In Chennai, southern India, housewife Nagireddy Sriramyachandra spends her days filming herself performing routine household tasks, such as slicing mangos, to help train artificial intelligence-powered robots designed for future domestic use. Equipped with a smartphone mounted on her head, Sriramyachandra records videos that are then used by global technology companies to teach machines how to replicate human movements in real-world settings. She earns just over $2 per hour for this work, which she describes as a reasonable income given that the tasks resemble her usual household chores.

Sriramyachandra is part of a growing workforce of thousands of AI trainers across India who provide first-person, egocentric video data crucial for developing humanoid robots capable of navigating and operating within human environments. While AI chatbots and image generators analyze vast digital datasets, the challenge of enabling robots to function effectively in physical spaces requires specialized training data captured through head-mounted cameras, video glasses, and motion sensors.

Some AI trainers operate from their homes, while others work in controlled studio environments, such as those run by Objectways, an AI data company based in India. In these studios, trainers like 21-year-old engineering graduate Rani N. perform repetitive household activities—such as folding towels—within fully furnished, mock apartment rooms that are periodically rearranged to offer varied visual backgrounds. Rani typically records around 90 videos daily, each about four minutes long, adjusting positions and angles to capture a comprehensive set of human movements.

These recordings extend beyond simple videos, as some contributors employ motion-sensor bands on their wrists, hands, and legs to deliver detailed spatial data. Qanat Consulting Services, a subcontractor in Andhra Pradesh, supplies such movement-rich recordings to about a dozen large data firms, drawing from a contributor base of roughly 2,000 individuals.

The humanoid robot sector is gaining momentum, with investment bank Morgan Stanley forecasting that more than a billion robots could be in use globally by 2050, predominantly in industrial and commercial settings. Ravi Shankar, head of Objectways, highlighted the range of household tasks robots are being trained to perform, including folding clothes, making coffee, preparing specific dishes, and assembling sandwiches.

Beyond physical task training, Bengaluru-based Humyn Labs focuses on capturing speech and conversations. Manish Agarwal, the company's founder, noted that contributors engage in discussions on diverse subjects, from politics to entertainment, to help train AI systems in processing speech patterns. Agarwal expressed optimism that rather than displacing human workers, robots will serve as collaborative tools, envisioning a future where roles such as welders in India could operate welding robots remotely in locations like Prague.

As AI technology advances, these human trainers play a critical role in bridging the gap between digital intelligence and physical interaction, providing essential data to fuel the growing capabilities of humanoid robots intended to assist in household and industrial tasks worldwide.