Israeli developers were able to teach the AI to defeat a human in Mortal Kombat Bashny.Net

The 3D environment is still difficult for the perception of a weak form of AI, which leads to computer problems with the passage of these games

With the help of modern video game AI is going to teach AI techniques to overcome obstacles and solve problems on the go. For example, employees of DeepMind in conjunction with Blizzard turned StarCraft II into the learning environment of weak forms of AI. Last year, the artificial intelligence system Google alone has mastered 49 old Atari games.

And it's not about the system, integrated into the game (like AI opponents in fighting games, soccer simulators or racing simulators), which are well known terms and conditions. AI, which taught the developers of computer games are now delivered in equal terms with the human condition. The system observes the image on the screen, learning methods of trial and error. And this program is able to find a solution not only in games, it is suitable to search for solutions in the broadest range of tasks, regardless of the rules or conditions.


A group of students from the Technological University of Israel recently announced its development, the system Retro Learning Environment (RLE). Is a software platform that allows you to train the AI for many games of the 90-ies, including those that came out for consoles Nintendo and Sega. This, for example, many well-known F-Zero, Wolfenstein, and Mortal Kombat. According to the developers, the AI for many of the games proved to be difficult, some system have not learned how to understand and pass. But RLE perfectly learned how to play Mortal Kombat. The results of their work, the experts stated in the article on arXiv. AI has repeatedly been able to outright defeat the enemy-man. And this enemy was not new. The article stated that the computer was opposed by an experienced player in Mortal Kombat.

In Wolfenstein, where volume levels, plus you need to navigate the maze and determine the number of objects, the system showed not very good result. In Gradius III RLE were able to examine technical aspects of the game, which include the need to destroy the enemies encountered in follow-up. But the system could not show a better result than a human player. Here it is necessary to improve the abilities of the character found artifacts. The more artifacts a player passes — the harder it is to pass the game. Computer paid little attention to the power-up objects, which greatly complicated the process of passing.

The fact that the program was able to learn how to play a computer game so well that was to win the man — an undoubted merit of the developer. For the computer to learn through the game by trial and error not so simple, this is a complex task, which overcome a few of the software platform. "If algorithms can play complicated games, we can start working on the implementation of such systems in the real world to solve real problems," said Shai Rosenberg, one of the authors of the study. "Just as a child learns to play games, the computer sees only the information on the screen. They (the child and the computer) learn to avoid obstacles and solve problems to obtain maximum reward," he continues.



AI learned to play the Boxing on the Atari, and in Mortal Kombat, just "looking at the screen" and evaluating the consequences of their actions in the game environment

In the real world, the ability of computer systems to learn from their mistakes and predict the consequences of certain "actions" can be useful in many areas. The robots can navigate difficult spaces (the corridors of the premises, for example) with lots of obstacles without colliding with them. Any small error made by the computer, they will be considered next time, while performing the same or similar tasks.

According to Rosenberg, RLE can learn to take more complex game system, and not just to play SNES games. The next stage of the project will be the development of gaming platform PlayStation. However, while Israeli developers are focused on how to teach your system to pass a large part of developing games. What the computer learned to play Mortal Kombat this is good, but not enough most games "left behind", RLE are unable to learn them.



The results through different games system RLE using a variety of algorithms for passing

"In subsequent phases, we believe it is possible and even relatively easy to adapt our training systems to more complex games, such as Grand Theft Auto," — said the developers. Now, unfortunately, games like Grand Theft Auto V AI unavailable — they are too complicated.

The source code for the system developers made open and posted on Github. To obtain the source code here.

Source: geektimes.ru/post/282560/