Constraint manifold filtering

Abstract

The effectiveness of particle filters for dealing with highly non-linear system dynamics and measurements has been well demonstrated [1] However, particle filters do have their own improvement to make: notably, how to overcome the trade-off between the efficiency represented by the number of particles for filtering and the robustness against the erroneous estimation of uncertainties as well as the ambiguities and outliers involved in measurements. This paper presents a new approach to filtering, referred to here as Constraint Manifold filter, or shortly CM filter. CM filter uses particles to represent constraint manifold. Comparative evaluations of CM filter against conventional particle filters indicate that CM filter is indeed far more robust yet efficient than conventional particle filters against the erroneous estimation of uncertainties.