- Galbot found an unprecedented way, called LATENT, to train robots
- Using “skill fragments”, they trained a Unitree G1 robot to play tennis
- The robot developed relatively robust tennis skills based on this minimal training
Future Wimbledons, where a fifth-seeded tennis pro battles a sixth-seeded robot, has just transitioned from the realm of science fiction to something that feels inevitable.
How did we get here? Blame it on Galbot and its LATENT innovation.
The best robotic athletes, those who can karate, box each other or parkour, are either remote-controlled or highly scripted to perform a pre-programmed set of actions. Real-time competition against e.g. a human opponent is thought to be difficult or impossible. But now Galbot and a team of researchers have done it: using minimal learning to teach a Unitree robot to play tennis against an unpredictable human opponent.
The article continues below
They call it “Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data” (which they jury-rigged to LATENT). Instead of highly detailed robotic training that captures the full range of human tennis skills, LATENT focuses on “movement fragments that capture the primitive skills used when playing tennis”.
Look at
Somehow, the researchers figured out how to use these pieces of tennis skill, or what they call “imperfect” data, to provide sufficient insight about “human primitive skills in tennis scenarios.”
The robot, a Unitree G1, can then draw on these fragments to make sense of live gameplay and, according to researchers, “consistently hit incoming balls under a wide range of conditions and return them to target locations.”
That’s a dry way to describe what’s happening in the sensational demonstration video, in which a Unitree G1 robot skillfully plays — and sometimes outplays — a human tennis player.
As they note in the research summary: “Our method achieves surprising real-world results and can stably sustain multi-shot rallies with human players.”
Now you might watch the video and assume that the human is taking the robot lightly or even aiming the ball in the direction of the robot. It is possible, but how do we account for the robot consistently returning the volley and deliberately placing the ball where the human player is not? It looks downright competitive.
Obviously, the robot could be better. It often appears to be teetering on the brink of disaster, and that racket appears to be fused to its right arm. I’m not even sure how the robot would handle a shot going over its metal and plastic head.
Still, the Galbot/LATENT robot puts on a big show from the ball return to a wicked backhand, the quick footwork and even the ability to stay upright.
It’s not too hard to imagine where this leads. Give it time and a robot like this could play exhibition matches against the likes of Rafael Nadal.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can also follow TechRadar on YouTube and TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
The best business laptops for all budgets
