Ping Pong Robot Ace Defeats Top Players in Tokyo, Creating a First in Sports Robot History

“Ace” was developed by Sony AI, Sony’s artificial intelligence research division, and is the first robot system to reach an expert level in a competitive physical sport that demands extremely high speed and precision. The project leader stated that Ace utilizes high-speed visual perception, artificial intelligence control algorithms, and an advanced robotic hardware platform to make extremely fast and accurate technical choices and execute shots in ping pong matches.

Since 1983, various ping pong robots have emerged, but none have been able to compete with high-level human players. Ace has changed this situation: it has already won against elite and professional players in matches adhering to official International Table Tennis Federation rules and enforced by certified referees. Peter Del, head of Sony AI Zurich and the project leader, pointed out that unlike “pure digital” AI that has comprehensively surpassed human experts in games like chess and electronic games, physical sports like table tennis, which require close-range, high-speed confrontation, remain a major challenge in the fields of AI and robotics.

Del emphasized that the project’s goal is not only to allow robots to compete with humans in ping pong but, more importantly, to explore how robots can achieve a human-like “perception-decision-action” closed loop in dynamic environments, completing prediction and control in a very short time. He stressed that Ace’s success in visual perception and learning-based control algorithms suggests that similar technologies could be extended to other scenarios requiring high-speed real-time control and human interaction, such as manufacturing, service robots, and a variety of applications including sports, entertainment, and even safety-critical physical systems. The related research results were published in the journal *Nature* on the 22nd.

The research team disclosed in the paper that Ace won 3 out of 5 matches against elite players in April 2025 and lost both matches against professional players. Sony AI added that Ace has since achieved victories against professional players in December 2025 and this March, with its performance continuing to improve. Meanwhile, companies around the world are constantly making breakthroughs in the robotics field, such as the scene at the Beijing Half Marathon this week where a robot runner outperformed human athletes.

Unlike chess or electronic games that operate in a virtual environment, ping pong requires the system to make instantaneous decisions while completing precise hits on high-speed flying balls and continuously adapting to unpredictable changes from the opponent. Ping pong’s high speed and complex spin push both humans and machines to the limits of their perception, prediction, and motion control capabilities. To this end, Ace’s architecture integrates 9 synchronized cameras and 3 visual systems to track the rapidly rotating ping pong ball in real-time with extremely high precision. Del said that the system’s response speed is sufficient to capture subtle movements that appear almost “blurred” to the human eye.

In terms of body design, Ace adopts a custom robot platform with 8 joints. Del explained that this is the “minimum configuration” for performing competitive shots: 3 joints control the racket’s position, 2 joints are responsible for the racket’s posture, and another 3 joints are used to adjust the hitting speed and force. This design allows Ace to complete a variety of high-quality shots, including fast loops, changes in direction, and responses to different spins.

From feedback from human opponents, Ace’s “style” on the court is also highly oppressive. Mayuka Hirayama, a professional player who lost to Ace in December last year, said that the robot is “very unpredictable and completely emotionless.” She admitted that it is almost impossible to find its weaknesses by “reading the opponent” because she cannot judge its preferences or vulnerabilities from its expressions or body language, which greatly increases the difficulty of the match.

Rui Takenaka, an elite player who has played against Ace multiple times with mixed results, believes that the robot’s “recognition ability” for different serve qualities is impressive. He said that when he serves with complex spin, Ace also responds with complex spin, making it very difficult for him to sustain the rally; however, when he uses a weaker “no-spin ball” (commonly known as a “float serve” or “sidespin serve”), Ace’s return is relatively simple, creating an opportunity for him to attack on the third shot, which he considers one of the key factors in defeating Ace.

Although Ace has demonstrated “superhuman” capabilities in several indicators, Del believes that the robot system still has considerable room for improvement. He pointed out that Ace has already reached a level beyond humans in reading ball spin and reaction speed, and because it does not learn by watching human matches but mainly through simulated self-play training, it often exhibits response patterns that are very different from humans, even creating situations that experienced players have not anticipated. However, conversely, professional athletes are extremely talented at adapting to opponents and finding weaknesses, which is one of the areas the research team is focusing on.