Nonlinear Gaussian Mixture Filtering with Intrinsic Fault Resistance

Abstract

Because of limitations in onboard computing, spaceflight navigation has long been relegated to estimation processes that promote computational efficiency over increased model accuracy. Such is the case with fault tolerance methods, where residual editing is the most common practice to screen out erroneous sensor data. This work investigates an alternative approach, where sensor returns classified as faulty are not simply rejected, but are accounted for within the measurement model of the sensor, ultimately leading to a filter with intrinsic fault resistance in lieu of supplementary extensions like residual editing. Several different faulty measurement models are examined by comparing the proposed filter to a baseline filter outfitted with residual editing, where analysis is performed to test relative performance, robustness, and approximation ability of the proposed filter. The proposed approach not only outperforms residual editing, but is also found to be exceptionally robust to both unknown and approximated faulty measurement distributions.