Outlier Rejection for GNSS-aided Inertial Navigation

Abstract

Motivated by the importance of accurate measurements in navigation-applications, this thesis presents the implementation of an algorithm utilizing the estimated uncertainty from a Kalman filter to determine the quality of the position-measurements coming from the GNSS. The algorithm rejects the position-measurements suspected to degrade the quality of the estimates. The algorithm was implemented together with a terrestrial observer consisting of an attitude observer and a Kalman filter. A proposed technique to improve the rejection algorithm, named "Dual Observers" was implemented using a secondary observer to increase robustness of the rejection algorithm in the presence of long-lasting periods of inaccuracy in the GNSS-measurements.

The implemented methods were tested on two data-sets generated from two flights, one with an aerobatic plane and one with a light weight UAV. As errors were added to the position measurements, the rejection algorithm detected and rejected the error induced measurements. During longer periods of error present in the measurements, the rejection-algorithm failed by wrongfully rejecting error-free measurements in some cases. The dual observer technique provided increased robustness for the rejection algorithm, succsessfully reducing the deviation of the estimates for the longer periods of GNSS-uncertainty.