This proposal deals with control of linear systems driven by poorly known Markov processes and white noise. Nonlinear filtering techniques can give the conditional probability distribution of the current state of the Markov process in terms of system's history. In this way the controller learns about the poorly known processes based on the observation of their past performance. Specific examples of applications of the type of control problems discussed in this proposal include models of solar electric generating stations, where the Markov process is the variation of the solar power due to cloud cover, and certain pursuit-evasion problems, where the pursuer bases its strategy on the past performance of the evader.
