India’s online food delivery sector has seen remarkable expansion in recent years, highlighted by the public offerings of Zomato and Swiggy and the surge in cloud kitchens. Concurrently, startups focusing on home services, particularly on-demand staffing platforms like Urban Company, Snabbit, and Pronto, are also gaining traction.
In this evolving landscape, Human Archive, a startup based in Silicon Valley, is capitalizing by collaborating with various companies. Their innovative approach involves equipping workers with specially designed caps that house cameras to gather egocentric (first-person perspective) video data aimed at training robots.
Although the startup refrains from disclosing specific collaborators, it has stated that it is actively gathering data within the home services, hotel, and restaurant industries. Currently, over 1,000 headsets have been deployed across diverse locations.
In light of this progress, Human Archive announced on Tuesday it has successfully attracted $8.2 million in funding, backed by Wing Venture Capital, NVP Capital, Y Combinator, along with investors from OpenAI, Nvidia, Google, among others.
The inception of the company was driven by a quartet of university students—Samay Maini, Rushil Agarwal, Shloke Patel from UC Berkeley, and Stanford’s Raj Patel, who is the CEO. Each boasts a robust research background in areas such as robotics, hardware, and tactile data.
The formation of Human Archive aligns with a critical juncture in the AI sector. As organizations endeavor to develop machines suited for physical tasks, they encounter a severe bottleneck—there simply isn’t enough high-quality training data showcasing human activity in real-world scenarios. Human Archive seeks to bridge this gap by leveraging the workforce of India’s burgeoning gig economy.
Even as Human Archive engages with various partners, notable Indian home services firms—including Pronto and Urban Company—have turned down proposals for collaboration.
This turn of events came to light recently, when Entrackr reported that Pronto is actively pursuing partnerships to amass worker data for robotics training, indicating prior discussions with Snabbit that ultimately did not progress.
Urban Company’s CEO, Abhiraj Singh Bhal, publicly declared on X that the company would refrain from participating in such arrangements, prompting a pointed response from Patel, who cautioned that Urban Company might be compelled to rethink its strategy to avoid losing market relevance. Co-founder Rushil Agarwal described a particularly candid moment with Pronto’s founder, claiming she laughed and dismissed his partnership suggestion. Pronto did confirm discussions took place but chose not to engage further, denying any derogatory remarks were made.
At the same time, a range of startups across India are also gathering egocentric data from various work settings, including factories. To stand out, Human Archive is enhancing its offerings with additional tools like tactile gloves, motion capture suits, and wrist cameras to gather comprehensive data that integrates motion and tactile feedback coupled with RGB-D (real-time color imagery with depth information), aiming to present this to AI laboratories. The firm asserts that simply having video data is insufficient; pairing it with other sensory input significantly increases its value.
Initially employing makeshift solutions to capture data, Human Archive now focuses on proprietary hardware that interconnects and collects various data types. The company has deployed over 50 distinct devices to gather different dimensions of data.
Patel explained their evolution: “We began with iPhones, then created custom rigs and caps. Now, we utilize more than seven different hardware products adaptively across different modalities. After collating data from varying devices, we’ve refined our methods to synchronize all sources.”
The startup is also aiming to enhance AI models with its proprietary data and test these models on robots to gauge performance effectiveness, thereby showcasing the value of its data to prospective clients.
Zach DeWitt, a partner at Wing VC, highlighted the startup’s distinctive advantage in multi-sensor data collection.
“No other organization has effectively synchronized and gathered headset RGB-D, force feedback, full-body motion capture, along with coordinated chest and wrist camera data on such a large scale. Their internal model training using this data has drawn interest from major laboratories and universities due to the innovative nature of their sensors and the extensive dataset they will soon release,” he elaborated.
Collecting data in India and expansion plans
Despite setbacks from major home service companies, Human Archive aligned with smaller enterprises to provide discounted services. When a worker visits a home, consumers can choose through the app: accept a lower price in exchange for allowing data collection or opt for the full rate for a non-recorded service.
Patel indicated that customers are generally willing to consent to data collection, primarily due to frequent disputes over service quality, as video evidence can help clarify issues.
The company compensates workers with a base rate of $1 per hour for participating in the data collection initiative. Recent reports suggest competitors typically offer ₹250 to ₹400 per hour (around $2.63 to $4.20). Patel acknowledged that while rivals may pay higher wages, Human Archive’s local presence enables it to maintain lower compensation costs.
“Human Archive’s network creates immediate, flexible earning opportunities worldwide, reducing barriers to entry into the AI economy. We view this as a vital bridge that not only supports immediate livelihoods but also lays the groundwork for a safer and more productive future,” DeWitt stated.
Alongside wage considerations, there are significant privacy issues regarding video data collection. It remains unclear what specific information Human Archive communicates to workers about the use of their footage. The company asserts it complies with India’s Digital Personal Data Protection (DPDP) Act, including displaying a privacy policy that explains data collection purposes and processing methods. They maintain that all data is anonymized with faces blurred in videos. Recently, Moneycontrol reported that India’s Ministry of Electronics and Information Technology is scrutinizing the consent mechanisms and data practices of companies like Human Archive.
While Human Archive primarily collects data within India, it has begun to extend operations into Southeast Asia and the U.S. The firm is also developing a platform to involve more participants in data collection efforts. Plans are in place to offer cleaning or cooking services in the U.S. in exchange for data collection, albeit these services are still in preliminary stages.
Numerous well-funded startups are competing to innovate physical AI. Satisfying this demand necessitates vast quantities of training data depicting human labor, positioning Human Archive as a key contender in this arena. Its capacity to scale will depend on forging partnerships and ensuring the data it collects is both unique and abundant enough to meet the needs of physical AI laboratories.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
#startup #betting #Indias #gig #economy #train #worlds #robots