Humanoid robots participating in the World Humanoid Robot Games in Beijing are already demonstrating speeds exceeding those of Usain Bolt. However, the main question now is whether these machines can perform real work with the required precision, autonomy, and some resilience to unforeseen situations in the real world.
The second edition of the competition gathered over 2,000 robots that participated in 51 disciplines. The tests include tasks more characteristic of a manufacturing environment than a running track, such as connecting cables and assembling components.
The robot Tien Kung Ultra covered the 100-meter distance in 9.39 seconds, surpassing Usain Bolt's world record of 9.58 seconds. In the 400-meter race, another robot, Tien Kung, finished in 38.15 seconds, which is faster than the human record of 43.03 seconds.
Another notable development is the significant leap in performance over the past year. According to SCMP, the winner of the previous 400-meter competition completed the distance in 88.03 seconds; this time, the top three finished the race in under 40 seconds.
Nevertheless, the track itself revealed limitations. Some robots fell and required staff intervention. During a football match, one of the machines knocked over its teammate.
It is outside the track that the competition takes on a different meaning. A cable might not be in the expected position, an object might be too far away, or packaging might be displaced. For a humanoid, such minor deviations require environmental perception, motion control, and the ability to correct errors. It is precisely these scenarios that may determine the practical usefulness of this technology.
As noted by Yu Chao, CEO of Lumos Robotics, simply making a robot run and jump is not enough to increase efficiency.
A demonstration by Galbot showed how the humanoid played against humans on a field. To retrieve the ball, the machine had to track its trajectory, anticipate the landing point, move, and coordinate arm and torso movements. The robot fell during the game but was able to get up and continue, according to SCMP. This is a small but telling moment: humanoids are already capable of reacting to some failures, but they are still far from flawless operation.
Autonomy became an important aspect in this edition. According to Reuters, over 40% of the tests require full autonomous operation.
All this intellectual activity requires significant energy consumption. The motors responsible for movement already consume a lot of battery charge. Adding more AI processing inside the machine increases the load on this resource.
One solution is to offload some computations to the cloud. This strategy saves energy for movement but makes the network connection more critical, as pointed out by SCMP. David Li, head of Huawei, told SCMP: 'If the AI computing power is also placed in the robot, battery consumption will increase even more.'
Records help demonstrate what humanoids have achieved. But it is in the least spectacular tasks—connecting a cable, moving an object, or correcting an error—that they will have to prove their readiness for work.