AI-powered robot can defeat elite table tennis players
Sony AI's Ace robot is capable of defeating elite human table tennis players.
An AI-powered robot is able to beat elite players at table tennis.
The machine, called Ace, has been built by experts at Sony AI - who feel that it is on a par with achievements such as IBM's Deep Blue beating chess grandmaster Garry Kasparov in 1997.
AI has long excelled at board games but the physical demands of table tennis - such as rapid ball speed and wicked spin - have been seen as particularly demanding for the technology to master.
Peter Stone, chief scientist at Sony AI, said: "This breakthrough is much bigger than table tennis. 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."
At the highest level of table tennis, the ball can travel at over 45mph, with players having less than half a second between shots.
To keep pace, Ace relies on nine high-speed cameras tracking where the ball is and three extra "event-based" camera systems that focus on how it spins.
The bot's special sensors zoom in on the ball and track markings on its surface, such as the logo.
Rather than taking normal pictures, Ace records tiny changes in light as they happen - allowing it to keep up with the ball even at extremely high speed.
By tracking how the markings move, the robot can work out the speed and direction of the ball's spin.
The information is then sent to a computer, which has learned how to play table tennis via access to a virtual version for thousands of hours.
The shot is carried out by an eight-joint robotic arm that is designed to move with the speed and reach of a human player.
In matches played under official table tennis rules, Ace defeated three out of five elite amateur players and returned over 75 per cent of high-spin shots.
The bot was also able to adapt to unexpected situations, such as the ball deflecting off the net.
Stone said: "It represents a landmark moment in AI research."
