The province of Henan has become the third region in China — after Beijing and Shanghai — capable of training heterogeneous humanoid robots, following the launch of a new facility in its capital, Zhengzhou.

At the Central China Heterogeneous Humanoid Robot Training Facility, 140 robots have been deployed across six major fields, including modern agriculture, industrial manufacturing and medical care.

Assigned to 27 real-world scenarios, the robots practice walking, grasping objects, avoiding obstacles and interacting. When they fail, they adjust and try again. Nearly 70 data collectors work alongside them, gathering and labeling up to 50,000 high-quality data entries daily.

The facility was jointly developed by the Henan Investment Group and the National and Local Co-Built Humanoid Robotics Innovation Center after they signed an initial agreement for the project in July 2025.

As the first facility in Central China focused on heterogeneous embodied artificial intelligence data training and scenario-based applications, it serves as a key platform for Henan's embodied intelligence industry.

Tailored to 12 pillar industries in Henan, the facility aims to open more than 100 real-world scenarios, allowing robots to quickly grow from "new hands" to "skilled workers", said Chen Yang, business manager at the Henan Embodied Intelligence Industry Development Co, which is responsible for construction and operations.

Chen, who described the facility as a "data factory" or "robot training school", said each robot is assigned a dedicated trainer who provides hands-on instruction in simulated scenarios. Motion data is uploaded to the cloud in real time to support model iteration.

The facility adopts a "front store, back factory" model, combining training operations with a robot 4S store where customers can interact with the machines.

When customers propose new tasks — such as organizing a wardrobe — the request is sent to the training facility for scenario design. After training, the robot returns to the store for testing, with customer feedback generating new data for further improvement.

"The facility supports the entire life cycle of a robot," Chen said.

Chen Yuecheng, business manager of the training division, said the hands-on teaching model — covering perception, decision-making and execution — is more complex than traditional training methods that rely on image annotation.

By breaking down tasks for segmented model training, robots can ultimately achieve autonomous execution, he said.

Even simple actions can be time-consuming. Basic flower arranging requires dozens of hours of data collection, while teaching robots to distinguish between flowers, vases and positions can take weeks.

The company mainly hires Generation Z trainers.

"These young individuals are flexible thinkers. They are innovative and patient, making them well-suited for robot training," Chen Yuecheng said.

Wang Hao, a 24-year-old trainer who has worked in the industry for three years after majoring in big data, said the focus has shifted.

"The core of training now is to evolve machines from 'understanding commands' to 'executing them precisely'," Wang said.

He said robots must be trained to replicate human actions such as picking up and placing objects. Even holding a glass of water can result in breakage or spillage without sufficient precision.

"To perfect such actions, they must accumulate vast amounts of data," he said.

In addition to common applications, Wang highlighted high-risk scenarios such as high-temperature welding and hazardous river environments.

"In places where the risk to humans is high, robots can safely replace human labor," he said.

Yan Chen, director of the Department of Industry and Information Technology of Henan Province, said the province aims to build embodied intelligence-related industries worth more than 100 billion yuan ($14.5 billion) by 2030.