Sony AI’s cutting-edge table tennis robot, dubbed “Ace,” has made a significant stride in the realm of physical AI, successfully challenging and defeating high-caliber human players in competitive matches. This development marks a crucial advancement in artificial intelligence’s application to real-world, dynamic environments that demand split-second decision-making and extraordinary motor control.
Ace is engineered to excel in the fast-paced, precision-driven arena of competitive table tennis. The system is a testament to the synergy between sophisticated high-speed perception and AI-driven control algorithms, enabling it to execute complex shots under intense match conditions. In trials conducted under the stringent rules of the International Table Tennis Federation and officiated by licensed umpires, Ace demonstrated its mettle. Documented trials in April 2025 saw the robot secure victories in three out of five matches against elite human competitors, while also experiencing losses against professional-level opponents. Subsequent encounters in December 2025 and early 2026 reportedly resulted in Ace triumphing over professional players.
While table tennis robots have been a fixture since the 1980s, none have previously approached the performance capabilities of advanced human athletes. Peter Dürr, Director at Sony AI Zurich and the project’s lead, highlighted this distinction: “Unlike computer games, where prior AI systems surpass human experts, physical and real-time sports like table tennis remain a major open challenge.” He elaborated that AI has achieved remarkable success in digital environments like chess and video games due to their fully simulated nature, a stark contrast to the unpredictable variables of a physical sport.
The development of Ace was motivated by a desire to explore how robots can achieve speed and accuracy in dynamic, real-world settings. This groundbreaking work was detailed in a study published in the prestigious journal *Nature*. The inherent challenges of table tennis, including the extreme speed, spin variability, and rapidly changing ball trajectories, necessitate rapid sensing and highly coordinated movements within exceptionally tight timeframes. Ace addresses these challenges through its advanced architecture, which comprises nine synchronized cameras and three vision systems. These components meticulously track the ball’s trajectory and spin, processing visual data at a speed that surpasses human visual acuity, effectively capturing motion that would appear as a blur to the naked eye.
Mechanically, the robotic platform employs eight joints to control the racket. Three joints manage positioning, two handle orientation, and three are dedicated to modulating shot force and speed. This configuration is meticulously designed to meet the minimum mechanical requirements for competitive play, allowing for a broad spectrum of shots.
Significantly, Ace’s training regimen diverged from the conventional human demonstration approach. Instead, the AI was trained in a simulated environment, a method that empowered it to develop its own unique strategies and play patterns, often differing from those of human opponents. Dürr explained, “The system learns to play not from watching humans but through self-training in simulated environments.”
Professional players who have faced Ace offer compelling insights. Mayuka Taira, who lost a match, found the robot difficult to predict due to its lack of visible cues during play. Rui Takenaka, an elite player who experienced both victory and defeat against Ace, noted its proficiency in handling complex spins but found it more predictable on simpler serves. Taira further observed that the robot’s absence of emotional signals made anticipation of its responses a significant challenge: “Because you can’t read its reactions, it’s impossible to sense what kind of shots it dislikes or struggles with.”
Dürr confirmed Ace’s strong capabilities in spin detection and rapid reaction, while acknowledging that ongoing research is focused on enhancing its in-match adaptability. The project team believes that the perception and control techniques refined for Ace hold considerable promise for applications in sectors such as manufacturing and service robotics.
### Humanoid Robots Tackle the Half Marathon Challenge
In a parallel development showcasing advancements in physical AI, the 2026 Beijing E-Town Humanoid Robot Half Marathon saw over 100 humanoid robots compete in a demanding 21-kilometer race. This event, which also included approximately 12,000 human participants on separate courses, served as a critical testbed for humanoid robots operating under large-scale, real-world conditions.
A robot named Lightning, developed by Honor, distinguished itself by completing the course in an impressive 50 minutes and 26 seconds, surpassing the Olympic runner Jacob Kiplimo’s recorded time. Despite a mid-race collision with a barricade, Lightning persevered to finish first, with other Honor robots securing second and third places. This performance represented a dramatic improvement from the previous year, where the fastest robot finished in over two hours and forty minutes.
Honor engineers indicated that the technological innovations developed for their robots, including advanced structural reliability and liquid-cooling systems, are directly transferable to industrial applications. Notably, another Honor robot reportedly completed the course in a faster time of 48 minutes under remote control. However, prioritizing autonomous navigation in line with race rules, Lightning was officially recognized as the winner, underscoring the paramount importance of independent robotic operation.
Original article, Author: Samuel Thompson. If you wish to reprint this article, please indicate the source:https://aicnbc.com/20937.html
