Automated Trajectory Planning: A Cascaded Deep Reinforcement Learning Approach for Low-Thrust Spacecraft Orbit-Raising

Efficient computation of orbit-raising trajectories for low-thrust propulsion spacecraft, especially within multi-body regimes of spaceflight, pose significant challenges due to complex dynamical environment, prolonged transfer times and reliance on initial expert solutions. To address these challenges, we propose a novel cascaded deep reinforcement learning approach to optimize the planning of low-thrust spacecraft trajectories. Focusing on transfers from launch-injection orbits, such as geostationary transfer orbit (GTO) and super-GTO, towards targets such as geosynchronous equatorial orbit (GEO) and near-rectilinear halo orbit (NRHO), our methodology, guided by a gradientaided reward function, outperforms existing automated methods in achieving time-efficient spacecraft orbit-raising. The results demonstrate that our approach is effective and time-efficient, achieving optimal or near-optimal solutions.