Sony’s AI-powered table tennis robot, named Ace, has demonstrated the ability to compete with and occasionally defeat elite human players in real-world matches, according to a study published in the journal Nature on April 22, 2026. The robot combines high-speed perception, reinforcement learning and precision robotic hardware to track and return balls with spin and velocity estimates comparable to those of professional athletes.
Ace was developed by Sony AI as part of ongoing research into autonomous systems capable of operating in unpredictable, fast-paced physical environments. Unlike earlier AI achievements in simulated domains such as chess or video games, Ace functions in real time using conventional cameras to monitor the ball’s position and three specialized gaze control systems to estimate its angular velocity — a critical factor in table tennis where spin dramatically affects trajectory.
The research team, led by Dürr et al., designed Ace to overcome the significant engineering challenge of closing the perception-action loop at speeds necessary to rival human reflexes. In matches against skilled amateur and professional players, the robot has shown it can sustain rallies and win points through a combination of predictive modeling and rapid mechanical response, marking what the researchers describe as a milestone in physical AI.
This development builds on Sony AI’s prior work with GT Sophy, an agent that surpassed human performance in the racing simulation Gran Turismo. However, Ace represents a step further by operating in the physical world with real rackets, a standard table, and an unmodified ball — variables that introduce noise, latency, and mechanical constraints absent in virtual environments.
According to the study published in Nature, Ace’s control system integrates reinforcement learning trained on both simulated and real-world data, allowing it to adapt to opponent behavior and refine its shot selection over time. The perceptual pipeline processes visual input at high frame rates to predict ball trajectory within milliseconds, enabling timely actuation of the robotic arm.
While the robot does not yet consistently defeat top-ranked players, its ability to challenge them in competitive play underscores progress in closing the gap between AI performance in digital and physical domains. Researchers note that such systems could inform future applications in industrial automation, prosthetics, and human-robot collaboration where split-second decision-making is essential.
The findings were reported by multiple outlets including Agence France-Presse (AFP), Lavanguardia, and the Associated Press, all confirming the April 2026 publication in Nature and detailing Ace’s public demonstrations in Tokyo during late 2025.
As of the date of publication, Sony has not announced plans to commercialize Ace or disclose technical specifications beyond those shared in the peer-reviewed study. The research remains focused on advancing fundamental capabilities in embodied AI rather than immediate product development.
For updates on Sony AI’s robotics research, interested readers can follow official publications from the company’s AI division or monitor future releases from Nature Robotics.
What do you suppose about the implications of AI systems mastering complex physical sports? Share your thoughts in the comments below and spread the conversation by sharing this article.