Make: Volume 93

HOW DOES PUPPER MOVE?

All of Pupper’s physical skills are learned using

reinforcement learning (RL), which is a kind of

artificial intelligence used to teach computers

how to make good decisions to accomplish some

task. It’s a general and flexible framework that’s

also perfect for teaching robots. If you’ve seen

any of the recent videos of robot dogs climbing

boxes, or humanoid robots flipping and sprinting,

you can bet they’re using reinforcement learning!

In our case, we use RL to teach Pupper how to

precisely move its motors in order to walk, climb

over obstacles, jump, and perform all its other

athletic capabilities.

For each task we want ...