We have already covered a bunch of robots that can solve Rubik’s Cube puzzles fast. Most of them use advanced algorithms to get it done. Researchers at OpenAI have trained a robotic hand to manipulate a cube with great dexterity. It uses a reinforcement-learning algorithm to work the cube. All the training was done in simulation.
The robot can handle situations that it never saw during training. As the researchers explain:
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.