Beyond Kalman Filtering - Evaluation of Filtering Techniques for Autonomous Relative Navigation

Abstract

Estimation problems have to be solved in several space applications. During rendezvous to a target spacecraft, e.g. in an on-orbit servicing mission, the relative pose (position and attitude) between the two involved bodies has to be estimated by navigation filters. The Kalman filter is a popular filter technique for linear estimation problems. A modification for non-linear problems is the Extended Kalman Filter which is based on local linearization of non-linear functions. In this paper, further filter techniques alternative to the standard Kalman filter like e.g. Particle Filter, Unscented Kalman Filter, H-infinity Filter and their application on relative navigation are investigated. Practical results where the methods were applied on a navigation problem are shown and discussed. Therefore a rendezvous process including non-linear dynamical system equations was simulated on the robotic based test facility EPOS 2.0 (European Proximity Operations Simulator). A mono, CCD camera was used as rendezvous sensor. The outer edges of a target satellite were detected by an image processing system and served as measurement for the navigation filters. As camera model a pinhole model with radial distortion was applied which results in a non-linear measurement equation. The resulting non-linear estimation problem was solved with different navigation filters. A comparison of the different estimation methods was drawn with respect to applicability, necessary statistical a-priori knowledge, computational effort, real-time capability and accuracy.