Images released by Sony show table tennis robot 'Ace' competing against a professional player during project tests, using a network of cameras and AI, to perform complex, real-time interactive tasks with accurate predictions of the ball's trajectory on return.

Researchers say the robot learned through reinforcement learning rather than fixed programming, allowing it to adapt during rallies and serve under official rules.

Tested on a full-size court in Tokyo and described in Nature, Ace combines speed with unusual perception.

It can read the logo on the ball to gauge spin and react within split seconds. While its physical reach and pace are kept close to human levels, its vision and consistency stand out.

The study suggests AI trained in virtual settings can now transfer into fast, physical tasks, with potential uses beyond sport.