A week after a Chinese humanoid robot stunned spectators by running 100 metres in 8.64 seconds, nearly a second faster than Usain Bolt’s world record, its lead engineer is already pursuing a more difficult target: turning extraordinary performance into machines that can reliably perform useful work.
TianGong Ultra achieved the time at the World Humanoid Robot Games following days of intensive testing, repairs and repeated attempts to push the robot’s running systems further. Jack Guo, the 34-year-old engineer leading development of Ultra and Omni, a smaller model used in 400m races, said he only relaxed after Ultra passed the halfway point without difficulty.
“When it started smoothly and got past 50 metres, that was when I finally relaxed,” Guo said.
But the achievement has not changed his priorities. “We want to turn high performance into high reliability and high usability,” he told Reuters in his first interview with foreign media.
The record offers a glimpse into China’s rapidly advancing humanoid robotics industry, supported by state-backed research, deep supply chains and intense competition. Yet the next stage may prove considerably harder than setting records. Developers must transform machines capable of impressive demonstrations into dependable systems that can operate in everyday environments.
X-Humanoid, the Beijing-based company behind TianGong, is located in Yizhuang, a high-tech development zone in the capital’s southeast that has become one of China’s main robotics clusters. The district, long associated with automobile manufacturing, now features a robot sales centre and a robot-themed restaurant as part of Beijing’s effort to showcase the sector.
Inside X-Humanoid’s facilities, engineers work on both physical machines and software. Robots are repaired and tested in research areas, while Guo’s team develops motion-control software before testing it on the machines. The company was established in November 2023 and is backed by state-owned and private investors including robotics company UBTech, as well as technology companies Xiaomi and Baidu.
X-Humanoid has not disclosed how many of its own robots have been sold or commercially deployed. Its customers include universities, research institutes and companies examining industrial applications. At Foton Cummins’ engine factory in Beijing, its robots have been tested moving boxes, with the Tianyi 2.0 lifting boxes weighing 8-12 kg onto shelves in what the factory described as an early-stage exploration.
Ultra itself was not originally developed to win races. Its purpose was to push the boundaries of humanoid movement. Guo compares the development of today’s humanoids with the earliest automobiles: first making machines controllable, then combining reliable physical systems with increasingly capable artificial intelligence.
For now, he said, the priority is making the robot body and movement-control system stable enough for everyday use. Developing the artificial intelligence required for more complex autonomous tasks will take longer.
The ability to run, however, can contribute to broader mobility. Training robots to move at speed helps develop generalised mobility and their ability to cope with different terrain. Omni has already demonstrated some of that potential, climbing from the first floor to the 17th floor of X-Humanoid’s headquarters during an internal test this year.
Guo expects smaller humanoids to find early applications in commercial services and light industry, while Ultra could eventually be used for outdoor inspection or emergency rescue. But the machine’s speed also creates new engineering challenges. Ultra weighs 75kg and can exceed 17 metres per second at top speed, making braking a major concern. During the Games, sprinting robots had only a short run-off area before a crash mat and barrier.
Future versions could be lighter, brake more forcefully and adopt safer postures, Guo said.
Ultra’s winning performance was the result of progressively refined running strategies. It arrived at the Games with three programmes trained through simulation, one proven in testing and two faster but less established. Each programme continuously adjusted the robot’s joints and balance, allowing it to sprint upright without an operator directing every movement.
Ultra recorded 9.39 seconds in the preliminary round, 8.86 seconds in the semifinal and 8.64 seconds in the final. Between races, engineers returned to their test site, pushing faster programmes and adjusting navigation and running controls. Some worked through the night, while robots repeatedly stumbled, lost balance and crashed.
“Legs broke, waists broke,” Guo recalled of earlier stages of development.
Even on the morning of the final, the team was testing a newly trained running programme. It performed repeatedly during trials but failed on the competition track, while differences between individual robots added further uncertainty.
Ultra’s heavier legs limited its early acceleration, but once it exceeded about 10 metres per second, it could continue building speed. By the semifinal, the team had reduced its starting response by roughly half a second, before shifting its attention to outright speed.
In the final, faster reaction and late acceleration enabled Ultra to make a decisive overtake. The achievement brought celebrations backstage and a row of TianGong humanoids wearing red “Champion” capes at the closing ceremony.
For Guo, however, the victory offered little time for rest. “Finally, I could get a good night’s sleep,” he said.
