Boston Dynamics Atlas Walking, Crawling with Agility Using RL

Boston Dynamics’ Atlas has gotten a lot of coverage in the media. The latest videos show how it can walk, run, and crawl with agility using reinforcement learning. This was done in partnership with RAI Institute. RL is used to speed learning in Atlas. A physics-based simulator is used to generate training data for a variety of moves.

As RAI Institute explains, each move is created with data from 150m runs of the simulator. It is transferred zero-shot to the hardware.

[HT]