Handling of USAR Void-Space Visual Conditions by SLAM and SfM Algorithms

Determine how widely used online mapping algorithms in robotics—specifically ORB-SLAM2, NerfSLAM, and RTAB-Map—and offline Structure-from-Motion pipelines (e.g., COLMAP) handle dim and inconsistent illumination, sparse or low-contrast visual features, and short camera-to-scene working distances encountered inside the void spaces of collapsed structures typical of urban search and rescue operations.

Background

Robotics mapping and perception algorithms such as ORB-SLAM2, NerfSLAM, RTAB-Map (online methods) and Structure-from-Motion (offline methods) have been applied successfully in indoor, outdoor, and various confined spaces like pipes, mines, and industrial facilities. However, they have not been systematically evaluated within the void spaces of collapsed structures due to the lack of suitable, publicly available datasets.

Collapsed-structure void spaces present extreme visual challenges—dim and inconsistent lighting, feature-poor surfaces, and very short working distances—that differ substantially from typical benchmark environments. The paper introduces RubbleSim to generate photorealistic, controllable rubble environments to begin assessing these algorithmic limitations, but explicitly notes that the community still lacks knowledge of how these classes of algorithms handle such conditions.