Shift, an AI training data startup, is offering free home cleaning services to residents in New York and is expanding to London. The catch: the company records the cleaning process to gather training data for future robot development. Workers scrub, vacuum, dust, and clean while cameras capture their movements and techniques.

This is a novel approach to gathering embodied AI training data—video footage of humans performing physical tasks in real-world environments. Rather than hiring contractors to clean labs in controlled conditions, Shift gets real household environments and varied cleaning approaches, potentially producing more generalizable training data.

What This Means for Your Business

Companies developing physical robots face a chicken-and-egg problem: you need real-world training data to build useful robots, but collecting that data at scale is expensive. Shift's model—incentivizing data collection through free services—may inspire similar approaches in other domains requiring embodied learning. If you're building robotics, monitoring data collection methods and costs could reveal new cost-reduction opportunities.