Artificial intelligence training data provider Human Archive Inc. today announced that it has raised $8.2 million in funding.

Wing Venture Capital, NVP Capital, Y Combinator headlined the consortium that provided the capital. The funds were joined by employees at Nvidia Corp., OpenAI Group PBC, Google LLC and other major players in the AI market.

AI models are usually trained on data that is similar to the information they will process in production. A robotic arm’s neural network, for example, might be shown clips in which other robotic arms assemble car parts. Sourcing such data for humanoid robots is difficult because there is a limited number of such machines on the market.

San Francisco-based Human Archive launched earlier this year to tackle the challenge. According to TechCrunch, the startup sources training data by partnering with gig economy companies that operate food delivery and household service platforms. Workers on such platforms receive devices that they can use to record how they perform day-to-day tasks. Human Archive organizes that footage, adds labels and sells it to AI developers.

The company initially relied on iPhones to record participants’ work. Today, it uses more than a half-dozen different devices, including a camera-equipped headset. Human Archive reportedly has a network of more than 1,000 active users who wear its headsets while carrying out their work.

The company compensates participating workers for the data they source. Additionally, it offers discounts to gig economy service buyers who agree to data collection. Currently, Human Archive mainly relies on gig economy companies in India to source data. It’s reportedly developing a cloud service that will enable users in other markets to submit footage.

Human Archive’s engineering roadmap also encompasses hardware development initiatives. The company is reportedly developing tactile gloves, wrist cameras and motion capture suits. It plans to integrate sensory logs from multiple sources to create higher-quality training datasets.

A job posting hints that some of Human Archive’s devices will include at least three cameras. Besides recording videos, they will also be capable of collecting “inertial, magnetic, acoustic and environmental” data. Additionally, Human Archive is working on a mechanism that will enable users to swap a device’s battery with a fresh one to avoid charging-related downtime.

Human Archive is the latest in a string of AI training data providers to have raised funding. If its approach to data sourcing proves successful, some of those rivals may launch similar offerings. Human Archive’s platform may also face competition from world models, neural networks that can generate highly realistic synthetic footage. Some world model startups are using their algorithms to generate robot training datasets. 

Image: Unsplash