The Hidden Gig Economy Training Humanoid Robots

What Is Egocentric Video Data and Why Does It Matter?

Unlike the text or static images used to train large language models, humanoid robots need continuous, first-person footage of real human movement to learn how to grasp objects, navigate kitchens, fold shirts, and scrub dishes . This is called egocentric data—video recorded from the perspective of the person performing the action. The camera is mounted on the head, pointed down at the hands, so the AI can learn the precise angles, grip pressures, and timing of each movement .

Companies pay workers to strap an iPhone to their forehead and perform tasks for sessions lasting tens of minutes each. The video is then uploaded, and both human and AI reviewers annotate it frame-by-frame, marking key moments like hand grasps, object releases, and task completions .

The scale is staggering: Figure AI alone manufactured approximately 150 humanoid robots in April 2026, up from single digits months earlier, and needs billions of hours of real-world motion data to train them .

Extreme Pay Disparities Across Countries

The compensation for this work varies dramatically depending on location, revealing a stark global labor arbitrage:

• Nigeria: Workers can earn around $15/hour — a strong local wage in a struggling economy . Zeus, a Nigerian medical student and Micro1 contractor, describes the pay as meaningful in a country with high unemployment .
• India: Reported rates range from ~$2.00–$2.60/hour (roughly 150–250 rupees), a fraction of the Nigerian rate . Nagireddy Sriramyachandra, a 25-year-old housewife in Chennai, earns just over $2/hour recording herself slicing mangoes .
• United States (Los Angeles): Some local crews are paid $80 for two hours ($40/hour) . Salvador Arciga, a veteran of the LA gig economy, earned that amount for footage of himself washing dishes and scrubbing toilets .
• Human Archive (India): One startup paid workers as little as $1/hour for camera-equipped headset data .

Micro1's founder Ali Ansari told the Los Angeles Times that the company collects data in Brazil, Argentina, India, and the United States . Objectways founder Ravishankar Rajalingam told CNN that roughly half the submitted footage is unusable .

Workers Lack Informed Consent and Knowledge of Footage Use

A recurring finding across multiple investigations is that workers have little to no idea how their recordings will be used, who will ultimately buy them, or that they are training the humanoid robots that could one day replace their own jobs .

The MIT Technology Review investigation highlighted that workers receive a filming guide and task list but rarely meaningful information about the downstream AI applications . Some workers understood the footage would train robots but did not know which company, robot model, or how their recordings would be stored or shared .

As one Indian factory worker told The Guardian, "Who is going to pay us when we're replaced by robots?" .

Unintended Surveillance of Family Members and Private Spaces

Cameras strapped to workers' heads inevitably capture spouses, children, other family members, and intimate home environments—kitchens, living rooms, bedrooms—without their explicit consent . The first-person perspective is inherently intrusive: it records not just the task but the entire visual context of a person's private home .

Micro1 requires workers not to show their faces, not to say their names or phone numbers, and claims to use AI plus human reviewers to filter sensitive information . But even with those safeguards, the recordings capture workers' home interiors, personal belongings, and daily routines . Some participants refuse to film in bedrooms, and labor activists have raised the question of whether people whose movements are encoded into commercial datasets should receive ongoing compensation or royalties .

Workers Are Training Their Own Replacements

A recurring sentiment among workers and commentators: these same recordings are being used to train humanoid robots for factory and domestic labor—the very jobs many of these workers currently hold . In a garment factory in Tamil Nadu, a worker folds a shirt with a small camera mounted on her head, recording the exact angle of her wrist, the grip pressure of her fingers, and the millisecond timing of each fold . That footage flows into data pipelines serving robotics labs linked to OpenAI, Nvidia, and Google .

The scale of the industry extends far beyond Micro1 and Objectways. India is fast becoming a crucial hub in the global race to collect egocentric data, with a growing ecosystem of firms including Humyn AI, FPV Labs, Egodata, Neocambrian, XP Robotics, Scale AI, and CynLr all building data pipelines for robotics companies .

Regulatory and Privacy Investigations Are Emerging

India's Ministry of Electronics and Information Technology has launched a privacy probe into startups (including Human Archive) that record inside workers' homes with camera headsets, focusing on consent practices . The review was opened after home-services startup Pronto confirmed it had tested opt-in video recording .

No comprehensive global labor or data-privacy framework yet governs this cross-border practice . The humanoid robot training data market operates in a regulatory gray zone, spanning dozens of countries with varying labor laws and privacy protections.

Instawork, best known for matching hourly workers with shifts at hotels and warehouses, launched a system called Instacore that includes five cameras mounted on the head, chest, and wrists, connected to a compute backpack for eight-hour shifts of recording tasks like chopping vegetables . Tesla's own glass-walled lab in California has workers in helmet-mounted cameras and backpacks, repeating actions hundreds of times during eight-hour shifts to train the Optimus robot .

The Bottom Line

This practice sits at the center of a triple tension: valuable, well-paying work for some (e.g., $15/hr in Nigeria) versus poverty-level wages for others (as low as $1–$2.60/hr in India); a gold-rush data market for humanoid robotics versus near-zero transparency about how footage is used; and the capture of intimate household moments versus weak or absent consent protocols. The scale is vast—spanning 50+ countries—and the regulatory response, most notably in India, is only just beginning.