People Filming Chores for Robots
A growing gig economy is paying workers in 50+ countries roughly $15/hour to film themselves doing everyday chores so that companies can build large, labeled datasets for humanoid robots. That distributed data collection is feeding the 'data flywheel' many robotics labs say they need to train generalist embodied models. (siliconcanals.com)
What makes this notable is where the training is happening: not in robot labs, but in ordinary apartments and kitchens. Micro1, a Palo Alto startup, has been recruiting people in countries including Kenya, the Philippines, India, Brazil, and Nigeria to wear cameras and slowly film chores like folding towels, loading dishwashers, and wiping counters, then sell that footage to robotics companies. (siliconcanals.com, technologyreview.com) The point is not to show a robot the finished result, but to capture all the messy little motions humans make along the way. A person opening a refrigerator, shifting plates to make room, adjusting grip on a slippery glass, or moving around clutter gives robot makers something they still struggle to get from clean lab demos: examples of how real homes actually work. (technologyreview.com, micro1.ai) That is the “data flywheel” robotics founders keep talking about: more human demonstrations create better robot behavior, better robot behavior creates more deployments, and more deployments create more data. Micro1 says its product is “high-quality human data” for factories, households, and warehouses, including “fine-grained action segmentation,” which means breaking a task into labeled micro-steps like reach, grasp, twist, place, and release so a model can learn not just what happened but how it happened. (micro1.ai, figure.ai) That helps explain why household chores matter so much to humanoid companies. Figure said in February 2025 that its Helix system — a “vision-language-action model,” meaning software that turns camera input and text instructions directly into robot movements — could handle unfamiliar household objects, and in January 2026 it said Helix 02 completed a four-minute dishwasher task across a full kitchen with no human intervention after training on more than 1,000 hours of human motion data. (figure.ai, figure.ai) The labor model is spreading beyond specialist data firms. DoorDash launched a separate Tasks app on March 19, 2026 that pays couriers to complete assignments aimed at improving “AI and robotic systems,” including filming themselves washing dishes with a body camera, which shows how fast this kind of work is becoming a standard gig platform category rather than a niche robotics job. (techcrunch.com) The unresolved part is ownership and control of the footage. MIT Technology Review reported that workers it interviewed understood they were helping train robots, but did not know exactly how their videos would be stored, shared, or reused by third parties, which means the raw material for home robots is being gathered inside private homes by contractors who are paid once and usually get no continuing claim on the data after that. (technologyreview.com)
