Adaptive control of a stochastic system with unobservable state under conditions of unidentifiability (Q1319795)

This paper is dedicated to establish a separation principle for locally optimal control laws. The authors attempt to avoid the fact that for stochastic systems, unidentifiability of the system prevents the realization of the filter required for optimal control. To this end, laws which can be realized directly from observations are used, bypassing filter construction. Application of the averaging method allows to distinguish between fast and slow variations and suggests that estimators tend to a manifold where every point corresponds to the true control law.