A Med Student in Nigeria Films Folding Shirts for $15/Hour. He's Critical Infrastructure.

Robotics companies training humanoid robots are relying on a global gig workforce of thousands in developing countries to collect real-world manipulation data by filming themselves doing household chores. This data—annotated and processed by companies like Micro1—sells for hundreds of millions annually to firms including Tesla, Figure AI, and Agility Robotics, who find real-world movement footage essential for training physical AI systems that simulations cannot replicate. The arrangement highlights a labor-intensive, low-paid dependency at the foundation of physical AI development, with limited transparency around data use and worker privacy.

• •Humanoid robot training remains fundamentally dependent on human-generated video data rather than simulation, with robotics firms spending over $100M annually on real-world movement recordings.
• •Gig workers in countries like Nigeria and India are paid $15/hour to film repetitive domestic tasks, creating a new invisible labor category that underpins physical AI development.
• •Data collection practices raise significant privacy concerns: workers film the interiors of their homes without clear disclosure of downstream clients, storage practices, or third-party sharing.