Signals Inbox·August 27, 2026·Humanoid Robotics

Are Chinese robots actually useful?

Yes, Chinese robots are already useful at serious scale. The surprise is that the strongest evidence still comes from specialized warehouse, factory, cleaning and mining machines, while humanoids are only beginning to prove themselves on narrow industrial jobs.

Q1Why do Chinese robots suddenly look much more useful?

Chinese robots look far more useful today because China has moved from producing impressive prototypes to shipping robots in volumes that were hard to imagine a year ago.

At the latest World Robot Conference in Beijing, an industry report from the Chinese Institute of Electronics said China shipped more than 40,000 humanoid robots in the first half of 2026, representing 97% of global shipments under its methodology. Omdia had counted only 13,318 humanoids shipped worldwide during all of 2025. The definitions are not perfectly comparable, but the jump is too large to dismiss as statistical noise.

Unitree gives us another reference point. The company says it delivered more than 5,500 humanoids to end customers in 2025 and produced more than 6,500. AgiBot shipped 5,168 according to Omdia. UBTECH sold 1,079 full-size humanoids. China has gone from dozens of prototypes competing for attention to several manufacturers producing four-digit volumes.

The robots themselves are also attempting harder things. At the latest World Humanoid Robot Games, more than 2,000 humanoids competed in sports, factory work, logistics, hotel tasks and household activities. Some of the viral performances were still entertainment, but the newer tests included plugging cables, stocking shelves, manipulating small objects and folding clothes.

So the question has changed. Building a humanoid robot is no longer the unusual achievement. We now need to know whether these machines can do enough useful work, for long enough and cheaply enough, that someone would keep paying for them.

Q2Are Chinese robots already doing useful work at scale?

Yes. Chinese-made robots are already doing useful industrial work at huge scale, even before we get to humanoids.

The International Federation of Robotics counted 2.027 million industrial robots operating in Chinese factories in 2024, the largest installed base in the world. China added another 295,000 during that year, equal to 54% of all new industrial robot installations globally.

Those figures include foreign brands, which used to dominate China's market. The more interesting change is who now supplies the machines. Chinese manufacturers took 57% of domestic industrial robot installations in 2024, up from 47% one year earlier and 30% in 2020. Applying that 57% share to annual installations gives roughly 168,000 locally supplied industrial robots installed in a single year.

Chinese suppliers are already particularly strong in some industries. IFR says they supplied 59% of robots installed in China's electronics sector and 85% in metal and machinery.

That gives us a useful baseline. If "Chinese robots" means robots made by Chinese companies, they are already performing real factory work by the hundreds of thousands. The uncertainty begins when we narrow the question to the new generation of humanoid and general-purpose robots.

Chinese industrial robotics at scale

Q3Are the most useful Chinese robots actually humanoid?

No. The Chinese robots creating the most measurable economic value today are still specialized machines built around one job.

Warehouse robots move shelves and pallets. Cleaning robots scrub huge commercial floors. Autonomous trucks move material through mines. Industrial arms weld, assemble and handle components. These systems sacrifice human-like versatility in exchange for speed, reliability and much simpler operating environments.

That trade-off has already produced large businesses. Geekplus has turned warehouse robotics into a profitable operation. Pudu has found much stronger demand for commercial cleaning than for the restaurant robots that first made the company famous. Chinese mines are running fleets of autonomous haul trucks around the clock.

Humanoids are interesting precisely because they could eventually automate jobs that remain awkward for those specialized systems. Today, specialized Chinese robots set the economic benchmark humanoids still have to catch.

Q4Do Chinese warehouse robots really save workers time?

Yes. Chinese warehouse robots already produce productivity gains large enough to explain why customers keep buying more of them.

Geekplus's early deployment at Walmart's Shenzhen distribution center gives us a clean example. Walmart said picking productivity went from 149 items per worker per hour to 533 after Geekplus introduced its shelf-to-person robots. That is roughly a 3.6-fold increase, mainly because workers stopped spending so much time walking between products.

A newer Schneider Electric deployment shows that the technology has moved beyond mobile shelves. Geekplus installed an AI robotic picking station at Schneider's Shanghai warehouse that the companies say roughly doubled picking throughput, exceeded 99.99% accuracy and went live within 48 hours.

The aggregate numbers are more convincing than either case study. Geekplus reported RMB3.17 billion in revenue for 2025, up 31.6%, while adjusted profit and operating cash flow both turned positive. By year-end, it had shipped more than 72,000 robots to roughly 950 customers across more than 40 countries. Its annual report put the customer repurchase rate at about 78%.

That 78% figure is particularly useful. Companies may buy one robot system because they want to experiment. Reordering after living with the first deployment is much stronger evidence that the robots are solving a real problem.

Q5Are Chinese cleaning and mining robots genuinely useful?

Yes. Cleaning robots and autonomous mining vehicles may be less exciting than humanoids, but they already handle repetitive work with very clear economics.

Pudu Robotics is a good example of customers pulling a robot company toward the boring task that works best. Pudu became famous for restaurant delivery robots, yet commercial cleaning generated more than 70% of its revenue in 2025. Revenue more than doubled that year, and the company has now shipped more than 130,000 robots across more than 85 countries and regions.

Cleaning is a good robot job because the environment is repetitive and easy to measure. A mall, supermarket, airport or warehouse has thousands of square meters of floor that need to be cleaned again tomorrow. Customers can quickly work out whether a robot covers enough area, requires too much supervision or saves enough labor.

Mining pushes the same logic much further. At the Yimin open-pit mine in Inner Mongolia, China Huaneng, XCMG, Huawei and other partners deployed 100 autonomous electric haul trucks. Each carries up to 90 tonnes and can operate continuously in temperatures below minus 40 degrees Celsius. Huawei, one of the project partners, says earlier autonomous operations went from 87% of human-driver productivity to 120% after several years of iteration.

Removing a driver from a dangerous mine creates immediate value even when the machine cannot do anything else. That is why specialized autonomy has moved faster than general-purpose robotics.

Q6Are Chinese humanoid robots actually working in factories today?

Yes. Chinese humanoids are currently doing real production work, although the deployments remain small compared with conventional factory automation.

AgiBot provided one of the clearest demonstrations at Longcheer Technology's tablet factory in Nanchang. Eight Lingxi G2 robots spent six days inside a live mass-production workflow. AgiBot reported more than 64 operating hours, 64,828 production-line actions across at least four workflows and a 99.99% task success rate. The robots contributed to output of 17,625 units.

Longcheer separately described the same deployment and said the robots worked inside the tablet-testing process, performing repetitive pick-and-place, transfer and sorting work. Human employees remained responsible for broader inspection, coordination and handling unusual situations.

CATL has gone further with Galbot. The battery manufacturer says the Galbot S1 is now in regular operation on its production lines, handling material movement and picking during module and battery-pack manufacturing. The robot has a 50-kilogram dual-arm payload and up to eight hours of battery life.

SAIC-GM has also put an AgiBot-developed wheeled humanoid into the Buick Electra E7 battery production line to pick and load battery cells. UBTECH's Walker series is being used or tested across automotive and electronics manufacturing for handling, sorting, parts work and inspection.

Factories have therefore moved beyond simply inviting humanoid companies in for a one-day demonstration. What remains limited is the number of robots, the range of jobs and the amount of independent decision-making those jobs require.

Chinese humanoids already working in factories

Q7What jobs can Chinese humanoid robots reliably do right now?

Chinese humanoids currently work best when the job involves moving known objects through a fairly predictable environment.

UBTECH's own annual report is unusually specific about the jobs it is targeting. Walker robots handle totes at different heights, sort parts, remove soft protective sleeves from bolts, inspect production kits, plug and unplug charging connectors and perform visual quality checks. These are real physical tasks, but the environment and expected actions are tightly defined.

Galbot's CATL deployment follows the same pattern. The S1 carries and picks industrial materials. SAIC-GM uses its AgiBot system to grasp battery cells and load them into a production process. Longcheer's robots move tablets between known testing positions.

The latest humanoid competitions also show where the frontier sits. Robots have become much better at folding clothes, stocking shelves and manipulating tiny objects over the past year. Yet judges at the latest games noted that only a small number could coordinate both arms well enough to hang clothes on a hanger. A task that feels trivial to a person can still expose weak perception, poor grip control or a failure to recover from an unexpected position.

The useful humanoid today is closer to a flexible piece of factory automation than to a human worker made of metal. Give it the right workstation and a narrow set of jobs, and it can contribute. Move it into a completely unfamiliar environment and performance drops quickly.

Q8Are Chinese humanoid robots as productive as human workers?

No. Chinese humanoids are still much slower than people on many of the factory tasks they are supposed to automate.

UBTECH gave the Financial Times a surprisingly candid benchmark. The company said its Walker S2 was achieving around 30% to 50% of human productivity on suitable jobs such as box stacking and quality inspection. UBTECH wants to reach about 80% by 2027.

That gap changes the economics. A robot costing less per hour only helps if it can complete enough work, stay online for long periods and avoid creating extra supervision for employees. A machine running at 40% of human speed needs major advantages elsewhere to compensate.

Chinese robot makers are trying to squeeze more hours out of the machine. UBTECH's Walker S2 can swap its own battery in roughly three minutes, which is meant to support continuous operation. Galbot says its CATL robot can run for up to eight hours between charges.

Speed is the part worth watching. Humanoid factories can look futuristic even when the worker standing next to the robot would finish the same job twice as fast. For now, the productivity numbers keep the humanoid story grounded.

Q9Why would a factory use a humanoid instead of a normal robot?

Factories would use humanoids when they need flexibility more than maximum speed.

A fixed industrial arm is hard to beat when a factory repeats the same motion millions of times. The machine can be positioned perfectly, fenced into a controlled area and optimized around that one movement.

A lot of remaining factory labor looks different. Workers carry bins, reach shelves at several heights, move between workstations, use equipment designed for human hands and switch jobs when a production line changes. Automating each of those tasks with a separate custom machine can become expensive.

Longcheer described exactly this problem during its work with AgiBot. Electronics manufacturing changes quickly, and fixed automation can struggle when products, testing equipment or workflows move. A humanoid that learns a new routine and works with existing human-designed equipment could make more sense even if it executes each individual movement more slowly.

That is the opening humanoids are chasing: the messy gap between highly automated processes, where industrial robots already dominate, and jobs factories still give to people because dedicated automation has never been worth building.

Q10Do China's huge humanoid shipment numbers mean workers are already being replaced?

No. China's exploding humanoid shipments show that manufacturing and experimentation are scaling much faster than actual worker replacement.

As mentioned earlier, the industry report released at the latest World Robot Conference put Chinese humanoid shipments above 40,000 in the first half of 2026. That would be more than three times Omdia's estimate for worldwide humanoid shipments during all of 2025.

The destination of those machines is the important part. Many go to universities, research labs, data-collection centers, exhibitions and developers. Unitree's relatively affordable robots are particularly popular as research platforms. Other machines enter factories in small fleets so manufacturers can collect real-world data and find out which tasks are worth automating.

Even China's robot industry leaders are careful about the distinction. At the latest World Robot Conference, Unitree founder Wang Xingxing said the company's humanoids were being tested in automotive factories and in Unitree's own facilities, while acknowledging that reliable general-purpose work remains difficult.

So 40,000 shipped humanoids does not mean China suddenly gained 40,000 robotic employees. It means tens of thousands of physical AI platforms are now out in the world generating data, testing tasks and exposing failures. That learning base could become extremely valuable later, but the labor replacement today is much smaller than the shipment number suggests.

Q11Are companies actually paying serious money for Chinese humanoids?

Yes. Humanoid robots have become a meaningful business for at least some Chinese manufacturers, which makes the market much harder to dismiss as a collection of subsidized demonstrations.

UBTECH's audited 2025 results are probably the cleanest evidence. Revenue from full-size humanoid robots and related services reached roughly RMB820 million, up more than twentyfold from the previous year. The company sold 1,079 full-size humanoids, and the segment accounted for around 41% of total company revenue.

UBTECH also finished the year with annualized production capacity above 6,000 full-size humanoids. Its customers and partners have included manufacturers such as BYD, Foxconn and Geely, while Airbus has also bought a Walker S2 for manufacturing research.

Those figures tell us that humanoids have moved into real procurement budgets. They still do not tell us whether every buyer earns an attractive return. A carmaker may spend heavily because it wants production data, fears falling behind competitors or expects the technology to improve while it learns.

Repeat fleet expansion will eventually answer that question. A factory buying its hundredth robot after measuring the first twenty would be much stronger evidence than a large initial order. We are starting to see serious spending today, while the proof of sustained return on investment is still catching up.

Q12Why are Chinese humanoid robots getting so cheap?

Chinese humanoids are getting cheap because China already has much of the manufacturing ecosystem a robot needs.

Unitree now lists its G1 at $13,500 and its lightweight R1 Air from $4,900. Those are extraordinary prices for machines with legs, motors, cameras, compute and dozens of moving joints. They also make clear that low hardware price and high labor productivity are separate achievements. A $4,900 robot still needs useful software.

The bigger advantage sits underneath the finished robot. Humanoids need motors, reducers, actuators, batteries, power electronics, sensors, machined components and increasingly dexterous hands. China spent years building enormous supply chains for electric vehicles, electronics and industrial automation, so robot makers can source many of those components locally.

Morgan Stanley estimated that the bill of materials for a Tesla Optimus-style humanoid using China's supply chain was about $45,500 in 2025. The same analysis put a non-China supply chain at roughly $131,800, close to three times as much. Actuators and dexterous hands explained most of the difference.

McKinsey has reached a similar conclusion. Its recent humanoid supply-chain work says China's dense network of electromechanical and EV suppliers gives local robot companies a major advantage in cost and iteration speed.

Cheap hardware creates another advantage: Chinese companies can put more robots into factories, labs and universities, discover what breaks, change the design and try again.

Humanoid robot cost markers

Q13Can Chinese humanoid robots work without humans babysitting them?

Only within fairly narrow boundaries. Chinese humanoids can now complete some long sequences autonomously, but they still struggle when the environment changes in ways they have not seen before.

The latest Robot Games gave us a surprisingly useful snapshot. Organizers added 21 scenario-based competitions covering factories, hotels, homes and logistics. Robots have clearly improved at manipulation, yet failures still appear when tasks require several coordinated actions, flexible objects or recovery from mistakes.

Factories reduce those problems by designing the job around the robot. Objects arrive in expected places. Routes are mapped. Equipment communicates with the robot. Engineers choose tasks with limited variation. Longcheer's deployment worked because AgiBot and the factory spent months adapting robots to an existing production process.

Unitree's Wang Xingxing is unusually open about how much remains unsolved. He recently said the breakthrough where a robot can receive a language instruction and reliably complete most tasks in an unfamiliar environment could still be five to ten years away, although he also gave a more optimistic two-to-three-year possibility.

His caution is useful because Unitree has some of the world's best humanoid hardware. The body can already run, jump, balance and recover from impacts. The harder problem now is deciding what to do when reality stops matching the training example.

Q14Are Chinese humanoid robots useful at home yet?

No. Chinese humanoids are currently much easier to justify in a factory or research lab than in someone's home.

Homes are chaotic. Clothes deform. Objects move. Children leave things on the floor. Different cupboards require different forces. A robot may need to understand a vague instruction, find an unfamiliar object, manipulate it safely and recover when the object slips.

Factories deliberately remove much of that uncertainty. That is why today's successful humanoid deployments revolve around battery cells, tablets, boxes, totes and inspection routines.

Unitree's R1 shows how quickly the hardware barrier is falling. A basic humanoid now starts below $5,000, which would have sounded absurdly cheap only a few years ago. Yet Unitree itself emphasizes open interfaces, simulation support, multimodal models and development. The product currently makes far more sense as a robotics platform than as a household employee.

Household robots may eventually be one of the biggest markets, and Chinese manufacturing could make them affordable earlier than many people expect. The autonomy needed to make one genuinely helpful around an unpredictable home is still missing today.

Q15So, are Chinese robots actually useful?

Yes. Chinese robots are already extremely useful, while Chinese humanoids are useful today in a much narrower set of jobs.

The first half of that answer is easy. Chinese companies supply industrial robots by the hundreds of thousands, warehouse robots already multiply picking productivity, commercial cleaning has become a large robot business, and autonomous mining fleets work around the clock in dangerous environments. These machines produce measurable economic value.

Humanoids have now crossed an important line as well. AgiBot is running robots inside live electronics production. Galbot is working on CATL battery lines. SAIC-GM uses an embodied robot for battery-cell handling. UBTECH generated RMB820 million from full-size humanoids and related services in one year. Calling the entire Chinese humanoid industry a collection of staged demos no longer fits the evidence.

But current humanoids remain weak substitutes for general human labor. UBTECH puts productivity at roughly 30% to 50% of a person on suitable tasks. Most deployments concentrate on handling, picking, sorting and inspection. Engineers still shape the environment around the robot, and unfamiliar situations remain difficult.

China's biggest achievement so far may therefore be the speed of the learning loop. It can build humanoid hardware cheaply, manufacture tens of thousands of units, put them into real environments and improve the next generation from the failures.

Today, a Chinese robot can absolutely be useful. A Chinese humanoid can also be useful if the job is narrow enough. The machine that walks into an unfamiliar workplace and works like another employee is still ahead of us.