An integrated planning of exploration, coverage, and object localization for an efficient indoor semantic mapping

Abstract

This paper describes an integrated viewpoint planner for indoor semantic mapping. Mapping of an unknown environment can be viewed as an integration of various activities: exploration, (2D or 3D) geometrical mapping, and object detection and localization. An efficient mapping entails selecting good viewpoints. Since a good viewpoint for one activity and that for another could be shared or conflicting, it is desirable to deal with all such activities at once, in an integrated manner. We use a frontier-based exploration, an area coverage approach for geometrical mapping, and object recognition model-based verification for generative respective viewpoints, and get the best next viewpoint by solving a travelling salesman problem. We carry out experiments using a realistic 3D robotic simulator to show the effectiveness of the proposed integrated viewpoint planning method.