A startup in New York will clean your apartment for nothing. A typical visit from a cleaning service in the city runs into the hundreds of dollars. This one is free. The only condition: the cleaner shows up wearing a camera strapped to the forehead and records the whole visit, from scrubbing the bathroom to washing the dishes to straightening up the mess. The recording is the payment. The company, called Shift, sends the footage to its German parent, MicroAGI, which collects first-person video of everyday tasks and licenses it to robotics labs that want to see how chores unfold inside a real home.

Shift blurs faces, names, and documents before the footage goes anywhere, and does not charge for a cleaning when the customer agrees to the recording. The company also says it will not sell the recordings to advertisers, and will use them only to train AI and robots. It also seeks out messier apartments on purpose, because a chaotic kitchen produces more varied footage than a tidy one. Shift has drawn heavy press coverage since launching, and is one of several companies now paying, in cash or in kind, for the same kind of footage.

DoorDash now runs a separate app called Tasks, paying its delivery couriers to film themselves doing ordinary chores: washing dishes by hand, folding laundry, repotting a plant for a higher rate. One listed assignment asks a courier to wash at least five dishes on camera, holding each clean plate up to the lens before moving to the next. DoorDash has said the goal is to help AI and robotic systems “understand the physical world,” the same aim behind Shift’s free cleanings.

Scale AI, a data company already supplying robotics firms with training material, has logged more than 100,000 hours of this kind of real-world video at its own collection site in San Francisco, gathered with help from contributors around the world. That is already a large amount of footage of ordinary chores. By the company’s own account, it is still not enough.

The company has explained why on its own blog. Text and images already sit on the internet by the billion, free for the taking. Robot training data does not. It has to be recorded one task at a time, inside real homes, because those homes vary endlessly in ways a lab has trouble reproducing. The dishes are a different shape from one kitchen to the next, the sink is a different width, a fork ends up hidden under a plate. A machine that has only ever practiced in one tidy demo kitchen has no way to guess what a stranger’s kitchen looks like, which is why it has to see enough real mess before it is ready to handle a version of the task it has never met.

The gap between a lab and a real home explains why companies are suddenly willing to pay for chore footage instead of asking people for money. Household robots built and tested in clean, controlled settings tend to struggle once they reach a real one, because the real version rarely matches what they trained on. A flawless demo in a showroom does not prepare a robot for someone else’s kitchen. Filming a stranger’s sink, a courier’s kitchen, or a cleaner’s daily round is a way of feeding a machine the mess it will eventually need to handle.

None of the three efforts has yet produced a robot that folds laundry unsupervised in a stranger’s home. Shift, DoorDash, and Scale AI are all still collecting, still comparing chores across many different kitchens and bathrooms, and still working from a dataset that is small next to what full-home robotics still needs. The variation inside even one house, let alone every other house, is close to endless, and none of the three has said its footage has caught up yet. For now, the fastest way to close that gap is to keep pointing cameras at people doing what they were already doing anyway.

Sources

Yahoo Tech, This AI Startup Will Clean Your NYC Apartment for Free—But There’s a Catch — tech.yahoo.com
TechCrunch, DoorDash launches a new ‘Tasks’ app that pays couriers to submit videos to train AI — techcrunch.com
Scale AI, Expanding Our Data Engine for Physical AI — scale.com