Robust Control, Feedback and Learning: Data-Driven Methods
Abstract
The research effort supported under this grant ended 31 July 2004. A highlight of progress during the final fiscal year was a clear theoretical explanation of and solution to. the model mismatch stability problem generally associated with state-of-the-art adaptive control design methods. The source of these model-mismatch instability problems was traced to the implicit use of absolute-error cost functions and/or use of assumption-driven cost functions. These cost-functions were shown via counter-example to he capable of giving preference to destabilizing control laws in some cases involving model mismatch. Having identified the model- mismatch problem, a fix was also identified, which is to employ a data-driven input-output-gain related cost function for adaptive control-law selection. This progress was enabled by theory that explains the synthesis of adaptive control processes in terms of control law unfalsification.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 31, 2004
- Accession Number
- ADA427715
Entities
People
- Michael G. Safonov
Organizations
- University of Southern California