Robots are already capable of solving magic cubes as long as they are given the right algorithm. OpenAI has managed to enhance the Dactyl robotic hand with self-learning to solve a Rubik’s Cube one handed. The neural networks are trained in simulation but allow the robot to handle situations never seen during training.
As the developers explain, this new technique called Automatic Domain Randomization (ADR):
begins with a fixed size of the Rubik’s Cube and gradually increases the randomization range as training progresses. We apply the same technique to all other parameters, such as the mass of the cube, the friction of the robot fingers, and the visual surface materials of the hand. The neural network thus has to learn to solve the Rubik’s Cube under all of those increasingly more difficult conditions.
Pretty promising, don’t you think?