Mobile Robot 6-D Localization Using 3-D Gaussian Mixture Maps in GPS-Denied Environments

The mobile robot six-dimensional (6-D) localization process in GPS-denied scenarios, e.g., in a cave or a mine, is a challenging problem. This article presents the modification of a well-known literature method using Gaussian mixture maps to determine the robot pose in rough terrain and outdoor nonurbanized environment. To improve the accuracy in rough 3-D terrain, I extended the terrain description from a 1-D Gaussian mixture model to a 3-D Gaussian mixture model (3DGMM) combining terrain height and terrain inclination in two orthogonal directions. Using this approach, one can maintain centimeter accuracy even in the most demanding workspaces. Moreover, I replaced the 3-D scan matching process combined with ground-based smoothing by 6-D scan processing (therefore, the 3DGMM method is capable for both land vehicles and flying robots) using the point-by-point extended Kalman filter based on the work by Niewola and Podsedkowski. Thanks to the point-by-point correction approach, it is possible to run the system in real time and process a full point cloud without GPU-based computing up to 300 000 points per second using small computers dedicated to mobile robots.