¿-Filter: An Evidence Theoretic Approach to Unmanned Ground Vehicle Localization

Abstract

In this paper, we present a novel evidence theoretic fusion filter, and its application to the Unmanned Ground Vehicle (UGV) localization problem. The various components of the sensor fusion framework such as the adaptive pre-processing unit, the evidence extraction and combination unit, and the extended Kalman filter are described in detail. The crux of this architecture is the evidence extraction and combination unit that employs a twopronged approach, one to switch between parametric models, and another to adaptively vary the measurement noise covariance matrix. The process of evidence extraction using fuzzy-type or rule-based techniques, and their subsequent combination using the Dempster's rule for combination are detailed. An experiment is conducted to demonstrate the merits of this UGV localization approach. Finally, we conclude with a brief summary of the results.

A Sensor Calibration Methodology for Evidence Theoretic Unmanned Ground Vehicle Localization

“…The results of this work are applicable to the -Filter system presented in [18]. Future research would include building parametric models for more operating scenarios, and creating mass assignment tables for additional sensors such as the digital compass.…”

“…A series of three designed experiments were conducted with the objective of fitting parametric calibration models that are geared to the requirements of the adaptive preprocessing unit of the -Filter system described in [18].…”

“…Taking into account the requirements of the -Filter presented in [18], velocity models are developed for two operating scenarios by taking weighted averages as in equation (7). Model A1 (for "normal" operating scenario):…”

“…The error variances along the x G (North) and y G (East) axes are also computed using the aforementioned rule (see Table III). Tables II and III are subsequently used in [18] for computing the values of the R matrix.…”

“…An evidence theoretic approach to UGV localization, the -Filter, presented in [18], has two key features: (1) it employs two models each for linear velocity and angular velocity, and switches between them to "adapt" to the operating scenario (trapped or normal) of the UGV, and (2) it uses a variable measurement noise covariance matrix, R, that depends on the number of visible satellites (#Sat) and the Horizontal Dilution Of Precision (HDOP). The -Filter leverages the Dempster-Shafer (DS) theory of evidence [16], for extracting evidence from the available information in the form of mass functions, and for combining the independent mass functions by Dempster's rule for combination.…”