How China's 'cyber-laborers' teach humanoid robots everyday skills

In a brightly lit office in Shanghai, Kim, a 20-year-old computer science student, repeatedly opens and closes a microwave door. Wearing a VR headset and a full-body sensor exoskeleton, a wearable frame fitted with motion sensors, his movements are mirrored by a humanoid robot in real time.

It is one of many tasks Kim performs at that humanoid robot training facility. On other days he stacks wooden blocks or folds clothes, repeating the same actions hundreds of times to generate data for robots to learn from. "We call ourselves cyber-laborers," Kim, who uses a pseudonym, told tech site Rest of World. "It’s a fine gig though a bit boring."

Workers like Kim are seen as essential to China’s ambition to become a global robotics powerhouse, an area viewed as the next frontier in the U.S.-China tech race. In early 2025 the Chinese government identified embodied intelligence, or AI in physical form, as a national priority, triggering a surge in investment, including in making humanoid robots. More than 150 such companies now operate nationwide.

A worker is training a humanoid robot to pick up items. Photo courtesy of Beijing Shijingshan center

While large language models have succeeded by training using massive internet datasets, experts say robots require far more complex data, including visual inputs, joint movements and rotations, which cannot easily be obtained from the internet or produced synthetically.

To tackle this challenge, China has turned to state-funded robot training centers. According to British market research firm Interact Analysis, more than 40 government-owned robot data collection centers were slated to open by December 2025, with over 20 already in operation.

Typically huge, they house dozens of robots and are run by robot manufacturers. In Beijing’s Shijingshan district, the Phase II Beijing Humanoid Robot Data Training Center, the largest of its kind in China, serves as a "school" where robots undergo scenario-based learning before deployment. It replicates real-world production and living environments, including tasks like coil sorting, parcel packing, cooking, and bedroom organization. Training takes place in "small-class" settings, with each robot assigned two human trainers. "Just as children require repeated practice to walk, robots must train extensively across diverse scenarios to build functional intelligence," Zhu Kai, head of the training center, told People’s Daily newspaper.

Trainers Shi Xuanyu and Han Weiqi, considered among the top instructors there, typically take seven to eight days to guide a robot through a single scenario. The center employs 110 staff, most of whom were born in the 2000s. "Training requires meticulous attention to detail," Shi said, explaining that teaching a robot to place a frying pan on a stove, for instance, involved 1,250 repetitions, with each movement precisely recorded to enhance its coordination and replicate human-like performance.

Robots trained at the center have now acquired more than 20 operational skills, such as material handling, inspection and delivery, with a task success rate surpassing 95%. After "graduating," the robots are sent to work in roles such as material handlers at electric vehicle factories or as couriers, power inspection officers and service guides in offices.

Trainers guide humanoid robots through task in kitchen environment. Photo from X

At another government-run data collection center in Hubei Province, nearly 100 humanoid robots are remotely controlled to practice tasks such as folding clothes, ironing and wiping tables hundreds of times each day. The robots, produced and trained by Optics Valley Dongzhi, operate across 14 scenarios, mainly for use in supermarkets, restaurants, residential communities, factories, and laboratories to serve consumers.

According to data published on the Wuhan East Lake High-Tech Development Zone’s website, the robots fold 300 pairs of shorts, iron clothes 500 times, wipe tables 500 times, and package 500 boxes of medicine daily under the control of data collection workers.

According to a report released on Jan. 8 by U.K.-based market research firm Omdia, global humanoid robot shipments reached 13,317 units in 2025 and are expected to grow to 2.6 million by 2035. Chinese robotics firms dominate the market, with 87% of all humanoid robots delivered in 2025 manufactured in that country. Its leading firms, Unitree, Galbot and AgiBot, are each valued at over $1 billion. U.S. rivals include Figure AI, recently valued at $39 billion, Tesla and Boston Dynamics.

Unlike the U.S., where momentum largely comes from Silicon Valley, China’s humanoid robotics industry is driven strongly by government support. In addition to data centers, local governments have set up AI development funds, prioritized robots’ access to computing resources and expanded university courses in AI, including embodied intelligence.

"The central government is sending clear signals to local governments, expecting them to provide on-the-ground, tangible support to local companies," said Pavlo Zvenyhorodskyi, an analyst at think tank Carnegie Endowment for International Peace. "Every local government is being creative [in coming up with] support measures."

However, the rapid expansion of robot training facilities raises concerns about overcapacity. Ccommercial deployment of humanoid robots is still some way off, with the largest orders currently coming from the public sector. The boom in data collection centers also comes as researchers continue to debate whether recording human motion is the best path toward fully intelligent robots, given the labor-intensive nature of the task.

Ken Goldberg, a robotics researcher at the University of California, Berkeley, in the U.S. said the cost-effectiveness of China’s data factories remains unknown, noting that while it is a noble effort, progress is "slow." "Even if you have hundreds of people working, it’s going to take a long time to get enough data."