A robot that plays table tennis like a pro

Sony AI specialists unveiled the Ace robotic arm. It plays table tennis on par with high-level players. The system is trained according to the official rules of the International Table Tennis Federation.

The robot sees the ball with nine cameras, reads its spin in real time, and adapts to the opponent's style during play.

What's striking is that there are no predetermined scenarios. Ace learns on the fly, analyzing the trajectory and choosing a counter. In tests, it won three out of five matches against "elite players" (with over 10 years of experience) and was even able to score points against professionals (playing in officially recognized professional leagues).

This is how Sony is developing technologies for the real world, where robots need to think and act quickly.

The robot's training process was fundamentally different from classical programming. It was carried out using an artificial intelligence method known as reinforcement learning.

Reinforcement learning is a method of machine learning in which a system learns through trial and error. It interacts with the environment, learning along the way, and receives rewards for performing actions. Over time, the system learns to avoid negative actions and performs only positive ones.