Humanoid robots during the large robot 400m race at the World Humanoid Robot Games in Beijing, this Sunday. EFE/Andres Martinez Casares
Beijing, Aug 23 (EFE) – A year later, the humanoid robots at the Beijing Games run faster, jump higher and hit harder, so much so that some are beginning to encounter a new problem: learning to brake, while Huawei and other companies work on the necessary infrastructure for them to react more quickly.
The second day of competition on Sunday showcased these advances and their limitations, from machines that overshot their targets after jumping to others that lost their balance in the most demanding tai chi postures, while the tests focused on real-world tasks shifted the challenge to hand precision.
In front of stands approximately three-quarters full, the competitions took place on a 5G-A infrastructure deployed by China Unicom together with Huawei, with which both companies are testing the role of low-latency networks in the future of humanoids.
Learning to brake
Speed is no longer the only problem. In the long jump, some humanoids accelerated, jumped, and landed, but then continued their run because they couldn’t stop in time.
Something similar happens when power comes into play: in kickboxing, punches displaced gloves and protective gear, while in football, robots collided with each other in disputes over the ball.
Even stillness has its complications. In tai chi, some of the most difficult and highest-scoring postures were also those that most tested balance and caused some machines to malfunction.
On the track, the Tiangong robot won both 400-meter finals, with times of 38.15 seconds among the larger robots and 45.66 seconds among the smaller ones.
Tianzhuo won gold in the 1,500 meters with a time of 2:21.64, while Tianjiao won the long jump with a best mark of 7.97 meters.
From the track to the hands
Just a few meters away, the challenge shifted to much smaller movements. Weighing dust, building with blocks, or handling small steamed buns serve to measure an essential skill for working outside these enclosed spaces: handling objects with precision.
The joint team from Wuhan University and AGIbot won the powder weighing in 27 seconds, Shanghai AGIbot needed 108 seconds to prevail in block building, and Chengdu AGIbot won in bean picking with tweezers.
The organization applies this expertise to scenarios designed to replicate real-world tasks. In the restaurant industry, the robot has twenty minutes to respond to a verbal command, pick up food with tongs, heat it in a microwave, serve a drink, and deliver the order.
In another test, the machines must respond to a simulated office emergency by removing obstacles, opening windows, and placing signposts.
Autonomy also has its rewards: completing tasks without human intervention applies a scoring coefficient of 1, compared to 0.5 for remotely operated robots.
“Judging by the preliminary rounds, the performance of the fully autonomous robots has exceeded many people’s expectations,” said Gong Xiao, a member of the organizing committee.
However, he warned that small variations in objects can disorient them: “This uncertainty when faced with the real physical world is precisely one of the most fundamental technical challenges we must overcome today.”
Intelligence, outside the robot
This development raises another question: how much intelligence should travel inside the humanoid itself and how much can be transferred to external systems.
China Unicom, in partnership with Huawei, has deployed a 5G-A infrastructure for the Games that separates and prioritizes robot communications over traffic from spectators, media, and other users.
The operator claims that during the opening ceremony it recorded a latency of 18 milliseconds on the network intended for the machines, while Huawei plans to move part of its processing capacity to the cloud in the future.
“If we can offload some of the computing to the cloud, we can save a lot of energy for the robot’s movement,” David Li, head of wireless technology at Huawei, told EFE, noting the limited battery life as one of the current obstacles for humanoids.
But taking that model beyond a prepared environment like the Games presents difficulties. Li himself acknowledged that reliably guaranteeing latencies of 20 or 30 milliseconds across an entire city like Beijing would currently be “extremely expensive.”
While Huawei is studying how to move part of the machines’ “brain” outside the robot, inside the stadium the problems remain much more physical: after learning to run, jump or hit faster and faster, some humanoids now face a much more basic lesson: knowing when to stop.
