An AI-powered robot named Ace, developed by Sony AI, has beaten elite table tennis players in a series of matches played under official competition rules. The robot won three out of five matches against elite players, but lost both matches against professional players, managing to win only one game across those two contests.
The achievement is hailed as a milestone in robotics, as table tennis requires lightning-fast reactions, perception, and skill. Ace demonstrated mastery of spin, handled difficult shots such as balls catching on the net, and executed a rapid backspin shot that a former Olympic player believed impossible.
A research paper on the robot was published in Nature on Wednesday. Peter Dürr, director of Sony AI in Zurich and project lead, said Ace had improved since the report was submitted, noting that it beat increasingly strong opponents. The robot's skills were honed through 3,000 hours of simulated games, while serves were drawn from expert players.
Ace uses an eight-jointed arm on a movable base and multiple cameras to track the ball's position and spin. By zooming in on the ball's logo, the system estimates spin and axis of rotation in milliseconds. However, the robot struggled with slow, low-spin balls, returning them weakly.
Elite player Rui Takenaka noted that Ace returned complex spin with complex spin, but simple serves led to simpler returns, making it easier to attack. Former Olympic player Kinjiro Nakamura said Ace's unusual backspin shot was previously thought impossible, but now humans could learn it.
Dürr highlighted that Ace lacks eyes or body language, and does not succumb to pressure. Jan Peters, a professor at the Technical University of Darmstadt, called the project impressive but noted that table tennis research does not solve all robotics challenges. He predicted a world-changing moment in the next decade, similar to ChatGPT's impact in 2022.
