_________

Published on Oct. 14, 2026

Changsha is opening its factories for hands-on physical AI training.a Zoomlion, the global construction machinery giant, is validating robots on production lines. Sany, another top-tier heavy machinery conglomerate, is opening industrial problems to outside developers.

Both initiatives reflect a broader strategy in Changsha, a city with five million peopel in central China’s Hunan province, to turn the city’s manufacturing base into a practical training ground for physical AI, the Chinese technology publication Huxiu reported on October 8, 2026 (in Chinese).

Physical AI enables machines to perceive their surroundings, make decisions, and perform physical tasks. This is not a one-directional relationship, however. While machines become more productive with AI, the vertical models and applications need machines to train.

Changsha is not only home to construction-machinery producers Zoomlion and Sany Heavy Industry. It is a manufacturing powerhouse. Five companies from the city feature among the 13 Chinese manufacturers in the global construction-machinery rankings cited by Huxiu for 2026.

“Many places talk about ‘robotics + industry’. We believe it should be ‘industry + robotics’,” Yang Yue, chairman of Siasun Cobot, was quoted by Xinhua as saying during a recent visit to the city.

Shenzhen or Shanghai build humanoid-robot clusters whose producers are searching for industrial use cases. Changsha starts with the industry it already has and gives humanoids a clear raison d´être.

Zoomlion has built an embodied-intelligence training facility with 120 workstations. Dozens of robots collect data and validate their capabilities across nearly 20 applications, including assembly, logistics and quality inspection, Huxiu reported. These trials draw on a manufacturing environment with more than 300 intelligent production lines and over 2,000 industrial robots.

Those industrial robots are different from the humanoid robots now being tested. Those existing industrial robots are separate from the humanoids now being tested. They include multi axis articulated robotic arms for precision welding, automated guided vehicles for workstation logistics, and automated paint spraying systems. It is a kingdom of cobots and legless but reliable workhorses that humanoids are now beginning to enter.

The companies believe that humanoids can increase productivity in environments designed for humans. “Drawing on the enormous volume of industrial scenario data we have accumulated over more than a decade, a group of humanoid robots are now ‘learning while working’ on the production line,” She Lingjuan, deputy general manager of the Robotics Division at Zoomlion’s Zhongke Yungu, was quoted by Xinhua as saying.

The attraction of the factory for developing humanoids, regardless of their final use case, is that it gives them a structured and repeatable environment to learn under tough operating conditions. Dust, vibration and glare can disrupt a machine’s work. Keeping it running reliably through a shift requires engineering knowledge that manufacturers have accumulated over decades.

At independent workstations, robots carry out specialized tasks while collecting production data. Zoomlion's robots operate night shifts and take over unfinished tasks left behind by the shift before. That ties training to a factory’s immediate production needs.

This, as well, is a two-way street. Humanoids are not simply “fitting in”, they are slowly changing the way factories are built.

“We are not developing humanoid robots simply to replace human labour. Based on a deep understanding of their characteristics, we are jointly redesigning manufacturing processes and management systems,” Zang Bing, deputy director of the Intelligent Manufacturing Center at Zhongke Yungu, was quoted by Xinhua as saying. (in Chinese)

Zoomlion plans to complete a humanoid-robot factory by the end of 2026 and deploy 1,000 to 2,000 embodied-intelligence robots in its smart factories worldwide over the next five years. These are company announcements and to be treated with caution, but the figures sound achievable.

Sany is pursuing a different connection between industrial demand and AI development. Its competition with Central South University attracted 134 teams across six fields, including construction machinery, smart mining and intelligent manufacturing. Successful projects could gain access to investment funds and industrial demand, which could ideally foster robotic startups with customers.

Local specialist companies in Changsha show how this highly pragmatic approach extends beyond humanoids, into various applicatons of physial AI. Speedbod develops industrial physical AI and 3D machine vision software that acts as the "eyes and brain" for heavy machinery and can be used for industrial inspection, among other things.

Its products are already used in automotive and construction-machinery manufacturing. Such applications also tie AI to an identifiable production task.

Changsha's well-established manufacturing expertise also feeds the robotics supply chain. Lens Technology, a big supplier of Apple, is has started to transfer its precision-manufacturing capabilities into a new business with robotics components. The company's joint modules, dexterous hands and controllers are entering volume delivery to humanoid makers such as AgiBot, Xinhua reported. To build a useful two-legged robots needs top hardware just as much as it needs intelligence.

In Changsha, entrepreneurs and official believe that this is their sweet spot. “Completing an action once in a laboratory and repeating it hundreds or thousands of times every day on a production line are two completely different things,” an engineer at Zoomlion said.