An Adaptive Controller Inspired by Recent Results on Learning from Experts

Abstract

In computational learning theory there have been some interesting developments recently on the problem of “learning from experts.”

In this paper, we “adapt” the learning problem to an adaptive control formulation. What results is an adaptive controller which is reminiscent of a certainty equivalence scheme using the “posterior mean” for the parameter estimator. We show that this scheme can be analyzed in a somewhat novel way, for ideal linear systems. The analysis techniques may be of some interest to researchers in the theory of adaptive control.

This research was conducted while the author was visiting the IMA in Winter/Spring 1993. The author wishes to express his appreciation for the extremely warm hospitality in an extremely cold climate. The research reported here was supported in part by the U.S. Army Research Office under Contract Nos. DAAL-03-91-G-0182 and DA/DAAH04-93-G-0197, by the National Science Foundation under Grant No. ECS-92-16487, and by the Joint Services Electronics Program under Contract No. N00014-90-J1270.