Fuzziness in knowledge-based robotics systems

Abstract

This paper addresses how fuzziness is employed in a robotics system, for the purposes of object representation, object or feature location and as a means of representing actions performed on objects. Many, perhaps most, robotics systems do not take advantage of either regular data bases nor Solid Geometry data bases. This paper describes a robotic system which would use a multi-modal representation knowledge base (KB). The focus of the present paper is on the use of fuzziness in this robotic system. Keyboard entered natural language phrases would be used by a human user of the robot. The robot system would, in practice, translate these phrases into executable commands and coordinates. There are several points of departure that would be employed and the first of these is the use of ‘fuzzy descriptions’ rather than the conventional binary distinctive feature matrix (such as SHRDLU) or a point cluster approach. A fuzzy descriptor is a generalization of a geometric description of an object from the system's geometric data base. How the fuzzy descriptor works is described. Object or (geometric) feature location is handled by means of extrapolating the object description method. Named concrete or specific coordinate locations, or named abstract locations which involve a range of coordinates, are used to do this; by means of the familiar ‘linguistic variable’ which we all use in everyday speech. Some means whereby this is done is explained. The user of this robotic system may semantically attach linguistic variables to a fuzzy mapping of knowledge base object descriptions. The paper provides some details as to how a robotic system might be operated in an ‘action abstraction’ mode.