CyberRunner AI is a robot designed to take advantage of model-based reinforcement learning, and now it has proven that AI can even be trained to overcome physical skill games. The creators of the AI tasked it with learning how to play and complete the widely popular labyrinth marble game, and it succeeded with a fastest time of 14.48 seconds.
Of course, this isn’t the first time that we’ve seen AI beat humans in something. The fact that AI can defeat humans in games like chess isn’t as impressive as it once was because it doesn’t feel all that surprising. However, by beating humans in this labyrinth game, the AI has proven that it can also best humans in a game of physical skill.
To complete the labyrinth maze, CyberRunner AI had to control two knobs, which allow the user to change the orientation of the board, letting the marble game piece move across it. There are walls and holes in the way, and the user needs to figure out how to balance the marble so that it doesn’t fall into the holes on its way.
Watching the video, it was really impressive to see how the AI learned the best way to complete the puzzle. The researchers behind it even mentioned that, at one point, it found ways to hop over the walls and skirt around holes, but they went back into the programming and changed it so that it wouldn’t do that.
The programming allowed the robot to continuously learn more about the game as it played, allowing it to complete the run in just 14.48 seconds. A paper on these findings and how CyberRunner AI was designed to complete the puzzle is currently available on the researcher’s website. Additionally, the YouTube video above showcases how the robot learned to play the game and then showcases how it completed it at such a quick pace.