China Invests $29 Million in Robot Training Center to Accelerate Humanoid Robot Development

In Wuhan, China, a new facility dedicated to training humanoid robots has opened, covering an area of 12,000 square meters. This center is part of a broader initiative to enhance the capabilities of robots that perform tasks such as serving steamed buns, wiping tables, and folding clothes. Chinese graduates work long hours operating these robots, which are equipped with cameras and sensors that meticulously record every movement.

The Hubei Humanoid Robot Innovation Center, built at a cost of 200 million Chinese yuan (approximately $29 million), is one of many state-funded training centers established across the country. These centers aim to create a comprehensive database specifically for robot training, addressing the challenges faced by the emerging humanoid robot industry.

Zhang Jia, a 21-year-old program manager at the center, highlighted the difference between teaching robots and humans. He noted that while humans can grasp concepts after a few repetitions, robots require hundreds, thousands, or even tens of thousands of repetitions to learn effectively. This data collection is seen as crucial for advancing artificial intelligence from software applications to physical implementations.

This initiative aligns with President Xi Jinping’s vision to position China as a global leader in science and technology. Recently, Beijing identified “embodied intelligence” as one of six key sectors to be developed in its 2026-2030 five-year plan, which includes the establishment of training centers, AI models, and hardware to facilitate the deployment of humanoid robots.

Challenges in Robot Training Data

Experts have pointed out that the lack of robot-specific training data presents a significant barrier to translating recent advancements in AI into practical robotics applications. Unlike large language models that are trained on vast amounts of text from the internet, the collection of data for robots is still in its infancy.

Both Chinese and American companies are exploring various methods for gathering training data, including practical applications, simulations, and AI-generated data. For instance, Tesla has investigated using human demonstration videos to train its “Optimus” robot, while the Silicon Valley startup 1X Technologies aims to deploy robots in homes, where they can be partially controlled by humans during the learning process.

Beijing’s multi-year plan for the humanoid robot industry emphasizes the expansion of robot training and data collection. Local governments, from affluent coastal cities like Hangzhou to smaller inland areas such as Mianyang, are investing heavily in new training centers. Hubei Province has announced a 10-billion-yuan government fund specifically for humanoid robots.

Jay Huang, head of Asian industrial technologies at Bernstein Research, remarked on China’s strategic approach to supporting emerging industries. He noted that the government promotes data sharing, which fosters collaboration and drives progress in the sector.

Training Operations in Wuhan

In Wuhan, Zhang Jia supervises a team of 70 young trainers who work eight-hour shifts to train 46 robots. They utilize remote controls and handheld devices equipped with sensors to operate the machines, repeating the same movements multiple times. Nearby, employees analyze video outputs, adding annotations every few seconds, such as “turn left” or “extend arm.” This facility generates approximately 100 hours of usable data each day. Zhang explained that the data is collected, organized, and uploaded to a platform for classification and processing, although they are still in the exploratory phase.

The training methodology involves feeding AI models—known as “Vision-Language-Action” models—with sensor readings and videos that track the movements of robot parts. This approach aims to replicate the successes achieved by large language models in robotics, potentially enabling machines to learn general skills, such as picking up objects, without explicit programming.

Zhao Xiang, co-founder of the startup Motviz, which has developed a platform for simulating embodied intelligence, stated that scaling data collection is a complex challenge. Young engineers in the state-supported lab in Wuhan use VR goggles to train robots, allowing for more efficient and cost-effective data collection. Zhao emphasized the importance of simulation, noting that achieving a breakthrough in intelligence may require hundreds of millions or even trillions of hours of data.

Data Portability Challenges

Experts have also identified a critical challenge: data collected from one robot often cannot be easily transferred to another robot with different components. Huang from Bernstein indicated that data portability is an active area of research, with advancements anticipated. For example, AI models developed by Google DeepMind have shown promise in transferring skills across different hardware platforms.

China’s strategy for developing the humanoid robot industry through 2027 has been outlined by the Ministry of Industry and Information Technology. The plan emphasizes the need for large-scale training databases and high-quality multimodal data as essential components for creating the “brain” of humanoid robots. Local governments are eager to support this initiative, aiming to generate jobs and bolster future industries in their regions.

Despite the ambitious goals, leading researchers and engineers involved in data collection have expressed uncertainty about whether these efforts will yield the expected technological advancements. However, there is a tangible benefit: the purchases of robots by data collection centers have provided a boost to humanoid robot manufacturers in China, even as actual demand for these devices continues to develop. The Wuhan center has acquired dozens of robots from Shanghai-based Agibot at a cost of 350,000 yuan each. Analysts from Bernstein estimate that sales from data collection contributed to approximately 20% of the over 20,000 humanoid robot shipments in China last year.

Follow the latest developments and breaking updates in the Latest News section.

Published on 2026-03-14 19:36:00 • By Editorial Desk