Exploring Transfers between Earth-Moon Halo Orbits via Multi-Objective Reinforcement Learning.

Area of Science:

• Aerospace Engineering
• Astrodynamics
• Artificial Intelligence

Background:

• Designing low-thrust trajectories for spacecraft transfers, especially for SmallSats between libration point orbits, presents complex multi-objective challenges.
• Traditional methods can be computationally intensive and may not efficiently explore the full design space.
• Balancing propellant usage, flight time, and trajectory accuracy is critical for mission success.

Purpose of the Study:

• To investigate the design space of low-thrust trajectories for a SmallSat in the Earth-Moon system using a multi-objective deep reinforcement learning algorithm.
• To efficiently train multiple policies simultaneously for distinct trajectory design scenarios.
• To autonomously construct the solution space for rapid insights into trade-offs.

Main Methods:

• Application of Multi-Reward Proximal Policy Optimization (MrPPO), a multi-objective deep reinforcement learning algorithm.
• Training multiple policies on three distinct trajectory design scenarios, each with a unique reward function.
• Evaluation of policies on perturbed initial conditions to generate performance metrics like propellant mass usage and flight time.

Main Results:

• Successfully trained unique control schemes for each trajectory design scenario and reward function.
• Generated data on propellant mass usage, flight time, and state discontinuities for various low-thrust trajectories.
• Examined a subset of the multi-objective trade space, revealing insights into transfer geometry and performance.

Conclusions:

• MrPPO enables efficient exploration of the multi-objective trade space for SmallSat trajectory design.
• The algorithm autonomously constructs solutions, providing rapid insights into propellant mass, flight time, and transfer geometry.
• This approach accelerates the understanding of complex orbital transfer dynamics and design parameters.