Learning Control by Using Kalman Filter

Abstract

The learning control syntheses are discussed for linear sampled-data systems with unknown parameters as follows.
1) The unknown parameters are additive: e.g. the means of additive gaussian noises in the dynamical system and the measuring system.
2) Some elements of the driving matrix are unknown beside additive parameters.
The learning control problem is reduced to the conventional stochastic control one with the Kalman estimation for a generalized state vector which is composed of the actual state and unknown parameters.
By this method, the optimal control can be obtained for case 1) and the sub-optimal control policy is obtained by using the same conditional probability distribution for the future values of the driving matrix element for case 2).