Particle filter underwater terrain-aided navigation based on gradient fitting

Abstract

Inertial navigation systems (INSs) are widely used for autonomous underwater vehicle navigation, with excellent short-term precision. However, the positioning error of an INS accumulates with time. It is imperative to adopt other approaches to mitigate drift errors, especially for long-term sailing tasks. To that end, underwater terrain-based navigation (TBN) is effective in bounding the drift errors. In particular, the particle filtering (PF) method is extensively employed in TBN to tackle the highly nonlinear measurement equation. In our previous work, the statistical properties of the digital terrain model gradient were exploited and we proposed the use of three probability density functions to characterize the normalized data, i.e. Gaussian, gamma and Weibull distributions. In this paper, the likelihood was modified according to the gradient fitting results, i.e. an optimal distribution selection. Moreover, to prevent the measurement data with large gradients from generating particles with lower weights, a pre-screen procedure was proposed to stabilize particle filter sampling. As shown and demonstrated in simulations, our proposed improved PF method outperforms the comparative ones in terms of root mean square error and standard deviation, as well as accuracy, especially for long-term sailing tasks. In terms of computational cost, a smaller number of measurement data are employed and the proposed method is faster than the standard PF method.