The training program emphasizes scenario-based data collection to generate high-quality, labeled datasets for large-model development, addressing a bottleneck in the robotics industry.

Each robot is guided by two human trainers who rotate roles to ensure accuracy, consistency, and comprehensive data gathering, mirroring how children learn through repeated practice.

Robots trained at the central facility have already been deployed in real-world roles, including material handling for factories like China FAW Group, courier services for Shenzhen Capital Group, and power-line inspections for China Southern Power Grid.

A replication and monitoring process uses two instructors guiding a robot in real time (for activities such as packaging with bubble wrap) while a third party records performance data, rotating roles every two hours to maintain consistency.

Centralization enables standardized data production and cost-effective datasets, with the center expected to generate several million high-quality data entries annually and to link with other sites nationwide.

The facility aims to generate several million high-quality data entries per year and connects with other centers in Suzhou, Jinan, Hefei, and Zhengzhou to build a national data network.

The center reproduces real-world production lines and living environments across two floors with modular training cells that can be reconfigured to simulate different scenarios.

The two-story facility uses modular, reconfigurable training areas to cover tasks from coil sorting and parcel packing to cooking and bedroom organization, mirroring production lines and everyday environments.

China has opened the Phase II Beijing Humanoid Robot Data Training Center in Shijingshan district to train humanoid robots for industrial, home, and public-space tasks.

Described as the largest facility of its kind in China, the Beijing center functions as a school for humanoid robots before deployment, addressing data shortages.

Robots are categorized into four broad skill areas, further divided into 16 disciplines, with over 20 operational skills trained to date and current task success rates above 95%.

The training program emphasizes major categories—industrial manufacturing, smart home applications, elderly care, and 5G-integrated scenarios—broken into 16 disciplines, with more than 20 skills trained and over 95% task success.