A regularised particle filter for context-aware sensor fusion applications

Particle Filters are the most suitable filtering techique for some problems where the prediciton and update models are extremely non-linear. However, they suffer some problems as sample depletion which can drastically reduce their performance. There are multiple solutions to this problem. Some of them make assumptions that invalidate the filter for the most difficult scenarios. Some others increase the computational cost far beyond the bounds of real time applications. Context is a very important source of information for those systems that must work flawlessly in changing scenarios, but it introduces strong nonlinearities and uncertainties that filtering algorithms must deal with. This paper analyzes the performance and robustness of a recently developed regularisation technique for particle filters. The proposed scenarios include a navigation problem where a map is used to provide contextual information, because the final target for the particle filter is a mobile robot able to navigate both indoors and outdoors.