Indeed: until now. The Swiss created a robot called CyberRunner to play marbles. It’s all about dexterity: players must move the marble all the way without it falling into strategically placed holes.
After a six-hour training session, the CyberRunner was able to beat the fastest, most “highly skilled” human player, with a 6 percent faster time. The AI method used is known as reinforcement learning. Unlike deep learning, such a system does not learn on the basis of large amounts of data, but on the basis of experiences in the environment.
“Just like us humans, robots learn through experience,” the researchers say. As you play, the system receives rewards based on its performance, all through the eyes of a camera overlooking the maze. The collected experiences are recorded in memory. Ultimately, the model learns which strategies work best. The video shows that the first attempts are still shaky and clumsy, but the wheels on the sides are later operated in such a way that the marble moves smoothly along the track.
Relevant research has run out Arxiv preprint server published.
“Thinker. Coffeeaholic. Award-winning gamer. Web trailblazer. Pop culture scholar. Beer guru. Food specialist.”