Robust Control

To quote 2021’s recitation:

Optimal control allows us to create controllers and state estimators that are mathematically optimal – provided our assumptions are valid, and our quadratic cost function accurately captures our real world needs. One issue in particular that optimal control suffers from is what happens when our model isn’t quite right. In some cases, the controller breaks down. These issues were famously documented by John Doyle in his 1978 paper Guaranteed Margins for LQG Regulators. The abstract succinctly states his findings: “There are none”.

This is the canonical text: https://www.ece.lsu.edu/kemin/essentials.htm,

| Essentials Of Robust Control Kemin Zhou, Louisiana State University John C. Doyle, California Institute of Technology Published September, 1997 by http://www.prenhall.com/ |

Our last controls topic will be ML and controls … but first, let’s talk reinforcement learning and other such things.