Sony's Ace beats pros

Sony AI’s Ace has played table tennis under official rules against elite and professional humans, and it has won some of those matches. (nature.com) Table tennis is a hard robotics test because the ball can travel faster than 20 meters per second and the gap between shots is often under half a second. Nature said those demands put play “at the edge of human reaction time.” (nature.com) Ace is not a humanoid robot. Sony built it around a high-speed robotic system with event-based vision sensors — cameras that register changes in a scene instead of full frames — plus a control system trained with model-free reinforcement learning, a trial-and-error method for choosing actions. (nature.com) Sony AI said the milestone evaluation behind the Nature paper took place in April 2025 in Tokyo, where Ace faced five elite players and two professional players. All matches followed International Table Tennis Federation rules and were refereed by licensed umpires from the Japan Table Tennis Association. (ai.sony) In those matches, Reuters reported, Ace won three of five contests against elite players and lost two matches against professional players. Reuters also reported that Sony AI said Ace later beat professional players in December 2025 and again in March 2026. (reuters.com) The Nature paper says Ace achieved “several victories” and returned high-speed, high-spin shots consistently. A separate Nature News & Views article said the result shows an autonomous system can compete with humans in “complex, fast-paced, interactive tasks.” (nature.com, nature.com) That distinction matters in robotics because earlier artificial intelligence systems had already surpassed people in chess, Go and racing simulations, but mostly in digital or tightly controlled settings. The Ace paper says real-time physical sports remained an open challenge because a machine has to sense, decide and move while a human opponent keeps changing the situation. (nature.com, reuters.com) Sony AI’s Peter Dürr told Reuters the project was meant not only to play table tennis but also to study how robots can perceive, plan and act with human-like speed in dynamic settings. He said the same approach could apply to manufacturing, service robotics, sports, entertainment and safety-critical physical work. (reuters.com) Some researchers are drawing a narrower lesson. The Telegraph quoted Jan Peters of the Technical University of Darmstadt saying the result was “truly impressive,” while adding that success at table tennis does not solve broader robotics problems such as everyday object handling and manipulation. (telegraph.co.uk) For now, Ace’s claim is specific: a machine built for one fast, adversarial sport has shown it can rally, return spin and win points against top human players on a regulation table. The paper calls it the first real-world autonomous system competitive with elite human table tennis players. (nature.com)