Robotic Hand Autonomously Solves Rubik's Cube

San Francisco-based OpenAI has trained a pair of neural networks to solve a Rubik's Cube with a single robot hand. The team places the cube into the palm of a dextrous 24-joint robot hand that mimics human anatomy, but they don't tell it how to solve the cube. Instead, it learns like a human through evolution, only faster. The team ran thousands of simulations to give the robot a base, and the AI takes it further. The team doesn't show the hand how to move the cube. It learns everything from the cube's weight (and how to manipulate it) to how to solve the puzzle. Most of the time.   The robot's track record isn't exactly perfect. It solves an easy cube about 60 percent of the time, and difficult cubes only 20 percent of the time. A "maximum difficulty scramble" implies that it requires 26 face rotations to solve. Simpler scrambles require 15 rotations.The algorithm is adaptable and learns in real-time as the team presents external obstacles to the hand. For example, they put a rubber glove on it, drop confetti on it, hit it with smoke, tangle bracelets around it, fondle it with a giant wooden hand, and peck at it with a stuffed giraffe. Basically, they teach it to withstand the rigors of Mardi Gras.The team has been working on the project since May 2017, and the demonstration could lead to general-purpose robots being capable of performing complex tasks. Oh, and if you're interested, they're hiring.