Rows of humanoid robots repeat the same motions over and over -- picking up water bottles, sliding trays into microwaves and scanning parcel labels. Nearby, human operators in motion-capture suits fold laundry and smooth out bedsheets, their movements captured by sensors and fed into AI training systems.

Scenes like these are unfolding in many places in China. Industry insiders call these facilities "embodied AI training grounds".

China is now home to more than 140 humanoid robot manufacturers. In 2025, the country shipped 14,400 humanoid robots, accounting for roughly eight in every 10 units shipped globally.

Yet impressive as these numbers are, the industry still faces a fundamental challenge: turning robots that can run, jump and dance in controlled settings into machines capable of reliably handling the countless, unpredictable tasks of everyday life.

A critical part of enabling them to handle everyday tasks is access to vast amounts of data.

Just as large language models learned from vast amounts of text to reshape the digital world, robots need vast amounts of real-world interaction data to master the physical one. They need to learn, through repeated encounters, how to grasp objects of different shapes, navigate unfamiliar spaces, respond to changes in their surroundings and complete everyday tasks.

But while the internet is overflowing with text, physical-world interaction data for robots has never been collected systematically. That is the gap these new training grounds, where robots learn to navigate and perform tasks in real-world environments through repetitive, data-intensive training, are being built to fill.

A recent research report shows that, as of the end of June 2026, China had opened more than 70 embodied AI training grounds. In addition, over 40 embodied AI training grounds are under construction or in the planning stage. Most are located in the Yangtze River Delta, the Beijing-Tianjin-Hebei region and the Pearl River Delta.

In south China's Guangdong Province, the provincial training ground has taken on a role that goes beyond teaching robots. Dubbed a "robot school" by its operator, it regularly brings robot makers together with potential users, helping manufacturers, healthcare providers and energy companies find suitable embodied AI solutions while giving robot developers access to real-world environments in which their robots can learn and be tested.

A similar model is taking shape in Hangzhou, capital of east China's Zhejiang Province. There, the National Pilot Base for Embodied AI Applications acts as a "super connector," bringing together state-owned enterprises, leading technology firms and application companies through an open cooperation mechanism.

"Previously, developing a single application scenario could cost a company tens of millions of yuan just in training data. Now, by moving into the base, you can access everything you need in just one place," said Zhang Haiwei, founder of a Hangzhou-based robot firm.

As these training grounds evolve from isolated facilities into a broader ecosystem, policy support is also moving in to accelerate their development.

In June 2026, Chinese authorities jointly launched a special action plan for real-scene training of humanoid robots and embodied AI, requiring each provincial region to identify at least 20 priority scenarios and create trainable, testable and verifiable real-scene training spaces.

At the same time, efforts are underway to improve not only the amount of training data, but also its quality. An industry standard on embodied AI dataset quality, drafted by a consortium of more than 40 organizations, is scheduled to take effect on Nov. 1. The standard is expected to help shift dataset development from a focus on scale toward greater quality, while addressing a long-standing gap in the field.

The business model is evolving, too. Wang Maolin, co-founder of an embodied AI ecosystem platform based in Shenzhen in southern China, said the next step is to move beyond simply supplying data and develop "training-as-a-service," creating diversified revenue streams that can help turn training grounds from heavy-asset investments into sustainable operations.

With policy support and growing momentum behind their development, embodied AI training grounds are no longer confined to China's major economic hubs. According to industry insiders, training grounds are now reaching into China's third- and fourth-tier cities, creating a multi-tiered network shaped not only by talent and capital, but increasingly by the availability of real-world scenarios and the data they can generate.