Sony AI's Ace Robot Defeats Professional Table Tennis Players

Table tennis is fast, physical, and deeply dependent on real-time adaptation. A player at the elite level can execute spin at more than 400 radians per second, react in milliseconds, and read an opponent's body language to anticipate the next shot. That kind of integrated physical intelligence has, until now, been exclusive to humans.

On April 23, 2026, Sony AI published research in Nature showing that their robot, Ace, has become the first autonomous system to win competitive matches against professional-level table tennis players. These were not simulated games or controlled lab demonstrations. They were real competitive matches, under International Table Tennis Federation rules, against opponents who were actively trying to win.

Among Ace's opponents was Miyuu Kihara, a top-25 ranked player. In sessions held in December 2025 and March 2026, Ace defeated two elite players and one professional player in December, then beat all three professional opponents it faced the following March. Over those sessions, Ace maintained a return rate above 75 percent on serves with spin up to 450 radians per second and scored 16 service aces against elite opponents, who collectively scored only eight against the robot.

The speed difference between robot and human is striking. Ace's latency, the time from perceiving a shot to beginning its physical response, is 20.2 milliseconds. Elite human players respond in approximately 230 milliseconds. The robot is not just faster; it operates on a fundamentally different timescale.

The engineering behind this is worth understanding. Ace uses nine high-speed cameras with IMX273 image sensors and three gaze control systems built on event-based vision sensors for real-time ball tracking. Its control system runs on model-free reinforcement learning, which means it adapts without pre-programmed models of how the ball or opponent should behave. It works things out as play unfolds.

Peter Dürr, Director of Sony AI in Zürich and the project's lead researcher, put it plainly: this research shows that an autonomous robot can win at a competitive sport, matching or exceeding the reaction time and decision-making of humans in a physical space. Peter Stone, Chief Scientist at Sony AI, described it as a landmark moment, noting that an AI system has for the first time demonstrated the ability to perceive, reason, and act effectively in complex, rapidly changing real-world environments.

Previous AI milestones have mostly lived in digital spaces with fixed rules: chess, Go, video game competitions. Physical sports add layers of complexity those environments cannot replicate. Lighting shifts. Ball spin varies. Players adjust their footing mid-rally. A robot competing in physical space must process continuous sensory input and make split-second mechanical adjustments against an opponent who can see and respond to it.

Sony AI says the implications extend well beyond the table tennis hall. The same real-time perception and control capabilities that let Ace return a 450-rad/s serve are relevant to any task demanding rapid, precise physical interaction. Applications in surgical robotics, manufacturing, and emergency response all require fast, adaptive decision-making that Ace has now demonstrated at a competitive level.