Sony AI’s Ace beats pros; Honor’s Lightning wins robot half

Sony AI’s table tennis robot Ace defeated professional players in matches conducted under International Table Tennis Federation rules. Trials first documented in April 2025 showed Ace winning three of five matches against elite opponents. Additional matches in December 2025 and early 2026 included further victories. Official umpires oversaw the matches and applied standard competitive rules.

Ace was built for regulated, high-speed sport conditions. The platform uses nine synchronized cameras and three vision systems to track ball movement and spin, and it processes visual data fast enough to capture motion that appears as a blur to the human eye. The robot controls its racket with eight joints: three for positioning, two for orientation and three to manage shot force and speed. A study on the project was published in the journal Nature.

The team trained Ace through simulation and self-play rather than by imitating human demonstrations. According to project lead Peter Dürr, director at Sony AI Zurich, the work focused on how robots can respond with speed and accuracy in dynamic, real-world settings. He described the technical challenge this way: “Unlike computer games, where prior AI systems surpass human experts, physical and real-time sports like table tennis remain a major open challenge.”

Professional players reported that Ace presented new patterns. Mayuka Taira described the difficulty of reading the machine: “Because you can’t read its reactions, it’s impossible to sense what kind of shots it dislikes or struggles with.” Rui Takenaka, who both beat and lost to Ace, praised its handling of complex spins while noting it was more predictable on simpler serves.

Sony AI researchers said ongoing work aims to improve Ace’s adaptability during matches. The project team also highlighted potential applications for the perception and control methods used in Ace in manufacturing and service robotics, where fast sensing and coordinated movement can be useful.

In Beijing, the E‑Town Humanoid Robot Half Marathon staged a large-scale test of bipedal robot endurance and autonomy. More than 100 humanoid robots ran on a separate track from roughly 12,000 human participants. Honor’s robot Lightning completed the 21-kilometer course in 50 minutes and 26 seconds. Lightning hit a barricade during the race but continued and finished first; other Honor machines took second and third places.

Organizers reported that performance improved from the previous year, when the fastest robot finished in two hours, 40 minutes and 42 seconds. Honor noted that another of its robots completed the course in about 48 minutes under remote control, but race rules prioritized autonomous navigation, so Lightning was recognized as the official winner. Honor engineers pointed to structural reliability and liquid-cooling systems tested in the event and identified potential industrial applications for those technologies.

Both projects are examples of “physical AI,” where artificial intelligence is embedded in machines that operate in the physical world. Each project focused on fast sensing, real-time decision-making and coordinated motor control in live environments.