A Lightweight ML-Based Pipeline for Optimizing Mission Planning of Earth-Orbiting Satellites

Abstract

Efficient mission planning for Earth-orbiting satellites is necessary for carrying out different astronomical studies. This work presents a lightweight, machine learning (ML)-based pipeline for the estimation of three mission-critical parameters: up time (periods when the satellite receives solar illumination for power generation), dark time (intervals ideal for astronomical or scientific observations, where the Sun is blocked by Earth), and ground station access windows (periods of uninterrupted line-of-sight communication with Earth-based stations). Unlike traditional orbital mechanics simulations, which are computationally intensive, this work uses a two-stage ML model, a binary classifier followed by a regressor model. The models are trained on high-fidelity simulation data generated through orbital propagation and line-of-intersection geometry (for eclipse calculation), spanning diverse satellite orbit configurations. Evaluation on unseen data shows good enough accuracy and highly reduced computational time. This developed pipeline significantly streamlines satellite mission planning, making it ideal for small research teams who lack access to expensive commercial tools like STK, enabling broader participation in the scientific exploration of the universe. © 2025 IEEE.