Travis Gibson, Nov 2013

Schedule

• 8:00 am: meet with Richard, 109 Steele
• 8:30 am: Free Time
• 10:00 am: NCS (networked control systems) group meeting, 243 Annenberg
• 12:00 pm: Lunch with some postdocs; talk by K. Dvijotham: "Convex Structured Controller Design" in Annenberg 107
• 1:00 pm: Seminar, Steele Lab 114
• 2:00 pm: John Doyle, 210 Annenberg
• 2:45 pm: Open
• 3:30 pm: Doug MacMartin (128 Steele)
• 4:00 pm: Eric Wolff (114 Steele)
• 4:30 pm: Meet with Richard, 109 Steele

Talk Abstract

Transients, Learning, and Stability in Adaptive Systems

One of the main features of adaptive systems is an oscillatory convergence that exacerbates with the speed of adaptation. Over the past two decades several attempts have been made to provide adaptive solutions with guaranteed transient properties. In this work it is shown that Closed-loop Reference Models (CRMs) can result in improved transient performance over their open-loop counterparts in model reference adaptive control. In addition to deriving bounds on L-2 norms of the derivatives of the adaptive parameters which are shown to be smaller, an optimal design of CRMs is proposed which minimizes an underlying peaking phenomenon. The analytical tools proposed are applicable to a range of adaptive control problems including direct control, composite control and partial states accessible control. If time is available the application of CRM based adaptive control to Very Flexible Aircraft will be presented.

Bio: Travis received his BS('06) from Georgia Tech, and MS('08) and PhD ('13) from MIT all in Mechanical Engineering. He has worked for NASA and Boeing, and was a visiting scholar at the Technical University of Munich in 2009.