DeepMind’s Robot Beats Humans in Half of Table Tennis Matches
Researchers at Google DeepMind have created the world’s first robotic table tennis player capable of playing at an amateur level. The system combines an industrial robotic arm, the ABB IRB 1100, with specialized software from DeepMind.
In a recently published paper titled “Achieving Human-Level Performance in a Competitive Table Tennis Robot,” the Google DeepMind Robotics team unveiled a robot “capable of competing in sports with humans at human levels, and represents a milestone in robot learning and control.”
During testing, the robot was able to defeat all beginner-level players it encountered. Against intermediate players, the robot won 55% of matches. However, against professionals, the robot lost all matches. Overall, the system won 45% of the 29 matches it played. The system’s biggest drawback is its delayed response to fast balls. The robot also has difficulty playing the backhand, receiving high and low balls, and judging the spin of the ball.
The system is not yet perfect. It has difficulty handling fast or high balls, as well as those with a lot of spin. It also has problems with backhands. For example, a video from DeepMind shows the AI failing to keep up with an opponent’s fast hit. The researchers emphasize that with further development, the system will be able to compete with professional table tennis players.
