Machine Learning Assisted Low-Thrust Orbit-Raising: A Comparative Assessment of a Sequential Algorithm and a Deep Reinforcement Learning Approach

The focus of the paper is machine-learning-assisted computation of low-thrust orbit-raising trajectories. We consider a sequential algorithm for computing multi-revolution trajectories, whose optimization cost function parameters can be updated through a high-level planner utilizing a suitably trained artificial neural network. Considering two different orbit-raising mission scenarios based on the final target orbit (geostationary and near-rectilinear halo orbit), we conduct numerical simulations to compare the results of this approach with that provided by a deep reinforcement learning framework.