AI-Powered Precision: Revolutionizing Lunar Landings with Deep Learning and Reinforcement Learning

Abstract

This paper investigates how a deep learning-based neural network can effectively enable precise and
safe autonomous navigation for space vehicles in various environmental conditions. With a recent increase in
interest in developing a lunar presence via the Artemis missions, the need for safe autonomous landing systems also
increases. Using a convolutional neural network trained on topographic maps of a simulated landing zone for a
robotic lander, the neural network is capable of identifying global regions of interest for flat landing spots along
with local solutions to simulate real-time response during a mission scenario. Via simulated results with various
initial landing trajectories, the neural network was capable of identifying if the initial landing spot can be deemed
safe and if not, able to identify nearby landing locations that are attainable in terms of velocity limitations. The
model efficiently calculates the best landing spots in the quickest time.