Chinese researchers have introduced a new approach to robot training, thanks to which a humanoid was able to master the game of tennis. According to New Atlas, the development could be a breakthrough in the field of machine learning and applied artificial intelligence.
Instead of complex motion analysis systems that require a large number of cameras and accurate data, scientists have created a method called LATENT. He works even with incomplete information, focusing only on the basic elements of technique, such as forehand and backhand, as well as moving around the court.
Only five hours of recordings of athletes' movements were used for training. Based on them, a set of basic "motor skills" was formed, which were then loaded into the Unitree G1 humanoid robot.
After that, the robot independently learned to adapt its movements to real conditions.: determine the trajectory of the ball, select the moment of impact and adjust the position of the racket. Most of the training took place in simulation, which significantly accelerated the process.
As a result, the robot has demonstrated impressive efficiency: about 90% of successful punches from the right and almost 80% from the left. At the same time, his movements look natural and smooth, approaching the human way of playing.
Despite the fact that the robot is not yet ready for full-fledged matches with humans, the researchers consider this achievement to be an important step forward. The technology can be used not only in sports, but also in more practical areas where rapid adaptation to complex and dynamic conditions is required. korrespondent.net.
In addition, LATENT software has already been published in the public domain, which will allow other developers to use and improve this technology.
