Orbit-Attitude Predictive Control in the Vicinity of Asteroids with In Situ Gravity Estimation

TL;DR

This work tackles autonomous orbit-attitude station-keeping near asteroids in the presence of an unknown and inhomogeneous gravity field. It proposes an integrated framework that combines unscented Kalman filtering for joint navigation and in-situ gravity identification with model-learning predictive control to adapt the dynamics online. The approach uses spherical-harmonics gravity models, gravity-gradient torques, and MRPs for attitude, and demonstrates improvements in tracking accuracy and control efficiency, including a constellation-based extension for faster gravity convergence. The results suggest practical autonomous operation without relying on Earth-ground data, and show that a multi-satellite constellation can significantly accelerate gravity estimation with favorable performance trade-offs.

Abstract

This paper presents an integrated model-learning predictive control scheme for spacecraft orbit-attitude station-keeping in the vicinity of asteroids. The orbiting probe relies on optical and laser navigation while attitude measurements are provided by star trackers and gyroscopes. The asteroid gravity field inhomogeneities are assumed to be unknown a priori. The state and gravity model parameters are estimated simultaneously using an unscented Kalman filter. The proposed gravity model identification enables the application of a learning-based predictive control methodology. The predictive control allows for a high degree of accuracy because the predicted model is progressively identified in situ. Consequently, the tracking errors decrease over time as the model accuracy increases. Finally, a constellation mission concept is analyzed in order to speed up the model identification process. Numerical results are shown and discussed.

Figures (14)

• Figure 1: Inertial, asteroid, orbit, body and camera frames of reference.
• Figure 2: Integrated model-learning predictive control scheme for a standalone spacecraft around the asteroid.
• Figure 3: Orbital radius. Blue: learning-based MPC; red: non-learning MPC; black: reference.
• Figure 4: Average tracking error per day (top) and fuel consumption per day (bottom). Dots: learning-based MPC; triangles: non-learning MPC.
• Figure 5: Semi-major axis (top) and eccentricity (bottom) for $i_0=90^{\circ}$. Blue: learning-based MPC; red: non-learning MPC.