LiDAR-based Obstacle Type Analysis and Costmap Design for an Unmanned Ground Vehicle : Obstacle Classification and Costmap Generation

Sammanfattning

Unmanned Ground Vehicles (UGVs) are increasingly employed in a wide range of applications, from industrial to military exploration. Industrial applications often offer predictable, structured environments, whereas navigating rough, unstructured terrain, such as a forest environment, introduces significantly greater complexity. With autonomous navigation in complex terrain comes the challenge of environmental perception. The vehicle utilised in this thesis, built by the authors and other students at Mälardalen University (MDU), was designed to operate in a typical forest environment. The custom UGV motivated the development of a system that evaluates environmental features and utilises them to classify traversible and non-traversable terrain, as well as a pipeline for generating a costmap tailored to its physical characteristics. The research employed the System Development Method (SDM), which enabled an iterative development process suited to the exploratory nature of this work. The research questions served as the foundation for formulating system requirements, which in turn guided the implementation. A Robot Operating System2 (ROS2)-based pipeline was developed that integrates Light Detection and Ranging (LiDAR) point cloud processing, open-source mapping libraries, and methods for estimating roughness to generate elevation and traversability costmaps. The results demonstrate that point cloud height, density, and distribution can be used to estimate obstacle traversability, and that a costmap customised for the MDU UGV can be generated using the system design implemented in this thesis. During the final validation, the system correctly classified 4 out of 5 obstacles according to ground truth. The discussion highlights limitations regarding the threshold-based system design and proposes future work for point cloud cluster analysis and cell sizing in the costmap.