Ace’s Performance in December 2025 Match
Credit: Sony AI
Ace, an advanced autonomous robot with AI and state-of-the-art sensors, plays competitive table tennis, defeating elite human competitors—a groundbreaking achievement in robotics.
While computers have dominated the strategic game of chess, Ace’s triumph suggests a pivotal moment for physical sports is near, reminiscent of the “Deep Blue” episode in 1997 when a machine bested chess champion Garry Kasparov.
“Games have always served as benchmarks for AI, akin to Deep Blue Chess and the game-changing AlphaGo,” says Peter Duerr, the mastermind behind Ace at Sony AI, Zurich.
Duerr emphasizes that unlike previous AI milestones achieved online, Ace competes directly against real-world professional table tennis champions, marking a significant progression.
“Ace offers a unique insight: the competition between robots and humans in genuine athletic events,” observes Duerr.
Ace incorporates three key advancements in autonomous robotics, according to Duerr. First, it uses “event-based sensors” that focus on specific image areas, detecting movement and brightness changes critical for tracking the trajectory of a table tennis ball.
Moreover, Ace’s table tennis skills are enhanced through “model-free reinforcement learning,” where it learns through practical experiences rather than pre-defined models of play. This process equates to an extensive training regimen of thousands of hours in simulated environments.
Finally, the high-speed robotic hardware enables Ace to exhibit “human-like agility.” Duerr notes that Ace’s response time is around 20 milliseconds, compared to the 230 milliseconds human athletes typically require, making it exceptionally agile.
Currently, Ace maintains a robotic appearance akin to those on factory floors, utilizing a network of cameras and sensors around the table tennis arena. Yet, researchers predict that with technological advancements, Ace may eventually evolve into a humanoid form.
In a recent match conducted under the regulations of Japan’s professional table tennis league, Ace faced five elite non-professional players, all with at least 10 years of experience and extensive training hours. Two professional players also participated in the event.
Ace secured victories in three out of five matches against elite players, although encountered losses against professional opponents, with a notable win against one professional participant.
Ace’s advantage lies in its unpredictability; while human players often read body language for cues, Ace relies solely on data, creating a distinctive challenge.
“Some athletes noted they usually gauge opponents’ expressions, yet Ace doesn’t exhibit such behavior,” Duerr explains.
A few players were surprised by Ace’s ability to interpret the spin on serves, which it countered adeptly despite attempts to disguise them. Ace even surprised its creators by returning a ball that hit the net, showcasing an unanticipated skill.
Since the research concluded over a year ago, the team has been continuously refining Ace’s capabilities.
In December 2025, Ace won its first match against a professional player, followed by victories against three professional players, including Miyu Kihara, currently ranked within the top 25 in the world, and two male pros, Touto Ryuzaki and Fumiya Igarashi.
“With further improvements, we aim to surpass even world champions,” states Duerr.
Duerr adds that the evolution of skills works both ways.
“Former Olympian Kinjiro Nakamura remarked that he believed certain shots were impossible until he witnessed Ace in action, leading him to believe that human athletes could emulate those techniques,” Duerr concludes.
Topics:
Source: www.newscientist.com
