How Sony's AI Robot Mastered Table Tennis to Beat Elite Human Players

For decades, artificial intelligence has dominated the digital realm. AI systems have routinely crushed human grandmasters in chess, outmaneuvered world champions in Go, and executed flawless laps in virtual racing simulators. Yet, when tasked with the physical world—where gravity, friction, and unpredictable human opponents dictate the rules—these same supercomputers have historically struggled to perform tasks as simple as catching a tossed ball.[1]

That boundary between virtual mastery and physical clumsiness has just been shattered. In a landmark paper published in the journal Nature, researchers from Sony AI unveiled "Ace," an autonomous robotic system that has successfully defeated elite and professional human table tennis players under official competitive rules.[1]

The achievement represents a holy grail for the field of "physical AI." Table tennis has long been considered one of the ultimate benchmarks for robotic agility. The sport demands split-second perception, complex aerodynamic calculations, and the physical dexterity to execute high-speed motor commands—all within a fraction of a second.[2]

"This research has shown that an autonomous robot can, in fact, win at a competitive sport, matching or exceeding the reaction time and decision making of humans in a physical space," said Peter Dürr, the director of Sony AI in Zurich and the project lead for Ace.[2][3]

To prove the system's capabilities, Sony AI subjected Ace to a grueling gauntlet of matches against highly ranked human opponents. In an initial testing phase in April 2025, Ace faced five "elite" players—defined as athletes with over a decade of experience who train an average of 20 hours per week. Ace won three out of those five matches.

During that same initial phase, the robot also squared off against two active professionals from Japan's premier table tennis league, Minami Ando and Kakeru Sone. While Ace ultimately lost both of those professional matches, it managed to win a single game, proving it could hold its own at the highest echelons of the sport.

But the system did not stop learning. Because Ace's underlying architecture is designed to continuously adapt, its performance improved dramatically over the following months. In subsequent matches held in December 2025 and March 2026, Ace successfully defeated professional players, including Miyuu Kihara, a top-25 player in the World Table Tennis women's singles rankings.

How does a machine track a sphere moving at 20 meters per second while spinning 160 times per second? The answer lies in a bespoke, multi-layered perception system that effectively allows Ace to see the world faster than the human eye.[2]

Ace's "vision" relies on a network of nine active pixel sensor (APS) cameras distributed around the court. These cameras track the ball's three-dimensional position at 200 hertz, operating with a latency of just 10 milliseconds.[2]

However, tracking position alone is insufficient for table tennis; understanding spin is the true differentiator between amateur and professional play. To solve this, Sony engineers integrated three event-based vision sensor (EVS) cameras equipped with pan-tilt mirrors and telephoto lenses. These specialized sensors measure the ball's angular velocity and spin at an astonishing 700 hertz, capturing aerodynamic data that would be a literal blur to a human opponent.[2]

All of this sensory data feeds into Ace's "brain," which is powered by model-free deep reinforcement learning. Rather than relying on hard-coded physics equations to calculate trajectories, the AI was trained entirely in a virtual simulation. It played millions of matches against itself, learning through trial and error how different spins, angles, and speeds interact.[3]

The true breakthrough was "sim-to-real transfer"—the ability to take an AI trained in a flawless digital simulation and deploy it seamlessly into the messy, noisy reality of a physical ping-pong table. Ace's algorithms proved robust enough to handle real-world anomalies, such as balls clipping the top of the net or taking unpredictable bounces off the table's edge.[2]

To execute its physical returns, Ace utilizes a custom-built robotic arm featuring eight degrees of freedom. Mounted on a linear track that allows it to slide side-to-side, the arm is constructed from optimized lightweight alloys. This design grants the robot human-level reach and the acceleration required to return balls at linear velocities of up to 19.6 meters per second.

Interestingly, the matches highlighted a unique psychological challenge for the human athletes. In professional table tennis, players constantly read their opponent's body language, eye movements, and subtle shifts in posture to anticipate the next shot.[1]

Ace offers none of these tells. "The players want to see the eyes of their opponent," Dürr noted. "And the eyes of Ace are all around the court and they don't show any intention or feeling." Furthermore, the robot is entirely immune to the psychological pressure of a tied game or a match point, executing its algorithms with cold, unyielding consistency.[1]

While the spectacle of a robot trading topspin rallies with Olympians is captivating, the underlying technology was never just about winning a paddle sport. Table tennis served as a highly constrained, extreme-stress test for physical AI.[3]

"This breakthrough is much bigger than table tennis," explained Peter Stone, Chief Scientist at Sony AI. "Once AI can operate at an expert human level under these conditions, it opens the door to an entirely new class of real-world applications that were previously out of reach."[3]

The ability to perceive a dynamic environment, process that data in milliseconds, and execute a precise physical action has profound implications. The same architecture that allows Ace to return a 450-radian-per-second backspin could eventually power surgical robots that react instantly to patient movements, or autonomous manufacturing systems that safely handle unpredictable materials alongside human workers.

For now, Ace remains a research platform rather than a commercial product. But its victories on the table tennis court mark a definitive turning point. The era of AI being confined to screens and servers is ending; the machines are finally learning how to move in our world.[1]