Cascaded Deep Reinforcement Learning-Based Multi-Revolution Low-Thrust Spacecraft Orbit-Transfer

Transferring an all-electric spacecraft from a launch injection orbit to the geosynchronous equatorial (GEO) using low thrust propulsion system presents significant challenge due long transfer time typically spanning several months. To address of determining such time-scale orbit-raising maneuvers GEO, this paper novel technique compute transfers starting geostationary (GTO) and super-GTO. The is complex, involving multiple eclipses revolutions. tackle challenge, we introduce cascaded deep reinforcement learning (DRL) model guide low-thrust towards desired by appropriate direction at each state. ensure mission requirements, gradient-aided reward function incorporating orbital elements, guides DRL agent obtain optimal flight time. obtained results demonstrate that our proposed approach yields or near-optimal time-efficient orbit-raising. implementation important for autonomy; in context, DRL-based trajectory planning provides significantly better as compared state-of-the-art approaches allow automated computation.