It's far more informative to put the robot to work and study what happens
In Beijing and Shenzhen, a quiet experiment is unfolding in the most intimate of human spaces — the home — where AI-guided robots now share the workday with cleaners like Lin Meiqiong, folding clothes with the uncertain hands of a beginner. China is not merely testing a service; it is harvesting the raw material that laboratory walls cannot provide: the irreducible complexity of real human life. Behind the modest price tag of 149 yuan lies a far larger wager — over 57 billion yuan committed this year alone — on the belief that the next frontier of artificial intelligence must be embodied, physical, and present in the world.
- A cleaning robot that folds pants 'very well' but still needs an engineer to enter the room signals just how wide the gap remains between demonstration and genuine capability.
- China's embedded AI sector has already surpassed last year's total investment in a single year, creating a race where data collection matters as much as the service being sold.
- Researchers warn that robots entering homes carry cameras and sensors into spaces where privacy standards, data ownership, and safety regulations simply do not yet exist.
- Companies like X Square and GigaAI are deliberately deploying imperfect machines — treating every apartment as a laboratory and every task as a dataset — to accelerate what no simulation can replicate.
- Experts across three continents agree: widespread adoption is still years away, and the human workers sharing their routines with these machines remain, for now, clearly ahead.
Lin Meiqiong, a 56-year-old cleaner in Beijing, now works alongside a white-and-silver robot guided by artificial intelligence. She still handles the floors herself, but the machine moves through rooms collecting debris and folding clothes. "It's lightened the load a bit," she said.
The service, a partnership between platform 58.com and robotics company X Square, costs 149 yuan for three hours and has been available in Beijing and Shenzhen since March. The Quanta X1 Pro enters apartments with an engineer's assistance, using cameras to map the space. Its folding work is tentative — one user compared it to a child learning — but about 200 households have signed up, drawn by curiosity as much as convenience.
For X Square, an imperfect service serves a deliberate purpose: gathering real-world data to develop embedded artificial intelligence. Unlike language models trained on internet text, robots lack comparable datasets from actual human environments. "It's far more informative to put the robot to work and study what happens than to leave it forever in a laboratory," explained researcher Christoforos Mavrogiannis of the University of Michigan. X Square engineers deliberately send machines into unknown environments, treating each home as a source of irreplaceable training data.
China is investing heavily in this direction — more than 57 billion yuan committed to embedded AI this year alone, surpassing all of last year. Beyond cleaning, robots are directing traffic in Hangzhou, and GigaAI plans to send 100 machines to Wuhan homes for free trials.
Yet the obstacles are substantial. Robots still lack human dexterity, no industry-wide safety standards exist, and privacy concerns are acute — these machines will access intimate details of people's homes, with no clear answers about where data goes or who controls it. Experts describe the current moment as a very early stage, with widespread adoption still years away.
Lin herself seems untroubled by the prospect of being replaced. "Compared to people, it's obvious he's not at the same level," she said. "After all, it's a robot." For now, each apartment these machines enter is less a workplace than a classroom — and the lessons are only beginning.
Lin Meiqiong, a 56-year-old cleaner in Beijing, now shares her workday with an unexpected partner: a wheeled robot painted white and silver, guided by artificial intelligence. When the machine arrived at her apartment, it changed the rhythm of her labor. She still handles the floors herself, but the robot moves through the rooms collecting debris and folding clothes while she works. "It's definitely different," she told reporters. "Before I did everything alone. It's lightened the load a bit."
The service, a collaboration between the online platform 58.com and robotics company X Square, costs 149 yuan—roughly 114 Brazilian reais—for three hours of work. It launched in March and is available in Beijing and Shenzhen. The Quanta X1 Pro enters each apartment with help from an X Square engineer, using cameras to map out which areas need attention. The folding work takes several minutes and produces results that resemble a child learning the skill for the first time. Engineers say future versions will respond to voice commands and hold conversations.
About 200 households have signed up since the service began. Tan Pei, who works in advertising, hired the robot partly out of curiosity—to see what it could actually do. "Although it's not perfect, some things surprised me," she said, noting that the robot folded a pair of pants "very well." Chinese robots have earned attention for their performances in dance and martial arts demonstrations, but their real-world capabilities remain limited. For companies like X Square, launching an imperfect service serves a larger purpose: gathering data to develop what researchers call embedded artificial intelligence.
Unlike the large language models trained on internet content, robots lack comparable datasets from actual human environments. "We don't yet have an internet for robots," explained Christoforos Mavrogiannis, a researcher at the University of Michigan. "It's far more informative to put the robot to work and study what happens than to leave it forever in a laboratory." X Square's engineers send their machines into "completely unknown environments," according to Hu Bowen, one of the company's engineers. "That's very challenging, but this data helps tremendously with robot development."
China is investing heavily in this direction. Investors have committed more than 57 billion yuan—approximately 43.7 billion reais—to the embedded AI industry this year alone, surpassing all of last year's investment. Beyond cleaning robots, the country is testing machines that direct traffic in cities like Hangzhou. GigaAI plans to send 100 robots to homes in Wuhan this year for free cleaning trials.
Yet significant obstacles remain before these machines become commonplace. The robots still cannot match human dexterity. "Many companies are working on autonomous robotic hands, but they haven't gotten there yet," Mavrogiannis noted. Regulatory questions loom larger. Privacy stands as a central concern—these machines will access intimate details of people's homes and lives. "We don't know where the data goes, where it's stored, or who will have access to it," said Valeria Alessandra Macalupu Chira from Queensland University of Technology. Safety standards for both residents and their property remain unresolved. "I think we're still in a very early stage," assessed Yang Jianfei from Nanyang Technological University in Singapore. Current robots require human supervision for emergency braking, and no industry-wide safety standards exist.
Experts agree that widespread adoption remains years away. When asked whether robots might transform her profession, Lin seemed unbothered. "Compared to people, it's obvious he's not at the same level," she said. "After all, it's a robot." For now, the machines are tools for data collection as much as they are tools for work—each apartment a laboratory, each task a lesson in what comes next.
Notable Quotes
Before I did everything alone. It's lightened the load a bit.— Lin Meiqiong, apartment cleaner in Beijing
It's far more informative to put the robot to work and study what happens than to leave it forever in a laboratory.— Christoforos Mavrogiannis, University of Michigan researcher
The Hearth Conversation Another angle on the story
Why does it matter that China is testing these robots in actual homes rather than just in labs?
Because a robot folding clothes in a controlled environment behaves very differently than one folding clothes in a real apartment with clutter, different lighting, unexpected obstacles. The data from real homes teaches the AI what actually happens in the world, not what engineers predicted would happen.
But the robots aren't very good yet. Lin's robot folds like a child learning. Why release something imperfect?
That's the whole point. X Square isn't trying to replace cleaners tomorrow. They're paying customers to let them learn. Every fold, every mistake, every surprise teaches the system something. It's cheaper and faster than running a thousand simulations in a lab.
What's the real risk here that experts keep mentioning?
Privacy, mainly. These robots see inside your home—your belongings, your habits, your layout. Nobody knows yet who owns that data, where it lives, what happens to it. And there's no safety standard. If a robot malfunctions, there's no agreed-upon way it should fail safely.
Is Lin worried about being replaced?
Not really. She sees the robot as a tool that helps her, not a threat. She's realistic about what it can and can't do. But that's one person's view. The larger question is whether this scales—whether in five years, cleaning robots are common enough to reshape the labor market.
How much money are we talking about here?
Fifty-seven billion yuan this year alone in embedded AI investment across China. That's more than all of last year. This isn't a small experiment. It's a major bet that robots in homes and cities are the next frontier.