ML/RL methods are often viewed as a magical black box, and while that's not true, learned policies are nonetheless a valuable tool that can work in conjunction with the underlying physics of the robot. In this video, Agility CTO Jonathan Hurst - wearing his professor hat at Oregon State University - presents some recent student work on using learned policies as a control method for highly dynamic legged robots.
This presentation was recorded on Oct. 20, 2020 as part of the of University of Washington Electrical & Computer Engineering seminar series:
https://www.youtube.com/watch?v=Q4h-g4Ncl8E