Sony AI's 'Ace' Shocks the World | First Ever Robot to Beat a Pro at Table Tennis
機械翻訳 / Machine-translated
A
AI Friends
2026年4月26日・32分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"Will the day ever really come when a robot beats a professional table tennis player?" — Sony has answered that question head-on.
On April 23, 2026, Sony AI unveiled its autonomous table tennis robot, 'Ace.'
The research paper appeared on the cover of the scientific journal Nature, cementing Ace's place in history as the world's first AI robot to defeat elite players in a real-world sport.
We'll break down exactly what makes this so impressive, how it works, and what ripple effects it will have on our lives and work.
Let's start by being clear about what makes this announcement a "world first."
The paper is titled "Outplaying elite table tennis players with an autonomous robot," and was published in Nature, Issue 8111, released on April 23, 2026.
Nature is the world's most prestigious multidisciplinary science journal — simply being published in it is a remarkable achievement.
Being featured on the cover is proof that the research is the flagship study of the issue.
Think of it as the equivalent of winning the annual MVP award in professional baseball. Sony Semiconductor Solutions' announcement explicitly confirms that its image sensors were adopted for the project.
Ace is the world's first robot to compete against humans under official ITTF rules. Previous table tennis robots came with limitations — they could only return balls, used non-standard court sizes, or only handled predetermined practice shots.
The difference is like "clearing a mini-game versus winning an actual professional tournament." What makes Ace groundbreaking is that it "played under the same conditions, the same racket specifications, and the same rules as professional players — and won." That distinction is the substance of the "world first" title.
"Physical AI" refers to AI that operates not just on a screen, but in the physical world — in robots, vehicles, and other embodied forms. If ChatGPT is "AI that's intelligent in the world of text," Ace is "AI that's intelligent in the world of movement."
Ace joins the ranks of landmark examples of physical AI, alongside cooking robots, laundry-folding robots, and self-driving vehicles. Gartner has designated 2026 as "the first year of Physical AI," and Ace has become its defining achievement.
Here is a chronological breakdown of the results. Ace underwent multiple updates throughout this period.
The first official evaluation was held in April 2025.
Ace faced "5 elite players (with 10+ years of table tennis experience)" and "2 professional players," winning 3 of 5 matches against elite players (taking 7 games, losing 6). Against the professional players, Ace went 0-2 (winning 1 game out of 7).
"Winning even a single game against a pro" was already a record-breaking performance in the history of robotics.
In additional matches held in December 2025, Ace faced 2 new professional players and 2 elite players. It defeated both elite players and went 1-1 against the professionals — recording its first-ever win against a pro.
This was a milestone similar to "the moment a child beats an adult for the first time," and an important result demonstrating that "AI can reach the pinnacle of human performance."
In the most recent matches in March 2026, Ace had evolved to the point of winning at least one game against each of three professional players.
"Faster shots, aggressive placement targeting the edges of the table, and faster rally tempos" were all observed.
"In one year, Ace went from being completely outmatched by pros to reliably beating them."
The pace of progress is like "a human player achieving in one year what would normally take ten."
Why is Ace so strong? The secret lies in the technology behind its "eyes," "brain," and "body."
Ace tracks the ball with a total of 12 cameras. At the core are three "IMX636" sensors (an event-based vision sensor jointly developed by Sony and Prophesee), which only transmit data from pixels where brightness has changed — they track "changes only" rather than capturing full images.
Think of it as "subtitles appearing only at the exact moment something moves during a live broadcast." Combined with nine standard image sensors ("IMX273 Pregius"), which capture high-resolution images at 200Hz, the system uses 3D triangulation to pinpoint the ball's position with precision.
Ace's perception latency — the time it takes to "see" the ball — is 10.2 milliseconds (0.0102 seconds). Even elite human players require around 230 milliseconds, meaning Ace perceives the ball more than 20 times faster.
This goes far beyond "a seasoned player for whom opponents seem to move in slow motion" — it's superhuman reaction speed capable of "reading the ball's spin at the 0.01-second level." Ace accurately measures ball spin exceeding 9,000 rpm, predicting ball trajectory and positioning itself optimally in advance.
Ace's decision-making is powered by "model-free Deep Reinforcement Learning" — an AI that "automatically learns the most effective movements through repeated trial and error." Before practicing on a real table, Ace plays through billions of rallies in simulation (a virtual table tennis environment).
"Because you can practice an unlimited number of matches in training mode, you can improve tens of thousands of times faster than in the real world." End-to-end latency — from seeing the ball, to deciding, to striking — is 20.2 milliseconds, less than one-tenth of a human's.
Ace's hardware consists of "2 linear joints + 6 rotational joints, totaling 8 joints."
This design combines "rails for rapid lateral movement" with "joints for swinging the arm in multiple directions."
The end effector (the robot's "hand") holding the racket also features a "cup" for holding the ball, enabling it to serve with one arm — a sophisticated setup.
Capable of returning balls at up to 19.6 meters per second (approximately 70 km/h), Ace can reproduce nearly all the techniques used by professionals, including topspin, slice, and smash.
When people hear "table tennis robot," many think of Omron's FORPHEUS. Here's a breakdown of the differences.
FORPHEUS is a table tennis robot that Omron has been developing since 2013 under the concept of "harmony between humans and machines." The latest ninth generation features a large language model (LLM), automatically adjusting its verbal encouragement and ball speed to match each player.
It's more like "a kind coach guiding its opponent," and it has been a star at trade shows for its ability to "adapt its pace to the player's level."
Ace is "a research robot that goes all-out with zero mercy."
"If FORPHEUS is a casual garden tennis game, Ace is a player sent into the finals of a national championship."
It's important to note that "comparing them by strength is meaningless because their purposes are fundamentally different." FORPHEUS is about "providing an enjoyable experience for anyone at a trade show," while Ace is about "breaking through the frontier of AI research" — their roles are clearly distinct.
"The strength of Ace" and "the gentleness of FORPHEUS" will both be needed in future physical AI development.
A future is coming into view in which robots equipped with "professional-level precision + the judgment to connect with humans" will be applied across a wide range of fields — sports coaching, rehabilitation, caregiving, and factory automation.
"A robot with the qualities of both a kendo master and a dance instructor" will become a full-fledged reality within the next ten years.
What looks like a one-off headline about "a robot winning at table tennis" actually carries much broader implications.
The IMX636 and IMX273 sensors used in Ace are Sony's flagship sensors already slated for installation in automotive cameras and industrial equipment.
"Proving world-class performance in Nature" directly translates into marketing power in the automotive, factory, and medical sectors.
"The sharp eyes that defeated a professional player," installed in a self-driving car, would make for an extremely persuasive safety system to prevent accidents. The fact that Sony Semiconductor Solutions issued its own separate announcement on April 23rd underscores the business intent behind this achievement.
"A robot on par with a pro" is revolutionary as a practice partner for up-and-coming players.
"A coach available 24 hours a day, never tiring, never complaining, always giving 100%" becomes a reality — and "the day when one Ace is placed in every top table tennis school" may not be far off. In manufacturing too, the realization of "industrial robots working with the same precision as veteran craftspeople" is seen as drawing near, with high expectations for solving the labor shortage.
The Ministry of Economy, Trade and Industry projects that the multipurpose robot market, including humanoids, will grow to approximately 60 trillion yen by 2040.
"On the scale of an entirely new major industry being born," Sony AI's Ace is a result that strongly marks Japan's presence.
It also demonstrates that "while the U.S. dominated conversational AI with ChatGPT, Japan — including Sony — still has a fighting chance in physical AI," making this a truly significant achievement.
Mr. Nakamura coaches the table tennis club at a rural public high school.
His frustration: "Unlike powerhouse schools, we only have 3 students to practice against, and our most advanced member has hit a wall."
Around 2028, a practice table tennis robot applying Ace's technology becomes available for lease at roughly 3 million yen, giving club members unlimited access to "balls 1.5× faster than a pro, every type of spin, and returns from any angle."
"One student's skill level skyrocketed in a single year — from the third round of the prefectural tournament to qualifying for nationals." With tears in his eyes, Mr. Nakamura reflects, "I was finally able to give that student a real chance."
Ms. Mori develops autonomous driving systems at a major automaker.
She had hit a technical wall: "We can't shave off that last 50ms between detecting a pedestrian jumping out and coming to a full stop."
The high-speed perception technology of the IMX636 event sensor — proven by Ace — is adopted in a 2027 model, "cutting detection latency to one-fifth of what it was."
"A safety system that can fully stop the vehicle more than 70% of the time when a pedestrian runs out while traveling at 50 km/h" goes into mass production. Mori reflects with deep emotion: "Ace's achievements have started saving lives."
Mr. Sato handles production technology at an electronics manufacturer.
His headache: "The fine assembly work that only veteran craftspeople could do is impossible to pass on, with so many of them retiring at once."
In 2029, his company introduces a next-generation industrial robot applying Ace's deep reinforcement learning and high-speed vision technology. "Skills it took veteran workers 20 years to master are acquired through simulation training in three weeks."
"Production line yield improved from 95% to 99.3%, and new employee training costs dropped by 70%."
"The light of Japanese manufacturing didn't go out after all," Mr. Sato says with a sigh of relief.
A. Ace is strictly "a research prototype" and there are currently no plans to sell it commercially.
Like a space rocket or an F1 car, it is a proof-of-concept machine not intended for general sale. However, the high-speed vision technology and reinforcement learning algorithms developed through Ace are expected to be progressively applied to Sony's automotive cameras, industrial equipment, and consumer products.
"You won't be able to buy Ace itself, but Ace's descendants will increasingly surround you in everyday life."
A. There is absolutely no reason to retire. After chess and shogi professionals lost to AI, the player base did not shrink — in some cases, it actually grew.
"AI will serve as a practice partner and analysis tool," accelerating the growth of the table tennis world.
"Just as F1 drivers practice on simulators, table tennis players will train with Ace" — that's all that changes. Human competitions will continue to be popular.
A. In terms of pure win/loss performance, Ace — designed to beat professionals — is far stronger. However, FORPHEUS is an educational and exhibition robot built to "entertain and teach humans," so they're competing on entirely different terms.
Comparing them is like "having a professional boxer fight a PE teacher — pointless." Rather than comparing them, the right understanding is that "each excels in its own domain."
A. The IMX636 is an image sensor provided for industrial and research use, and individuals can purchase it through industrial camera manufacturers.
Companies such as "IDS Imaging" sell EVS cameras with the IMX636 for around 500,000 yen.
It's in the same position as "a professional-grade high-end camera" — specialized equipment for robot developers and researchers. Integration into consumer electronics is expected to expand over the next few years.
A. This result demonstrates that "while the U.S. dominates language AI (ChatGPT, etc.), Japanese companies still have a fighting chance in physical AI." Sony, Toyota, Fanuc, Kawasaki Heavy Industries — Japan has long been a top-tier player in robotics and sensors.
If "AI that operates in the physical world is the next frontier," it aligns perfectly with Japan's strengths.
The combination of "Japan's forte in manufacturing and robotics + AI" offers a path to reclaiming global leadership in the second round.
"AI leaping off the screen and beating a professional in a real-world sport" — ten years ago, anyone would have called this science fiction.
"A table tennis robot beating national championship-level players," "a self-driving car detecting danger faster than a human," "a factory robot mastering a veteran craftsperson's technique in an instant" — these futures lie squarely on the trajectory of the technology Ace has proven.
"ChatGPT changed the way we work" — next, "physical AI will change the physical world."
2026 will be remembered as "the first year AI stood shoulder to shoulder with humans in the physical world."
Start by watching Sony AI's official Ace match footage and seeing for yourself "the moment a robot beats a professional."
That is the fastest route to not missing the massive industrial transformation of the next decade.
This article is a cross-post from AI Friends.
