 |
| Close-up of a futuristic humanoid robot with metallic armor and blue LED eyes.(For reference only) |
After outrunning Bolt, China’s TianGong Ultra targets real-world work
Article
BEIJING — September 10, 2026
TianGong Ultra has moved from a record-setting sprint to a more difficult challenge: becoming a reliable humanoid robot for real-world work. The Chinese machine recently completed 100 metres in 8.64 seconds, beating the time set by human sprinting record-holder Usain Bolt by almost one second.
The result came after intensive testing and repeated repairs at the World Humanoid Robot Games. Jack Guo, the 34-year-old engineer leading development of Ultra and the smaller Omni model, said the team's focus is now shifting from maximum performance to reliability and practical use.
Ultra was developed at the Beijing Humanoid Robot Innovation Centre, known as X-Humanoid, in the Yizhuang high-tech zone in southeastern Beijing. The area has become one of China's major robotics centres as the country expands investment and competition in humanoid machines.
Inside X-Humanoid, engineers work on both physical robots and the software that controls their movements. During a recent visit, technicians tested walking systems, adjusted robot legs and carried out repairs while other team members worked on software before transferring it to physical machines.
X-Humanoid was established in November 2023 and has backing from state-owned and private investors, including robotics company UBTech and technology companies Xiaomi and Baidu. The centre develops robot hardware, movement-control systems and artificial intelligence while also producing its own machines.
The company has not disclosed how many of its robots have been sold or deployed commercially. Its customers include universities, research institutions and businesses testing possible industrial uses.
At a Foton Cummins engine factory in Beijing, X-Humanoid robots have been tested for moving boxes. Its Tianyi 2.0 robot has lifted boxes weighing between 8 and 12 kilograms onto shelves as part of an early exploration of industrial applications.
Ultra itself was not initially built as a racing machine. The project began as an effort to test the limits of humanoid movement, with running seen as a way to improve a robot's ability to control its body and deal with different surfaces.
Guo compares the development of humanoids with the early stages of automobiles. He said engineers first need to create machines that can be controlled reliably, while more capable artificial intelligence can later give them the ability to perform complicated tasks with greater independence.
For now, making the robot's body and movement-control system stable is a bigger priority than developing highly advanced autonomous decision-making. Guo said the AI required for more complex independent work will take longer to develop.
The smaller Omni model has already demonstrated some of the potential for everyday mobility. In an internal test this year, it climbed from the first floor to the 17th floor of the X-Humanoid headquarters.
Guo expects smaller humanoids to take on early jobs in commercial services and light industry. He said Ultra could eventually be used for outdoor inspection work or emergency rescue.
Speed also creates new engineering problems. Ultra weighs 75 kilograms and can travel at more than 17 metres per second at full speed, making it difficult to stop safely.
At the robot games, sprinting machines had only a short run-off area after the finish line, with a crash mat placed beyond it. Some robots lost balance or hit barriers after races, highlighting the difficulties involved in controlling heavy machines travelling at high speed.
Future versions of Ultra could be lighter, use stronger braking systems and adopt a safer body position while slowing down, Guo said.
Ultra competed at the games with three running strategies developed through computer simulation. Each strategy was a trained movement-control system that continuously adjusts the robot's joints and balance so it can run upright without being directly controlled by an operator.
The robot recorded 9.39 seconds in its preliminary race, improved to 8.86 seconds in the semifinal and reached 8.64 seconds in the final. Engineers worked between races to test faster control programs, adjust navigation and improve the robot's running system.
The work often continued through the night. When pushed beyond their limits, robots stumbled, fell and suffered mechanical failures, requiring teams to bring out replacement machines while technicians repaired damaged equipment.
Guo said failures had been even more severe during earlier development, when components including legs and waist sections broke during testing.
The team was still experimenting with a new running program on the morning of the final. Although the system performed repeatedly during practice, it failed to work as expected on the competition track. Differences between individual robots also made it difficult to predict how the machine would perform.
Ultra's relatively heavy legs restricted its acceleration at the start. But once it reached speeds of about 10 metres per second, it was able to continue accelerating and close the gap with other competitors.
The engineers estimated that they had improved Ultra's starting response by about half a second before the semifinal. In the final, faster initial movement combined with stronger acceleration later in the race helped the robot overtake its rivals.
The victory brought a brief celebration for Guo and his team, but the longer-term objective is no longer simply to make the robot run faster. The developers now face the harder task of making humanoid machines dependable enough to perform useful work outside controlled demonstrations.
