Machine learning-aided adaptive control of spacecraft formation path planning and collision avoidance

In this study, a machine learning-assisted adaptive control approach for multi-spacecraft formation flying (multi-SFF) trajectory planning and collision avoidance is proposed. Composed of a Radial Basis Function Neural Network (RBFNN)-assisted adaptive nonlinear feedback control for multi-SFF path planning and a fuzzy-learning-aided adaptive collision avoidance (FACA) control. The effectiveness and reliability of this novel approach, namely the RBFNN&FACA algorithm, are demonstrated through the Lyapunov method and a detailed fuzzy analysis. Proving that the developed RBFNN-assisted control drives the system exponentially faster to the desired state and rapidly attenuates the oscillation in the steady state. The developed FACA algorithm incorporates a new collision probability distribution function to integrate the relative velocity information into collision avoidance decisions, overcoming the shortcomings of classical Artificial Potential Field (APF)-based collision avoidance methods; further analysis shows that the FACA algorithm is reversible and independent of path planning control, which simplifies the control system structure and reveals potential applications in spacecraft noncooperative orbital games. Furthermore, a novel time-dependent operator is introduced to transform the multi-SFF nonlinear dynamics into a new simplified equation, without compromising its accuracy. Numerical simulations with five spacecraft in a circling configuration under Earth’s J2 and virtual disturbances are conducted. Comparisons with the APF-based methods confirm significant improvements in time response, effectiveness for nonlinear systems and perturbations, and autonomy in complex scenarios. Therefore, this research serves as a reference for the application of machine learning in spacecraft control.