7/26/2025–|Last update: 14:36 (Mecca time)
In the summer of 2010, tennis John Isner and Nicolas played one of the most drained confrontations in the history of Wimbledon, the match lasted 11 hours over a period of 3 days. After more than a decade, two opponents of the last type of match are playing no less stubborn, but this time in Google’s Deep Mind laboratories, and without an audience.
According to a report of the Popular Science website, two robotine arms are moving in an endless table tennis match at the Research Center South London, as part of a project launched by Deeb Mind in 2022 to develop the capabilities of artificial intelligence through continuous self -competition. The goal is not limited to improving playing skills, but rather to training algorithms capable of adapting to complex environments, such as those facing robots in factories or homes.
Whoever is a winner without a winner to non -stop training
At the beginning of the project, the exercise was limited to simple reciprocal strikes between robots, without seeking points. Over time, using augmented learning techniques, every robot learns from its opponent and developed its strategies.
When the score goal was added, the regime faced difficulty in adapting, as the arms were losing some of the movements that we had previously mastered. But when facing human players, there began to appear remarkable signs, thanks to the diversity of playing methods that provided more learning opportunities.
According to the researchers, robots won 45% of 29 games against human beings, and surpassed medium players by 55%. The total performance is classified as an amateur player, but it is more complicated with time, especially with the introduction of new techniques to monitor and improve performance.
When the video teaches artificial intelligence
Improvements did not depend on the actual exercise, as the researchers used the Gemini model for the vision and language from Google to generate notes from the matches.
The robot can now adjust its behavior based on text orders, such as “hit the ball to the far right” or “close the network”. This linguistic visual feedback enhances robot’s capabilities to make accurate decisions during play.
Table tennis gateway for future robots
Table tennis game is an ideal environment for artificial intelligence test, because of the balance between speed, accuracy and decision -making. It allows robotics training on skills that go beyond mere movement, to include real -time analysis and response, which are necessary skills for future robots in realistic environments.
Although advanced robots still stumble in simple tasks for humans, such as linking the shoe or writing, the recent developments – as a success of Deep Mind in teaching a robot linking the shoe, or the new “Atlas” model presented by Boston Dynamix – indicates a gradual rapprochement between the performance of the machine and the human.
Towards adaptable general intelligence
Deep Mind experts believe that this approach to learning, based on competition and self -improvement, may be the key to developing a multi -use general artificial intelligence. The ultimate goal is to enable robots to perform various tasks, not only in industrial environments but also in daily life, in a natural and safe manner.
Until then, Deep Mind’s arm will remain in an open match, exchanging balls and skills, on a long way towards a more intelligent and flexible roboty future.
