Terrain-Structure-Invariant Range-Visual-Inertial Odometry

We propose a novel range-visual-inertial odometry (range-VIO) algorithm compatible with any terrain structure. Unlike previous approaches, our approach does not make any assumption - not even local - about the planarity of terrain between the 1D range measurement and the visual feature observations. Range information is critical to prevent metric scale drift in the absence of visual-inertial excitation. This paper is motivated by state estimation for the new mid-air helicopter delivery maneuver being developed for the next-generation Mars helicopters. Not only does the descent trajectory feature low VIO excitation, but scientists are also considering highly non-planar targets in the largest canyon in the Solar system Valles Marineris. Through extensive testing in image-based simulations using actual terrain structure and textures collected in Mars orbit, we demonstrate that only our new range-VIO approach is able to estimate terrain-relative velocity within the mission requirements compared to state-of-the-art approaches from both the robotics and planetary exploration literature.