A robot built by Japanese electronics company Sony has done something remarkable: beaten professional table tennis players at their own game.
The robot, called Ace, is a robotic arm with eight joints that let it swing a paddle, position shots, and react quickly to incoming balls. It also has nine cameras positioned around the court to track the ball, including its spin.
Ace learned to play using a method called reinforcement learning, where it practices over and over until it figures out what works. Sony researcher Peter Dürr explained that you simply cannot program a robot by hand to play table tennis. It has to learn from experience, just like humans do.
Sony says this is the first time a robot has reached expert human level in a commonly played competitive sport in the real world, which researchers consider a major milestone for both AI and robotics. To keep things fair, the team designed Ace to play within human limits. Its speed and reach were kept similar to a skilled athlete who trains at least 20 hours a week, and it follows official table tennis rules. The goal was not to build a machine that just smashes the ball faster than any human could return, but one that wins through smart tactics and skill.
Competing against Japanese professional players, Ace defeated all but one of four high-level players in recent trials. One former Olympian watching Ace play a particular shot said he did not think it was possible for anyone to pull off, but seeing the robot do it made him wonder if a human could learn it too.
