Grounded representations for a robots team

摘要

Realistic applications of autonomous robotics face a lot of difficulties in real environments. To navigate, self-localize and cooperate in such contexts, a multi-robot system has to find a way to represent and share knowledge about the world it is living in. We discuss the necessity and the advantages of grounding representations and communications of a robot team in real perceptions. We describe a technical framework designed for this purpose, which is based on the classification of the perceptions of a robot population. Percepts must have specific properties to enable grounding, and we present particular percepts well suited to this process. Finally we discuss the possible usages of grounded classes and propose a grid representation for fusion and exchange.