This paper presents a novel hyperparameter tuning strategy for a decentralized Model Predictive Control (MPC) framework based on the quasi-nonsingular Relative Orbital Elements (ROE) for satellite formation flying. We present the MPC architecture and the simulation framework used for validation. Our key contribution is a crash-aware Bayesian optimization algorithm that systematically tunes the MPC horizons and control weights to minimize relative-position RMS error across different initial conditions. Simulations incorporating the Earth’s J 2 perturbation and a range of initial along-track separations evaluate the controller’s ability to converge to specified relative distances and to station-keep. The results show that the controller’s hyperparameters are strongly influenced by the initial conditions, and that the controller offers a robust solution for autonomous satellite formation flying given well tuned hyperparameters.
Moen et al. (Wed,) studied this question.