Real‐Time SLAM with Octree Evidence Grids for Exploration in Underwater Tunnels

• DOI: 10.1002/rob.20165
• Publication Date: 2007-02-13T21:56:31Z
• Authors: Nathaniel Fairfield; George Kantor; David Wettergreen
• Keywords: Octree; Representation

Abstract We describe a simultaneous localization and mapping (SLAM) method for hovering underwater vehicle that will explore caves tunnels, true three‐dimensional (3D) environment. Our consists of Rao‐Blackwellized particle filter with 3D evidence grid map representation. procedure dynamically adjusting the number particles to provide real‐time performance. also how we adjust prediction step accommodate sensor degradation or failure. present an efficient octree data structure makes it feasible maintain hundreds maps needed by accurately model large environments. This can exploit spatial locality temporal shared ancestry between reduce processing storage requirements. To test our SLAM method, utilize collected manually deployed sonar vehicles in Wakulla Springs cave system Florida Sistema Zacato´n Mexico, as well DEPTHX tank at Austin Applied Research Laboratory. demonstrate approach these real‐world datasets. © 2007 Wiley Periodicals, Inc.