Nui, Galway Faculty of Engineering Research Day 2004

1 1 I. INTRODUCTION In this paper we consider an approach to scene understanding based on multiple cooperative robots. Object recognition is combined with cooperative mapping and exploration such that a team of robots can map a complex environment in terms of the objects within it. Since the ability of a robot to carry out chores such as fetching specified objects in domestic environments is becoming increasingly important due to aging populations [1], the demand for intelligent robots with an understanding of their environment is growing. For a robot to act in an intelligent and flexible way, it will have to be able to recognize objects and learn the identity of unknown objects, as well as the properties of these objects. The most suitable sensory model on which to build an understanding of a scene is vision. Vision provides both 2D data and colour information, which can be used to extract 3D data. An important step towards this is the ability to recognize objects and model their location in the environment. Using a team of robots equipped with cameras offers greater flexibility than a single robot, as they can be used to view objects from different angles. Stroupe et al. [2] state that right angles should be used to view an object in order to optimally estimate its location. As real world objects, e.g. a stapler, often appear different when viewed from different sides, multiple cameras allow for the robust use of object recognition techniques that deal with the appearance of objects. Cooperative robots have been widely recognized as a more suitable alternative to single robots in many problem domains, providing concurrency, improved reliability, and more accurate map building. This paper outlines how coordination of a team of single-camera robots can also be applied to identification and modeling of objects.