KNOWLEDGE ARCHITECTURE FOR ENVIRONMENT REPRESENTATION IN AUTONOMOUS AGENTS
Dario Maio, Stefano Rizzi · 2000
Abstract- Representing knowledge of the environment is a primary research issue in designing intelligent autonomous agents. In order to reach a satisfactory level of autonomy in executing repetitive or hazardous tasks, agents should be provided with a compact though effective method for modelling the environment at different abstraction levels. The paper proposes a knowledge architecture for the representation of environments where distinctive places can be identified. Knowledge is structured according to a taxonomy of layers, where each layer represents an abstraction of the environment which can be profitably used to carry out specific tasks. Different formalisms may be adopted for representing the different layers, so that the specific properties and advantages of each formalism can be exploited to best advantage; in particular, we adopt an analogic representation to achieve motion between distinctive places, and a symbolic representation to support high-level planning of paths in terms of sequences of distinctive places. The paper briefly discusses some solutions to problems of knowledge representation and task decomposition, and presents some experimental results. 1.