China is stepping up efforts to move humanoid robots beyond eye-catching demonstrations into workplaces, as policymakers seek to accelerate the commercialization of embodied AI.

A nationwide action plan launched on June 9 by the Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission aims to put humanoid robots and related technologies into routine use across a wide range of real-world scenarios by the end of this year.

The initiative supports the deployment of around 10,000 humanoid robots in more than 100 high-value scenarios, marking one of China's most ambitious efforts yet to bring the technology into practical use.

"With the implementation of these application scenarios, we will accumulate more data, and that data will help train more capable and generalized robots in the future," Wang Zhongyuan, president of the Beijing Academy of Artificial Intelligence, told CGTN.

According to Wang, one of the industry's biggest challenges is improving robots' ability to operate reliably in complex environments.

"The key bottlenecks are still generalization, task execution accuracy, battery endurance and operational safety," he said.

Some humanoid robots have already begun working in controlled environments such as hotels and factories. According to Xinhua News Agency, China shipped about 17,000 humanoid robots in 2025, produced by more than 140 companies nationwide.

But industry experts say large-scale deployment will require more than advances in hardware. Robots must learn to understand and interact with the physical world — perceiving their surroundings, making decisions autonomously and adapting to new tasks in dynamic environments.

These challenges are increasingly described within the industry as building a "physical world model."

"To help robots better understand and interact with the real world, the data and training environments need to be more dynamic rather than purely static," Sui Yanan, an associate professor at Tsinghua University, told CGTN.

Unlike large language models, which are trained primarily on internet-based text data, humanoid robots must learn from constantly changing real-world environments, where people, objects and conditions are continuously in motion.

While six months is a relatively short timeframe, experts say the initiative could accelerate the industry's development by encouraging companies to test technologies in real-world settings, accumulate valuable data and converge more quickly on viable engineering solutions.

Videographer: Wang Peng