Introduction
In a milestone achievement for artificial intelligence and robotics, Sony AI has successfully developed the world's first robot capable of defeating top-tier human players in table tennis. The robot, named Ace, recently graced the cover of Nature, where the team published a comprehensive study detailing its development and real-world performance.
In April 2025, Ace played a series of matches against elite and professional players, winning roughly half of its games. When the experiment was repeated in December, Ace's performance had improved so dramatically that it defeated nearly all challengers, including an athlete from the Japanese professional league. The matches were strictly officiated by licensed arbiters from the Japan Table Tennis Association. Impressively, Ace had no prior knowledge of its opponents' playstyles, reacting entirely on the fly.
The Need for Speed and Precision
Table tennis is an incredibly fast-paced sport, making it notoriously difficult for robotic systems. The high velocity of the game—where balls can travel at speeds up to 150 km/h—and the unpredictable trajectories caused by intense spin create a monumental challenge. Every single shot must be perceived, processed, and returned within milliseconds.
Historically, this level of speed and precision was considered a stumbling block for robotics. However, Sony engineers overcame this by achieving an astonishing end-to-end latency of just 20.2 milliseconds. To put this into perspective, elite human table tennis players have an average reaction time of approximately 230 milliseconds. Ace reacts roughly 11 times faster than the best human athletes.
How Ace Achieves Expert Performance
Sony's success with Ace is the result of three critical breakthroughs spanning hardware, mechanics, and advanced artificial intelligence:
1. Ultra-Precise Perception Systems
Ace relies on high-speed cameras and sensors that track the ball at a frequency of 200 Hz with millimeter-level accuracy. Crucially, the system doesn't just track the ball's velocity and trajectory; it also monitors the logo printed on the ball. This allows the AI to accurately deduce the ball's rotation, a vital factor since spin dictates the flow of table tennis.
2. Optimized Hardware Mechanics
The physical design of the robot was honed to the smallest detail. Constructed using optimized lightweight alloys and featuring an 8-joint mechanical arm, Ace possesses the physical agility necessary to execute lightning-fast returns and dynamic swings without structural failure.
3. Simulation-Based Reinforcement Learning
Ace was trained entirely in a simulated environment using a three-tiered Reinforcement Learning (RL) approach, mimicking how humans learn the sport:
- Level 1: Learning the fundamental strokes.
- Level 2: Mastering tactics (deciding how, where, and with what force to hit the ball).
- Level 3: Formulating match strategy (structuring gameplay across multiple rallies).
To perfect its predictions, Sony employed the "privileged critic" (or physics distillation) approach, the same technique used to train their champion AI in Gran Turismo. In the simulation, a "teacher" model has perfect access to all physical data about the ball. The "student" model—which only receives the limited data a camera would see in the real world—learns to predict the ball's trajectory by attempting to match the teacher's perfect understanding.
The "Move 37" of the Physical World
The sheer capability of Ace has left professionals astounded. Witnessing one of Ace's unique shots, ex-Olympian and table tennis expert Kinjiro Nakamura remarked:
"No one else could have done that. I didn't think it was possible. But since it turned out to be possible—then there is a probability that a human could do it too."
This achievement draws parallels to AlphaGo's famous "Move 37," a moment where an AI demonstrated profound, creative insight that transcended human understanding of the game. However, unlike AlphaGo's triumph on a digital board, Ace's victory takes place in the real, physical world.
Conclusion
Sony AI's Ace represents the first time in history that an artificial intelligence system has reached the level of a human expert in an active, physically demanding sport. By combining ultra-low-latency hardware with sophisticated simulation-based reinforcement learning, Sony has shattered previous limitations in robotics. The development of Ace not only revolutionizes our understanding of sports AI but also paves the way for future physical robots capable of hyper-fast, dynamic interactions in unpredictable real-world environments.