China’s Robots Move From Showpieces to Factory Work

China’s robotics industry is entering a more demanding phase. At the 2026 World Robot Conference (WRC) in Beijing, companies still displayed dancing and athletic robots, but the strongest commercial message came from machines sorting parcels, loading production lines, restocking goods and performing inspection tasks. The shift is from asking whether a robot can perform an impressive movement to asking whether it can complete useful work repeatedly, safely and at an acceptable cost.

WRC’s focus shifted from spectacle to workflow

The conference’s scale reflected the growing ambition of China’s robotics sector. More than 300 companies exhibited at WRC, with reports variously describing over 2,000 exhibits and, in broader counts, more than 3,000 products and over 300 debuts. The difference appears to reflect how organizers and media counted products and exhibits, but the direction was consistent: WRC was increasingly organized around applications and buyers rather than isolated technology showcases.

The most revealing demonstrations connected several steps into a working process:

• UBTECH presented a coordinated logistics line involving about 10 humanoid robots. The machines lifted and stacked boxes, sorted goods, and replenished materials, illustrating how multiple robots could be assigned complementary tasks.
• Galaxea showed a robot locating, retrieving and packaging a bottle ordered through a QR code. The company said the system took just over a minute and was already operating in Beijing and other cities. That is a company claim about deployment, rather than independent proof of large-scale operation.
• Other booths featured parcel sorting, mobile-phone packing, welding, loading, shelf restocking, cooking and household chores. These tasks test manipulation and workflow integration more directly than dancing or racing does.

The conference also divided applications across manufacturing, food service and retail, healthcare and elder care, and emergency rescue. In practice, the nearer-term opportunity is likely to be concentrated in structured environments—factories, warehouses and inspection routes—where tasks, safety boundaries and operating conditions can be defined.

Manufacturing and logistics are the first proving grounds

China’s automakers are among the most important potential customers. Reuters reported that nearly 10 major automakers, including BYD and SAIC, were testing embodied-intelligence systems for material handling and parts loading on production lines. Unitree also said its robots were being tested on automotive assembly lines and in its own factories.

These applications matter because they offer a clearer commercial test than a one-off public demonstration. A factory robot must repeat a task over long operating periods, recover from errors, work alongside existing equipment and justify its purchase or operating cost. A warehouse robot faces similar requirements, including reliable object recognition, safe navigation and integration with ordering and inventory systems.

WRC’s exhibitor mix reflected that systems challenge. Algorithm developers, 3D-vision and battery suppliers, robot manufacturers and industrial users were brought into the same marketplace, pointing to an effort to shorten the path from components and training data to customer-ready deployments.

Training infrastructure is being built around real tasks

Physical robots need more than a large language model and a polished prototype. They need data from cameras, force sensors and human demonstrations, plus repeated practice in environments that resemble the places where they will work.

China has more than 70 embodied-intelligence training grounds, with another 46 reported to be under construction or planned. These facilities are intended to collect teleoperation and sensor data and let robots repeat practical tasks. A national pilot base in Hangzhou, for example, was reported to be training more than 140 robots across over 40 job-oriented scenarios, including kitchens, warehouses, farms and a mock mine tunnel.

This infrastructure supports a feedback loop: robots attempt tasks, operators correct failures, new data improves the models, and updated systems are tested again. It does not guarantee general-purpose autonomy, but it addresses one of the central bottlenecks in embodied AI—the shortage of high-quality real-world data linked to measurable task performance.

Policy and capital are pushing the industry toward deployment

Beijing has moved embodied intelligence from a specialist research topic into a national industrial priority. The government included “embodied intelligence” in its 2025 work report, while the 15th Five-Year Plan for 2026–30 calls for forward-looking development of embodied AI and mechanisms to fund emerging industries and share their risks.

A national action plan for humanoid robots and real-world embodied-AI training aims to identify more than 100 high-value application scenarios. Related reporting describes a focus on manufacturing, public services and specialized tasks, including production, inspection and maintenance.

The market is expanding alongside that policy support. China’s robotics-sector revenue was reported to have exceeded 300 billion yuan in 2025, following average annual growth above 20% over the previous five years. First-half 2026 revenue was reported at 165.5 billion yuan, up 24.5% year on year. These figures describe sector revenue, not proof that humanoid robots themselves are already profitable at scale.

Investment enthusiasm was particularly visible in Unitree’s August 2026 listing on Shanghai’s STAR Market. It became the first humanoid-robot maker listed on China’s A-share market, and its shares opened 629.44% above the initial public offering price before closing 460.34% higher. The debut demonstrated investor appetite, but a strong listing is not the same as evidence of sustained customer demand or operating economics.

The “ChatGPT moment” remains a forecast

The industry’s long-term goal is not simply a robot that performs one task well. It is a machine that can enter an unfamiliar environment, understand a natural-language instruction and complete most of the required work without extensive task-specific programming.

Unitree founder Wang Xingxing described that capability as the robotics equivalent of a ChatGPT moment. He suggested an optimistic timeline of two to three years and a latest window of five to 10 years. Other industry commentary has placed a similar milestone around the end of 2028, but these are forecasts from company and industry figures, not an established consensus.

The evidence from WRC supports a narrower conclusion: China is moving quickly from technical demonstrations toward pilots and scenario-based deployment. It does not yet establish that general-purpose humanoid robots can operate reliably and economically across the open-ended range of tasks people expect from them.

What to watch next

The next phase should be judged by operational evidence rather than choreography. The most useful questions are:

1. Can the robot work repeatedly? Look for disclosed uptime, task-success rates and performance over extended periods.
2. How much human intervention is required? A robot that needs continuous teleoperation is not equivalent to an autonomous worker.
3. Can it recover from failure? Real workplaces contain misplaced objects, changing layouts and unexpected obstacles.
4. Does the economics work? Purchase price, maintenance, energy use, training and supervision all affect the business case.
5. Is there a paying customer? A demonstration, pilot and repeat commercial order are different stages of adoption.

China’s robotics transition is therefore real but incomplete. WRC 2026 showed a sector building the industrial partnerships, training infrastructure and policy mechanisms needed for scale. The decisive test will come when robots must deliver dependable output outside the exhibition hall—and continue doing so after the novelty disappears.