but they’ll undergo a transformation.
- Surviving Centers: Those with leading technology (e.g., using UMI/Ego), real-world scenarios, and diversified revenue sources.
- Implications for Investors: Be cautious of companies that rely heavily on government subsidies and have poor data quality.
- For the Public: If you invest in robotics, look for companies that generate real value from their data, not just from government-affiliated sales.
- The Rebirth of Beijing’s Center: It signals a shift from hype to practical application in the industry.
---
In summary, the rebirth of Beijing’s humanoid robot data training center is a microcosm of the embodied intelligence industry’s transition from conceptual hype to practical implementation.
- Past: Focusing on building robots, creating scenarios, and selling stories to rely on subsidies.
- Present: Emphasizing technology, cost reduction, data quality, and real-world applications.
This incident reminds us that the tech industry can’t rely on internal cycles; it must stand up to market scrutiny. Data is not just about collection; it’s about how it’s used. Robots aren’t just for show; they need to be useful in real-world contexts.
Next, we’ll focus on who has the most valuable data and whose robots can actually make a difference in real-life applications, such as in factories or caring for the elderly. That’s where the true potential of embodied intelligence lies.
Thank you for listening to this analysis.