Spatial reasoning and semantic environment understanding is a fundamental ability of robots navigating in unstructured dynamic environments. Since spatial and semantic reasoning is tightly linked to the sensor perception information, it is desirable that it is directly integrated into the environment model. This paper presents the interplay of a novel environment representation called Semantic Rtree (SRTree) and Markov Logic Networks for reasoning. The SRTree is a semantic occupancy grid based on the hierarchical Rtree data structure that models the probability of occupancy of each grid cell and additionally assigns a class label to it. The main advantages of the proposed approach are (1) a hierarchical representation of large scale outdoor urban environments, which (2) captures both quantitative (metric) and qualitative (semantic) aspects of the environment and allows reasoning in a single data structure, and (3) the capability of dealing with higher-order spatial relations. The proposed methods are experimentally evaluated on a large scale 3D point cloud dataset of downtown Munich enhanced by RGB image data. The dataset is being made publicly available.     «

Spatial reasoning and semantic environment understanding is a fundamental ability of robots navigating in unstructured dynamic environments. Since spatial and semantic reasoning is tightly linked to the sensor perception information, it is desirable that it is directly integrated into the environment model. This paper presents the interplay of a novel environment representation called Semantic Rtree (SRTree) and Markov Logic Networks for reasoning. The SRTree is a semantic occupancy grid based o...     »