Claim: reinforcement learning AI beats humans in physical domain, playing the “labyrinth” game.
In six hours of training, it solves faster than any human.
Video is fun to watch.
We have created an AI robot named CyberRunner whose task is to learn how to play the popular and widely accessible labyrinth marble game. The labyrinth is a game of physical skill whose goal is to steer a marble from a given start point to the end point. In doing so, the player must prevent the ball from falling into any of the holes that are present on the labyrinth board.
CyberRunner applies recent advances in model-based reinforcement learning to the physical world and exploits its ability to make informed decisions about potentially successful behaviors by planning real-world decisions and actions into the future.
The learning on the real-world labyrinth is conducted in 6.06 hours, comprising 1.2 million time steps at a control rate of 55 samples per second. The AI robot outperforms the previously fastest recorded time, achieved by an extremely skilled human player, by over 6%.
Interestingly, during the learning process, CyberRunner naturally discovered shortcuts. It found ways to ’cheat’ by skipping certain parts of the maze. We had to step in and explicitly instruct it not to take any of those shortcuts.
A preprint of the research paper is available on the project website, https://www.cyberrunner.ai/. We believe that this is the ideal testbed for research in real-world machine learning and AI. Prior to CyberRunner, only organizations with large budgets and custom-made experimental infrastructure could perform research in this area. Now, for less than 200 dollars, anyone can engage in cutting-edge AI research on the physical world. Furthermore, once thousands of CyberRunners are out in the real-world, it will be possible to engage in large-scale experiments, where learning happens in parallel, on a global scale.
0:00 Intro.
0:25 Humans playing the game.
0:36 Meet CyberRunner.
1:08 CyberRunner playing the game.
1:27 Cheating.
1:49 CyberRunner vs humans.
2:06 Credits.
Researchers:
Thomas Bi.
Raffaello D’Andrea.
Comments are closed.