AI startups are recruiting people to film themselves doing household chores like folding laundry and loading dishwashers, paying hourly rates of $25–$150, to generate real-world training footage for robotic systems.
AI and robotics firms are increasingly turning to crowdsourced video from everyday people to generate the vast amounts of data needed for teaching robots basic physical tasks. Startups such as Encord, Micro1, and Scale AI say they are expanding their robotics operations and require footage of humans performing chores like folding laundry, loading dishwashers, or making espresso.
Unlike language models, which can learn from existing text data, robots require real-world visual and motion datasets—and those datasets do not yet exist at scale. “Unlike LLMs, robotics doesn’t have the internet as a ready-made dataset — you have to generate training data from scratch in the real world, which is far harder,” Ulrik Hansen, cofounder of Encord, told Business Insider.
To incentivize participation, the startups offer monetary compensation. Micro1’s CEO, Ali Ansari, said the company pays “incredibly well,” citing rates between $25 and $50 per hour for basic chores, while Encord may go as high as $150 an hour for videos of complex tasks such as handling surgical equipment.
Still, sourcing suitable training video remains challenging. One stealth-mode startup placed ads on Craigslist offering $10–$20 per hour to film chores, with its founder admitting that “There are no major datasets you can buy … The biggest ones are around 5,000 hours, which is not nearly enough.”