"Sony AI," the artificial intelligence research and development division of the Sony Group, announced that it has developed a table tennis robot utilizing AI. According to the announcement, the robot won multiple matches against high-level players with national tournament experience in a realistic setting. Although it was defeated by current professional players, it put up a strong fight by returning high-speed spinning balls, demonstrating high performance. Sony AI stated, "This is the first time a robot has played on par with top-tier players in a real-world sporting setting." A paper detailing its performance and match results was published in the British scientific journal Nature on April 22.
To win a match between high-level table tennis players, error-free responses during fast-paced rallies are essential, and the robot must predict the complex trajectories of balls spun by a player's unique serves. Trading shots with high-level players is highly challenging for an AI robot. Previous robots were only capable of maintaining a simple rally and lacked the performance required to compete in actual matches.
The research group, including Sony AI, developed the autonomous robot "Ace" by combining Sony's proprietary advanced technologies, AI "reinforcement learning" techniques, and precision hardware.
Ace features an arm with eight specialized joints capable of agile movements. A control system utilizing a method called "model-free reinforcement learning" allows the robot to respond rapidly to balls struck by high-level players, with the high-speed movements of the ball tracked by nine cameras equipped with specialized sensors installed around the table tennis table. After validating Ace's capabilities, the research group decided to confirm its performance through actual matches against high-level and professional players.
Provided by Sony AI
According to Sony AI and the paper published in Nature, the matches against the players were conducted in a realistic format in April 2025, adhering to the rules of the International Table Tennis Federation (ITTF). The opponents consisted of five high-level players, each with over 10 years of experience as active players, national tournament appearances, and an average of 20 hours of training per week, and two current professional players active in Japan's professional league.
As a result of the matches, Ace won 3 out of 5 matches (3 out of 5 individuals) against the high-level players. In terms of games, it won 7 out of 13 total games. While it lost both matches against the two professional players, it managed to take 1 game out of 7 total games played. Analysis revealed that Ace could handle various high-speed spins, achieving a return rate of over 75%. The maximum speed of the ball struck by Ace was 16.4 m/s, but it demonstrated high return capabilities by successfully returning even faster high-speed shots clocked at 19.6 m/s struck by the players.
Regarding examples of robots excelling in the field of sports, a Chinese humanoid robot previously made headlines when it was reported to have significantly shortened the men's world record time in a half marathon. Commenting on the impressive performance of this AI table tennis robot, a representative from Sony AI stated, "This demonstrates for the first time that an AI system can perceive, reason, and act effectively in a complex, rapidly changing real world where precision and speed are required," adding that "the breakthrough [of the AI robot] extends far beyond the boundaries of table tennis."
Provided by Sony AI
Original article was provided by the Science Portal and has been translated by Science Japan.