Neural Networks as a Guidance Solution for Soft-Landing and Aerocapture

This paper presents guidance algorithms based on neural networks and illustrates their performance for both aerocapture and soft-landing applications. When tackling guidance problems that do not admit a complete analytic solution, this neural network approach makes it easier to determine a satisfactory command law without making strong simplifying assumptions. Thanks to genetic algorithms, we successfully trained feed-forward neural networks with one hidden layer for both missions. And using comprehensive simulation tools, we then performed Monte Carlo analyses to compare our algorithm with classic guidance methods (Apollo E guidance for soft-landing and an extension of the Cerimele-Gamble scheme for aerocapture). The results show that neural networks can be an interesting alternative as a more optimal guidance scheme.