A new omnidirectional vision sensor for the spatial semantic hierarchy

Abstract

We propose a new sensor for the spatial semantic hierarchy created by Kuipers (2000). To prove the effectiveness of this new sensor it has been used as a sole sensor for a robot. The task of the robot is to build a topological map of an unknown environment using the spatial semantic hierarchy. In the paper, we present the strict link that it is possible to create between the spatial semantic hierarchy and the omnidirectional images. We propose a set of topological events that it is possible to identify in the sequence of images acquired while the robot is moving. These topological events can be used to pose a discrete set of places that will be the nodes of the topological map. The results of simulated experiments, and the experiments we carried out with a real robot, show the feasibility of this approach. We are also testing a new multi-part mirror expressively designed for this application.