This Robot Taught Itself to Walk in a Simulation—Then Went for a Stroll in Berkeley

It's the first time a machine learning approach known as reinforcement learning has been so successfully applied in two-legged robots.

Link to Full Article: Read Here