First, they learned how to play tennis on two legs. Then they came for our half marathon world record.
Robots are now on the verge of overtaking us in the sport of table tennis. In a recent paper published this week in the journal natureThe researchers detailed how they used AI to teach a robotic arm developed by Sony called Ace to repeatedly beat “elite and professional athletes under official competition rules.”
After decades of developing table tennis robots, Sony claims that this is the first robot to achieve expert-level performance in any competitive sport.
A promotional video put together by Sony’s AI division shows a paddle-wielding robot bouncing back and forth at breakneck speeds to counter aggressive attacks. The human experts don’t seem to be holding back either, and continue to hit challenging shots with all their might.
project ace
This is an impressive technological feat and a major milestone in the application of AI to robotics, bringing together two research fields that are at the core of the boom in humanoid robotics that has been going on for several years.
Teaching a robot arm to play table tennis may not be the start of the next industrial revolution, but many of Ace’s achievements could eventually spill over into other fields of research.
“The success of Ace, with its recognition system and learning-based control algorithms, suggests that similar technology can be applied to other fields that require fast real-time control and human interaction,” said Peter Duerr, lead author and project leader at Sony AI. Reuters“Applications across manufacturing and service robots, as well as sports, entertainment, and safety-critical physical domains, etc.”
According to the company’s paper, as of April 2025, Ace has won three out of five matches against elite players with more than 10 years of experience, but lost two matches against top-level professionals. However, Sony claims that Ace continued to beat even more pro players in December, just as he did last month.
The level of complexity is staggering. The robot not only tracks the ball, but uses nine cameras and three vision systems to determine its trajectory in real time, requiring extremely fast reaction times.
The entire system is “capable of measuring spin at up to 700 Hz while tracking the ball at 200 Hz with millimeter accuracy and approximately 10 milliseconds of latency,” the accompanying Sony AI blog post explains. “This is fast enough to capture motion that would be blurry to the human eye.”
Deep reinforcement learning allows you to seamlessly predict the behavior of the ball and choose how to counter your opponent.
On the other hand, many high-level players were confused by this device. Table tennis professional Mayuka Taira said, “Basically, it’s impossible to detect what kind of shots you don’t like or what kinds of shots you struggle with, and that makes the match even more difficult.” Reuters.
Despite this, Dürr said professional players still have an advantage, at least for now, because they are “very good at adapting to their opponents and finding weaknesses, and that’s something we’re working on.”
More information about table tennis: This video of a humanoid robot playing tennis is very impressive
