Motion and Inertia Estimation for Non-Cooperative Space Objects During Long-Term Occlusion Based on UKF-GP

Area of Science:

• Spacecraft dynamics and control
• Robotics and autonomous systems
• Machine learning for aerospace applications

Background:

• Estimating motion and inertia of non-cooperative space objects is crucial for debris removal and servicing.
• Long-term sensor occlusions pose a significant challenge for traditional estimation algorithms.
• Data-driven approaches are increasingly explored to overcome limitations in modeling complex dynamics.

Purpose of the Study:

• To develop a robust method for estimating motion and inertia parameters of tumbling space objects, even with long-term occlusions.
• To leverage Gaussian processes (GP) for simulating sensor measurements and handling periodic trends in non-periodic data.
• To propose and validate a fusion algorithm combining unscented Kalman filter and Gaussian process (UKF-GP) for enhanced estimation accuracy.

Main Methods:

• Utilized multi-output Gaussian processes (GP) with product kernels (two periodic kernels) to predict stereo-camera projection measurements.
• Employed fast Fourier transform (FFT) analysis to derive initial hyper-parameter guesses for GP periodicity.
• Developed an unscented Kalman filter-Gaussian process (UKF-GP) fusion algorithm, using GP predictions as pseudo-measurements during occlusions.

Main Results:

• The UKF-GP algorithm demonstrated accurate estimation of target motion variables over extended periods (hundreds of seconds).
• Performance was validated across varying tumbling frequencies through Monte Carlo (MC) simulations.
• The proposed method significantly outperformed the conventional UKF algorithm in handling long-term occlusions.

Conclusions:

• The UKF-GP fusion algorithm provides a robust solution for motion and inertia parameter estimation of tumbling space objects under challenging occlusion conditions.
• Gaussian processes effectively model and predict sensor data, enhancing estimation capabilities.
• This approach offers a significant advancement for autonomous operations involving non-cooperative targets in space.