Thesis Defense: Robustness Analysis for Identification and Control of Nonlinear Systems

Speaker: Mark Tobenkin , MIT CSAIL, LIDS

Date: Monday, November 25, 2013

Time: 4:00 PM to 5:00 PM Note: all times are in the Eastern Time Zone

Refreshments: 3:45 PM

Public: Yes

Location: 32-G449 (Patil / Kiva)

Event Type:

Room Description:

Host: Russ Tedrake, MIT CSAIL

Contact: Mark M. Tobenkin, mmt@mit.edu

Relevant URL:

Speaker URL: None

Speaker Photo:
None

Reminders to: seminars@csail.mit.edu

Reminder Subject: TALK: Thesis Defense: Robustness Analysis for Identification and Control of Nonlinear Systems

Abstract:

Nonlinear dynamics are at the heart of many engineering and scientific endeavors such as the design and analysis of robotic systems, experimental and theoretical electrophysiology, and computer-aided design of high speed electronics. Modern computing power has revolutionized the role that simulations play in these disciplines, but finding models whose simulations accurately match experimental data on long time horizons remains a challenging task. First, it is crucial to ensure that the identified dynamics are stable, but general techniques for testing stability do not exist for nonlinear models. Second, while it is natural to desire a model whose open-loop simulations match the experimental data, minimization of such simulation errors leads to challenging nonlinear programming problems for both linear and nonlinear models.

This dissertation provides new computational methods for identification and analysis of nonlinear dynamical systems using convex optimization. I provide a framework for system identification extending previous work based on convex parameterizations of stable nonlinear systems and convex upper bounds for simulation error. These existing methodologies generate severely biased estimates in the presence of measurement noise. Two algorithmic improvements are introduced to overcome this problem: (i) an instrumental-variable strategy based on the availability of a pair of repeated experiments, and (ii) a family of tighter upper bounds for simulation error applicable to state-affine systems. The effectiveness of these techniques is examined through application to several experimental and simulated data sets.

Research Areas:

Impact Areas:

This event is not part of a series.

Created by Mark M. Tobenkin at Saturday, November 23, 2013 at 3:30 AM.