Nonlinear Filtering Using Sparse Grids

This paper presents a new nonlinear filtering algorithm applicable in realtime. Nonlinear filtering problems are mostly solved with the Extended Kalman Filter which due to the nonlinearities is a suboptimal estimator. Optimal estimates are provided by Fokker-Planck-Equation in combination with Bayes rule. Conventional approaches for the numerical solution of this equation suffer from the “curse of dimension” and are therefore not applicable in higher dimensions. We use sparse grids for solving the Fokker-Planck-Equation and present a six dimensional nonlinear problem solved in real-time with this new approach.