10.26.2009 - Ferst Symposium on Controls

Grizzle

• Mabel - controls leg length and leg position (LIKE Yueng-He!)
  • Uses "virtual holonomic constraints"
  • Assume there is a "crank" that drives all of the kinematics forward, then when various segments diverge minimize the error on the constraints.
• Utilizing mechanics gives a 4:1 improvement in efficiency based on:
  • PositiveWork / Mass / Distance
• For control utilize hybrid zero-dynamics
  • When controlling stiffness adding a deadbeat improves robustness

Petros Ioannou

• Robust adaptive control
• Identifier based control
  • Try and find appropriate control based on measured output
  • This makes the open-loop LTI system into a non-linear time-varying closed loop system
• Use feedforward control to move the position of the zeros of the closed loop system
• Using ARMA or backward-difference Euler is equivalent

Kahlil

• High gain observers
• Two problems in high gain observers
  • Peaking phenomenen
  • Saturation
• In high gain you have the possibility of destabilizing the system with faster control than measurement
• Have the possibility of recovering the actual trajectory of the system
• Can you use these techniques for system identification?

Krstic

• Stochastic averaging and extremum seeking
• You can reach the global optimum through using stochastic averaging