PREVIEW

Scientists at the ETH Zurich in Switzerland university have created an AI robot, CyberRunner, that is capable of guiding a marble through a labyrinth in record time. The labyrinth marble game is traditionally seen as a domain of human expertise as it requires subtle physical skills and intuition for a person rather than the cold calculations of chess or the strategy game Go. However, even this task has now been conquered by AI using deep reinforcement learning. Remarkably, the machine cost under $200 - raising hopes of the research being widely replicable for those looking to build on it. The CyberRunner learns through experience, capturing observations and receiving rewards based on its game performance. It employs two knobs to control the orientation of the labyrinth board, demonstrating fine motor skills and spatial reasoning. Remarkably, CyberRunner completed the learning process on the real-world labyrinth in just 6.06 hours, outperforming the fastest recorded time by a skilled human player. During the learning phase, the AI robot did something even more interesting - it discovered shortcuts, prompting researchers to intervene and instruct it not to cheat by exploiting these. Lead researcher Prof. Raffaello D’Andrea, says: “Now, for less than $200, anyone can engage in cutting-edge AI research, as applied to the physical world”. The research paper is available on the project's website, www.CyberRunner.ai.