The artificial intelligence called CyberRunner was developed by Swiss engineers. Her accuracy increased dramatically during training, and eventually she became invincible in a game that is not as demanding as in chess – you also have to understand the laws of physics.

The first wooden game called Maze was published nearly six decades ago. The gameplay is very simple: you have to move a metal ball from the start to the finish line by moving the board so that it does not fall into the holes. If the ball falls into the hole, the game must be restarted from the beginning.

Although the rules are simple, it's actually a difficult game: in addition to a lot of patience, you need fine motor control, manual dexterity, and real-time problem-solving skills.

See also  Nvidia DLSS 3.5 debuted this year, and it's becoming clear how it improves the gaming experience

CyberRunner, an artificial intelligence, has learned how to play this game, reports A TakradarWhich requires only six hours to master the rules and understand how to move the board to achieve success.

And then let's savor the above sentence a little: an AI that knows nothing at all, starting from scratch, needs six hours

Correctly control the board in real space, understanding the physical laws necessary to achieve a successful outcome of the game.

Engineers from the Zurich University of Technology solved the problem with the help of motors so that the robot could rotate the panel, on top of which they installed a room. The computer driving the engines was running CyberRunner, which not only learned the rules, but also ran the game the entire time.

The AI ​​uses model-based reinforcement learning technology, so it was able to make an informed decision to succeed. What's really impressive is that each attempt at CyberRunner has been better than the last. This is simply not better:

The success rate has increased dramatically.

The AI's best track time was 14.8 seconds. It may take several minutes to get from the start to the finish line with the same board.

Interestingly, cheating was detected in CyberRunner at times, and the AI ​​was so involved that it found mouse paths and shortened the path. The engineers discovered this and reprogrammed him, so that he could then only complete the task by following the maze.

If you want to know about similar things at other times, like it HVG Tech Department Facebook page.

See also  2022 phone with 3.5 jack input? Yes!