BEIJING, (Xinhua) — On a crisp Sunday morning, the starting gun rang out in an innovation zone on the southeastern outskirts of Beijing. More than 100 humanoid robotic runners clattered across the starting line, their servos humming with a staccato whir.
The novelty had worn thinner than the previous April, when the inaugural humanoid robot half marathon drew large crowds. This time, however, the machines had more to prove than simply putting on a show.
Over the past year, Chinese robots have wowed global audiences with dynamic feats of speed and agility, including sprinting, martial arts, and gala dancing. Once the cheers faded, a key question emerged in the era of embodied intelligence: how can these machines move beyond being seen as remote-controlled toys for tech enthusiasts?
The event organizers made their intent clear with new rules for the 2026 edition of the 21.1-kilometer race. Robots unable to navigate autonomously were still allowed to compete, but their finishing times would be multiplied by 1.2. This rule aimed to push the field toward genuine autonomy in motion. As a result, nearly 40 percent of the competing robots took on the challenge of running the race entirely on their own.
For Chinese engineers, the priority is sharper perception and smarter “brains,” not just stronger “muscles.” Progress depends on advanced AI at the core, rather than solely on more stable frames, tougher joints, or better batteries.
Unlike last year’s wide variety of designs, this year’s competitors largely converged on a few standard robotic platforms. Many teams used models such as Unitree, Tien Kung (with a UBTECH background), or devices from Huawei’s spin-off, Honor. The real contest lay in the software—a head-to-head race to develop the best optimization algorithms on largely identical hardware.
Leading the pack was a mecha warrior–like Honor robot with a red-and-black metallic finish. It moved with a distinctive gait: its legs formed an inward-pointing V shape, while its upper body swayed side to side with each stride.
In contrast, Tien Kung’s running form more closely resembled that of a professional long-distance runner, featuring a high arm carriage and an upright posture.
Honor proved to be the day’s top performer. While one of its models was the first to cross the finish line—with an operator trailing in a golf cart—the true self-navigating champion was another Honor robot.
Finishing in a blistering 50 minutes and 26 seconds, the winning robot shattered the human world record, cutting more than six minutes off the best human time. “In our training, we emulated top human runners,” said Du Xiaodi, an engineer at Honor, standing beside the 1.69-meter-tall champion after the award ceremony.
Second- and third-place prizes also went to autonomous Honor robots. In one dramatic moment, an Honor humanoid, carried by its momentum, failed to stop in time and collided with a safety barrier, drawing a mix of cheers and gasps from the audience.
A Tien Kung Ultra robot, fitted with blue protective guards, did not make the top three but still impressed spectators. It executed a smooth S-curve in the final 50 meters to edge out a rival, finishing in 1 hour and 15 minutes.
This marked a stunning improvement for the defending model, which posted a time of 2 hours and 40 minutes in winning the 2025 race. The leap underscores the rapid advancement of China’s humanoid robotics technology in just one year.
Zhao Wen, a control algorithm engineer behind Tien Kung, attributed the progress to smarter embodied AI. “With major upgrades to running control algorithms, communication, and perception systems, the ‘big brain’ and ‘small brain’ have integrated to enable dynamic, intelligent adjustment of stride frequency and length,” Zhao explained.
“Last year, robots struggled just to stand up. This year, they are all stable and racing,” noted the leader of the Paris-Saclay University team. The 2026 event featured global participation, with engineers from Germany, France, Portugal, and Brazil.
International interest in Chinese robotics was also highlighted in February, when German Chancellor Friedrich Merz observed humanoid robots performing various tasks at Unitree Robotics during his visit to China.
Sunday’s half-marathon course put robotic systems to a rigorous test. It included more than 10 terrain types, slopes reaching an 8 percent incline, and a total climb of 100 meters—challenging both power control and energy efficiency.
Narrowed paths and a traffic island obstacle simulated urban complexity, while a series of 22 turns—including near 90-degree angles—demanded centimeter-level precision and balance.
“Autonomously navigating a track with this level of complexity is a major test of a robot’s agility,” said Zhao Mingguo, an automation researcher at Tsinghua University in Beijing.
The race may be over, but the real-world journey is just beginning. There is growing consensus that robotics should not merely replace human labor, but take on tasks people are unwilling to do—especially dangerous work in remote areas or high-risk rescue scenarios.
While AgiBot, another leading Chinese robotics firm, skipped the long-distance race, it demonstrated its precision in a factory setting this week, live-streaming robots working an eight-hour shift on a tablet assembly line. The demonstration is part of China’s broader push to integrate AI with manufacturing.
Meanwhile, Shenzhen, a major technology hub in southern China, is already deploying humanoid robots in diverse roles—from directing traffic to handling household chores.
Positioned as a key future industry in China’s five-year plan, embodied intelligence is set for rapid growth. One forecast estimates the country’s humanoid robot market could reach around 870 billion yuan (approximately $127.6 billion) by 2030.
“Humanoid robots don’t run just for the sake of running,” said Liang Liang, deputy secretary-general of the Chinese Institute of Electronics. The race aims to identify top engineering teams, spur technological competition, and bridge the gap between innovation and real-world applications, he explained.
