Some people find it cool, while others leave after just one month. Robots are still learning how to work, and the people who act as their "teachers" have already started their jobs. They are robot data collectors. Folding clothes, making beds, washing dishes, and sticking labels on packages—these everyday actions that people are used to performing are the tasks they must repeatedly demonstrate every day. The recorded data, after review and filtering, is then used to train robots.

Currently, there are roughly several approaches to collecting data: one is teleoperation, where collectors wear VR headsets to "operate" real robots; another is using handheld grippers like UMI, where a person holds a device that mimics a robotic hand to perform tasks, recording motion trajectories without needing an actual robot; and yet another is using cameras to record a person's first-person perspective, known in the industry as Ego data, where collectors wear a camera on their head or chest and do household chores as usual. Many collectors may not be able to clearly explain which category their work falls into. What they care about more is how many minutes one data entry takes, how many hours they need to accumulate each day, how much money they can earn from repeating these actions daily, and how long they can keep doing it. "Dingjiao One" spoke with five practitioners. Some do household chores repeatedly at home, earning about 5,000 yuan a month, treating this job as a transitional phase; some lead teams collecting data, where a single parameter set incorrectly can invalidate hundreds of hours of data; and some have collected data for robots for months but have never even seen a robot at their workplace. Headsets pressing against glasses, hands sweating inside gloves, having to start over when clothes fall—these are the daily work realities they described. Some feel like they are "screwing bolts," while others have developed a "feel" for the machine and derive a sense of accomplishment from it. What has this new job brought them, and why have they made different choices? Here are their stories.

01. Earning 120 Yuan a Day, I Was a 'Hand' for Robots for a Month

Xiao Sun | 22 years old, Shanxi, Part-time at an embodied intelligence data collection company. This summer, I originally wanted to find an internship, but after searching around and finding no suitable positions, I decided to find a part-time job to pass the time. When I saw "robot data collection" on a recruitment app, I was curious about how people provide data to robots, so I applied. Before going, I thought it would be like an experience project in a science museum. Put on the equipment, do a few actions—novel and not too tiring. It was only when I actually started working that I realized it was more like assembly-line work requiring long hours of repetitive labor.

After joining, we had two days of training, mainly learning how to use the equipment and the requirements for actions. When formally on duty, we had to wear a VR headset, carry a charging backpack, and hold a gripper in each hand. Every day at work, we first claimed tasks in a group chat. The work mainly involved doing household chores while wearing the equipment, including folding clothes, sweeping floors, and washing dishes. It looked simple, but there were many actual requirements. For example, when folding clothes, you couldn't keep folding the same item—T-shirts, skirts, and jackets all had to be included, and different clothes required different folding methods. You also couldn't keep folding in the same scene—living rooms and bedrooms had to be alternated. The company wanted data that was as diverse as possible; recording the same action from the same angle too many times would reduce its value. Each video segment had to be at least two minutes long, and after completion, they were batch-uploaded for backend review. Segments with irregular movements, unclear footage, or inadequate lighting would be rejected and not counted toward effective hours.

We were divided into day and evening shifts, both 8 hours, working six days and resting one, with daily wages—120 yuan for the day shift and 135 yuan for the evening shift. The assessment mainly looked at effective hours, requiring at least 3 hours a day, with an 85% accuracy rate to pass. Each month, the company also ranked by cumulative effective hours, with the top few receiving extra bonuses. At first, I was always worried about not meeting the standards, but after getting familiar with the equipment, I found the requirements weren't as difficult as I had imagined. The hardest part was wearing the equipment for long periods. The grippers weren't very flexible, and the angles were hard to control. When dealing with soft, thin clothing, they often couldn't grip properly, and if something fell, you had to re-record. Folding one piece of clothing often required repeating it a dozen times, and the movements couldn't be completely identical—you had to think about what angle or method to use next time. My arms were raised the whole time, and after a day, my shoulders, wrists, and neck were all sore. After the first two days, I felt like I never wanted to touch this job again. After about a week, my body gradually adapted, and I developed a better feel for it.

I later discovered that some people's movements weren't very standard but still passed review—there seemed to be some room for "slacking off" in actual execution. Even so, repeating similar actions every day was still very tiring, not much different from tightening bolts in a factory. Staff turnover here was high. Most people who came were young, basically under 30, and many left after two weeks. The company's requirements for newcomers were also getting higher. When I first arrived, there were still two days of training; later, training was shortened, and newcomers were required to reach the level of experienced workers after just one day. After completing a full month, I still left. For me, the daily wage wasn't high enough to offset the repetitive labor and physical toll. I think if you just want a part-time job as a transition, or you're curious about robot data collection, you can give it a try. But if you want to broaden your horizons or find a long-term career direction, this job probably won't bring you much.

02. I Taught Robots to Fold Towels, and It Even Learned My Unconscious Little Habits

Zheng Zheng | 26 years old, Beijing, Data collector at Lingyu Intelligence, an embodied intelligence company. When my family asks what I do, I usually say: "You can think of me as a kindergarten teacher—I'm teaching robots." I do robot data collection in Beijing and have been in the industry for over half a year. I previously worked in IT. When I first saw this position on a recruitment app, it even asked whether I usually play video games. I found it novel, so I submitted my resume. The interview required operating a machine on-site to assess coordination; after two or three days of onboarding training, you could get started. Every day I wear a headset to teleoperate robots, collecting the process of completing actions as data to hand over to colleagues who train the models.

This job pays a base salary plus performance bonus, earning somewhat less than my previous job. But I'm genuinely curious about this industry and willing to keep doing it. When I first joined and saw so many robots, I thought it was awesome and impressive, but repeating the same thing every day gets boring over time. Recently, I've mainly been teaching robots to fold towels and boxes—relatively simple, each taking under two minutes. What's harder is putting on pillowcases. The pillow insert is very soft, and you have to operate the robot to stuff it bit by bit into the pillowcase, which is very time-consuming. For some simple tasks, 200 data entries can be collected in an afternoon; for difficult tasks, it might take a week. My work hours are from 9:30 AM to 6:30 PM, with a 15-minute break every 45 minutes of work. Some tasks require holding your arms up the whole time, and after a while, your arms get tired. When it's boring, I can listen to music or chat with colleagues nearby.

What frustrates me most is when the machine suddenly has problems—like a loose cable or poor network causing recording failure—which immediately affects collection efficiency. After doing it repeatedly, I gradually developed a "feel" for the machine. Some movements are awkward with the forehand but easier with the backhand because the machine's range of motion is limited. Some look simple but fail if the starting position or wrist angle is off. This kind of experience can't be learned from documents alone. Once I'm familiar with the machine, I know what angles it can achieve and can adjust accordingly. If some prescribed steps don't work smoothly, I'll give feedback upward: a different approach might be simpler. When new colleagues find folding boxes difficult, I'll say, "It's okay, practice more, it'll get better." Many tasks only seem simple after you've practiced them and look back. What's more interesting is that I can see some familiar little movements in the robot. Sometimes I unconsciously make the gripper snap shut on empty air, and later in the trained model, its gripper also twitches unconsciously. Our prescribed initial position is both hands held level in front, but some people habitually hold them a bit higher, and the robot learns a correspondingly higher starting position. A couple of days ago, the company made several demonstrations: the robot could continuously fold several towels, and also sort and assemble shapes. These demonstrations also used data I collected. Watching them, I still felt it was awesome, with a sense of "I contributed too."

The robot is still like a child now, not a very smart child, needing us to teach it over and over. When it becomes more capable, we collectors may also need to do more complex and targeted tasks. I also worked on JD Qixian's robot food-tasting project, which involved interacting with customers—more interesting than repeatedly folding towels. In the future, I want to move toward marketing, attending exhibitions and training clients. After more than half a year, I already know how boring this job can be, but I still think it's cool and fun.

03. After Months of Data Collection, I Never Saw a Single Robot

Li Zhe | 23 years old, Guangzhou, Data collector at an embodied data company. I previously worked in video editing. After this year's Spring Festival, I wanted to see if there were any new opportunities. While job hunting, I came across a "data collection" position. The job description in the posting looked pretty easy: wear data collection equipment and do things you'd normally do at home. I thought, what's so hard about that, so I signed up. The work location was in Guangzhou. Onboarding training was in the company's factory building, which was partitioned into many small rooms simulating various home scenarios. At first, I thought this job was just wearing equipment at home and doing some actions, but it turned out I had to commute here every day—I was already having second thoughts.

In the first phase, my main task was wearing a headset and doing three things: washing dishes, arranging plates, and repeatedly moving a tea set from one table to another. Each data entry was less than a minute. There was a big screen in the workroom showing in real time how many entries I'd done today and how much effective data time there was. Typically, after a day's work, you could accumulate three to four hours of effective time. The collection equipment could only last two hours on a full charge, so every two hours you had to stop, charge the device, and take a break. Later, the company provided external power banks, which could keep the equipment running for over four hours at a stretch.

After getting proficient, I was transferred to the express delivery scenario, sticking labels on cardboard boxes. This time, besides the headset, I had to wear an extra glove on my dominant hand. The glove was heavy, and the work became more complex: hand movements had to appear completely in the recorded frame for the data to qualify. Sometimes I had to repeat it several times to record one valid data entry. The pay also increased. Initially it was 100 yuan a day, but you had to meet both work hours and effective hours to get the full amount. After switching to express labeling, it was 150 yuan a day. This job is far from as easy as it looks. I'm nearsighted, and my regular glasses have relatively large frames. The headset would press right on the frames, which was very uncomfortable after a while. The equipment itself is also heavy, and wearing it for long periods makes your neck sore. The gloves aren't breathable, and after wearing them for a while, your hands get sweaty. Sometimes the device would signal weak connection, and I'd have to stop, take off the gloves, wipe my hands, and put them back on to continue. If off for too long, the connection between the gloves and the head device would disconnect, requiring re-pairing.

Sometimes friends ask what I've been up to, and I say robot data collection. They think I get to see many humanoid robots at work. Actually, after months of working, I've never seen a robot at the job site. I've always treated it as a part-time job, not planning to do it long-term—mainly to have something to do and earn some income along the way. How should I put it? It feels only slightly better than working on an assembly line. Compared to my previous video editing job, the income level is about the same. At least based on my few months of experience, there's still plenty of data collection work, so I don't need to worry about running out of work in the short term. If you want a part-time job as a transition, you can try it, but for long-term work, I wouldn't recommend it.

04. Making the Bed 50 Times a Day, Similar to 'Screwing Bolts' in a Factory

Xiao Ran | Post-85s, Chongqing, Embodied intelligence data collector. I previously worked as a headhunter. During those years, I came into contact with quite a few robotics companies. After talking with them a lot, I learned about this industry. In August this year, I joined a data collection company in Chongqing, full-time, working from home. It's been over a month now. My job is very simple: do household chores at home with a DJI camera device on my head, recording the whole process. Tasks include folding clothes, folding blankets, making beds, and so on—I choose what I'm good at. The daily effective time requirement is 3 to 4 hours, and you just need to accumulate task time to meet it. For example, making the bed is a simple action, about 3 minutes per entry, but you must complete 50 entries.

The device weighs about half a jin (250 grams), and wearing it for a long time makes you dizzy. Generally, after 10 entries you need a rest; if each entry takes longer, you need a break after 5. There are also quality inspection colleagues in the backend checking whether you appear on camera, the range of motion, and whether the footage and lighting meet requirements. Once proficient, basically all recorded content counts toward effective hours; if not proficient, maybe only 70-80%. This project currently only has one day of training. When I did another collection project before, it used VR headsets, charging backpacks, and a gripper in each hand, with two days of training. The grippers were the kind used in factories, each 0.5-0.6 kg, very clumsy for doing actions. I did that for about ten days before switching to the current project. To be honest, doing that long-term would really take a toll on your body. That project also had high turnover—after a day, your back ached and your hands hurt.

As for pay, honestly, it's not high. Base salary is four to five thousand, plus performance bonus, totaling about five thousand. Performance mainly depends on two things: first, whether you've completed the hours—exceeding them adds points; second, the validity of the recordings. If output doesn't meet standards, you don't get the full performance bonus. Part-time is paid by the hour. For simple tasks like ours where we don't hold grippers, it's about 20 yuan an hour. I also learned about some data collection apps where the most basic tasks pay 12 yuan an hour, and the more you do, the higher the hourly rate. I've seen people in our group doing tasks on platforms, earning 80 to 100 yuan a day.

Compared to being a headhunter, the labor intensity now is much lower. With 3 to 4 hours of effective time daily, if you're proficient, you finish in 3 to 4 hours, at most 5 hours. The rest of the time is yours to arrange. If you have something to do today, you can do it earlier—very flexible. Some people think, once robots learn, won't you be unemployed? I think that won't happen so quickly. My understanding is that even though simulated synthetic data is developing rapidly, at this stage robots still need real human demonstrations to learn these everyday actions. So I'm not too worried about this job being replaced anytime soon. However, I myself don't plan to do it long-term. It doesn't have much technical content—essentially just earning some living expenses, making the transition period feel more secure. Like screwing bolts, repeating many times a day, very boring. At our company, full-time employees get social insurance, which I think is quite practical for unemployed people and stay-at-home moms. So my positioning is clear: although I'm doing it full-time now, I only treat it as a transition. In the remaining time, I can write, do some crafts, and slowly explore where to go next.

05. One Wrong Parameter, Hundreds of Hours of Data Invalidated

Song Yu | Post-00s, Jiangsu, Head of an embodied data collection company (Dongtai Guyi Smart Technology). A while ago, a batch of data I handled was entirely invalidated—hundreds of hours—because one parameter wasn't set correctly. This batch of data was the actions we demonstrated to robots over and over while wearing collection equipment. I used to be a university teacher and also worked as a designer. Now I lead a team doing robot data collection. Getting into this industry was purely coincidental. A friend introduced me to data annotation, which naturally led to data collection. The longer I did it, the more I felt this industry matched the resources at hand. From the start, our team aimed to take on contracts, and we've been doing it for over half a year now.

My first formal training was at JD, lasting three days. This industry has a faster learning curve than outsiders imagine. People familiar with computers can understand what they're doing in a day or two; complete beginners in three to five days. The first time I teleoperated a robot, the human-machine interaction amazed me, but ultimately, as someone without a science background, I got used to it after a few operations. The collection equipment our team uses mainly includes Ego, UMI, and PICO headsets. Ego is a camera worn on the chest or head, recording the entire work process from a first-person perspective; UMI is a handheld gripper—hold it and demonstrate once, and the trajectory data is recorded; PICO uses a VR headset to teleoperate robots, where you see the robot's perspective—your hand moves left, and the robotic arm follows.

The PICO headset we use weighs about one jin (500 grams), and how long you can wear it varies by person. Some can't last an hour; others adapt well and are fine all day. Some people in the team left due to dizziness, so learning to operate the equipment doesn't necessarily mean you can adapt to this job. We currently mainly do Ego collection for home and commercial scenarios, and take on fewer projects that directly teleoperate physical robots. How many times an action needs to be repeated has almost no upper limit—it mainly depends on project requirements, and sometimes you need to collect for a long time. Work hours also vary by project: some are 6 hours a day, some are 8 hours per shift with two shifts, and some projects collect 10 hours a day. In the Ego projects we do, the best collector can produce over 9 hours of effective data in a 10-hour workday.

After data is collected, it still needs backend review and cloud platform cleaning. Rejections are mainly due to movements or footage not meeting requirements. Currently, the collectors we retain have all developed a "feel"—they know the effective range of the collector, do more meaningful actions, and avoid ineffective ones. The common problem for beginners is stiff movements. Some people, once they put on the equipment, unconsciously imitate robots. Actually, it's the opposite—when collecting, just work like a normal person. What really sets people apart is communication skills and learning speed. I think the shoulder and neck strain is somewhat heavy, but overall it's not too tiring. We also allow listening to audiobooks and chatting. The team had relatively high turnover early on, but gradually stabilized, and everyone's acceptance has increased. Here, full-time is base salary plus commission, and part-time is paid per effective hour. The unit price fluctuates with project difficulty—lower for simple projects, higher for difficult ones. The income gap among collectors isn't very large.

Leading a team involves quite a bit of extra work. After collectors finish their shifts, we often spend two or three more hours organizing data and managing equipment. Some of our projects can be done at home, and some stay-at-home moms use scattered time to earn extra income. We also cooperate with the Women's Federation and Disabled Persons' Federation to help some people achieve re-employment. People often ask, once robots learn, will they no longer need us? I'm well aware that body collection and simulated synthetic technologies are progressing rapidly, but I'm not anxious. Based on the projects we've handled, I think real human data collection still has its advantages in balancing data validity and cost. When new technologies emerge, the old methods won't immediately disappear, so I'm still fairly optimistic about this industry.