Inside the multi-modal data testing and collection center of semi-human robots in Zigong city, Sichuan province (China), trainer Ouyang Yuanbin is wearing virtual reality glasses.
A few meters away, a Walker S2 robot simulated his movement in real time. The metal arm reached out to grab a package and coordinated with the other hand to place it on the next table.
This seemingly simple action is actually a process of mining "gold". Pressure, vision and torque sensors in the joint continuously record every micro-change. The system is collecting high-quality real-world data to make "food" for artificial intelligence.
The goal is to combine human remote control processes with automated collection," Ouyang explained.
This 6,000m2 facility officially opened on January 8, 2026 and is expected to reach its maximum capacity in March with 3 million data items per year. This is a key step in China's strategy to shift from manufacturing to robot intelligence.
Last year marked a milestone of mass production of semi-human robots in China when more than 140 manufacturers launched more than 330 different models. However, the scarcity of actual data is still a bottleneck hindering commercialization.
Mr. Wang Feili, an analyst at UBS Securities China, commented: "Even if a child is born intelligent, without real-world datasets to nurture them, they cannot grow up.
This data collection process is extremely costly. Training robots to classify a battery requires tens of thousands of grip operations. Moreover, the biggest bottleneck lies in the quality of tactile perception. Mr. Peng Zhihui, co-founder of AGIBOT, said that nearly 80% of industrial tasks that traditional automation faces difficulties are related to the sense of touch, due to the industry lacking standardization roadmaps.
To solve the challenge, China has announced the first national standard system for semi-human robots. At the same time, large-scale training facilities are springing up across the country from Beijing, Shanghai to Guangzhou.
In Sichuan alone, as of February 24, 2026, the locality has recorded 1,138 businesses related to robots. Through simulating real-world industrial scenarios, the Tu Cong center helps robots improve grip skills, providing a solid foundation for optimizing algorithms.
When Ouyang Yuanbin finishes his working day, robots like Walker S2 will soon be deployed to factories and warehouses across China.
