3D Global Localization of a UAV Using 2D Monte Carlo Localization and Ground Plane Extraction

Abstract

Estimating the 3D pose of an aerial robot is more computationally expensive than estimating it’s 2D pose as there is significantly more data to process with the additional dimension. As such, it is more difficult to achieve real-time updates using a 3D adaptive Monte Carlo localization (AMCL) algorithm than a 2D AMCL. The ideal solution would be a combination of the additional dimension of a 3D AMCL with the lower processing time of a 2D AMCL. In this paper, we propose a method for reducing the cost of using AMCL for estimating 3D pose of an aerial robot by applying a point cloud conversion. This conversion involves extracting the ground plane from a scan and using it to estimate vertical position, while flattening the remainder of the scan into 2D to be used for odometry and a 2D AMCL. Through a simulated test flight it is shown that with this approach it is possible to estimate 3D pose accurately with low processing time.