Sony AI's robotic system Ace has defeated elite table tennis players in official competition matches, winning three out of five contests against highly trained athletes in what researchers are calling a breakthrough for physical artificial intelligence.
The robot, announced Wednesday and detailed in a study published in Nature, represents the first AI system capable of competing at expert human levels in a fast-paced physical sport under official International Table Tennis Federation rules. While AI has long dominated strategic games like chess and Go, physical competitions have posed far greater challenges requiring split-second reactions, precise motor control, and real-world adaptability.
How Ace works
Ace uses an eight-jointed robotic arm mounted on a movable base, equipped with a sophisticated vision system comprising nine traditional cameras positioned around the court and three specialized "gaze control systems." The cameras track the ball's position in 3D space while the gaze control systems measure angular velocity and spin to calculate trajectory.
"Table tennis is a game of enormous complexity that requires split-second decisions as well as speed and power."
Peter Dürr, director of Sony AI and project lead for Ace, said in the company's press release. The system can determine ball spin by zooming in on the ball's logo within milliseconds of it reaching Ace's end of the table.
The robot's decision-making relies on model-free reinforcement learning, allowing it to adapt without preprogrammed responses. Sony researchers trained the system through 3,000 hours of simulated gameplay, while serving techniques were modeled on expert human players.
Performance against human opponents
During testing in April 2025, Ace faced seven human competitors: five elite players with over 10 years of training and two professionals who compete in professional leagues. The robot won three matches against elite players but lost both contests against professionals, managing only one game victory across seven total matches.
- Won 3 of 5 matches against elite players
- Lost 2 of 2 matches against professional players
- Scored 16 direct points while serving versus elite players' 8
- Demonstrated ability to handle complex spin and net shots
Sony says Ace has continued improving since the Nature study was submitted, with the robot going on to defeat professional players in December 2025 and March 2026, according to Reuters reporting cited by the Guardian.
Elite player Rui Takenaka noted that Ace returned complex spin shots with equally complex spin, making rallies difficult. However, the robot struggled with simple "knuckle" serves, returning weaker balls that allowed human players to attack on the third shot.
What makes this different
Previous table tennis robots, including Omron's FORPHEUS demonstrated at CES 2017, were designed to play against amateur competitors. The Verge notes that Ace is the first robot capable of holding its own against top-ranked human players while following official ITTF rules.
The robot displayed unexpected capabilities during testing. Former Olympic player Kinjiro Nakamura said he initially thought one of Ace's unusual backspin shots was impossible, but now believes humans could learn the technique. The robot also demonstrated remarkable adaptability when balls caught the net, responding extremely quickly to altered trajectories.
Unlike human opponents, Ace provides no visual cues about its intentions and shows no emotional response to pressure situations. "The players want to see the eyes of their opponent. And the eyes of Ace are all around the court and they don't show any intention or feeling," Dürr explained to the Guardian.
How the outlets are framing it
CNET positioned Ace as a glimpse into the future of physical AI applications beyond sports.
The Verge emphasized the technical achievement of creating the first robot competitive against top human players.
The Guardian framed the story as a robotics milestone while noting the robot's continued improvement since initial testing.
Peter Stone, chief scientist at Sony AI, said the achievement "opens the door to an entirely new class of real-world applications that were previously out of reach." However, Jan Peters from the Technical University of Darmstadt cautioned that while the project is "truly impressive," table tennis research won't solve broader robotics challenges like object manipulation, and "a lot of good old-fashioned engineering is needed" for robots to be useful to the general public.
Conversation 0 comments
Join the discussion ↓