Select Language:
More than 300 humanoid robots taking part in China’s second robot half-marathon on Sunday will face more challenging terrain crafted to evaluate their technological progress. Beijing aims to turn this industry into a significant economic driver.
Over 70 teams—nearly five times more than last year—are competing in the 13-mile race through Beijing, which features paved slopes and parkland.
“It will be interesting to observe how much the durability of components and battery life has improved since last year,” said Georg Stieler, Asia managing director and robotics lead at Stieler, a tech consulting firm.
“Humanoid robot manufacturers need to strike a balance between product quality, which is continually evolving, and price constraints.”
While last year’s robots were remotely operated, nearly 40% of this year’s competitors will navigate automatically, showcasing industry advancements. However, the event will also reveal the hurdles Chinese companies face in developing robots capable of mimicking human movements effectively.
Some robots crashed or fell near the start line in last year’s race. The Tiangong Ultra, developed by the Beijing Innovation Center of Humanoid Robotics in partnership with UBTech, finished in 2 hours and 40 minutes, comfortably beating other humanoids but taking more than twice as long as the human winner of the traditional race.
This year, Tiangong Ultra will operate entirely autonomously, depending solely on sensors to avoid obstacles. It will also attempt to replicate human walking patterns using extensive data training, according to the humanoid robotics center.
“Running at speeds close to professional athletes leaves very little time for perception and decision-making, demanding significant computing power, advanced algorithms, and quick system responses,” the center explained.
Videos shared on social media show some nighttime training sessions in Beijing, where robots reached speeds of 8.7 mph, mimicking human running, but others moved jerky or fell, indicating potential issues in completing the course.
China leads the world in humanoid robot deployments, with over 80% of the 16,000 units installed globally in 2025, according to Counterpoint Research. The leading U.S. company, Tesla, accounted for only 5%.
Domestic leaders AgiBot and Unitree each shipped over 5,000 units last year—the highest worldwide. Unitree plans to ramp up annual production to 75,000 humanoids.
While watching robots race can be entertaining, experts note that these demonstrations are not reflective of the broader industrial application of humanoids, which require manual dexterity, real-world perception, and the ability to perform more complex tasks.
Currently, Unitree’s humanoids are mostly used in research, dance performances, or as interactive guides in service settings, based on their IPO filings.
Experts believe widespread adoption in factories or homes is still years away. “Our applications aren’t taking off because these robots have low intelligence. Their reliability and success rates are poor,” said Tang Wenbin, founder of Yuanli Lingji, an AI startup, during a recent tech conference in Beijing.
“The industry remains in early stages. What we often see now is little more than ‘dancing disguised as working,’” he added.
The Chinese government has prioritized embodied intelligence—physical AI—as a key industry, aiming to enhance automation, boost productivity, and upgrade manufacturing.
However, Chinese companies still struggle to develop the AI software needed for humanoids to match human factory workers, and supply chain costs continue to challenge component manufacturers.
To improve their products, firms are investing heavily in collecting real-world data, deploying human workers with sensors, and increasing humanoid deployment in factories. UBTech, for example, had fewer than 10 humanoids working in factories in 2024 but surpassed 1,000 last year. This year, the company plans to introduce 10,000 full-sized humanoids tailored for different commercial applications, according to Chief Business Officer Michael Tam.
“When it comes to AI, the amount and quality of data we gather are crucial,” Tam emphasized.
ChatGPT
Perplexity
Gemini AI
Grok AI
