Silicon Valley Startup Uses Indian Gig Workers to Train Robots

Human Archive, a startup founded by graduates of the University of California, Berkeley, and Stanford, has found an unconventional solution to one of robotics' key challenges: the lack of high-quality training data. The company collaborates with Indian service platforms, such as food delivery and household services, to collect video recordings and sensor data from the workers' perspective. This is achieved using specialized cameras mounted on headgear, as well as tactile gloves and motion-capture suits.

Human Archive has already raised 586,7 млн ₽ from leading venture capital firms, including Wing Venture Capital, NVP Capital, and Y Combinator, as well as investors from OpenAI, Nvidia, and Meta. According to the founders, the startup has deployed over 1,000 data-collection devices across various regions in India. However, not all companies are willing to collaborate: for instance, Urban Company and Pronto declined partnerships, sparking public debate on social media.

To differentiate itself from competitors, Human Archive uses not only video recordings but also synchronized data from multiple sensor types, including depth cameras (RGB-D), force sensors, and motion-capture systems. This enables the creation of more valuable datasets for robot training. The company also tests its own AI models on the collected data to demonstrate their effectiveness to potential clients.

Despite criticism and privacy concerns, the startup continues to expand. In India, clients are offered discounts on services in exchange for consenting to data recording. Workers receive compensation for participating in the project, though the amounts are lower than those offered by competitors. Human Archive also plans to launch a global data-collection platform, targeting markets in the U.S. and Southeast Asia.