Google just trained a robot to beat humans at table tennis

In the past year, AI has gone from rewriting our texts on our smartphones to creating music, planning vacations and so much more, and it still feels like we have barely scratched the surface with artificial intelligence. Now, Google’s latest showcase in AI training proves this, by teaching a robot how to play table tennis, well enough to actually beat a human at it!

The experiment comes from Google’s DeepMind wing, where researchers managed to train a robotic arm with a 3D-printed paddle to not just play table tennis but get better at it. As per a report published in an Arxiv paper, the robot managed to win 13 out of 29 games against human opponents of varying abilities.

ALSO READ: 10 ways AI is being used in laptops in 2024

Of course, different playing levels is something that can change the results here. Most of the players the robot beat were amateurs, while advanced players were still able to best the robotic arm.

Still a long way from beating professionals

The robot was trained to account for a lot of parameters, including direction and strike speed. It could use this to accurately plan and execute its return shot.

However, some aspects like the spin of a table tennis ball could not be accounted for. As a result, advanced players who can spin the ball were able to easily take advantage of it.

ALSO READ: I tried Claude AI, and it’s now my go-to chatbot app

Senior staff software engineer at Google DeepMind, Pannag Sanketti said that the team “projected that realistically the robot may not be able to win against people it had not played before.”

“The system certainly exceeded our expectations,” Sanketi added. While the robot doesn’t make robots better than humans at table tennis, it is a step in the right direction for the advancement of machine learning.