This paper presents a neural-aided Guidance, Navigation & Control algorithm for reconfiguration of distributed space systems. The guidance algorithm is based on Artificial Potential Fields (APF) in the Relative Orbital Elements (ROE) space. Since the relative orbit determination measurements are typically referred to the Cartesian metrics (e.g. range or range rate), a linear mapping between the set of ROE and the Cartesian coordinates expressed in the Local-Vertical-Local-Horizontal (LVLH) reference frame is derived. The navigation and control algorithms rely on the relative dynamics expressed in the same ROE set of coordinates. To cope with uncertainties and nonlinearities of the system, a Radial Basis Function Neural Network (RBFNN) is employed to reconstruct the perturbed dynamics. The Artificial Neural Network (ANN) is coupled with an adaptive Extended Kalman Filter for state estimation. A feedback control is designed to track the desired state, whose stability is analyzed using Lyapunov theory. The guidance, navigation and control algorithms are tested in a high-fidelity numerical orbit propagator. Moreover, the algorithm is tested in relevant Processor-In-the-Loop (PIL) simulations using a TI C2000-Delfino MCU F28379D. The results demonstrate the effectiveness of the algorithm for relative reconfiguration maneuvers involving relative distances ~102 m with limited fuel consumption and constrained available thrust (⩽1mN). In particular along-track maneuvers, relative plane change and formation enlargement are analysed in the paper, showing the comparison between the proposed algorithm and the legacy one without the neural network. The benefit of implementing a neural network is particularly highlighted when the nonlinearities or unmodelled terms in the on-board dynamics become prominent.
Read full abstract