At a pioneering robot academy in Hangzhou, east China’s Zhejiang Province, four out of thirty robotic trainees have completed a rigorous two-month training program and transitioned into formal employment. Their assignments are diverse: one serves as a museum docent in Hangzhou, another leads tours across the robot school itself, a third oversees industrial inspections at a Nestlé manufacturing plant in Tianjin, and the fourth has journeyed to the Republic of Korea to stage live piano performances.
These deployments underscore a defining challenge in operationalizing embodied AI: preparing robots for dynamic, physical environments demands far more resources than individual developers or enterprises can shoulder independently. Instead, it necessitates centralized, shared infrastructure where autonomous systems can be trained, evaluated, and refined at scale.
In Xiong’an—the emerging “city of the future” located roughly 105 kilometers southwest of Beijing—an array of robotic models is mastering everyday tasks, including waste disposal, shelf stocking, and material transport, within a purpose-built facility. Human trainers guide the units through 1:1 replicas of actual commercial and public environments, operating them via virtual-reality headsets and wearable exoskeletons.
Upskilling a robot demands patience comparable to training a human worker. “Teaching the robot the baseline mechanics of grasping an object required over 1,000 repetitions,” explains Liu Rui, an instructor with a digital urban technology subsidiary of China Xiong’an Group. Executing the complete operational chain—retrieving waste, returning to an upright posture, and depositing it into a receptacle—demanded an additional week. Each execution logs vital training data; instructors subsequently label the frames, deconstruct continuous actions into discrete sub-steps, and annotate physical entities to bolster the machines’ spatial recognition and situational decision-making.
Data from the China Academy of Information and Communications Technology indicates that as of late June, China had commissioned over 70 dedicated robot training facilities, with more than 40 additional centers currently in development or pre-construction.
This physical network is becoming an integral pillar of China’s broader artificial intelligence and robotics ecosystem. According to the Ministry of Industry and Information Technology, the nation hosted more than 6,200 AI enterprises in 2025, with its core sector valued in excess of 1.2 trillion yuan (approximately 173.9 billion U.S. dollars). Official records also reveal that China engineered over 400 complete humanoid models in 2025, representing more than half of the worldwide total. Furthermore, figures presented at the 2026 World Robot Conference revealed that domestic manufacturers accounted for 97 percent of global humanoid robot shipments in the first half of 2026.
Jiang Lei, chief scientist at the National and Local Co-Built Humanoid Robotics Innovation Center, notes that while embodied AI has successfully advanced from “zero to one,” scaling operations from “one to ten” poses ongoing friction. The primary bottlenecks reside in standardized protocols, engineering governance, and pilot testing—anchored fundamentally by how effectively machines adapt across varied physical settings.
China’s vast manufacturing baseline affords robotics firms not only an expansive domestic market, but also a rich spectrum of real and simulated testing environments to stress-test their hardware and algorithms. Consequently, specialized training facilities are turning behavioral datasets into a high-value economic asset necessary for embodied AI adoption.
While large language models and multimodal frameworks have advanced swiftly on web-scale text and media archives, embodied AI requires data generated through physical interaction. Because operational demands vary significantly across factories, logistics hubs, and culinary venues in terms of applied torque, precision, and range of motion, acquiring high-quality physical datasets remains costly. Industry observers emphasize that much of this operational data has historically lacked systematic collection, processing, and curation.
In 2025, as China Xiong’an Group aligned its digital framework with the physical municipality, leading foundational model developers sought to acquire tens of millions of hours of its high-grade interaction datasets. At the Xiong’an facility, these datasets train advanced models and populate a centralized urban data repository available for future sharing and commercial exchange, notes Cao Yue, a managerial lead at the group’s subsidiary—effectively establishing the training center as a high-volume data refinery.
Similarly, at a national pilot base in Hangzhou, enterprises can access pooled computational capacity, data repositories, foundational model services, and real-world scenario validation via an open technical infrastructure. “Previously, building a single physical application meant spending tens of millions of yuan on dataset collection alone. Participating in this hub enables startups to access national innovation resources under one roof, slashing preliminary pilot R&D expenditure,” observed Zhang Haiwei, founder of a local robotics venture.
The underlying rationale is practical: rather than requiring every firm to independently develop private training grounds, pipelines, and evaluation sandboxes, shared platforms distribute capital expenditures and accelerate the translation of laboratory prototypes into production-ready machinery.
As robotic systems increasingly enter the workforce, questions regarding the role of human personnel inevitably follow. Yu Jiang, an academic at the Chinese Academy of Sciences, highlights that technological inflection points regularly displace certain occupations while generating entirely new ones: “When automobiles superseded horse-drawn transport, coachmen transitioned into motor vehicle drivers.”
Professor Xiong Rong of Zhejiang University projects that machines will progressively assume repetitive, physically demanding, and hazardous responsibilities, freeing personnel to prioritize managerial, technical, and creative oversight. Reflecting this evolution, China formally added 11 new roles to its national occupation registry earlier this month, including embodied AI robot application technicians and AI agent developers. Analysts conclude that continuous learning and targeted reskilling will prove indispensable as workers transition into an integrated workplace model that fuses human judgment, empathy, and innovation with robotic endurance and precision.
https://english.news.cn/20260928/fb09138c57f24c18bf896c09e9b36070/c.html