How Companies Are Using Regular Workers to Teach Robots to Do Chores

Companies are hiring people to wear cameras on their heads and chests while doing everyday household tasks like washing dishes and folding laundry. The videos they record are being used to teach robots how to do the same work. These aren't quick recordings — researchers are collecting thousands of videos showing the same task done different ways, all to help robots learn the small movements and adjustments humans make without thinking.

This approach is different from how robots were trained in the past. For decades, robots learned from carefully controlled demonstrations in laboratories. Now companies are filming real people doing real tasks in normal conditions, capturing the kinds of details that only matter when a robot actually has to work in a home.

Why Companies Need So Much Video

Teaching a robot to pour a glass of water sounds simple, but it actually requires thousands of video examples. Each video might show a different type of glass, a different liquid, or a different pouring speed. A robot's brain — the machine learning software that controls it — needs to see all these variations before it can reliably do the task in the real world, where nothing is ever exactly the same twice.

Researchers at MIT have discovered that by combining lots of different videos from many sources, robots can learn skills faster and better. This is similar to how companies trained large language models like ChatGPT by feeding them billions of words from across the internet. The more diverse examples the software sees, the better it learns.

The Robot Industry Is Organizing Around This Work

In 2024, the International Federation of Robotics — essentially the main trade group for the robot industry — set up dedicated programs for collecting video data. Companies are starting to partner with each other to build these video libraries more efficiently. One example: a company called Omnipresent Robotics is buying robots from another company and opening a data collection center in Michigan specifically to film household tasks.

Universities are also building standardized tools to help researchers film these videos more cheaply and consistently. Having these shared standards means different companies and research groups can work together on the same problem.

The Robots Themselves Are Improving

Meanwhile, the actual robots are getting better. Boston Dynamics, the company behind the famous Atlas robot, recently stopped using a version powered by hydraulics — the same technology that powers construction equipment — and switched to a fully electric design. Their new robot will be tested in car factories by Hyundai before moving on to other applications.

The shift to electric power matters more than it might sound. Hydraulic systems need a lot of support equipment and regular maintenance, which makes them difficult to operate in homes or offices. Electric robots are simpler and safer to work alongside people.

Looking at the Bigger Picture

From covering technology for three decades, I have seen this pattern before. When companies were developing self-driving cars, they started by testing them on closed courses with controlled conditions. The real breakthroughs only came when they began collecting millions of hours of actual driving video from real roads, different weather, and unpredictable traffic. What happened with cars is now happening with household robots. The companies assembling the biggest video libraries of real household tasks today may have a significant advantage once robots become affordable and reliable enough for homes and small businesses.

The Real Challenge: Learning From Mistakes

There is more to robot training than just showing videos of tasks going right. Robots need to see what happens when things go wrong too — spills, drops, and near-misses. They need to learn how to feel when something is too fragile, how to notice when a surface is slippery, and how to fix their own mistakes while they are happening.

First-person video from a human's perspective captures something that a camera watching from across the room cannot: the tiny adjustments people make based on what they feel in their hands, how they sense resistance, and how they correct course in real time. That kind of sensory awareness is hard to teach to machines, but video from a worker's point of view helps bridge that gap.

The Real Question Is Timing

Several things have come together recently that make robot training possible now. Computers have become powerful enough to process the complex video and movement calculations that household robots need. Batteries last long enough to power a robot for hours of work. And building robots has become cheaper, so companies can afford to sell them to regular people and small businesses, not just factories.

The fact that companies are investing so heavily in video collection right now suggests they believe household robots will be commercially available within the next few years. The video libraries they are building today could give certain companies a real edge once robots become affordable enough for homes and offices.

The question worth considering: the quality and variety of video training data may turn out to be what separates the companies that succeed in household robotics from those that do not. Building comprehensive video libraries of real household tasks is not just a research project — it is a strategic investment that could be difficult for competitors to catch up on once the hardware becomes cheap enough for mass use.