Mayuka Taira, a professional table tennis player, lost to a robot called Ace in December 2025 and described the experience: "it is very hard to predict, and it shows no emotion. Because you can't read its reactions, it's impossible to sense what kind of shots it dislikes or struggles with." Table tennis at the competitive level relies heavily on reading your opponent — their body language, their hesitation, the micro-adjustments they make before a difficult return. Ace has none of that to give.
Sony AI designed Ace around a custom 8-joint platform — three joints to position the racket, two to orient it, three more to control shot power and speed. Nine synchronized cameras plus three vision systems handle perception, running fast enough that ball motion the human eye would see as a streak is, for Ace, a trackable object. Through April 2025, the robot went 3-2 against elite-level players but dropped both matches against professionals. It cleared that bar in December 2025, beating Taira, and then went on to beat more professionals in March 2026. Sony published the results in Nature on April 22, per Reuters.
The paper claims no robot has previously beaten professionals at a commonly played competitive physical sport under that sport's official rules (here, ITTF regulations). Peter Dürr, director of Sony AI Zurich, framed the real goal as less dramatic than sports dominance — robotics capable of handling fast, unpredictable physical events well enough for manufacturing lines, service applications, and safety-critical environments.
One elite player who won matches against it spotted an exploitable gap: deliveries with no spin. Competitive table tennis is largely a conversation in spin — the server imparts it, the receiver reads it, and both adjust. A "knuckle serve," thrown without spin, apparently scrambles Ace's read of the situation in a way a human opponent would figure out how to hide.
Previously:
