Radar-Inertial Localization for Unmanned Ground Vehicles in GNSS-Denied Off-Road Environments

Radar is a growing candidate for localization in GNSS-denied scenarios that are challenging or restrictive to LiDAR. However, applications are still largely limited to indoor or urban environments. To bridge this gap, we look at the performance of radar-inertial-odometry in distinctly offroad environments that include a dense forest and underground mine. We develop a localization pipeline which uses an adaptive EKF for unstructured forested regions, and fuses both EKF estimates and point-to-plane radar ICP measurements via factor graph for structured underground environments. We observed high quality localization results using radar in these environments as well as improvements in accuracy when using our pipeline compared to current methods, particularly in the forested region where our AEKF was seen to be highly suitable for such terrains.