Award

Intelligent Feedback Systems

Description: The theme of the proposed research is to advance the state of the art in the design of feedback controllers. This basic research will integrate concepts from deterministic and stochastic optimal control theory and intelligent control methodologies to arrive at superior intelligent feedback designs. More specifically it is proposed to utilize neural network technology to implement the design of optimal, robust, and adaptive nonlinear controllers in the presence of significant un certainty. The emphasis will be on basic theoretical and algorithmic results. Specific research directions include the use of optimal control theory as a means for training artificial neural network implementations of linear, nonlinear, and adaptive controllers. One approach will involve the use of numerical data from the solution of the Two-Point- Boundary-Value (TPBV) problem arising from the application of the Pontryagin Maximum Principle to firs train artificial neural network feedback controllers and then evaluate their performance. In particular, we shall examine whether Cerebellar Motion Articulated Controller (CMAC) structures can be viewed from this perspective. In the beginning phase we plan to use bang-bang time- optimal nonlinear controllers for low order systems and investigate the degree that a trained multi-layer neural network approximates the switching curves and surfaces associated with these bang-bang controllers.

Funder: National Science Foundation

Amount: $200,000

Funding type: grant

Funder scheme: Standard Grant

Start date: 1992-09-15

End date: 1996-10-31

Investigators: Michael Athans (lead), John Tsitsiklis

Institution: Massachusetts Institute of Technology

Topic: Neural Networks and Applications

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences