A robot trained with AI already manages to beat elite ping-pong players

Nature magazine dedicates its cover this Wednesday to describing Ace, a new robotic system based on artificial intelligence that Sony developed to play ping-pong matches (and win them) against the best.

With table tennis, robotic science goes one step further than with athletics as it is a complex sport that requires, in real time, quick responses, continuous adaptation to the opponent and very precise prediction of the shots and the trajectory of the ball, including complex shots.

Ace consists of a high-speed perception system that uses a network of nine cameras with image sensors and gaze control, an AI-based control system, as well as a high-speed robotic arm with eight joints for agile physical interaction.

Ace was tested in Japanese professional league matches following the rules of this sport.

Among its rivals described in the article are five elite players (each with more than 10 years of experience and an average of 20 hours of weekly training) and two professionals (Minami Ando and Kakeru Sone), reported the EFE agency.

the robot won three of the five games against elite players, and lost both matches against professionals, although managed to snatch a set from one of them.

Scientists describe how Ace demonstrated a sophisticated range of abilities, including managing the effects, obtaining points through various types of plays instead of just faster punches, and the quick reaction to unusual situations, like balls bouncing off the net.

Although robots had already been built to play ping-pong, most had only proven effective in cooperative exchanges: none had beaten human experts in a real-time game.

Its inventors consider it a milestone in AI-based robots that compete with humans and outperform them on complex, interactive real-world tasks.

The authors even report that after observing a blow executed by Ace, former Olympic player Kinjiro Nakamura commented that He never thought such a move was possible. and that it would be very effective if humans could replicate it.

“It is very significant that athletes like Nakamura They could learn new skills just by playing against Ace and watching him, indicating that robotic systems controlled by AI could constitute an area for human development beyond table tennis,” say Brazilian researchers Carlos Ribeiro and Esther Colombini in a reaction to the study reported in Nature.

“The implications of the finding go beyond sport. We are demonstrating that an autonomous robot can win in a competitive sport, matching or surpassing the reaction time and decision making of humans in a physical space,” says Peter Dürr, director of the Sony AI division in Zurich and head of the Ace project.

“By solving a problem that requires exceptional real-time detection and control, this research lays the foundation for create robots with broader applications in fast and precise human interactions“he adds.

The researchers explained that they organized more competitive matches after delivering the documentation to Nature for the article published this Wednesday. The robot was measured against four new players: two professionals and two elite in December 2025.

In these matches, Ace defeated the two elites and one of the professionals, while he lost against the second highest level opponent.

With improvements, “the robot achieved even higher hitting speeds, more aggressive placement near the edge of the table, and faster exchanges of blows, reflecting continuous improvement of performance in competitive conditions“Dürr details.

Ribeiro and Colombini agree: “Ace is an important milestone, which shows the potential of the next generation of competitive robots of high capacities that interact with physical environments”.