This startup is betting India’s gig economy can train the world’s robots

In recent years, India’s online food delivery landscape has seen remarkable growth, with companies like Zomato and Swiggy going public and a rise in cloud kitchens. Concurrently, home service startups such as Urban Company, Snabbit, and Pronto are gaining traction in the gig economy. Capitalizing on this momentum, Silicon Valley-based startup Human Archive is partnering with these platforms to have workers don specialized caps equipped with cameras. This initiative aims to gather first-person video data of daily tasks that can be utilized for training robots. While the startup has not disclosed specific collaborators, it claims to be working with companies in sectors like home services, hospitality, and restaurants, boasting over 1,000 active headsets in various locations. Recently, Human Archive announced it has secured $8.2 million in funding from investors including Wing Venture Capital, NVP Capital, and Y Combinator, alongside contributions from notable names in technology such as OpenAI and Nvidia. Founded by a group of four students from Berkeley and Stanford—Samay Mani, Rushil Agarwal, Shloke Patel, and Raj Patel—Human Archive is strategically positioned within the evolving AI landscape. As the race intensifies among robotics labs and AI firms to create machines capable of performing real-world tasks, a significant challenge remains: the lack of high-quality, real-world training data illustrating human activity. Human Archive believes that India’s burgeoning gig economy workers represent an underutilized and scalable source of this critical data. However, the startup has faced rejections from several major players in the home services sector, including Pronto and Urban Company, which has led to public discussions about the potential for collaboration. Despite setbacks, Human Archive is innovating by developing additional devices, including tactile gloves and a full-body motion capture suit, to enhance data collection. The startup emphasizes that video data alone isn't enough; combining it with sensor data greatly increases its value. As the founders engaged with various labs, they recognized a growing market for sensor-based data and decided to focus their efforts on this niche. Initially relying on makeshift setups and standard rigs for data collection, Human Archive is now creating custom hardware to capture diverse data types. With over 50 distinct devices deployed, the company is paving the way for advanced data synchronization and collection. Zach DeWitt, a partner at Wing VC, highlighted the startup’s competitive edge in gathering data from various sensors. Human Archive is currently in the process of refining AI models using its proprietary data, which will be tested on robots to assess task performance. In response to its challenges, Human Archive has partnered with smaller startups to provide cost-effective services to consumers. When a worker arrives for a home service, customers are given the option to pay a reduced fee in exchange for allowing data collection, which many have opted for due to the potential for resolving service quality disputes with recorded evidence. While the startup compensates workers at a base rate of $1 per hour for participating in data collection, it faces scrutiny regarding privacy and data usage. Human Archive asserts that its operations comply with India’s Digital Personal Data Protection Act, ensuring that all collected data is anonymized and that faces in recordings are blurred. Recently, India’s Ministry of Electronics and Information Technology has begun investigating the consent processes and data practices of companies like Human Archive. Although primarily operating in India, the company is exploring expansion into Southeast Asia and the U.S., aiming to create a platform for wider participation in data collection. This initiative could potentially allow U.S. customers to receive services like cleaning in exchange for data gathering. As the demand for physical AI continues to surge, Human Archive finds itself in a competitive race to provide the essential training data needed. The success of its model will depend on the partnerships formed and the unique, scalable data it can collect to meet the needs of physical AI laboratories.