Thousands of contract workers in over 50 countries are recording themselves doing household chores to train humanoid robots. Micro1, a Palo Alto-based data company, pays workers like Zeus, a Nigerian medical student, $15 an hour to film tasks such as folding laundry and ironing clothes. The footage helps robotics companies like Tesla and Figure AI teach robots to navigate and manipulate objects in real-world environments.

This gig economy approach has emerged as companies race to build humanoid robots capable of factory work and household tasks. The rise of large language models has inspired a similar shift in robotics, where massive datasets of human movements could train robots to interact with the physical world. Investors have poured over $6 billion into humanoid robots in 2025, with companies spending more than $100 million annually on real-world training data.

Workers face challenges including privacy concerns, limited space for varied tasks, and the repetitive nature of the work. While companies like Micro1 use AI and human reviewers to screen footage, questions remain about data usage, storage, and long-term implications. Some roboticists worry about the reliability of home-recorded data for training safety-critical systems. As this approach matures, the industry must balance the need for diverse training data with concerns about worker consent and data quality.