This paper applies the reinforcement learning (RL) to multi-stage impulsive trajectory design of the manned lunar mission. The manned spacecraft is modeled as the agent and the nonlinear dynamics in the cislunar space is the interaction environment of the agent. Then the Markov decision model is mathematically built for the spacecraft and the optimization goal is to minimize the total impulses in the whole mission. The Q-learning method is used to learn the optimal policy and to better train the agent to explore and exploit the environment, a reshaping reward is designed to improve the convergency and effectiveness of the algorithm. Numerical results give the integrated designs of the lunar transfer orbit, Earth transfer orbit and three-impulsive lunar injection/ejection orbit. The proposed reshaping reward function shows better performance compared with the non-reshaping reward. The conclusions could provide a reference for China’s future manned lunar project.
A Reinforcement Learning Method to Trajectory Design for Manned Lunar Mission via Reshaping Rewards
Lect. Notes Electrical Eng.
International Conference on Guidance, Navigation and Control ; 2022 ; Harbin, China August 05, 2022 - August 07, 2022
2023-01-31
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
NTRS | 1963
|AIAA | 1963
|The manned lunar landing mission.
NTRS | 1964
|MIDCOURSE NAVIGATION FOR A MANNED LUNAR MISSION
SAE Technical Papers | 1962
|The Manned Lunar Landing Mission
SAE Technical Papers | 1964
|