С 22 по 26 августа на Национальном овале по конькобежному спорту в Пекине соберутся 666 команд и 2 056 гуманоидных роботов из 16 стран. Эти цифры выглядят как круг почета. Более сложный вопрос заключается в том, будут ли эти ...
Сообщение Planet-Today.com. Перевод заголовка и краткого описания выполнен автоматически.
From August 22 to 26, Beijing’s National Speed Skating Oval will host 666 teams and 2,056 humanoid robots from 16 countries. The numbers look like a victory lap. The harder question is whether these machines can leave the arena and do a job that pays for itself.
 |
| Source: Video Screenshot |
Key Takeaways by Planet Today
Participation surge: Teams jumped 138 percent from the 2025 inaugural Games (280 teams, 500-plus robots) to 666 teams and 2,056 robots this year. China accounts for 641 teams and 1,975 machines.
Format shift: Organizers added long jump, weightlifting, tug-of-war and table tennis, and expanded scenario events from a handful to 21 tasks in factories, hotels, homes, logistics sites and a real fire brigade.
Industry context: Chinese officials expect domestic humanoid output above 100,000 units in 2026. Market data cited for the first half of 2026 put Chinese makers at more than 97 percent of global shipments.
Commercial test: Cost, reliability and supervision remain the gap. Brokerage estimates say an industrial humanoid needs to land near 160,000 yuan all-in to recover against a typical worker wage; many units still cost several times that.
Geopolitical overlay: The United States restricted new authorizations for certain foreign advanced robotic devices in July 2026. Volume leadership and software/reliability leadership are not the same contest.
Four days from now, the Ice Ribbon in Beijing will fill with machines that walk, kick, lift and, in some cases, try to put out a fire. The second World Humanoid Robot Games run from August 22 to 26. Organizers report 666 teams and 2,056 robots from 16 countries and regions on six continents. Last year’s first edition drew 280 teams and more than 500 robots. The jump is not a rounding error.
China will field the overwhelming majority: 641 teams and 1,975 robots, drawn from 157 companies plus universities and research institutes. International entries include teams from the United States, Germany and Japan. Brazil is sending five RoboCup squads as a national group. Those facts come from the organizing committee and have been repeated by Anadolu Agency, Xinhua, China Daily and Global Times.
The curiosity is not whether China can stage a large event. It can. The curiosity is what the event is actually measuring — and what it cannot measure.
How Big Is This, and Why Did It Grow So Fast?
Jiang Guangzhi, director of the Beijing Municipal Bureau of Economy and Information Technology and a senior organizing official, has said the team count rose 138 percent year on year and that the robot count roughly quadrupled. The program itself grew from 26 events and a few hundred matches to about 51 events and more than 1,300 matches across five days instead of three.
A dedicated “Robot Home” will store more than 1,500 robots, charge at least 1,200 batteries and complete digital check-in and retrieval for a single unit in about 30 seconds, according to CCTV Finance reporting carried by Global Times. For some team sports the committee supplies bodies and nearby pre-charged batteries. That is logistics at industrial scale, not a university demo hall.
Tickets are on sale from 98 to 670 yuan. The venue is the National Speed Skating Oval. Co-hosts are the Beijing municipal government and China Media Group, with involvement from the World Robot Cooperation Organization and RoboCup Asia-Pacific structures. Official Beijing pages confirm the dates and the two-track design: competitive events and scenario-based events.
Scale is real. Scale is also the easiest thing to count. Counting medals is easier than counting hours of unsupervised work on a factory floor.
Sports Events: Balance, Strength and the New Rules
Competitive events now include athletics, football, gymnastics, martial arts, street dance, sport dance, plus newly added long jump, weightlifting, tug-of-war, table tennis and pitch-pot (a traditional precision game). Freestyle combat has weight classes. Football rules have tightened: more autonomy, more contact requirements before a goal counts in some formats.
Organizers shortened time limits. The 100-meter race moved from a three-minute window last year toward a one-minute autonomous standard. Many track and confrontation events now ban remote control. Jiang has described the new strength and precision events as extreme tests of “robotic brain, cerebellum, physical structure and core components.”
That language is useful. Humanoid robots fail in ordinary ways: they fall, they miss a grasp, they drain a battery, they freeze when the lighting changes. A tug-of-war or a long jump does not prove a robot can pack boxes for eight hours. It does prove whether the mechanical stack — actuators, reducers, balance controllers, power — can take a beating in public.
Dance and martial arts remain crowd events. Their value is different. They stress coordination and timing. They also remain closer to last year’s spectacle. The more interesting shift is the move away from scripted or heavily teleoperated routines toward autonomy under time pressure.
What a public contest actually stresses
A stadium is still a controlled environment. Floors are known. Lighting is known. Judges are watching. That is not a warehouse at 2 a.m. It is, however, a harder environment than a lab video. Falls will be filmed. Missed grasps will be filmed. That public failure is part of the point for developers who need data and for officials who want standards.
Real-World Tasks: Factories, Hotels, Homes and a Fire Station
Scenario events expanded sharply. Reports put them at 21 contests across settings modeled on factories, hotels, homes, logistics centers, hospitals, retail and emergency sites. Tasks listed by organizers and state media include assembly, housekeeping, book sorting, material handling, guest services and emergency response.
This is the part that matters more than the 100-meter dash if the policy goal is commercial robots.
On August 16, before the official opening, 23 teams ran a firefighting final at a real Beijing fire brigade rather than a mock room. Robots had 30 minutes to identify two simulated hazardous substances, close three valves and operate an extinguisher. Global Times and Interesting Engineering both covered the exercise. The change from last year’s more theatrical setups is deliberate: “realistic simulation.”
A dedicated dexterous-hand contest is new. Fine tasks — cable work, weighing, opening containers, tool use — sit closer to the “last meter” problem that still separates a walking torso from a useful worker.
Jiang has said competition rules could feed technical standards for industrial use. That claim should be read as an ambition, not a completed standard. Standards require repeatability, safety cases and third-party verification. A games score sheet is a start, not a certification.
“Once robots master these ‘last-meter’ skills, they can turn medals into orders and move directly from the competition arena to real-world workplaces.”
The sentence is the official hope. The commercial record so far is more mixed.
Latest: The Same Week, a Different Test
On August 18, Reuters reported that China’s humanoid makers are moving from breakdance and marathon headlines to a harder question: can the machines work reliably enough to create economic value. The World Robot Conference is running in Beijing this week with more than 300 companies, over 2,000 exhibits and more than 150 product launches. Unitree, one of the highest-volume humanoid makers, is making its Shanghai market debut after an IPO that retail investors oversubscribed by more than 8,000 times. Founder Wang Xingxing is scheduled to speak on the industry’s next decade. See Reuters.
That timing is not accidental. The Games and the conference sit next to each other on the calendar. One is a public stress test. The other is a sales floor.
Reuters also relayed a cost problem that does not appear in medal tables. Guotai Securities estimated an industrial humanoid would need to cost about 160,000 yuan including maintenance to pay for itself in two years against a worker earning about 80,000 yuan a year. MERICS, a Berlin think tank, put typical prices in a much higher band of 300,000 to 500,000 yuan. Analyst Georg Stieler estimated that 50 to 70 percent of humanoids produced this year could end up in “data factories” collecting training data rather than generating customer value.
Those figures do not cancel the shipment numbers. They explain why shipment numbers and deployed, unsupervised labor are different statistics.
China’s Production Machine — and What 97 Percent Means
Chinese officials have said domestic humanoid production should exceed 100,000 units in 2026. That target was restated at the World Artificial Intelligence Conference and in later official comments. Independent market trackers are in the same order of magnitude even when they disagree on exact splits.
Smart Analytics Global, cited by Bloomberg and others, put global humanoid shipments at about 19,100 units in the first half of 2026, more than triple the same period a year earlier, with Chinese makers above 97 percent of that total. AgiBot (Zhiyuan) and Unitree have traded the lead; mid-year figures had AgiBot around 8,400 units and Unitree around 5,900. Industrial and commercial uses were reported as a rising share of shipments. Full-year global forecasts cluster around 50,000–60,000 units, with China the bulk. See Bloomberg and follow-on coverage in China Daily and TechNode.
A 97 percent shipment share is a manufacturing fact. It is not automatically a software fact, a safety fact or a profit fact. Western firms such as Figure, Tesla, Agility and others remain far smaller in volume. They also argue, with some evidence from public demos, that the hard remaining problems are generalization, dexterity and reliability rather than the ability to ship a walking chassis.
China’s advantages are well documented: electric-vehicle supply chains, motors, reducers, batteries, rare-earth magnets, local government subsidies, a large domestic market and a policy push into “work mode.” MIIT and SASAC directed a push to put more than 10,000 humanoids into commercial use by the end of 2026 — working, not only demonstrating. Aging demographics and factory labor costs give the policy a domestic logic that is independent of any race with Silicon Valley.
The United States in late July 2026 restricted new FCC authorizations for certain foreign advanced robotic devices, citing security concerns. Prior models can still move in some channels. The restriction is a reminder that volume leadership can collide with market access. It is also a reminder that Washington treats this category as more than a toy.
For a wider look at how technology scale and geopolitics now travel together, see Planet Today’s earlier piece Why Trump’s China trip signifies the end of American primacy. On how science and industry rewrite everyday products, see Is Your Chocolate Going Biotech?
International Teams: Presence Versus Weight
Sixteen countries is a genuine international field. The United States, Germany and Japan are not spectators. Brazil’s RoboCup group adds a football pedigree that predates Beijing’s Games. RoboCup’s own calendar lists the World Humanoid Robot Games as a major 2026 humanoid gathering.
The arithmetic still shows a home event. Roughly 96 percent of the robots are Chinese. That does not make foreign teams irrelevant. It does mean the leaderboard will mostly compare Chinese platforms against each other unless a foreign team wins a high-visibility autonomous event. That outcome would be newsworthy precisely because it would cut against the volume story.
Universities matter here as much as companies. Beijing Information Science & Technology University is bringing 49 teams across a wide set of events, according to China Daily. That pipeline — students, datasets, failed runs — is how a national industry compounds. Western labs have deep research culture and weaker manufacturing scale. Both sides say they have the harder problem. Both can be partly right.
What the Games Cannot Decide
A five-day tournament cannot settle battery endurance across a shift, maintenance cost over a year, or whether a hotel manager will keep a robot after the novelty week. It cannot settle liability when a machine drops a guest’s luggage or misidentifies a chemical. It cannot settle whether open-source Chinese vision-language-action models, which have posted strong 2026 benchmark results in some rankings, transfer from a scored course to a messy kitchen.
It can do three useful things.
First, it can force public comparison under shared rules. That is rarer than press-release videos.
Second, it can generate failure data at volume. Every fall is a training sample.
Third, it can pull officials, standards writers and buyers into the same room as engineers. Beijing has said it wants the Games to support commercial standards. If those standards become real documents with test protocols, the event will have done more than sell tickets.
Skeptics will note that state media coverage is uniformly upbeat, that many “deployments” are still pilots, and that a robot that wins table tennis may still need a human to reset it after a slip. Supporters will note that no other country is putting this many bipedal machines into the same building at the same time, and that industrial robotics history often looked clumsy one year before it looked inevitable.
Both observations can stand at once.
Labor, Cost and the Quiet Question
China’s working-age population is shrinking. That is not a slogan; it is a census trend that has already pushed factories toward conventional industrial robots at a scale no other country matches. Humanoids are sold as the next step for work that still requires human-shaped access — stairs, existing tools, mixed spaces.
If costs fall toward the brokerage breakeven band, the labor question becomes practical rather than science-fictional. If costs stay high and reliability stays low, the machines stay in showrooms, research labs and data-collection loops. Reuters’ reporting this week is essentially that fork, stated without ceremony.
None of this requires picking a political team. A reader in Detroit, Stuttgart or Osaka can look at the same shipment tables and the same cost tables and reach different conclusions about who is ahead. Volume is China’s lead today. General-purpose reliability is still an open contest. Export controls and security reviews are now part of the market, not a footnote.
What to Watch Between August 22 and 26
Watch autonomy, not choreography. A robot that completes a scenario without a hidden operator is more informative than a robot that dances well.
Watch recovery after error. The firefighting drill already required identification, valve work and extinguisher use under a clock. That sequence is closer to work than a sprint.
Watch which Chinese firms convert medals into purchase orders in the following quarter. Unitree’s listing and AgiBot’s shipment lead make them obvious names. Dozens of smaller teams will disappear if they cannot show a customer a payback period.
Watch whether any non-Chinese team wins a fully autonomous strength or scenario event. That result would travel farther than another Chinese 100-meter heat.
The Ice Ribbon will look like the future. Parts of it will be. Parts of it will be a very expensive dress rehearsal. The honest reading is that Beijing has built the largest public test of humanoid hardware on the planet this year, while the economic test — cost, uptime, unsupervised hours — is only now being asked in public, including by Reuters on August 18.
The medals will be counted by August 26. The more useful score will arrive later, in factories, hotels and balance sheets.
Original source: Anadolu Agency, August 14, 2026 — Over 2,000 humanoid robots to compete in Beijing games. Core figures also reported by Xinhua, China Daily, Global Times, CGTN and official Beijing government English pages.
Disclaimer for fact-checkers: Participation counts, event lists and dates are taken from organizing-committee briefings as relayed by state and international wire services. Shipment and production figures come from Smart Analytics Global, official MIIT comments and secondary reports (Bloomberg, China Daily, Global Times). Cost and commercialization estimates come from Guotai Securities, MERICS and Reuters interviews. Chinese official media emphasize progress; independent reporting emphasizes cost, reliability and the gap between shipments and unsupervised work. Readers should treat medal results as public tests, not as certified industrial performance, and should verify any standard-setting claims against published technical documents when they appear.
