Feedback control of discrete linear systems with random state and input dependent noise

A generalization of the O.L.F.O. technique is proposed for the design of feedback controls of discrete linear systems with random input and state dependent noise. The control strategy is optimized at any time by assuming for the future, instead of an open loop, a feedback structure with given state estimator. The choice of the optimum estimator and of the control strategy for the future is approached with an original recursive algorithm which generates a monotonic convergent sequence of upper bounds of the cost-to- -go (CUB).