Local Chorus
Local news from local sources, read in your language.
Settled

Robot data collectors in China describe 120-yuan days teaching robots household chores

🇨🇳 China 15:57 Society Business2 Tech updated 3 d ago first reported by 虎嗅

In short

Huxiu, Jiemian and TMTPost published a feature on robot data collectors, workers who repeatedly perform tasks such as folding clothes, washing dishes and labelling parcels so the recorded data can train robots. One part-time collector earned 120 yuan a day and quit after a month, while Huxiu put full-time pay at about 5,000 yuan a month in base pay plus performance pay.

Read the full story 2 min read

Huxiu, Jiemian and TMTPost on 9 October 2026 published a feature, titled "120 yuan a day, I am a 'teacher' for robots", on robot data collectors. The Jiemian and TMTPost versions credit Dingjiao One (定焦One). The collectors repeatedly demonstrate tasks such as folding clothes, making beds, washing dishes and labelling parcels; the recorded data is reviewed, filtered and used to train robots. Huxiu said the job has emerged with the development of embodied intelligence and that the article interviewed five workers. [ 1 , 2 , 3 ]

According to the feature, data is collected in three main ways: teleoperation, in which collectors wear VR headsets to operate real robots; handheld grippers such as UMI, which record movement paths without a real robot; and first-person camera footage, known in the industry as Ego data, recorded while collectors do housework. [ 1 , 2 , 3 ]

A part-time collector who took the job during the summer holiday described two days of training, then wearing a VR headset and a charging backpack with a gripper in each hand. Shifts were 8 hours, six days on and one off, at 120 yuan for the day shift and 135 yuan for the night shift. Workers had to log at least 3 hours of valid footage a day at 85% accuracy, and clips had to be at least two minutes long. The collector said folding one garment often had to be repeated more than a dozen times, compared the work to tightening screws in a factory, and left after a month. Huxiu identified this worker as Xiao Sun (小孙) in Shanxi. The collector said turnover was high, with most workers under 30 and many leaving after two weeks. [ 1 , 2 , 3 ]

Zheng Zheng (郑郑), 26, works at Lingyu Zhineng (灵御智能) in Beijing. A former IT worker, Zheng Zheng has done the job for more than half a year, operating robots remotely through a headset to teach them to fold towels and boxes, and named putting on pillowcases as harder. Huxiu said Zheng Zheng finds the work rewarding and that the robot imitates even unconscious movements. [ 1 , 2 , 3 ]

Huxiu also described Li Zhe (李哲) in Guangzhou, who earns 150 yuan a day washing dishes and applying labels and has not seen a robot in months; Xiao Ran (小然) in Chongqing, who works from home wearing a camera and earns about 5,000 yuan a month; and Song Yu (宋宇), head of a data collection company in Jiangsu, whose team once lost several hundred hours of data because of a parameter error. Huxiu put full-time pay at about 5,000 yuan a month and part-time pay at about 20 yuan an hour, and said some workers worry about being replaced. [ 1 ]

Why it matters

Huxiu links the rise of the job to the development of embodied intelligence (具身智能). It says the work has high turnover and suits people in transition rather than long-term careers. It also reports that some workers worry about being replaced, though human data collection currently still has advantages.

Key facts

  • Robot data collectors record movements using VR headsets, UMI-type handheld grippers or first-person cameras, and the data is used to train robots. [ 1 , 2 , 3 ]
  • Data from collectors is reviewed and filtered before it is used to train robots. [ 2 , 3 ]
  • A part-time collector described 8-hour day and night shifts, six days on and one off, paid 120 yuan for a day shift and 135 yuan for a night shift. [ 2 , 3 ]
  • Collectors at that workplace had to log at least 3 hours of valid footage a day with 85% accuracy; rejected clips did not count. [ 2 , 3 ]
  • Zheng Zheng (郑郑), 26, works for Lingyu Zhineng (灵御智能) in Beijing, teaching robots by teleoperation. [ 1 , 3 ]
  • A collector working from home earns about 5,000 yuan a month and treats the job as a transition. [ 1 , 2 , 3 ]
  • A team lost several hundred hours of data because one parameter was set wrongly. [ 1 , 2 , 3 ]
  • Huxiu put part-time pay at about 20 yuan an hour. [ 1 ]

Confirmed by several sources

  • Collectors use teleoperation with VR headsets, handheld grippers such as UMI, or first-person cameras to record data for training robots. [ 1 , 2 , 3 ]
  • One collector works at home doing housework and earns about 5,000 yuan a month, treating it as a transition. [ 1 , 2 , 3 ]
  • A collector leading a team lost several hundred hours of data because of a wrong parameter. [ 1 , 2 , 3 ]
  • A collector worked for several months without seeing a robot at the workplace. [ 1 , 2 , 3 ]
  • A part-time collector was paid 120 yuan a day for the day shift. [ 1 , 2 , 3 ]
  • Zheng Zheng works at Lingyu Zhineng in Beijing and teaches robots to fold towels by teleoperation. [ 1 , 3 ]

Still unclear

  • Pay of 150 yuan a day for Li Zhe (李哲) in Guangzhou, and the location of the part-time worker Xiao Sun (小孙) in Shanxi. Only Huxiu's summary gives these names and places; the Jiemian and TMTPost texts available here are cut off or do not name them.
  • Huxiu's figures for typical pay: about 5,000 yuan a month for full-time staff and about 20 yuan an hour for part-time work. Only one outlet gives these general figures.
  • How widely the conditions described apply across the industry. The feature rests on interviews with five workers.

What local media are saying

Business mediaHuxiu summarised the five workers' accounts and concluded the job suits people in transition rather than long-term careers; Jiemian ran the Dingjiao One feature with first-person accounts of pay, shifts and physical strain. [ 1 , 2 ]
Technology mediaTMTPost ran the same Dingjiao One feature, including the account of Zheng Zheng at Lingyu Zhineng and the comparison of the work to tightening screws on a factory line. [ 3 ]

Timeline, local time

  1. Huxiu publishes its account of five robot data collectors. [ 1 ]
  2. Jiemian publishes the Dingjiao One feature. [ 2 ]
  3. TMTPost publishes the same feature. [ 3 ]