Robust regularized particle filter for terrain navigation

In this paper, we address the multimodality of the filtering distribution in the context of ambiguous measurements. In terrain navigation, the similarities in the elevation profiles can lead to multiple modes in the posterior distribution. Typical experiments based on standard particle filters show that the filter tends to lose the true mode as time goes on. Using a clustering algorithm, a new filter based on a mixture representation with local regularization is introduced and displays good robustness properties.