China's Robot Schools Train AI to Navigate Physical World Through Hands-On Learning

2026-03-10 data

Beijing, Tuesday, 10 March 2026.
China’s specialized robotics training centers are revolutionizing AI development by teaching robots physical skills through human demonstration and data collection. With over 50 government-backed facilities established since 2025, these schools focus on ‘embodied intelligence’ - AI that can interact with the real world. Chinese startup Spirit AI recently raised $280 million to scale this technology using ‘dirty data’ from real-world scenarios, while training centers in cities like Hefei and Guangzhou prepare robots for manufacturing, retail, and maintenance tasks through repetitive human-guided practice sessions.

China’s Strategic Focus on Physical AI Development

China’s commitment to physical AI technology reflects a broader strategic initiative outlined in the country’s new five-year plan, adopted during the week of March 9, 2026 [1]. This national strategy emphasizes technological innovation with a specific focus on ‘fysieke AI’ (physical AI), which refers to robots equipped with artificial intelligence brains [1]. According to Jeroen Groenewegen-Lau from the Mercator Institute for China Studies (Merics), physical AI represents ‘one of the most useful applications’ under this technological umbrella [1]. The initiative aims to reduce China’s reliance on foreign technology, particularly from countries like the United States, as the nation seeks greater technological self-sufficiency [1]. This strategic positioning comes at a time when the global market for humanoid robots, currently valued at several billion euros, is projected by Barclays to reach €170 billion by 2035 [1].

Training Centers Emerge as Data Collection Hubs

The robotics training facilities, officially designated as data collection centers, have proliferated rapidly across China since 2025 [1]. More than 50 such centers now operate nationwide, with most receiving direct or indirect government support [1]. The training center in Hefei, established by the government in May 2025, serves as a collaborative hub where eight companies, including VSTC, jointly collect training data [1]. Twenty robot trainers currently work at the Hefei facility, using innovative methods to teach robots basic physical tasks [1]. Huang Haoyu, a 20-year-old robot trainer at the center, employs Meta VR headsets and joysticks to guide robots through fundamental movements such as picking up bags of chips [1]. The training methodology addresses current limitations in robot capabilities, as Huang explains: ‘It’s not their fault if their movements are slow or limited. That’s simply because they don’t have enough data’ [1].

Real-World Applications and Industrial Implementation

The practical applications of China’s physical AI development are already demonstrating measurable success in industrial settings. Spirit AI’s technology has been deployed in production lines at CATL, one of the world’s largest battery manufacturers, where AI-controlled robots handle complex tasks such as processing flexible cable harnesses [3]. These applications address manufacturing challenges that are difficult to automate due to material properties, with the AI systems achieving a success rate exceeding 99 percent while operating with precision and cycle times comparable to experienced production workers [3]. The Guangdong Province Embodied Intelligence Training Field, located in Guangzhou’s Haizhu District, represents another significant development in this ecosystem [6]. This 5,400 square meter facility focuses on systematic robot training across multiple scenarios, including automotive assembly SPS sorting, building and workshop inspections, and retail operations [6]. The facility’s seven technical platforms create a comprehensive support system covering data collection, simulation, training evaluation, and application deployment [6]. As one technical staff member explained, the goal is to ‘systematically teach robots to work and get better at it’ through this integrated approach [6]. The center plans to expand in 2026 with additional scenarios covering commercial services, industrial manufacturing, intelligent inspection, home care, and medical services across more than 20 sub-scenarios [6].

Bronnen

physical AI embodied intelligence