Summary of Key Points
For robots to learn household tasks such as making beds and tidying up desks, they need to be “taught” by humans through physical actions—these individuals are known as robot data collectors. They wear helmets and gloves equipped with sensors, repeatedly performing these actions in slow motion to record the data, which is then sold to robot companies. The entry barrier for this job is low (as long as one has decent physical capabilities), but the daily salary is only 200-250 yuan, placing them at the bottom of the value chain. Most collectors are people who have been laid off or are in a transitional period with mortgages; few see this as a long-term career. Although there is high demand for these services (with the number of jobs increasing by 769%), the future is uncertain: once robots learn basic tasks, either the collectors need to upgrade their skills or they will be replaced.
I. The Robot’s “Teachers”: Making Beds in Slow Motion Every Day
The essence of this job is to serve as a model for the robots. You have to repeat actions like folding towels, making beds, and tidying up desks in a slow-motion manner, similar to what you would see in a time-lapse movie. For example, when folding a towel, you cannot just fling it out flat; you need to carefully lift, unfold, and press it down, with even the smallest movements of your fingers being precise for the sensors to record.
The equipment is quite complex: the helmet comes with a camera, the gloves have built-in sensors, and all three components are connected to a computer via wires. Putting on and taking off the equipment takes 15 minutes, which can be inconvenient, even for using the bathroom (it could waste half an hour). Moreover, the equipment has specific requirements—fingers that are too long, too short, or too thick won’t fit properly, and people who are too overweight may not be able to use it. Novices can only generate 2-3 hours of useful data per day, while experienced collectors can manage about 4-5 hours, but they also need to avoid repetitive movements or blurry footage that would render the data unusable.
After a day of work, your head hurts from the tight helmet, your gloves are soaked with sweat and wrinkled, and your shoulders, neck, and back ache—by the end of the day, you’ve become as stiff as a robot.
II. Data Is Valuable, but the Collectors Don’t Profit Much
Robot companies pay a high price for this data: human data costs 200-500 yuan per hour, while remotely operated robot data can cost up to 1000 yuan per hour. However, the daily salary for collectors is only 200 yuan (with an additional 50 yuan for night shifts). Where does all that extra money go?
The value chain looks like this: Collectors → Labor Companies → Data Service Providers → Robot Manufacturers. For instance, the employer may pay the labor company 300 yuan per day, but the labor company only gives the collectors 200 yuan. Data service providers then perform quality control, data cleaning, and annotation before selling the data to robot companies at a higher price. Collectors are essentially “fueling” the robots with their physical efforts, yet they receive the lowest pay.
Why is the data so expensive? Because robots need to learn practical skills that cannot be learned from online resources. While large language models can process text and images online, robots need to learn how to grasp objects or open doors through real-world examples. Currently, there are only 500,000 hours of such training data available globally, whereas training general-purpose robots requires tens of millions of hours, creating a significant gap.
III. Who Are the “Fuel”? People in Transitional Roles
Most of the collectors working in this industry are in temporary positions:
- Li Chenchen, an IT maintenance worker who was laid off and is looking for a new job after losing thousands in business;
- A former real estate agent with a mortgage, buying a house in Langfang and renting one in Beijing while waiting for a stable job;
- College students with degrees in IoT who have won awards but can’t find suitable jobs;
- Newly married women who have left home due to family conflicts and are taking part-time jobs.
No one sees this as a long-term career: the work is monotonous (repeating the same tasks hundreds of times), the employment is unstable (orders have deadlines), and the pay is low. Some people leave after just three days without even receiving their full salary. The team leader is a 21-year-old graduate, and the labor company takes 20% of their weekly wages—even the lowest-level managers in the industry don’t feel secure.
IV. How Long Will This New Career Be Popular?
On the surface, this seems like a “hot new career” with job listings increasing by 769%, and the number of positions in the entire chain is expected to grow by millions over the next five years. However, the reality is:
- There is a clear hierarchy within the jobs: high-paying roles involve remotely operating robots (daily salaries in the thousands), while most people perform the low-level data collection tasks (200 yuan per day);
- The demand will evolve as robots learn more complex tasks (such as cooking or caring for the elderly), and collectors will need more specialized skills (e.g., knowledge of medical care); the current group of collectors may be replaced.
- This is still in the early stages of the industry; robots are not yet capable of performing tasks reliably, so data collection is only a temporary need. Once robots learn enough tasks, the work for collectors will either become more difficult or disappear, just like how AI annotators were phased out when simpler annotation tasks became obsolete.
Company executives can only promise that “salaries will be paid on time in the next two years,” but the collectors know this is just a temporary stop in their lives.
V. Will Robots Liberate Humans? Not Yet…
The vision of robots is to serve humans, but for now, humans are acting as their “fuel.” You repeat the same actions millions of times, turning your physical experience into data for the robots to learn from. Li Chenchen’s question hits a nail in the coffin: “When the robots learn, will we have to find new jobs again?”
This contradiction is heartbreaking: we train robots to make our lives easier, but in the process, we end up doing the most mechanical labor ourselves. We hope robots will replace tedious tasks, yet we become those tasks themselves. When robots truly enter households, will these “teachers” find their place in the new world? No one knows the answer.
This report serves as a mirror, revealing the “invisible workers” behind technological progress and the powerlessness and confusion of ordinary people in the midst of changing times. The future of robots is promising, but on that path, some are paving the way with their own bodies.