A robot can 'fall' 10,000 times in simulation, but one fall on the production line is worth a thousand simulations.

At 0:00 AM, Zoomlion Smart Industry City in Changsha. Dozens of self-developed humanoid robots shuttle between logistics, assembly, machining, and quality inspection stations on an excavator assembly line, which produces one excavator every six minutes on average.

At 6:00 AM, Sany Group's Plant No. 18. AGVs and robotic arms operate on an unmanned production line, where a steel plate cutting, blanking, and sorting line runs automatically. Hundreds of industrial cameras complete thousands of inspections in milliseconds with a precision of 0.15 millimeters.

At 10:00 AM, in front of the IFS Tower in Changsha's Wuyi Commercial District. Inside a 16-square-meter intelligent service pod, a humanoid robot that once appeared on the Spring Festival Gala is busy picking and delivering goods. Residents place orders via QR codes, and the robot completes the entire process of identification, grasping, and delivery within 48 seconds.

Late at night, at Liuyang Lanshi Intelligent Yong'an Park. Quadruped robot dogs and wheeled robots roll off the production line, awaiting shipment to the next city. The park is designed to produce 500,000 embodied intelligent robots annually.

Individually, none of these scenes are remarkable.

Humanoid robots undergoing trial runs in factories can be found in Shenzhen, Shanghai, and Hangzhou; unmanned delivery vehicles navigating streets are common in Beijing and Suzhou; service pods delivering beverages are ubiquitous in Chengdu's new retail experiments. But what makes Changsha unique is that all these scenarios coexist in one city.

While cities nationwide are vying for algorithm talent, financing, and large-model infrastructure, what is Changsha competing for?

The answer lies in a document recently issued by the Hunan Provincial Government Office: cultivating intelligent robots as an emerging pillar industry during the '15th Five-Year Plan' period, aiming for industrial chain revenue to exceed 100 billion yuan by 2028 and 200 billion yuan by 2030. Changsha is explicitly designated as the 'provincial technology hub and innovation leader.'

This article dissects Changsha's strategic logic of transforming the entire city into a 'testing ground,' following a unique path of 'exchanging scenarios for data, nurturing intelligence with data, and feeding manufacturing with intelligence.'

01 Five-Ring Interlocking: Turning Heavy-Duty Scenarios into a Robot 'Driving School'

To understand Changsha's layout, consider its 'interlocking' industrial map: complete machine manufacturing, core components, scenario applications, research talent, and policy support—five rings tightly connected, each reinforcing the others.

In complete machine manufacturing, Lens Technology, a giant in Apple's supply chain, established robots as a core sector in 2025. Instead of building its own brand, it focuses on mass production and delivery.

By 2025, Lens aims to ship over 3,000 humanoid robots and 10,000 quadruped robot dogs, with annual revenue exceeding 1 billion yuan. Its Yong'an Park, launched in November 2025, has a designed annual capacity of 500,000 robots.

This gives Changsha a rare label (label): one of the few cities in the industry with both 'complete machine capacity and consumer electronics-level quality control.'

In core components, upstream robot enterprises account for 59% of Hunan's industrial output, according to the provincial Department of Industry and Information Technology. Changsha is not an 'assembly economy' but a complete ecosystem—Wanxin Precision (Ningxiang) breaks foreign monopolies in harmonic reducers, passing over 10,000 hours of continuous fatigue testing; Jingjia Micro is developing China's first physical AI chip for robots; Goke Micro's smart vision chips have shipped over 300 million units, holding the largest market share; Inovance Technology's (Yueyang) base produces 7 million servo motors annually, nearly 30% of the national total.

Scenario applications are the most critical part of Changsha's layout. Zoomlion has developed eight humanoid robots, with dozens deployed in trial runs for production line logistics, assembly, and quality inspection. It has built a 120-station training ground, with a humanoid robot factory set to launch by the end of 2026, planning to deploy 1,000–2,000 units over five years. Sany Group opens its scenarios to tech companies, with VisionBit's steel plate sorting line, launched in 2020, tripling productivity. Xingshen Intelligent's unmanned delivery vehicles operate routinely, while SuperRobot focuses on elderly care services.

In research and policy, the National Engineering Research Center for Robot Vision Perception and Control Technology, led by Academician Wang Yaonan, collaborates with National University of Defense Technology, Central South University, and Hunan University to ensure a steady talent supply.

On the policy side, provincial implementation plans deploy 'Ten Key Projects' and launch regular 'Intelligent Robot+' initiatives, establishing a 10 billion yuan future industry sub-fund. The Hunan Embodied Intelligence Innovation Center (Changsha Economic Development Zone, jointly operated by Lens and Zhiyuan) has attracted nearly 30 enterprises, while the Xiangjiang New Area hosts 532 AI and sensor companies with an output value exceeding 30 billion yuan.

The interlocking of the five rings lies in each ring creating value for the others—complete machine manufacturers provide orders, scenario companies supply data, universities provide talent, and policies provide support. This distinguishes Changsha from 'single-breakthrough' cities, requiring the entire chain to operate within one city.

Why can Changsha build this chain? It relies on its 'heavy-duty legacy.' The training data most needed for embodied intelligence comes from real-world scenarios, and Changsha is one of China's cities with the densest heavy-duty applications.

Changsha's construction machinery cluster is among the first national advanced manufacturing clusters, ranking first in scale for 16 consecutive years. Five enterprises—Sany, Zoomlion, China Railway Construction Heavy Industry, Sunward Intelligent, and Xingbang Intelligent—ranked among the top 50 global construction machinery manufacturers in 2026, accounting for 9.5% of the global market and 45% of China's share.

This means mines, construction sites, ports, and production lines surround Changsha, providing the most sought-after yet hardest-to-obtain training grounds for embodied intelligence—scenarios characterized by 'dirt, fatigue, danger, and non-standardization,' precisely where traditional industrial robots fail and general-purpose large models fall short.

This is the raw material for Changsha's strategic moat. Simulated environments can be built with money, but 'real-world physical data' cannot be accelerated by capital—it takes decades of 'dirty, tiring, and dangerous' work. While others 'feed' models in data centers, Changsha turns excavator production lines and mining sites into a robot 'driving school,' where the first product produced daily is not equipment but data.

02 Data is 'Earned,' Orders are 'Chosen'

With each rotation of the production line, Changsha accumulates more data. Every time a humanoid robot at Zoomlion completes a handling cycle, it leaves a complete data chain containing trajectory, force, and visual information. Sany Heavy Industry's steel plate sorting line records tens of thousands of irregular part sorts daily, feeding VisionBit's 'Kunwu' platform for cross-industry reuse.

However, data does not become intelligence on its own. A 'data flywheel' drives the flow, circulation, and value appreciation of this raw material, addressing an industry-wide pain point: the sim-to-real gap.

Friction, tolerances, lighting, and unexpected failures—these physical details cannot be learned in simulation. Hence, the industry adage: 'A robot can fall 10,000 times in simulation, but one fall on the production line is worth a thousand simulations.'

Changsha's solution is to turn production activities themselves into data generation, precipitate (precipitating) physical details like tolerances, lighting, stress, and failures into training samples—'working is data.'

Two typical (typical) cases illustrate this.

Zoomlion's self-developed humanoid robots start at 'support stations' like pre-assembly and quality inspection, gradually collaborating with humans rather than replacing them outright. Supported by its self-developed RobotOps operating system and 'Yungu' embodied intelligence large model, the 120-station training ground collects thousands of real-machine data points daily, completing nearly 20 scenario validations.

Every additional station the robot masters means more 'earned data.'

The second case is VisionBit's '10-month factory immersion.'

In 2019, startup VisionBit tackled Sany Heavy Industry's steel plate sorting challenge—tens of thousands of irregular parts that traditional industrial robots could not handle. The team spent 10 months debugging inside Sany's lighthouse factory.

In August 2020, the first fully automatic sorting line launched, tripling productivity and reducing sorting labor by over 60%. This capability was later replicated across industries, securing nearly 200 million yuan in shipbuilding smart manufacturing orders in Q1 2026 and completing a 100+ million yuan B++ round. VisionBit's 'Kunwu' platform reduced project debugging cycles from six months to under two weeks.

In both cases, data was not collected—it was 'earned.'

Scenario usage generates data, which trains models. Improved capabilities lead to more complex tasks, producing more scenario data. The production line is the only place forming a self-reinforcing loop, distinguishing a 'city as testing ground' from a 'park stacked with enterprises.'

Whether this chain can keep turning depends on a 'self-sustaining mechanism,' evidenced not by promotional PPTs but by corporate choices—many external firms are betting on Changsha with orders and investments.

Why did Zhiyuan Robotics entrust orders to Lens? Headquartered in Shanghai, Zhiyuan has deepened its collaboration with Lens, co-operating the Hunan Embodied Intelligence Innovation Center. In July, they signed a smart production line project agreement in Huizhou, commissioning Lens for manufacturing.

Changsha's production capacity and quality control have become options for top players—the first layer of the 'self-sustaining mechanism.'

The same logic applies to component suppliers. As mentioned, Changsha boasts a complete ecosystem of complete machines, components, and scenarios, with short supply chain radii—the most fundamental logic in manufacturing.

Dissecting Changsha's development logic reveals it attracts not 'pure algorithm teams' but 'teams needing scenarios and manufacturing.' Conversely, scenario providers like Zoomlion and Sany Heavy Industry need external intelligence capabilities, forming a two-way demand that sustains the 'self-sustaining mechanism.'

Meanwhile, policy amplifiers take effect. Provincial implementation plans guide construction machinery and other fields toward core components through 'industrial chain closure' and 'advantageous manufacturing capability transformation' projects. The 10 billion yuan sub-fund invests 'early, small, long-term, and in hard tech' across the chain.

Thus, Changsha's 'self-sustaining mechanism' relies not on sentiment but on interlocking interests—each ring profits from and learns data from the others.

03 Testing Ground or Assembly Site? Misalignments in Capacity, Data, and Scenarios

Before judging this layout, place Changsha in the national context. Objectively, Changsha has scenarios and manufacturing but cannot match first-tier cities in algorithm talent density, capital activity, or service scenario data reserves.

Strengths and weaknesses are two sides of the same coin. Flipping the coin reveals three structural cracks, amplified by the metaphor of 'city as testing ground.'

The first crack: testing ground or 'assembly site'? Are firms coming to Changsha for 'scenarios' or 'capacity'?

The most direct test occurred in July, when Zhiyuan and Lens signed a smart production line project, commissioning Lens in Changsha but locating the project in Huizhou's Embodied Intelligence Industrial Park. Coupled with Zhiyuan's May launch of a Southwest base in Chengdu's Pidu District, leading players are expanding capacity nationwide—Changsha is just one option.

The gap between 'attracting' and 'retaining' is real. Whether firms 'pass through' or 'take root' is the true test of the testing ground's quality.

The second crack: who owns the data?

The premise of interlocking rings is data flow, but Zoomlion's production line data belongs to Zoomlion, and Lens's manufacturing data belongs to Lens. Will corporate data 'islands' prevent the 'city-level data flywheel' from becoming reality? Who builds the data mid-platform? Currently, Changsha lacks a clear production line data-sharing mechanism.

If data flows only within firms, 'city as testing ground' becomes merely a collection of 'enterprise as testing grounds,' missing the critical link in the five-ring interlock.

The third crack: heavy-duty 'specialization.'

Changsha's scenario portfolio focuses on heavy industry—excavators, cranes, mines, ports. This is an advantage but also a limitation.

As the industry shifts toward household, commercial, and service scenarios, how versatile are robots 'trained for heavy work'? Xiangjiang New Area bets on breakthroughs in dexterous hands, bionic joints, and other 'Changsha-made robot components.' Zhongke Huisi launched three dexterous hand models in the New District (new area), with Academician Wang Yaonan stating, 'Dexterous manipulation is the last centimeter for robots to enter the real world.' This corrects 'specialization,' but service scenario data density remains a short-term weakness for Changsha.

Notably, these three cracks are interconnected, pointing to the same root: the 'testing ground' follows public good logic, while firms operate under private property logic.

Governments want data to flow city-wide into a flywheel, but firms treat production line data as core assets. Governments want firms to take root and build capacity, but firms naturally compare locations nationwide. When public and private logics collide, the testing ground's success depends on Changsha's governance design to bridge gaps—e.g., establishing data-sharing mechanisms and providing rooting certainty.

No ready answer exists, but it is the biggest variable for Changsha's 2028 100-billion-yuan goal. Can Changsha upgrade the 'testing ground' into a 'mass production base + data hub,' convincing firms to build capacity locally, data to flow freely, and diversifying scenario portfolios beyond heavy-duty applications? The answer will not appear in policy documents but in each negotiation between firms and the city.

04 Conclusion

China's urban industrial competition in embodied intelligence is emerging along two strategic logics.

One is the 'talent + capital' aggregation model in Beijing, Shanghai, Shenzhen, and Hangzhou, betting on general-purpose intelligence breakthroughs through high valuations and dense algorithm talent. The other is Changsha's 'scenario + manufacturing' interlocking model, betting on faster data accumulation in the physical world through a solid manufacturing base and dense real-world scenarios.

If the 'scenarios-for-data' path is validated, the competition will shift from 'talent grabbing' to 'scenario grabbing,' offering opportunities for all manufacturing cities. Changsha is merely an early exemplar.

Changsha's report card will be graded in 2028 (the 100-billion-yuan target year). The question remains: if general-purpose large models achieve generalization capabilities tomorrow, will the 100,000 hours of operational data accumulated on Changsha's production lines be assets or waste?

The answer may better define the 'city as testing ground' than any industrial plan.

* Image sourced from the internet. Please contact for removal if infringement occurs.