Human beings have a curious relationship with machines: we create them to help us, but also to challenge us. We have been doing it for decades, from large industrial systems to artificial intelligence systems and robots that today are beginning to move in more complex, more demanding environments and with less margin for error. And when those machines surpass us, we don’t just see a defeat: we see a clue as to where the technology is going. It already happened in chess and Go. What we are seeing now points to something different: the challenge begins to jump to sports where it is not enough to calculate the next play.
The robot that plays ping pong. The last signal comes from Sony AI and is shaped like a ping pong table. Its Ace robot, developed within Project Ace, has been presented by the company as the first AI system capable of competing in a real physical environment with elite university players and table tennis professionals under official rules. The firm illustrates this with a recent scene in Tokyo: Japanese professional player Taira Mayuka launched a shot that, under normal conditions, would have decided the point. On the other side of the net, Ace read the trajectory, adjusted the angle of the paddle and returned the ball to keep the exchange alive.
A notable jump. Ping pong adds something much less friendly than table games: a ball that moves, spins, bounces and changes direction in a very short time. That’s why Sony insists on Ace’s reaction speed, with an end-to-end latency of 20.2 milliseconds compared to about 230 milliseconds in elite human players. As we can see in the video that accompanies this article, the robot not only has to “see” the ball. You have to anticipate what he will do next and get the paddle at the right angle before it’s too late.
How do you get it? The key is that Ace does not depend on a single technology, but on a very tight chain between perception, control and movement. The system integrates nine synchronized conventional cameras and three event-based vision systems, capable of recording movement changes very quickly. With that set, the robot tracks the ball at 200 Hz with millimeter precision and measures the effect up to 700 Hz. An eight-degree-of-freedom robotic arm then executes the returns based on policies learned through reinforcement learning in simulation.
Ace didn’t get to that point overnight either. Sony places the start of the project in 2020, within the first works of Sony AI, and describes an evolution in stages: first juggling the ball, then maintaining cooperative exchanges with a person and, later, facing increasingly stronger players. This journey also served to discover limits that do not always appear in a simulation.
The limits. Ace’s merit lies in having reached an expert level, not in having turned table tennis into a solved problem. Sony recognizes that there are still humans above the system. In any case, the robot mainly excels in skill, where you decide how to move the robot and how to hit the ball in real time. What happens point to point, and what is planned during a match, can still improve.
Images | Sony AI (1, 2)
In | A young man has solved a mathematical problem that lasted 60 years in 80 minutes with ChatGPT. That’s the least interesting thing about the story.
