Mitigation of Vision Measurement Effects on Lunar Descent Navigation

Abstract

The navigation system of lunar lander is required to have a precision and autonomous navigation capability without the aid of the Earth-based ground tracking system and global navigation satellite system (GNSS). The vision-based terrain relative navigation (TRN) camera integrated with inertial measurement unit (IMU) is known to be the key technology for achieving the mission requirement. In this dissertation, the methods to mitigate the effects of measurement from the vision sensors that can be encountered during lunar descent navigation are studied. The effect of vision measurement geometry is analyzed by employing the dilution of precision (DOP) used as performance measure for GNSS. The problem of extended Kalman filter (EKF) divergence when integrating the IMU and nonlinear vision-based TRN camera is solved by applying EKF underweighting using the vision DOP. The vision DOP can be used as figure of merit for vision-based navigation and may vary by the geometric distribution of the matched features with respect to the vison sensor. The effect of measurement geometry can be different for vision-only, inertial and vision-based TRN camera, and that with star tracker, thus the geometric effect is investigated for each cases by analytic derivation of vision DOP and covariance analysis of integrated navigation system. It is shown that the vision DOP is a function of distance and line of sight angle from sensor to matched features and the number of features. Also, covariance analysis results show that the navigation accuracy can be improved and its dependency on line of sight angle can be reduced if accurate attitude information is provided by star tracker. EKF underweighting is known to be used extensively for practical applications to prevent filter divergence when the magnitude of state uncertainty is large compared to the nonlinear measurement accuracy by adding additional term proportional to the a priori state estimation error covariance to the residual covariance. Generally, constant value of underweighting coefficient is applied which is determined by complicated tuning process for the system under consideration. The underweighting for inertial and vision-based TRN camera integrated navigation of lunar lander using vision DOP is proposed to adaptively determine its coefficient and condition whether to apply or not. The underweighting coefficient is defined as the percentage of decrease of the state estimation error covariance and the predicted accuracy of vision-based navigation using vision DOP is used as the a posteriori covariance. The performance of the proposed EKF underweighting is demonstrated through Monte Carlo analysis.