More than 70 embodied-intelligence training grounds were operating across China by the end of June, with another 46 under construction or in the planning stage, according to a report from the China Academy of Information and Communications Technology. Yicai report on China’s embodied-AI training centers

The idea is straightforward but important. Humanoid robots need more than large language models and internet-scale datasets to operate in the physical world. Training them also requires data showing how bodies, objects and environments interact, including movement, visual information, joint torque and tactile feedback. The CAICT report said this kind of multimodal physical-world data has not previously been collected systematically. CAICT training-ground findings via Yicai

The twist is that robots can perform the same household tasks repeatedly in controlled environments, while human trainers manipulate the machines and monitor their actions. These facilities are designed to collect the physical-world data needed to train embodied AI. TechNode report on China’s training grounds

One such facility in Beijing’s Shijingshan district opened in September 2025. The center, which covers more than 10,000 square meters, uses human trainers to remotely control humanoid robots while they undertake different activities. The Business Times report on the Shijingshan training center

The expansion comes alongside China’s rapidly growing humanoid robot industry. A 2025 industry report counted more than 140 humanoid robot manufacturers in the country and about 14,400 shipments during the year. That figure represents 84.7% of global shipments under the report’s methodology. 2025 humanoid robot market report coverage The figure should not be treated as a directly comparable global market total without considering the report’s definitions, particularly because the International Federation of Robotics reported about 7,000 humanoid robots sold globally in 2025 under its own definition. Reuters on global humanoid robot sales

However, increased numbers of training grounds do not necessarily mean that the toughest robotics problems have been solved. Reuters on the challenges facing Chinese humanoids

China’s government is now pushing the same transition from controlled training to real-world deployment. In June, the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission launched a program focused on real-scene training, physical robot data and validation across industrial, service and specialized applications. Xinhua on China’s real-scene humanoid robot training program

The emerging training grounds are therefore less about teaching a robot one particular task than building the physical datasets needed to make embodied AI more capable. The open question is how much of that experience can ultimately translate into robots that can handle unfamiliar work without a human guiding every move.