Beijing, China - Recent (CCTV - No access Chinese mainland)
1. Lightwheel AI executive Du Xiqi showing embodied AI data platform on screen
2. Various of platform displaying multiple video feeds of robots undergoing training; synthetic simulation data, human video data
3. Various of robotic hand opening microwave oven door
4. Platform displaying video feeds of robots undergoing training
5. SOUNDBITE (Chinese) Du Xiqi, executive, Lightwheel AI (starting with shot 4/ending with shots 6-7):
"We are building infrastructure that allows robots to keep learning. In practice, that means feeding the experience they gain after deployment back into the system so it can support further training."
6. Plug insertion and removal trial in progress
7. Screen showing video clips, diagrams comparing real-world, simulated plug insertion and removal tests, data curve

FILE: China - Exact Location and Date Unknown (CCTV - No access Chinese mainland)
8. Computer screen displaying test data, researcher operating testing device
9. Various of robotic hands, humanoid robots undergoing training

Shenzhen City, Guangdong Province, south China - Recent (CCTV - No access Chinese mainland)
10. Various of wheeled humanoid robot undergoing training, pressing buttons on electrical control panel
11. SOUNDBITE (Chinese) Dong Dianbiao, executive, Shenzhen Cyborg Robotics (ending with shot 12):
"After training the robot on tens of thousands of data samples, it can now carry out a given action with a success rate of 80 to 90 percent."
12. Various of wheeled humanoid robot undergoing training

FILE: China - Exact Location and Date Unknown (CCTV - No access Chinese mainland)
13. Various of robot dogs undergoing training
14. Various of humanoid robots operating in factory setting
15. Various of robotic hands undergoing training
16. Humanoid robot undergoing training at subway security checkpoint
17. Various of humanoid robot tidying room
18. Humanoid robots working on factory assembly line

China is expanding training infrastructure, developing industry standards and increasing real-world trials to move embodied AI from laboratories into commercial use, as growing volumes of human and simulated data help robots learn complex workplace tasks.

Embodied AI refers to artificial intelligence integrated into physical systems such as robots, enabling them to perceive and interact with the real world.

More than 70 embodied AI training centers are operating across China, while over 40 others are under construction or being planned.

Three major clusters have emerged in the Yangtze River Delta in the east, the Beijing-Tianjin-Hebei region in the north and the Pearl River Delta in the south.

At Lightwheel AI in Beijing's Zhongguancun technology hub, a data platform pools videos of people performing tasks with synthetic data generated through simulations at training centers nationwide.

The company said it has delivered more than 1.5 million hours of human video data, a volume it describes as among the largest in the industry worldwide.

"We are building infrastructure that allows robots to keep learning. In practice, that means feeding the experience they gain after deployment back into the system so it can support further training," said Du Xiqi, an executive at Lightwheel AI.

The growing pool of training data is already being used to improve robots performing real-world tasks. In Shenzhen, south China's Guangdong Province, wheeled humanoid robots trained on such data can operate electrical control cabinets and open or close industrial valves.

"After training the robot on tens of thousands of data samples, it can now carry out a given action with a success rate of 80 to 90 percent," said Dong Dianbiao, an executive at Shenzhen Cyborg Robotics.

The development comes as China's broader artificial intelligence industry continues to expand. The value of the country's core AI industry exceeded 1.2 trillion yuan (about 180 billion U.S. dollars) in 2025, with more than 6,200 companies operating in the sector.

In June, China's Ministry of Industry and Information Technology and state assets regulator launched an embodied AI program that covers industrial production, services and specialized operations, including workplace safety and emergency response.

The program seeks to create a cycle in which real-world training generates data, the data drives product improvements and upgraded systems are deployed at scale. It aims to validate the regular deployment of key products in representative settings and identify more than 100 high-value applications by the end of 2026.