Traversability_generator3d: 3D Traversability from MLS Maps (JOSS 2026)

Developer haider8645 has published a significant open-source contribution to robotic navigation with traversability_generator3d, a C++ library featured in the Journal of Open Source Software (JOSS 2026). The tool addresses a critical gap in autonomous systems by generating detailed 3D traversability maps from Multi-Level Surface (MLS) data, which preserves multiple surface hypotheses per map cell. Unlike traditional 2.5D elevation maps that flatten terrain, this approach enables robots to reason about complex overlapping structures like vegetation covering rubble, tunnels beneath bridges, or multi-story environments. The library represents a practical implementation for real-world robotic deployment where understanding vertical space is essential for safe navigation.

The library's technical pipeline includes RANSAC-based plane fitting for slope estimation, axis-aligned bounding box (AABB) checks for step height evaluation, orientation-dependent motion constraints, and frontier detection for bounded map expansion. A particularly advanced feature is optional soil-aware traversability, where cells maintain probability distributions for soil types (sand, gravel, concrete) that influence traversal costs through Gaussian spatial propagation models. The system outputs a unified 3D map structure that classifies each cell as TRAVERSABLE, OBSTACLE, FRONTIER, INFLATED_OBSTACLE, or UNKNOWN. This enables seamless integration with existing SLAM systems and path planners, providing a foundation for next-generation unmanned ground vehicles (UGVs) operating in construction sites, disaster zones, or agricultural environments where terrain complexity demands true 3D reasoning.