Optimal control of nonlinear probabilistic systems on the basis of an incomplete state vector (Q796488)

The authors consider the optimal control of a partially observed stochastic system. The control functions are piecewise continuous in t for each value of the observed components and it is assumed that probability densities of the state process exist. Thus, the problem is reduced to discussing the optimal control of the Fokker-Planck equation. It is shown that an optimal control will maximize a certain Hamiltonian and a form of Bellman's equation is obtained for the optimum cost. The results are illustrated by considering the linear quadratic case.