Investors are feverishly pouring money into solving this challenge, and spending more than necessary 6 billion dollars on humanoid robots in 2025. Data logging at home has become a booming economy around the world. Data companies like Scale AI and Encord are recruiting their own armies of data loggers, while DoorDash Delivery drivers are paid To photograph themselves doing household chores. In China, workers at dozens of state-owned robotics training centers wear virtual reality headsets and exoskeletons to teach humanoid robots how to open a microwave and clear a table.
“There is a lot of demand, and it is growing very quickly,” says Ali Ansari, CEO of Micro1. He estimates that robotics companies now spend more than $100 million each year buying real-world data from his company and other similar companies.
A day in the life
Micro1 workers are screened by an AI agent called Zara who conducts interviews and reviews samples of routine videos. Every week, they send videos of themselves doing chores in their homes, following a list of instructions about things like keeping their hands visible and moving at a normal speed. Videos are reviewed by both AI and humans and are accepted or rejected. They are then annotated by artificial intelligence and a team of hundreds of humans who label the actions in the footage.
“There’s a lot of demand, and it’s growing very quickly.”
Ali Ansari, CEO of Micro1
Since this approach to training robots is still in its infancy, it is not yet clear what makes good training data. However, Ansari says, “You need to introduce lots and lots of variation so that the robot can generalize well to basic navigation and manipulation of the world.”
But many workers say creating a variety of “routine content” in their tiny homes is a challenge. Zeus, a cantankerous student living in a modest studio, struggles to record anything other than ironing his clothes every day. Arjun, a private tutor in Delhi, India, takes one hour to produce a 15-minute video because he spends a lot of time thinking of new homework assignments.
“How much content (can be made) at home? How much content?” He says.
There’s also the thorny issue of privacy. Micro1 asks workers not to show their faces to the camera or reveal personal information such as names, phone numbers and dates of birth. It then uses artificial intelligence and human reviewers to remove anything that sneaks through.
But even without faces, the videos capture an intimate slice of the workers’ lives: the interiors of their homes, their possessions, and their routines. Understanding what kind of personal information they might be recording while they are busy performing routines in front of the camera can be difficult. Reviews of these snapshots may not filter sensitive information beyond the most obvious identifiers.
