Self-organizing visual perception for mobile robot navigation

Abstract

Visual navigation for mobile robots is almost always performed on the basis of complex CAD models which have to be given to the system in advance. There are some approaches that map visual information more or less directly to robot actions in order to implement some basic capabilities like road following or docking. Complex navigation task, e.g. in buildings, however, are usually solved using three dimensional models of the environment or the objects within it. This paper presents a method for self-organizing the visual perception of a mobile robot in order to adapt it to the surroundings without the need to define and model the relevant aspects of the environment. The system uses a self-organization process to transform the continuous pour of images into a limited number of discrete perceptions which can be used far navigation purposes. Results obtained in an indoor office environment are presented. Nevertheless, the approach is suited also for outdoor navigation.