Did you know that a robot played table tennis against twenty-nine people? DeepMind's study published in two thousand twenty-four describes thirteen games won. It is not the same robot as Sony's presented elsewhere, nor a result obtained with these opponents. The robot must follow the ball and choose a gesture fast enough to return after its rebound. The racket remains attached to the arm and the base to the ground. Our exchange illustrates this mechanism, without pretending to reproduce a point actually played during the evaluation of this team.
Thirteen wins out of twenty-nine correspond to about forty-five percent of the matches. Performance changes depending on the level of the players. It is therefore not necessary to tell a machine that beats every professional, nor to confuse a spectacular demonstration with a success in all possible situations.
Real progress is to make perception and control work in a physical sport, with humans. The result keeps its limits. A failed ball does not cancel the search, and a successful ball does not establish a level of world champion on its own.
What we know
The 2024 DeepMind robot won 13 of the 29 games reported, about 44.8%. It is distinct from Ace and performance varies by player.
To go to the source
References consulted on October 12, 2026.
Signaler une erreur ou apporter une précision
