9. • PID controller is simple and generic approach
• PID controller is not optimal and in some
complicated cases can be unstable
https://github.com/tsybulkin/drone_stab
13. Three modes
• Mode 1: Pumping energy until E < Ethr
•
• U1(t) = sign(da/dt)
• Mode 2: No control until |a| > a0
• Mode 3: Balancing control |a| < a0
14.
15. LQR
• Q, R - positive-definite matrices
• u = - Kx; Riccati equation
• scipy.linalg.lqr(A,B,Q,R)
Wikipedia: https://en.wikipedia.org/wiki/Linear-quadratic_regulator
20. Optimal control by ML
• Agent explores different strategies by trial and error
method
• Optimal policy: maps states —> actions,
maximizing total reward