I sold videos of myself doing housework to robots for AI training, earning $6.6 per hour — below minimum wage

Wired reporter Reece Rogers strapped an iPhone to his forehead and spent a week recording first-person perspective videos of himself doing chores, uploading them to the data collection platform Luel to train humanoid robots. (Previous coverage: Unitree Robotics prospectus revealed: Humanoid robot shipments rank first globally, commercialization relies on schools and research institutions) (Background: Can you get a robot to wash dishes and sweep the floor just by asking? Figure launches AI model "Helix" aiming for a household chore revolution) Reece Rogers spent a full week with an iPhone strapped to his forehead, recording first-person perspective videos of himself doing household chores in exchange for an hourly wage of $6.60, which is below the U.S. federal minimum wage of $7.25. The employer of this Wired reporter is a data collection platform called Luel. Every action he took—slicing cucumbers, pouring drinks, loading dishes into the dishwasher—would eventually be packaged and sold to AI companies training humanoid robots. Rogers himself is well aware that he might be training something that will sooner or later replace humans. In his report, he described the entire process as both embarrassing and monotonous. According to Rogers, Luel’s technical requirements are strict: the recording device must be head-mounted, handheld is not allowed; the angle must be a horizontal wide-angle; and the resolution must be at least 1080p. The most critical rule: both hands must be visible in the frame for more than 95% of the time. His first 5-minute video was rejected by the platform because his hands were not prominent enough in the frame. He then adjusted his filming method and began repeating the same action: tying shoelaces, which he did 20 times. Making salads, washing dishes, slicing cucumbers—every action required his hands to continuously occupy the main position in the frame. This type of video is known in the industry as "first-person perspective data." Simply put, it involves mounting a camera at eye level, allowing robots to observe the physical world from a "human perspective": how hands approach objects, how fingers apply force, and how the gaze shifts to the next task after completing an action. Text descriptions cannot convey this information; only continuous first-person footage allows robots to understand how actions occur. This also explains why the platform's requirements are so precise: having both hands visible 95% of the time is not for aesthetics, but because robot models need to extract the complete sequence of hand movements from the video. Once the hands disappear, the value of the video for training purposes drops significantly. Luel offers an hourly wage of $6.60, with each video paying about $1 to $2. This figure is on the low end of the gig economy, but it is not the only reference point. Larger platforms offer much better conditions. Rogers mentioned in his report that companies like DoorDash pay an hourly rate of about $20 for similar videos, more than three times that of Luel. The gap between these two figures reflects how different buyers judge the urgency of the data: robot companies closer to commercial mass production are willing to pay a higher price for data acquisition costs. Behind Luel's low prices lies another layer of structure. The data it collects is not for its own use, but is packaged and resold to robot and AI companies as an intermediary. In this chain, the workers providing the videos receive the lowest compensation, while the licensing fees paid by the final buyers may be several orders of magnitude higher. What Rogers recorded was not just his own hourly wage, but the way labor is priced in an emerging data market that is currently taking shape. This type of work is currently scattered across multiple platforms, with no unified pricing standard and no labor protection framework to include it. $6.60 is below the federal minimum wage, but this is not illegal because the money is defined as "video sales revenue" rather than "wages." The irony of the whole situation is that the robot that will replace you in the future may have been partially trained by you yourself. And the fee you are currently receiving is even lower than the legal minimum wage...