Freetures: Localization in Signed Distance Function Maps

Localization of a robotic system within previously mapped environment is important for reducing estimation drift and reusing built maps. Existing techniques geometry-based localization have focused on the description local surface geometry, usually using pointclouds as underlying representation. We propose geometrybased that extracts features directly from an implicit representation: Signed Distance Function (SDF). The SDF varies continuously through space, which allows proposed to extract utilize describing both surfaces free-space. Through evaluations public datasets, we demonstrate flexibility this approach, show increase in performance over state-of-the-art handcrafted surfaces-only descriptors. achieve average improvement ~12% RGB-D dataset ~18% LiDAR-based dataset. Finally, our localizing LiDAR-equipped Micro Aerial Vehicle (MAV) map search rescue training ground. Our system, called freetures, available open source $^{1}$ .