Traversability of Uneven Terrain : Perception for an Unmanned Ground Vehicle

Abstract

This thesis explores traversability analysis for an unmanned ground vehicle (UGV) operating on uneven terrain. Several open-source traversability estimation packages are evaluated, ranging from geometric heuristics and probabilistic approaches to appearance-based and multimodal/hybrid techniques, including supervised, self-supervised, and unsupervised learning paradigms. The chosen solution is deployed on a robotic UGV platform equipped with LiDAR and camera sensors. Online tests demonstrate its ability to effectively identify traversable areas in real-time.