Lightweight visual SLAM and traversability mapping for space robotics: Real-time operation in challenging environments

Exploration of extreme environments, such as planetary surfaces and subterranean lava tubes, poses significant challenges for space robotics, particularly in navigation and traversability analysis. These tasks require high autonomy under stringent computational and power constraints, which limits the applicability of many modern Visual SLAM methods developed for terrestrial applications.In this paper, we present a lightweight Visual SLAM system for Unmanned Ground Vehicles (UGVs) tailored to space-relevant scenarios. The system integrates a computationally efficient front-end with a sparsification-enhanced back-end and incorporates dense 3D reconstruction to support traversability mapping.We evaluate our system alongside state-of-the-art navigation methods using multiple sensing modalities (Camera, Camera/IMU, LiDAR) on two representative datasets: the Martian Terrain dataset and the Mine Museum dataset, which emulates lava tube environments. Results demonstrate that our system achieves localization accuracy and 3D reconstruction quality comparable to existing approaches, while running in real time on low-power embedded platforms, such as a Raspberry Pi 4, and enabling autonomous navigation of our experimental UGV.All datasets and code are publicly available, including the Martian Terrain and Mine Museum datasets, as well as the Visual SLAM and Traversability Mapping implementations, facilitating reproducibility and supporting further research in planetary robotics.