State estimation of linear dynamic system with unknown input and uncertain observation using dynamic programming

The paper is devoted to deriving a novel estimation algorithm for linear dynamic system with unknown inputs when observations contain outliers. The algorithm is derived for arbitrary input signals and does not require a priori statistical information concerning input signals. The filtering problem is considered as a control problem in which the unknown input is regarded as a controlling signal for system dynamics, which is described by Kalman equations. In this case, optimal control using Bellman dynamic programming can be calculated. The problem is complicated by the presence of outliers in the observations. To cope with this problem the Lainiotis' partitioning theorem has been used. The nonlinear algorithm of state estimation is obtained. Presented approach can be used both in control systems and decision procedures in tracking systems.