We’ve educated a couple of neural networks to unravel the Rubik’s Dice with a human-like robotic hand. The neural networks are educated totally in simulation, the usage of the similar reinforcement finding out code as OpenAI 5 paired with a brand new method referred to as Automated Area Randomization (ADR). The gadget can care for scenarios it by no means noticed throughout coaching, corresponding to being prodded through a filled giraffe. This displays that reinforcement finding out isn’t only a device for digital duties, however can remedy physical-world issues requiring exceptional dexterity.
