Sony AI: Ace, the Robot Challenging Table Tennis Champions

Sony AI Revolutionizes Table Tennis with Ace

The Ace robot, developed by Sony AI, recently made headlines by defeating professional table tennis players during competitions held under ITTF standards. This project, which graced the cover of the prestigious journal Nature, demonstrates that an autonomous system can now compete with human athletes in terms of perception and responsiveness in the physical world.

Among Ace's opponents is Miyuu Kihara, a table tennis player ranked among the top 25 in the world in women's singles. Ace also faced Yamato Kawamata, a professional coach who has been training regularly with the robot since late 2023. During these matches, the ball reaches impressive speeds of over 20 meters per second, with spins exceeding 160 revolutions per second. Ace is capable of perceiving its position in three dimensions at a frequency of 200 Hz, and it adjusts its striking trajectory in real-time using event-based sensors. Ace's total perception latency is only 10.2 milliseconds, allowing it to react almost instantaneously to the ball's movements.

A Dynamic and Adaptive Learning Process

The Ace robot does not merely repeat pre-programmed movements; it learns and adapts to the unpredictable dynamics of a live match. Inspired by Sony's previous project, Gran Turismo Sophy, Ace applies similar learning principles to manage the fast and complex exchanges of table tennis. Its control policy operates at a frequency of 1 kHz, enabling it to make strategic decisions while precisely controlling its movements.

Ace utilizes a hierarchical architecture that separates strategic decision-making from low-level motor control, allowing it to vary its strikes and keep its opponents guessing. For example, when the ball hits the net and deviates unexpectedly, Ace can correct its trajectory in a fraction of a second. It is important to note that no specific data about opposing players was used to prepare Ace before the matches. According to Peter Dürr, director of Sony AI in Zurich, real-world table tennis requires quick decisions based on noisy sensor data and adversarial human interactions, unlike simulated environments where AI has perfect information.

Impressive Competitive Performance

In March 2026, Ace won three out of five matches against elite players, scoring 16 direct service points compared to just 8 for its human opponents. Ace also maintained a return rate exceeding 75% for spins of up to 450 rad/s. These performances were observed during additional matches organized by Sony AI in December 2025 and March 2026, where Ace defeated three professionals, including Miyuu Kihara. Ace's strikes were not only faster but also more aggressive, with precise placement near the edge of the table.

Peter Stone, chief scientist at Sony AI, emphasized that significant advancements were made between April and December 2025, allowing Ace to hit the ball as hard and with as much spin as its human opponents. He asserts that this demonstration proves an AI system can perceive, reason, and act in complex real-world environments, which could have applications beyond sports.

Kinjiro Nakamura, a table tennis expert and former Olympian, was impressed by one of Ace's shots during the matches. He stated, "No one else could have done that. I didn't think it was possible. But the fact that it was... means that a human might be able to do the same." Before mixed matches become common, Ace could already serve as an excellent coach for human players.