Neural-Network-Inspired Machine Learning for Autonomous Lunar Targeting

Abstract

Early work on a sequential machine learning technique for autonomous spacecraft navigation that is inspired by artificial neural networks is presented in this paper. This technique uses sequential function approximation, a method that has been used to analyze and solve a variety of dynamic aerospace systems, to develop a multiple-input/single-output neural map. Sequential function approximation, which has previously been used for direct function approximation and solving systems of governing equations meshlessly, uses the error between a set of current and desired system output objectives (in this case, future state vectors) based upon current system inputs: here, the thrust profile of an engine burn, to improve the accuracy of the inputs. A preliminary test case of a lunar flyby in the circular restricted three-body problem is presented as a proof of concept for sequential function approximation as an autonomous navigation and mission-management technique.