Grounding semantic maps in spatial databases

Abstract

Semantic maps add to classic robot maps spatially grounded object instances anchored in a suitable way for knowledge representation and reasoning. They enable a robot to solve reasoning problems of geometrical, topological, ontological and logical nature in addition to localization and path planning. Recent literature on semantic mapping lacks effective and efficient approaches for grounding qualitative spatial relations through analysis of the quantitative geometric data of the mapped entities. Yet, such qualitative relations are essential to perform spatial and ontological reasoning about objects in the robot's surroundings. This article contributes a framework for semantic map representation, called SEMAP, to overcome this missing aspect. It is able to manage full 3D maps with geometric object models and the corresponding semantic annotations as well as their relative spatial relations. For that, spatial database technology is used to solve the representational and querying problems efficiently. This article describes the extensions necessary to make a spatial database suitable for robotic applications. Especially, we add 3D spatial operators and a tree of transformations to represent relative position information. We evaluate the implemented capabilities and present real life use cases of SEMAP in different application domains. (C) 2018 Elsevier B.V. All rights reserved.