Online Dynamic Modeling and Localization for Small-Spacecraft Proximity Operations

Abstract

Proximity operations between spacecraft allows for docking, inspection, and repair of a target vehicle. Very small spacecraft, under 20 kg, are well-suited for proximity activities and are of growing interest in the aerospace community. However, due to size and power constraints, small vehicles cannot carry traditional precision navigation systems and have generally noisy sensor and actuator options. This paper presents two techniques for improved autonomous, on-board navigation that account for noisy and poorly observable states. First, an Unscented Kalman Filter is implemented for localization which incorporates orbital dynamics and quaternion rotation. Second, two online regression algorithms, Bayes Linear Regression and Gaussian Process Regression, are used to learn the time-varying thruster dynamics. These techniques have been demonstrated successfully on a simulated small inspector vehicle and are being integrated on the Washington University in St. Louis Bandit inspector spacecraft.