¥120 a day: inside the job of collecting data for robots
Dingjiao One spoke with five people working as embodied-AI data collectors, who spend their days teleoperating robots with VR headsets or recording household chores — folding clothes, making beds, washing dishes, labelling parcels — via UMI handheld grippers or Ego first-person cameras. Pay is tied to valid recorded hours: interviewees cited ¥120 for a day shift and ¥135 for a night shift, ¥150 after moving to parcel-labelling, about ¥5,000 a month for full-time at-home Ego collection and roughly ¥20 an hour for part-time work, with some platform tasks as low as ¥12 an hour. Workers described repetitive, physically tiring work with high turnover, daily quotas of at least three valid hours and around 85% accuracy; one team lead said a single misconfigured parameter voided several hundred hours of data.
Human-demonstration data remains a core input for training embodied AI, so the cost and churn of this labour-intensive collection work shape how quickly robot learning can progress.