China Pays Its Citizens to Film Their Chores — And It Might Win the Robot Race Because of It

China controls 90% of the humanoid robot market. But its real advantage isn't cheaper hardware. It's an army of ordinary people teaching robots how to be human

The Housekeeper, the Robot, and the $22 Session

Daniel Wang came home to his Beijing apartment and found a humanoid robot waiting for him. He opened the door, and the robot got to work — slowly. Developed by Shenzhen-based X Square Robot, it spent an hour folding about three pieces of clothing and another arranging Wang's shoes. Most of the actual chores were done by a human housekeeper accompanying it. But the robot's real job was not to clean.

It was to watch, and to learn. Wang paid 149 yuan — about $22 — for the three-hour session. "I feel like I made some contribution to physical AI," he said.

This small scene captures the single most underappreciated factor in the global robotics race. Robotics development worldwide is constrained by a shortage of training data that combines complex visual and movement information — the physical equivalent of the text that trained ChatGPT. And China has found a way to generate that data at a scale, and a cost, that no other nation can currently match: by paying its own citizens, in their own homes and workplaces, to teach robots how the physical world actually works.

The Data Neighbourhoods

The scale of China's data-collection strategy is staggering, and it is deliberately industrial. E-commerce giant JD.com is working with the local government in Suqian — the hometown of its founder — to generate 10 million hours of robotics training data over the next two years. In a dedicated "data collection neighbourhood," residents are paid to film themselves doing household chores. Workers at a local elderly care centre and a kiwifruit farm have been recruited to wear head-mounted cameras that record their hand movements while they work. A stay-at-home mother named Gao Bo gets paid simply to record herself doing housework.

The strategic logic behind this approach is what makes it so formidable. The bottleneck in embodied AI is not processing power or algorithms — it is diverse, real-world data showing how humans manipulate objects, navigate spaces, and recover from mistakes. While US companies, facing high labour costs, have outsourced this data collection to workers in developing countries, Chinese companies can collect massive data sets locally — which may help them build robots better adapted to domestic environments.

China is not just cheaper. It is generating higher-quality, more culturally specific, more diverse physical training data than anyone else — and it is doing so with the enthusiastic participation of its own population.

90% of the Market — and the EV Playbook All Over Again

The data advantage sits on top of a manufacturing dominance that is already close to total. Chinese companies control 90% of the humanoid robot market. The top humanoid makers by 2025 shipments were led by China's Agibot and Unitree, followed by UBTech, Leju, EngineAI, and Fourier. To grasp the gap: Unitree shipped roughly 36 times more units last year than US rivals Figure and Tesla, each of which sold only around 150 robots. Unitree and Agibot each individually sold more units than Tesla's entire 2025 production target.

The pattern is one watchchina.ai readers will recognise immediately, because it is the exact strategy that made China the world's dominant force in electric vehicles. China has a more robust hardware supply chain — much of it built up through the EV sector, from sensors to batteries — and the world's strongest manufacturing base, allowing companies to iterate far faster than Western competitors. The same EV supply chains that contain batteries, sensors, and actuators are now feeding directly into humanoid robots, which is why so many Chinese EV makers — BYD, XPeng, Chery — have added robotics divisions.

And Beijing is converting this momentum into deployment with characteristic speed: the government is targeting the deployment of over 10,000 humanoid robots and more than 100 high-value application scenarios by the end of 2026, moving robots from demos into "work mode" across manufacturing, logistics, healthcare, and disaster relief.

There is one genuine caveat, and watchchina.ai will not omit it. The AI model that controls the robot — not the body — may ultimately decide the winner, and here the picture is less clear: Nvidia still leads in end-to-end humanoid software, and most Chinese humanoid startups still run on Nvidia's Orin chips. "The robot will only be as useful as its model," as one analyst put it, and the underlying vision-language-action models remain early-stage everywhere. But if the model race is still open, the data race that will decide it is one China is running with a strategy no democracy could easily replicate: a nation of citizens, cameras strapped to their heads, teaching machines to take their jobs.

Sources: Rest of World, TechCrunch, CNBC, eWeek, Merics, People's Daily, Robotics Center of Silicon Valley, Omdia, Morgan Stanley