Startups pay for chore footage to train robots: Shift, Pronto lead data collection

Training robots for physical tasks requires vast amounts of real-world data, which is harder to obtain than text or images scraped from the internet. Startups like Shift are getting creative: they clean New Yorkers' homes for free in exchange for video footage of every chore - scrubbing, dusting, mopping. This data helps teach robots about force, friction, and awkward shapes. In India, Pronto faced backlash for using client homes to train AI, though it claims recording only happens with explicit consent. Other companies like Human Archive pay gig workers to wear camera hats that capture first-person (egocentric) video of daily activities, providing the perspective robots need to navigate spaces.

Beyond real-world recording, staged data farms pay workers to perform repetitive tasks like folding towels or stacking boxes while sensors capture every movement. This turns rote labor into valuable training material. Some companies also deploy robots prematurely and rely on remote workers to step in when they fail, using that data to improve. The trade-off is clear: consumers get free or discounted services (cleaning, home maintenance) while companies collect intimate footage of their homes. As physical AI advances, the demand for such data will only grow, accelerating the need for privacy safeguards.