The emerging industry centers on collecting so-called “egocentric data” — first-person recordings captured through head-mounted cameras, smart glasses and motion sensors that allow AI systems to learn how humans interact with physical environments.
In Chennai, 25-year-old homemaker Nagireddy Sriramyachandra spends part of her day filming routine household chores, including food preparation, for AI data company Objectways.

She earns about 250 rupees ($3) an hour for the recordings, which are uploaded through a specialized application and used to train robotic systems.
Developers believe that exposing AI models to large volumes of human behavioral data will help robots perform real-world tasks more effectively, particularly in domestic and industrial settings where navigation and object manipulation remain significant technological challenges.

Objectways, which operates in India and the United States and works with machine-learning platform Amazon SageMaker, collects videos requested by corporate clients seeking to train robotics systems for activities such as folding clothes, preparing beverages and performing basic kitchen tasks.

The business reflects growing investor interest in humanoid robotics. Morgan Stanley has projected that more than one billion humanoid robots could be in operation worldwide by 2050, primarily in commercial and industrial applications.
Objectways Chief Executive Ravi Shankar said automation would eventually assume responsibility for some routine jobs, allowing workers to focus on more advanced activities.

The company’s operations extend beyond homes. At a textile factory in Tamil Nadu’s Karur district, workers wearing smart glasses and head-mounted cameras record manufacturing processes while carrying out routine production work.

India has increasingly positioned itself as a global center for AI data collection, annotation and processing services, leveraging its large labor force and established technology sector.

Aditi Surie, a digital labor researcher at the Indian Institute for Human Settlements in Bengaluru, said demand for data collection services linked to AI development is likely to expand as companies seek increasingly sophisticated training datasets.

The rise of such work has also intensified debate over automation’s impact on employment in a country where informal labor remains a major component of the economy.

Government policy think tank NITI Aayog has warned that discussions surrounding artificial intelligence often focus on highly skilled workers while overlooking implications for India’s estimated 490 million informal workers.

Ahead of an international AI summit hosted by India this year, the institution examined how artificial intelligence could affect a wide range of occupations, including agricultural laborers, street vendors, cobblers and sanitation workers.

Among those participating in data collection projects is Ponni, a 55-year-old flower garland maker in Bengaluru who has spent years working in the informal economy. She has also been paid to wear a recording device while carrying out her daily activities.
She expressed concern that future workers engaged in similar occupations could face growing competition from automated systems trained on human-generated data.

At an Objectways recording studio, workers repeatedly film themselves performing household tasks in fully furnished mock apartments designed to generate varied training material. Employees record dozens of short clips daily while changing positions, angles and environments to increase the diversity of datasets supplied to clients.

Rani N., a 21-year-old engineering graduate employed as an AI system trainer, said she records roughly 90 videos each day, often repeating the same activity in different locations within a room.

Elsewhere, workers arrange objects such as water bottles, crayons and office supplies while specialized depth-sensing cameras capture movement and spatial relationships.

Qanat Consulting Services, a subcontractor based in Andhra Pradesh, supplies recordings to larger AI data firms through a network of around 2,000 contributors. Some participants wear motion-tracking bands on their hands, wrists and legs to provide more detailed movement information.

Other companies are expanding beyond visual data. Bengaluru-based Humyn Labs gathers audio recordings and conversations on assigned topics to help clients train systems capable of understanding speech patterns and human interaction.

Humyn Labs founder Manish Agarwal said he expects future workplaces to combine human expertise with robotic capabilities rather than replace workers entirely.

He said advances in robotics and artificial intelligence could eventually allow workers in one country to supervise automated systems operating thousands of kilometers away, creating new forms of cross-border employment.

Over 70 teams—almost five times the number in 2025—are set to participate in the event, which will feature a more demanding course including paved slopes and parkland terrain designed to test improvements in durability, balance and battery performance.“It will certainly be interesting to see the progress in durability of components and battery lifetime compared to last year,” said Georg Stieler, Asia managing director at a technology consultancy.

He added that manufacturers continue to face pressure to balance product quality with cost as the technology evolves.Organizers said the race marks a shift from last year, when all participating robots were remotely controlled. In contrast, a significant share of entrants this year will rely on onboard sensors and algorithms to complete the course independently, highlighting gains in perception and decision-making systems.

Among the contenders is Tiangong Ultra, developed by the Beijing Innovation Center of Humanoid Robotics in collaboration with UBTech. The robot, which won last year’s race in 2 hours and 40 minutes, is expected to run fully autonomously this time, using sensor-based navigation and data-driven gait modeling.

Developers said achieving human-like running speeds presents significant technical challenges due to the limited time available for real-time perception and response. Training footage shared on Chinese social media shows some robots reaching speeds of up to 14 km per hour, though others displayed instability, with occasional falls and collisions.

China remains the dominant player in humanoid robotics deployment, accounting for more than 80% of the roughly 16,000 units installed globally in 2025, according to Counterpoint Research. By comparison, U.S.-based Tesla held about 5% of installations.

Domestic firms including AgiBot and Unitree each shipped over 5,000 units last year, with Unitree planning to scale annual production capacity to 75,000 robots.Despite rapid growth, industry experts say humanoid robots remain far from widespread commercial adoption in industrial environments, where precision, adaptability and complex task execution are required.

Current applications are largely limited to research, demonstrations and service roles such as interactive guides.“The reason our applications aren’t taking off is that the robots’ IQ is too low. The models are poor, their success rates are low,” said Tang Wenbin, founder of embodied intelligence startup Yuanli Lingji, speaking at a recent Beijing forum.

The Chinese government has identified embodied intelligence, or physical AI, as a strategic sector to enhance productivity and modernize manufacturing. Companies are investing heavily in data collection and machine learning, often using human workers equipped with sensors to train robotic systems.

UBTech said it expanded the number of humanoid robots deployed in factories from fewer than 10 in 2024 to more than 1,000 last year, and aims to launch 10,000 full-size units in 2026, including models tailored for commercial use, according to its chief business officer Michael Tam.