Google DeepMind is now venturing into the world of robotics and has shown a robot that plays ping pong like an amateur player. In a video shared by DeepMind, these robots can beat human players.
It doesn't just hit the ball back to the human opponent, but can perform more difficult shots such as backhands, spin returns and serve returns. According to DeepMind at the initial stage artificial intelligence (AI) is trained to play ping pong in the virtual world with real world physics players.
Then the robot was instructed to compete with 26 human players with four different levels of expertise. The data taken is then used to teach the robot to further improve its ability to play until it is now at the level of an amateur level player. It is also taught to learn how human opponents play and change game strategies to defeat them.
It is still not able to reach the level of professional players because humans are more agile, faster and the physics of the rubber surface on the ping pong pedal is not yet able to be produced in the virtual world. So at this point humans still have an advantage.