Autonomous Navigation with Feature Points Map for Self-position Estimation for High-Accuracy Landing on Unknown Celestial Bodies

This paper presents a novel method for the highly accurate landing of small unmanned space probes on unknown celestial bodies. In this method, a map of feature points is generated, and the self-position is estimated precisely. Creating high-precision maps is a challenge owing to numerous space-specific disturbances, limited computational resources, and the need for high reliability. Here, we demonstrate highly reliable map generation and self-position estimation using only generic feature points (AKAZE and SURF) detected in images captured by a space probe for high-precision landing on unknown celestial bodies. This study focuses on visual simultaneous localization and mapping (V-SLAM) in space. We prepared four image datasets consisting of 80 images with randomly added space-specific disturbances. We tested the reconstruction using the dataset in a low computational resource environment, such as a Raspberry Pi 4B and Mac Studio, for repetitive simulations. The simulations indicated that a balance between computation time and accuracy can be achieved, even with limited computational resources, by adjusting the number of feature points. We also demonstrated that our proposed method can operate on a resource-limited Raspberry Pi 4B, meeting both accuracy and time requirements. Our experiments highlight that the feature point map is capable of guiding space probes to achieve highly accurate landings, and this algorithm raises the possibility of practical use for space probe operations.