Understanding the three-dimensional working environment is one of the most challenging tasks in robotics. Only by labeling perceived objects with semantics, a robot can reason about its environment, execute high-level plans and interact autonomously with it. A robot can perceive its enviroment by using 3D LIDAR systems, which generate 3D point cloud images of the environment. This data is perceived in a spatial domain, i.e. the raw data gives only positions of the measured points. The transfer from the spatial domain to the semantic domain is known as the gap problem in AI and one of the hardest to solve. In this paper we present an approach on how to extract spatial entities from unorganized point cloud data generated by a tilting laser scanner. Additionally, we describe how the extracted spatial entities can be mapped to entities in the semantic domain using feature descriptors. We also discuss, how a-priori knowledge about typical indoor environments can be used for semantic labeling. I
