Add to Chrome
Add to Firefox
Add to EdgeRange-Visual-Inertial Odometry: Scale Observability Without Excitation
Mar 28, 2021
Traveling at constant velocity is the most efficient trajectory for most robotics applications. Unfortunately without accelerometer excitation, monocular Visual-Inertial Odometry (VIO) cannot observe scale and suffers severe error drift. This was the main motivation for incorporating a 1D laser range finder in the navigation system for NASA's Ingenuity Mars Helicopter. However, Ingenuity's simplified approach was limited to flat terrains. The current paper introduces a novel range measurement update model based on using facet constraints. The resulting range-VIO approach is no longer limited to flat scenes, but extends to any arbitrary structure for generic robotic applications. An important theoretical result shows that scale is no longer in the right nullspace of the observability matrix for zero or constant acceleration motion. In practical terms, this means that scale becomes observable under constant-velocity motion, which enables simple and robust autonomous operations over arbitrary terrain. Due to the small range finder footprint, range-VIO retains the minimal size, weight, and power attributes of VIO, with similar runtime. The benefits are evaluated on real flight data representative of common aerial robotics scenarios. Robustness is demonstrated using indoor stress data and fullstate ground truth. We release our software framework, called xVIO, as open source.
