Humanoid robots are being trained by an invisible army of gig workers recording themselves folding laundry, washing dishes, and ironing clothes in their homes across the developing world. Companies like Tesla, Figure AI, and Agility Robotics are racing to build robots that can work in factories and homes, and they need massive amounts of real-world movement data to teach these machines how to interact with physical objects. Rather than relying on expensive in-house teams or simulations, robotics companies are turning to a booming gig economy where workers in over 50 countries are mounting iPhones on their heads and submitting videos of everyday tasks .

Why Can't Robots Learn From Simulations Alone?

The challenge facing robotics engineers is fundamentally different from training large language models (LLMs), which are AI systems trained on vast amounts of text data. While AI chatbots like ChatGPT learned to generate text by analyzing internet text, humanoid robots need to understand how the physical world actually works. Virtual simulations can teach robots to perform acrobatics, but they struggle to model physics with perfect accuracy, making it nearly impossible for robots to learn how to grasp and move real objects .

This gap between simulation and reality has created an urgent demand for authentic movement data. Investors are pouring money into solving this problem, spending over $6 billion on humanoid robots in 2025 alone . Robotics companies are now spending more than $100 million each year to purchase real-world data from companies like Micro1, Scale AI, and Encord .

Who Are These Data Recorders, and What Does the Work Actually Look Like?

Meet Zeus, a medical student living in central Nigeria who returns home from the hospital each day and straps an iPhone to his forehead to record himself doing chores. He is one of thousands of contract workers hired by Micro1, a Palo Alto-based company that collects real-world movement data to sell to robotics firms. Zeus earns $15 per hour, which is solid income in Nigeria's economy where unemployment rates are high . Yet despite the decent pay, he finds the repetitive work tedious. "I really do not like it so much," Zeus explained. "I'm the kind of person that requires a technical job that requires me to think" .

The work process is highly structured. Workers submit videos each week showing themselves performing household tasks, following specific instructions like keeping their hands visible and moving at natural speed. An AI agent named Zara conducts initial interviews and reviews video samples to vet new workers. Once videos are submitted, they are reviewed by both artificial intelligence systems and human reviewers, then either accepted or rejected. Accepted videos are annotated by AI and a team of hundreds of humans who label the actions in the footage .

The challenge for workers is creating enough variety in their content. Arjun, a tutor in Delhi, India, spends an hour brainstorming new chores just to produce a 15-minute video. "How much content can be made in the home? How much content?" he asked, highlighting the difficulty of generating diverse training data from limited home environments .

How Are Companies Using This Movement Data to Train Robots?

• Real-World Variation: Workers in different homes with different layouts and possessions provide the robot training systems with thousands of variations in how basic tasks can be performed, helping robots generalize their skills across diverse environments.
• Complex Manipulation Training: Videos of workers handling objects like dishes, laundry, and kitchen items teach robots the subtle physics of grasping, lifting, and moving items without dropping or damaging them.
• Safety and Efficiency Data: By analyzing how humans naturally perform household tasks, AI systems can learn efficient movement patterns and safety considerations that simulations cannot accurately model.

"You need to give lots and lots of variations for the robot to generalize well for basic navigation and manipulation of the world," said Ali Ansari, CEO of Micro1.

Ali Ansari, CEO of Micro1

This approach represents a fundamental shift in how robotics companies train their systems, moving away from expensive controlled environments toward crowdsourced real-world data.

What Privacy and Consent Issues Are Emerging?

While the work provides income to thousands of people in developing economies, it raises serious questions about privacy and informed consent. Micro1 asks workers not to show their faces or reveal personal information like names and phone numbers, and the company uses AI and human reviewers to remove such details. However, even faceless videos capture intimate details of workers' lives: the interiors of their homes, their possessions, their daily routines, and sometimes their families .

Sasha, a banker turned data recorder in Nigeria, must carefully position herself when hanging laundry outside in a shared residential compound to avoid recording her neighbors, who watch her work in bewilderment. Arjun, a father of two daughters, struggles to keep his small children out of frame during recording sessions. "Sometimes it is very difficult to work because my daughter is small," he said .

Perhaps most troubling is the lack of transparency about how data will be used. None of the workers interviewed by MIT Technology Review knew exactly how their data would be stored, shared with third parties, or used by the robotics companies purchasing it from Micro1. For confidentiality reasons, Micro1 does not disclose to workers which companies are buying their data or the specific nature of the projects they are contributing to .

"It is important that if workers are engaging in this, that they are informed by the companies themselves of the intention and where this kind of technology might go and how that might affect them longer term," explained Yasmine Kotturi, a professor of human-centered computing at the University of Maryland, Baltimore County.

Yasmine Kotturi, Professor of Human-Centered Computing at the University of Maryland, Baltimore County

Occasionally, workers have asked on company Slack channels whether Micro1 could delete their data, but the company declined to comment on whether such deletion requests are honored .

What Do Experts Say About Data Quality and Safety?

Not all roboticists are convinced that home-recorded data is reliable enough for safe robot training. Aaron Prather, a roboticist at ASTM International, raised concerns about the quality of data collected from workers' homes. "How we conduct our lives in our homes is not always right from a safety point of view," Prather noted. "If those folks are teaching those bad habits that could lead to an incident, then that is not good data" .

The sheer volume of data being collected also makes quality control challenging. Micro1 says it rejects videos that do not meet standards, but with thousands of workers submitting content weekly, ensuring consistent quality across all submissions remains an ongoing challenge. As the robotics industry races to scale humanoid deployment, the question of whether crowdsourced home data can meet safety standards for real-world robot operation remains unresolved.

Steps to Understand the Gig Economy Behind AI Training

• Recognize the Scale: Over 50 countries now have workers participating in this gig economy, with thousands of contract workers recording household tasks to generate training data for robotics companies.
• Understand the Economics: Workers earn $15 per hour, which represents significant income in developing economies like Nigeria and India, but the work is often repetitive and requires constant creativity to generate diverse content.
• Know the Data Flow: Videos are submitted weekly, reviewed by AI and human teams, annotated with labels describing actions, and then sold to robotics companies for training their humanoid systems.
• Consider the Implications: This model raises questions about worker privacy, informed consent, data ownership, and whether home-recorded data is safe enough to train robots that will eventually work in real-world environments.