Impact of the Trajectory on the Performance of RGB-D SLAM Executed by a UAV in a Subterranean Environment

This paper empirically investigates the influence of trajectory design for autonomous Unmanned Aerial Vehicles (UAV) on the performance of Simultaneous Localization and Mapping (SLAM) in a subterranean environment that exhibits visual features similar to man-made and natural rock caverns. This was investigated by flying an autonomous UAV in a simulated cave environment, and also by deploying an actual UAV equipped with a RealSense L515-LiDAR to map a pillar inside a limestone mine. A popular open source SLAM software package – Real-Time-Appearance-Based Mapping (RTAB-Map) – was used. RTAB-Map has the ability to detect loop closures and has its approach to estimate an odometry solution. It was found that as the image overlap percentage increased, so did the number of loop closures. In average, we observed a 4.84% loop closure acceptance at 50% overlap and 49.57% loop closure acceptance rate at 90% overlap. Not only did the loop closure acceptance rate improve, there was evidence that lower overlap tended to lead to incorrect SLAM maps.