Fiducial Marker SLAM for Extraterrestrial Localization

Abstract

Satellite-based navigation is currently unavailable around the Moon or Mars due to prohibitive initial costs. For early robotic construction in these environments, absolute site localization is essential for coordinated operations and precise execution of building plans. This study investigates the feasibility of using AprilTags in combination with a Kalman Filter to enable full SLAM capabilities. For this research, an OpenGL simulation and custom tools were developed for evaluation and testing. They allowed the establishment of a baseline for the AprilTag state-estimation accuracy, against which algorithmic improvements can be measured. This approach enabled a controlled evaluation of the algorithm. In future work, the same framework can be used to benchmark additional improvements. The final method uses the AprilTag visual state estimation provided by the Infinitesimal Plane-Based Pose Estimation algorithm, combined with graph-based mapping and localization, and a Kalman filter based on parametrized odometry. This approach is computationally simple and requires little memory. The study found that this approach performs moderately well when the tags are relatively large and close to the camera, but as the scale increases, it performs worse. In its current form, the simple approach to this problem does not seem to provide sufficient precision for practical use.