At the World Humanoid Robot Games in Beijing, held from August 22 to 26, 2026, humanoid robots produced striking results and equally striking failures. The Tiangong Ultra robot ran 100 m in 8.64 seconds, and the Tien Kung Ultra recorded 38.15 seconds in the 400 m event—times that substantially undercut the human world records (Usain Bolt’s 9.58 s for 100 m; Wayde van Niekerk’s 43.03 s for 400 m). According to the South China Morning Post, more than 2,000 robots competed across 51 events at the meeting.
Impressive speeds, frequent malfunctions
Alongside headline‑grabbing times, the Beijing event showed the technology’s limits: several robots stumbled, some caught fire during demonstrations, and a recurring problem was that sprinters failed to brake after their runs, colliding with foam barriers placed beyond the finish line. Csong Munhjon (Chungnam National University), cited by Reuters, said it is a notable engineering achievement for a humanoid to maintain balance at such high speeds, but the media footage paints a mixed picture of reliability.
Market scale and forecasts
Analysts see significant economic potential in humanoid robots. Morgan Stanley’s 2024 forecast suggests the market could grow to as much as $5 trillion by 2050 and that more than a billion such devices could be operating worldwide by then. The bank also raised its projection for China in 2024: it estimates 50,000 humanoid robots could be shipped in China in 2026 and that the Chinese market could reach $15 billion by 2030. Reports from the Boston Consulting Group and the Bank of America Institute likewise highlight that embodied AI—the integration of perception, adaptation and physical action—could drive the next phase of robotics, although projections for 2030 vary widely.
China’s state backing and geopolitical implications
China is providing strong state support for humanoid robotics, promoting standards, test centers and industrial pilots as part of its broader industrial strategy. The International Federation of Robotics reported that 54 percent of newly installed industrial robots in 2024 were deployed to China—about 295,000 units in a single year—raising the country’s installed industrial robot population above two million. The MERICS research institute notes that Beijing treats embodied AI as a key technological priority in its five‑year planning.
Chinese momentum has prompted policy responses in the West. In the United States, private companies and investors have urged a national robotics strategy, and lawmakers have proposed bipartisan measures to create a national robotics commission and to strengthen robotics supply chains. Washington has also begun treating some Chinese robotics firms as potential national security risks—adding companies such as RoboSense and Unitree to a list maintained by the U.S. Department of Defense.
How autonomous are the robots seen in Beijing?
Researchers caution that many public demonstrations—videos of humanoids dancing, playing ping‑pong or performing delicate hand motions—can be misleading because the machines are often executing preprogrammed sequences or being remotely controlled. The MERICS analysis and other expert commentary emphasize that, while Beijing’s event applied stricter rules this year (most events required full automation except the 100 m and 400 m sprints), the machines remain largely task‑specific and not general‑purpose agents.
Paruncshaja Dzsámkrádzsang (Mahidol University), speaking to Reuters, argued that future milestones will be less about raw speed and more about on‑the‑fly judgement: recognizing an unexpected obstacle and adjusting motion or braking in time. Achieving that requires advances in cognition and vastly richer datasets than those commonly used to train language models.
Collecting and processing the combination of visual, auditory, tactile and spatial data needed for real‑time physical interaction remains a major challenge. Rodney Brooks (co‑founder of iRobot and former MIT professor) has pointed out that there is no well‑established practice for gathering large volumes of tactile data comparable to the datasets available for text, audio or images.
Conclusion: spectacle today, wider usefulness still out of reach
The Beijing competition was simultaneously a dramatic showcase and a reality check. In narrow, well‑defined tasks some humanoid robots already exceed human performance, particularly in speed. Yet consistent, reliable and autonomous operation in complex real‑world settings will require much more progress—better data, stronger cognitive capabilities, and robust integration of perception and control. Investors and governments are placing big bets on the field, and China’s state support may accelerate development, but the technical gaps mean broadly capable humanoid robots remain a future prospect rather than an immediate reality.
