AI solves Rubik's cube in a second

It finds the solution in a fraction of a second, without any specific domain knowledge or in-game coaching from humans

An AI system developed by researchers at the University of California has managed to solve the Rubik's Cube in just over a second. And while, humans able to solve the unique puzzle quickly generally take about 50 moves, the deep reinforcement learning algorithm solved it in an average of 28 moves without any specific domain knowledge or in-game coaching from human.

"It learned on its own," said report author Prof Pierre Baldi, commenting on the achievement on the so-called DeepCubeA - an algorithm, combining deep learning with classical reinforcement learning and path finding methods.

He also noted that DeepCubeA  works on other combinatorial games such as the sliding tile puzzle, Lights Out and Sokoban and its strategy is very different from the way humans tackle such puzzles, especially the cube invented by the famous Hungarian architect Erno Rubik.

According to the study, published on 15 July in Nature Machine Intelligence, the algorithm was given 10 billion different combinations of the puzzle, with the target to decode all of them within 30 moves. It was then tested on 1,000 of these and managed to solve all of them, finding the shortest path to the solution about 60% of the time.

"The generality of the core algorithm suggests that it may have applications beyond combinatorial puzzles, as problems with large state spaces and few goal states are not rare in planning, robotics and the natural sciences," the study stresses.

Similar articles