Numerical Solution of Filtering Problem with Multimodal Densities

Authors:	Simandl Miroslav, University of West Bohemia in Pilsen, Czech Republic Kralovec Jakub, University of West Bohemia in Pilsen, Czech Republic
Topic:	1.4 Stochastic Systems
Session:	Linear and Nonlinear Filtering in Stochastic Systems
Keywords:	stochastic systems, state estimation, nonlinear filters, probability density function, estimation algorithms

Abstract

Numerical solution of filtering problem for nonlinear stochastic systems is treated. The aim is to improve the point-mass method for multimodal probability density functions of state. The main innovation items concern grid update, namely covering a nonnegligible probability density function support and merging grids in multigrid design. Comparing to the standard point-mass algorithm, the new boundary-based grid placement technique maintains estimation quality and the merging technique decreases computational demands for multimodal densities.