DeepMind, a subsidiary of Google, taught artificial intelligence-powered virtual players to play football from the very beginning. The gradual transition of artificial intelligence robots, which we have known for chess until now, to other sports branches can be considered as a sign of the development in this field.
Simulated players who initially undergo training individually; He gained skills such as running, dribbling and hitting the ball. Then, the virtual football players, who learned the moves to be done when the opponent came up against them, started team game training this time as their experience increased. Finally, the research team succeeded in creating a virtual football environment consisting of teams of 2 people.
The most curious element may be from whom artificial intelligence-supported simulations learned football techniques: Of course, from real football players.
Engineers have been working diligently for several years to develop physical robots that can play football. Research has also created a competitive environment among various groups, as it has given rise to the ambition of who can design the best robot players. The interesting part of the subject is; This competition we mentioned even led to the establishment of the league called RoboCup, in which both real and virtual robots compete fiercely. The importance of the work carried out by Google DeepMind is hidden in the fact that it carries the progress in this field to a completely different point.
All the players you see in the video MuJoCo Thanks to an advanced physics engine called, it has the ability to make any movement desired. The main focus is to leave the control of this engine to artificial intelligence that knows how to play football. Considering that DeepMind trained an artificial intelligence for nuclear fusion control before that, it may not be difficult to teach football.
The researchers completed the process of teaching artificial intelligence how people play football by watching it as a one-on-one video in a software environment, just like we do. The volatile situation here was that the training was taken almost from scratch. Because the artificial intelligence learned to walk, run and then hit the ball for the first three days. In proportion to the increase in skill, the AI was constantly shown videos of real-world football players. In this process, the system not only learned the basics of football, but also began to imitate the original movements of professional athletes.
Virtual players went into training alone after learning some basics. In the new stage, one-on-one match was started. The Google DeepMind team, which increased the number of players as the skills developed, achieved small teams playing two-on-two against each other at the end of the 50th day. At this point, artificial intelligence; In addition to individual skills such as dribbling, passing and blocking, he had a grasp of coordination and his ability to adapt to the team greatly improved. In addition, it was determined that the reward system integrated into the simulation every time a goal is scored contributes positively to the learning curve.
The results obtained by Google DeepMind are truly impressive. Although the study resembles a standard computer game, the distinction is that artificial intelligence performs a continuous learning action in the process and puts it into practice. After all, all the effort was to show such accumulation.
On the other hand, in the team’s article Some shortcomings were also mentioned. For example; For now, there is no foul and no off and balls are prevented from being thrown around the field, because virtual players need quite a long time frame to comply with the football rules. The biggest disadvantage of this long period, which corresponds to years, is that it prevents the achievements of simulation to be brought to real robots. So what do you think about the subject? Could we face bigger robot leagues in the future? Do not forget to share your thoughts in the comments.
Source: Technopat by www.technopat.net.
*The article has been translated based on the content of Technopat by www.technopat.net. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!
*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.
*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!