Intelligent Fusion Autonomous Navigation Method for Mars Precise Landing

Abstract

To overcome the difficulty of absolute optical navigation in unknown environments, an intelligent fusion autonomous navigation method for Mars precise landing was proposed. Considering the difficulties of the inability to detect features and the low efficiency of recognition brought by high texture similarity in the extraterrestrial environment and perspective scaling between images, an unsupervised homography network was constructed to estimate the inter frame motion of the lander. Based on the inertial measurement information, a recursive model of the lander state was established. Using the established measurement model and state recursive model, real-time estimation of the lander position, velocity, and attitude was achieved through UKF. The simulation results verify the effectiveness of the proposed method without the need of feature detection and matching.

Keywords

Mars landing / intelligent navigation / multi-source fusion / unsupervised learning / deep neural network