Yet another example of the value of simulated training, researchers at OpenAI trained a robotic hand using hundreds of years of  object manipulation inside a computer simulation. The resulting hand movements are significantly more dexterous than anything developed so far.
https://www.technologyreview.com/s/611724/artificial-intelligence-driven-robot-hand-spends-a-hundred-years-teaching-itself-to-rotate/
A reinforcement-learning algorithm allows Dactyl to learn physical tasks by practicing them in a virtual-reality environment.