China has opened a school where humanoid robots learn real-world tasks

design sem nome
For years, artificial intelligence existed mostly on screens. Chatbots generated text, algorithms created images, and ever more advanced systems dominated the tech conversation. But away from traditional computers, a quieter race has emerged. The new challenge is no longer just teaching machines to think — it’s training humanoid robots to move, handle objects, and operate in unpredictable environments without constant human control. And China appears determined to lead this next phase.
Image Credits:© Centro de imprensa do distrito de Shijingshan, Pequim

For years, artificial intelligence existed mostly on screens. Chatbots generated text, algorithms created images, and ever more advanced systems dominated the tech conversation. But away from traditional computers, a quieter race has emerged. The new challenge is no longer just teaching machines to think — it’s training humanoid robots to move, handle objects, and operate in unpredictable environments without constant human control. And China appears determined to lead this next phase.

In China’s Fujian province, an experimental facility is drawing attention for its unusual purpose. At first glance, it resembles a blend of a robotics lab and an industrial workshop. In reality, it operates almost like a school for humanoid robots.

Inside, dozens of robots spend hours repeating simple physical tasks under human supervision. They practice picking up objects, cleaning surfaces, moving boxes, adjusting positions, and interacting with different environments while sensors capture every detail.

From AI Data Center to Real-World Robotics Training Hub

The project was developed by Fujian Jufu Technology at the Fuzhou Software Park. Officially described as an AI data collection center, the facility points to a far bigger ambition: building massive physical datasets to train robots capable of functioning in the real world — a challenge far more complex than it sounds.

For humans, actions like holding a glass or pushing an object require little conscious effort. Our brains automatically calculate force, balance, distance, and coordination. For humanoid robots, however, every small movement is an enormous technical challenge.

Operators use virtual reality equipment and specialized controllers to direct the robots’ movements. When a human moves an arm, the robot mirrors the motion instantly while cameras and sensors capture data such as pressure, speed, angles, and movement.

The objective goes far beyond completing basic tasks. The main aim is to generate millions of data points showing how human motion functions in real-world conditions.

This explains why industrial robots perform exceptionally well inside factories but struggle outside them: real environments are always changing.

Why Real-World Variability Makes Robotics So Difficult

Tables are never positioned exactly the same way. Objects appear in different places. Lighting shifts constantly. Items may be partially hidden or farther away than expected. Humans adapt naturally to these situations, but for machines, they create major challenges.

To solve this, engineers are focusing on what’s known as “generalization.” Rather than programming fixed, repetitive actions, they want robots to understand physical principles that can be applied across different situations.

In practice, that means constantly changing the training exercises. Different cups, larger boxes, varied surfaces, and unpredictable object placements force the robots to adapt continuously.

The approach closely resembles how children learn. Instead of memorizing one specific object, they develop broader patterns that help them recognize and handle thousands of different objects throughout life.

And this is exactly where China appears to see a major strategic opportunity.

Much of AI’s recent progress has been driven by the internet. Text, images, and videos have helped train systems that can converse, write, and generate digital content.

Robotics, however, faces a different challenge: there are very few large-scale, detailed datasets of real human physical movements.

Building the Physical Data Infrastructure for Humanoid Robot Training

Projects like the one in Fujian aim to address that gap. The goal is to create a massive infrastructure for training humanoid robots physically — something that could become as valuable as today’s data centers powering language models.

The company behind the initiative is still in its early stages. Founded in 2025, it has only recently begun operations. Even so, the project already hints at the direction of China’s broader strategy.

The ambition goes beyond simply building humanoid robots. China appears focused on controlling the entire ecosystem required to train them: data collection, artificial intelligence, physical learning, and adaptation to real-world environments.

Perhaps the most striking aspect is how AI itself is evolving. A few years ago, artificial intelligence seemed confined to the digital world. Now, it increasingly depends on something far more human: practice, repetition, physical training — and even something resembling a classroom.

The difference is that this time, the students are not human.

whatsapp image 2026 03 21 at 15.37.18 1 768x384

Read the original article on: gizmodo

Read more: AI-powered smart rings can now translate sign language instantly

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top