After outrunning Bolt, China's robot champion races toward real-world work

“When it started smoothly and got past 50 metres, that was when I finally relaxed,” said Guo, who leads the development of Ultra and Omni, a smaller model that competed in the 400 metres, at the Beijing Humanoid Robot Innovation Centre, known as X-Humanoid. Just a week later, the bespectacled 34-year-old engineer was already looking beyond speed.

“We want to turn high performance into high reliability and high usability,” Guo interview with foreign media. Ultra’s record-breaking run offers a glimpse into China’s growing strength in humanoid robotics, an industry being propelled by state-backed research, extensive manufacturing supply chains and intense competition.

But Guo’s next goal highlights a more difficult challenge: turning spectacular demonstrations into robots capable of performing useful work reliably. Inside the workshopX-Humanoid is based in Yizhuang, a high-tech development zone in southeastern Beijing and one of China’s major robotics clusters.

Once dominated by automakers, the area now features a robot sales centre and even a robot-themed restaurant, reflecting Beijing’s drive to promote the industry. The company’s first-floor showroom displays successive generations of Tian. Gong robots, including one still wearing the vest from a winning half-marathon run last year.

On the ninth floor, white Ultra and Omni humanoids hang from metal frames. One wears a red “Champion Tian. Gong” headband.

During a Reuters visit, engineers repaired and tested the machines. One removed a robot’s head cover, another tightened a leg, while others conducted walking trials. In a nearby office, Guo’s team worked on software before testing it on physical robots.

Founded in November 2023, X-Humanoid is backed by state-owned and private investors, including robotics company UBTech, as well as Chinese technology giants Xiaomi and Baidu. The company develops robot hardware, motion-control software and AI for the broader industry while also commercialising its own machines.

X-Humanoid has not disclosed how many of its robots it has sold or deployed commercially. Its customers include universities, research institutes and companies exploring industrial applications. At Foton Cummins’ engine plant in Beijing, the company’s robots have been tested moving boxes.

Its Tianyi 2.0 humanoid has lifted 8-12 kg boxes onto shelves as part of what the factory described as an early-stage exploration. From speed to useful work. Ultra was not originally built to win races.

It began as an experiment designed to test the limits of humanoid movement. Guo compares today’s humanoids with the earliest cars: first, engineers must make the machines controllable; eventually, reliable physical bodies will need to be paired with increasingly capable AI “brains” that can perform complex tasks autonomously.

For now, he said, the priority is making the robot’s body and movement-control system stable and reliable enough for everyday use. Developing AI capable of handling more complicated autonomous tasks will take longer. Training a robot to run also helps develop more general mobility, improving its ability to handle different surfaces and environments.

The smaller Omni has already demonstrated some of that potential. In an internal test this year, it climbed from the first floor to the 17th floor of X-Humanoid’s headquarters. Guo expects smaller humanoids to find early applications in commercial services and light industry, while Ultra could eventually be used for outdoor inspections or emergency rescue.

But speed creates its own problems. Ultra weighs 75 kg and can exceed 17 metres per second at top speed. At the Games, a crash mat was placed beyond a short run-off area, meaning some sprinting robots slammed into the barrier after finishing their 100-metre races.

Future versions could be lighter, brake more aggressively and adopt safer running postures, Guo said. The race for speed. Ultra arrived at the Games with three running strategies trained in simulation.

One had been tested extensively, while two others were theoretically faster but less proven. Each strategy is a trained motion-control system that continuously adjusts the robot’s joints and balance, allowing it to sprint upright without an operator controlling every step.

Ultra used a different program in each round, recording 9.39 seconds in the preliminary, 8.86 seconds in the semifinal and 8.64 seconds in the final. Between races, engineers returned to their testing facility, pushing the faster programs harder and adjusting navigation and running controls.

Some worked through the night, testing into the following day before returning to the competition venue. As the robots were pushed to their limits, some stumbled, lost balance or crashed. Teams brought out backup machines while repair crews rushed to fix damaged robots.

Earlier in Ultra’s development, the failures were even more severe. “Legs broke, waists broke,” Guo said. A late surge.

Even on the morning of the final, Guo’s team was testing a newly trained running program. It performed repeatedly during trials but failed on the competition track. Differences between individual robots also created problems, leaving Guo uneasy before the race.

Ultra’s heavier legs limited its acceleration at the start. But once it passed roughly 10 metres per second, it was able to keep accelerating and gradually catch its rivals. The team estimated that it had cut about half a second from the robot’s starting response by the semifinal.

After that, the focus shifted almost entirely to top speed. In the final, a quicker reaction and stronger late acceleration helped Ultra make a decisive overtake and claim the fastest time of the competition. Backstage, the team celebrated.

At the closing ceremony, a line of Tian. Gong humanoids appeared in red capes emblazoned with “Champion”, with some waving pom-poms. “Finally, I could get a good night’s sleep,” Guo said.