On Deep Reinforcement Learning for Spacecraft Guidance

This paper introduces a novel technique, named deep guidance, that leverages deep reinforcement learning, a branch of artificial intelligence, that enables guidance strategies to be learned rather than designed. The deep guidance technique consists of a learned guidance strategy that feeds velocity commands to a conventional controller to track. Control theory is combined with deep reinforcement learning in order to lower the learning burden and facilitate the transfer of the trained system from simulation to reality. In this paper, a proof-of-concept spacecraft pose tracking and docking scenario is considered, in simulation and experiment, to test the feasibility of the proposed approach. Results show that such a system can be trained entirely in simulation and transferred to reality with comparable performance.