The people training the robots are not in the factory.
Zeus is a medical student in central Nigeria. Every night he sets up a ring light in his studio apartment, straps an iPhone to his forehead, and records himself folding laundry, ironing clothes, and making his bed. He moves slowly and carefully, keeping his hands in frame at natural speed. He is paid $15 an hour. He finds the work boring.
He is also, according to the robotics industry, one of the most important data collectors in the supply chain.
A feature published by MIT Technology Review this week describes how Micro1, a Palo Alto-based data company, has hired thousands of gig workers in more than 50 countries to record themselves doing household chores. The footage is annotated, processed, and sold to robotics companies including Tesla, Figure AI, and Agility Robotics. The workers mount iPhones on their heads, film themselves completing the same tasks hundreds of times in slightly different ways, and submit the recordings weekly for review by an AI agent named Zara and a human oversight team. The company estimates robotics companies are spending more than $100 million each year on real-world movement data of this kind.
The business model is straightforward. Humanoid robots are hard to train because physical manipulation requires genuine real-world data, and simulations cannot model the full complexity of how objects feel, break, or respond to force. The answer the industry has converged on is to record thousands of humans doing the same tasks and use that footage to train the robots. The work is repetitive, low-paid by American standards, and invisible.
The privacy questions are real. Micro1 instructs workers not to show their faces, names, or phone numbers. But the recordings still capture the interiors of workers' homes, their possessions, their daily routines. Workers told MIT Technology Review they do not know how their data will be used beyond the broad statement that it trains robots, or who it is shared with. Micro1 does not disclose its robotics company clients to workers. One worker, a banker turned data recorder in Nigeria, said she tiptoes around her residential compound so neighbors do not appear in frame. Another, a father in Delhi, has to keep his two-year-old daughter out of shot constantly.
"While the workers understand that their data is being used to train robots, none of them know how exactly their data will be used, stored, and shared with third parties," MIT Technology Review reported.
The labor economics are also worth noting. In Nigeria, where Zeus lives, $15 an hour is good money in an economy with high unemployment. The work is boosting local incomes. It is also weird, isolating, and by all accounts tedious. The promise of training robots that will eventually do these tasks for a living is not lost on the workers. Zeus, who wants to become a doctor, told MIT Technology Review he finds ironing clothes boring but believes the work matters. Arjun, a tutor in Delhi, said he spends nearly as much time thinking about what to film as he does filming.
Ken Goldberg, a roboticist at UC Berkeley who studies manipulation and dexterity, told MIT Technology Review: "It is going to take longer than people think."
The $6 billion that investors poured into humanoid robotics in 2025 is real. The race to build robots that can fold laundry, do dishes, and navigate a home is real. The workers making that possible are also real. They just are not in the factory.
They are in a studio apartment in Nigeria, an iPhone strapped to their forehead, hands in frame, filming the same motion for the fortieth time.