If you’re anything like me, the World Humanoid Robot Games are a highlight of your sporting calendar.
The event, held in China, showcases the latest humanoid hardware, lots of algorithmic innovations, and a good dose of robot slapstick. This year’s iteration, which wrapped up in Beijing on August 26, did not disappoint. Besides record-breaking feats—the fastest humanoid on the planet!—and the occasional calamity—weightlifting gone wrong!—the games also included some new challenges that hint at the event’s real purpose: pushing humanoid companies to develop systems that inch closer and closer to practical deployment.
First, though, let’s marvel at some of the more dramatic happenings, which have been all over the internet since things kicked off on August 22.
The opening ceremony showcased the sheer size of the event—the organizers say that more than 600 teams showed up with over 2,000 robots. There are more than 100 humanoid robot manufacturers in China alone, and participants also traveled in from 15 other countries.
The games featured a mix of contests and competitions, and some saw judges scoring a robot’s performance. Witness the Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center, running the 100-meter dash in 8.86 seconds. Another robot called Lightning, from the smartphone maker Honor, ran the distance in 9.47 seconds a few days earlier.
Both times beat the (human) world record set by Usain Bolt. The Tiangong robot also won the long jump with a sprawl of 7.97 meters. Renaissance robot! Another company’s robot, meanwhile, played Ping Pong against an Olympic champ.
Elsewhere, bots punched each other, scored soccer goals, lifted weights (or tried to), and performed cheerleading routines in silly costumes. Part of the fun stems from the fact that it’s easy to anthropomorphize a humanoid robot’s actions, which tends to make their mishaps particularly funny. Take, for instance, the robot that appeared distracted by a cheerleading display, then tumbled over a balustrade.
The fact that the robots on show struggled to stop—sometimes blowing out their motors in the process—also shows how the systems are being pushed to their physical limits.
Practical Applications
Humanoids have been around for decades, but the past few years have seen the development of compact, high-power motors capable of keeping robots dynamically balanced at high speeds.
“China has had amazing success building robot hardware,” says Stefanie Tellex, a roboticist at Brown University. Unitree, which spun up in the high-tech city of Hangzhou and recently went public, rose to fame by focusing on these innovations.
Besides hardware advances, running and jumping robots rely on algorithms trained in simulation to keep them from falling over. This approach most likely explains why some robots have distinctly nonhuman running gaits: An algorithm lands on an optimal strategy by exploring different movements; they may not necessarily match the way humans have evolved to run.
Some of the new events this year would be too simple for human athletes, but they were designed to tax the robots’ brains as much as their brawn. These included plugging cables into sockets and slowly depositing a piece of garbage in a trash can inside a mock bedroom.
Tellex explains that training robots to manipulate objects within a more complex environment is far more difficult than having them sprint or leap through the air. Advanced manipulation is something of a holy grail in robotics, and something that some startups in the US are trying to solve using more advanced AI.
Fiddly finger skills are difficult for robots partly because they involve plenty of hard-to-predict stuff, like friction and slippage. Humans have a finely honed sense of touch, an innate understanding of physics, and the ability to adapt on a microscopic scale at lightning speed. Robots? Not so much.
“It's an example of Moravec's paradox,” Tellex says. “Where things like backflips that seem more impressive than plugging in a cable are actually easier than plugging in a cable.”
One key caveat with the event: Some images and videos seem to show humans remotely operating some of the robots, suggesting the bots cannot handle them autonomously. Even so, the fact that robot hands are now deft enough to grasp tiny objects using tools, even under human control, is pretty impressive.
The full schedule for the games listed other events designed to test manual dexterity, including “block building” and “tweezer bean picking.” These tasks seem to reflect China’s eagerness to automate work in factories and warehouses, like wiring through an automobile’s chassis or assembling electronics.
A company called Galbot took part in several events, including a Supermarket Scenario Competition, which involves stocking shelves in a retail store. The company sees them as a way to push commercial capabilities forward. “Galbot will continue advancing humanoid robot capabilities by taking on increasingly challenging scenarios,” Yvonne Yuan, a representative for the company, told me in an email. “Moving from structured environments toward more open, unpredictable real-world applications where robots need to understand, adapt, and collaborate with humans.”
Despite a good deal of teleoperation across the various events, Galbot did demonstrate an impressive feat of autonomy—a robot capable of playing tennis with a human opponent. This requires advanced perception as well as high-speed reflexes.
At a time when AI is stirring up fears about mass job displacement, perhaps the biggest purpose of the World Humanoid Robot Games is getting the public excited about the potential of robotics.
"I think it is great that a country expresses excitement about science and engineering and has a positive view of a technological future,” says Chris Atkeson, a renowned roboticist at Carnegie Mellon University. “I am sorry it is not America.”
This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.
Comments
Back to top