LP-based accuracy improvement for UAVs

The problem of accuracy improvement in the context of an unmanned air vehicle (UAV) flying above a flat Earth with no terrain perturbations is considered. The six UAV degrees of freedom coordinates and two optical axis degrees of freedom angles are available periodically at some constrained sampling interval. Systematic and random errors accompanying these data corrupt computed results based upon them and the problem of improving their accuracy in the presence of concomitant random errors is the subject of this paper, where it is assumed that the systematic errors have been removed. A particular example is described, allowing no possibility of tracking, which is chosen for evaluation with the median of the probability density function of error taken as the measure of success. A typical UAV trajectory and dynamics including wind are taken along with representative values for the variances of the error distributions of mean zero, consistent with the assumption of no systematic errors, but having various degrees of asymmetry. We compare the proposed linear prediction (LP) algorithm using M_LP samples with the use of the current sample set and with a simple average using M_LP samples. It is shown that the LP method significantly improves the accuracy for the set of moderate parameters taken in the example over a wide range of asymmetry