FOLLOW US

The push to bring humanoid robots into everyday homes is giving rise to an unexpected new line of work, one that requires little more than a head-mounted camera, a smartphone and a checklist of household tasks.

As artificial intelligence advances, humanoid machines have become a focal point in the global technology race. Companies are steadily introducing robots capable of walking, dancing and even fighting with increasing precision.

Yet the industry’s ultimate goal — a versatile robot that can function across homes, offices and retail spaces — depends on enormous amounts of training data. To meet that need, companies are turning to ordinary people to record themselves performing routine chores.

This demand has fueled rapid growth in so-called “egocentric data,” or first-person footage capturing human activity. In recent months, startups have begun recruiting thousands of contract workers worldwide to film and label these videos for use in robotic training systems.

“Manufacturing, factory warehouses, retail, nursing homes, hospitals – you’re going to need this type of data in basically every single environment, and that’s because the movements are all different,” said Arian Sadeghi, vice president of robotics data at Micro1, which began building its global network of videographers last year.

Participants are equipped with camera headsets, detailed instructions and assigned tasks ranging from cooking and cleaning to gardening and caring for pets. Workers rotate assignments and are typically required to submit at least 10 hours of footage each week.

Although most recordings focus on household routines, Sadeghi said contributors are encouraged to broaden their scope to help robots adapt to unfamiliar settings and responsibilities.

“The thing we tell them is, ‘If you think you want a robot to do this for you, go ahead and record it,’” Sadeghi said.

‘Billions of hours’

Despite the scale of current efforts, the available data falls far short of what experts say is required. Micro1, headquartered in Palo Alto, California, works with about 4,000 contractors across 71 countries, generating more than 160,000 hours of footage monthly — a fraction of what is needed.

“You need probably billions of hours,” he said. “We haven’t even gotten to human interactions. This is just simple household chores.”

The surge in robotics data demand mirrors the early development of AI chatbots, which were trained on vast amounts of text scraped from the internet. Similar approaches later enabled the creation of AI-generated images and videos.

Robotics, however, requires more specialized data that cannot easily be sourced online. This gap has opened a lucrative market for companies that both collect and annotate footage, allowing robots to recognize objects, distances and motion. Analysts estimate the data labeling sector could grow roughly 30% annually, reaching at least $10 billion by 2030.

Ravi Rajalingam, founder of Objectways, previously worked on datasets for virtual assistants and autonomous vehicles before pivoting to robotics. Since launching human data collection efforts, he said only about half of submitted footage meets usability standards.

Even so, demand remains strong. With most clients based in the United States — where early adoption of home robots is expected — companies are often willing to pay higher rates for U.S.-based data compared to lower-cost labor markets like Vietnam or India.

“The India kitchen is very different from the US kitchen. A broomstick in India is very different from a broomstick in US. So variety is important, but it depends where you are going to place your robots first,” said Rajalingam. “That’s the reason we are collecting all over the world.”

How robots are trained

Traditionally, robots have been trained through manual control systems, a method that relies on expensive hardware. More recently, developers have turned to virtual simulations, though those often fall short when it comes to handling real-world objects.

“With data it’s always a trade-off between quality and quantity,” said Alicia Veneziani, vice president of market expansion at Sharpa, a Singapore-based robotics company focused on robotic hands.

Governments are also investing heavily in the sector. China, for example, has announced plans to establish at least 60 robot training centers. Most humanoid robots produced there so far have been used primarily for research and training purposes, according to industry analysts.

In recent months, companies have increasingly embraced human-generated footage as a cost-effective alternative. The approach requires only basic recording equipment — such as action cameras, smart glasses or smartphones — and hourly wages that vary widely by region.

“The idea here is: Okay, I don’t want the robot doing the task. I want the people doing the task,” one analyst said. “This way, you don’t need to pay for the robots, you just need to pay for the equipment and the people.”

Different regions are experimenting with similar strategies. Japan and South Korea have adopted models resembling China’s data collection centers, often outsourcing work to Southeast Asia to reduce costs. Meanwhile, companies in the United States and Europe tend to favor simulation-based approaches supported by advanced chipmakers.

Still, combining methods appears to yield better results. A February report found that integrating more than 20,000 hours of first-person video significantly improved robot performance in tasks such as folding clothes, sorting cards and opening containers.

“If you rely on just one way of data collection, it’s probably not the best approach,” the analyst said. “In the future, it will be a mixture of different approaches.”

The final hurdle

A major breakthrough in robotics came several years ago with the rise of large language models, which enabled systems to translate visual input into physical actions. This allowed robots to move beyond repetitive programming and begin interacting with their surroundings more dynamically.

Companies are now expanding human data collection into industrial settings, particularly in countries like India. Experts say prioritizing such data is essential in the near term, though its dominance may eventually decline as simulation tools improve or new methods emerge.

“Even robotics labs are feeling like they don’t know what data will be needed 12 months from now,” one industry executive said.

One of the biggest challenges remains the unpredictability of home environments, where objects, furniture and people are constantly shifting.

“What’s really missing is a human-like intuition of forces, friction, and uncertainty that people acquire throughout their lifetime,” said Rutav Shah, a researcher at the University of Texas at Austin. “Making robots generally useful for everyday household tasks like cooking, cleaning, that is going to be the last mile of automation.”

So far, humanoid robots have largely been limited to controlled settings such as factories, where success rates can reach 99.9%, according to the International Federation of Robotics. However, even relatively simple tasks like folding clothes still fall short of commercial reliability.

“The probability that it will succeed is usually around 70 or 80%. Coming from manufacturing, that’s really not something that our industry partners want to use,” said Alexander Verl, the organization’s research chairman.

Safety concerns also remain a significant barrier. Rajalingam noted that robots must be able to distinguish between objects with precision, particularly in sensitive environments.

“If the robot takes my baby and puts it in a bin, here comes the million-dollar lawsuit,” he said.

While such scenarios remain hypothetical, early testing has already begun in less risky settings.

Testing robots with babies is still a long way off, Rajalingam said. However, he added, they have already started with dogs.