Designing robust pose estimator for non-cooperative space targets for visual servoing during approach maneuvers

Kurzfassung

For on-orbit autonomous grasping of an uncooperative spacecraft using visual servo control of a robotic manipulator, it is imperative that the pose estimation algorithm provide accurate estimates of relative motion parameters from the noisy vision measurements. These non-uniformly sampled measurements have variable noise characteristics and represent a past state owing to the processing time. In this thesis, an event-driven and Out of Sequence Measurement (OOSM)-capable Extended Kalman Filter (EKF) observer with adaptive behavior is derived for estimating motion, inertial and geometric characteristics of an uncooperative Target spacecraft with an objective of grasping while using the measurements of the kind mentioned above. Observability and stability analyses have been presented with conclusions about target inertia and geometry that affect the estimation process. Special focus has been laid on the vision sensor’s noise and time-response characteristics to improve the estimator’s robustness and optimality. Robustness is analysed in terms of convergence, immunity towards outliers and adaptive behaviour in the face changing noise characteristics. The adaptive behaviour in the EKF is achieved using a Variational Bayesian (VB) approach and an assessment is presented for a step-change in noise characteristics. A nonlinear state-space model for relative dynamics between the OOS’s end-effector and the tumbling target have been derived which incorporate orbital dynamics and manipulator’s servoing motion. In order to avoid rank-deficiency variances for the attitude quaternion, a Multiplicative Extended Kalman Filter (MEKF) approach with a reduced state-vector is used. The small angular rotation and the Gibbs vector were used as candidates and an evaluation of both of these representations is provided. A Software In Loop (SIL) has been developed which allows fast prototyping of estimation/control algorithms for the grasping problem using an eye-in-hand topology for position-based servo control.