Evenly Weighted Particle Filter for Terrain-referenced Navigation using Gaussian Mixture Proposal Distribution

Abstract

The irreversible problematic situation of bootstrap particle filter that it is subject to the weight collapse, is tackled with an evenly weighted setup especially in application to the terrain-referenced navigation problem of unmanned aerial systems. The paper is featured with the Gaussian mixture proposal density taking multimodal noise characteristics of terrain clearance sensor into account. Each particle explores further towards the region of high likelihood in addition to its original motion model, while the amount of transition of the introduced proposal density is calculated from a superposition of a couple of optimal data assimilation methods. Numerical local terrain elevation gradient in conjunction with the parameters that describe the multimodality realize the calculation of transition gain by which the innovation is multiplied. The proposed approach significantly reduces the variance of particle weight and reinforces the diversity of particles by locating them exploiting both the terrain measurement and its noise characteristic.