The folks at OpenAI have created a robot hand that can solve a Rubik’s cube one-handed. Well, it has to, because it is only one hand.
The human hand is a marvel of creative engineering, and the folks at OpenAI want to study what makes it so special. To that end, they have trained a pair of neural networks to work in concert in order to get a robotic hand to solve a Rubik’s cube.
One Youtube commenter said it best: “The progress in dexterous manipulation over 1 year has been amazing. The success of these sim2real domain randomization algorithms makes me think that it is more important for these physics simulators to have a diverse space of potential dynamics, rather than to exactly model the real world.” The progress made here with Open AI’s project points toward a bright future for algorithm-based technologies.
