Robert E. Wray, Sean Lisse, and Jonathan Beard
Ontologies provide useful technology for organizing and managing large-scale knowledge bases and enabling interoperability in heterogeneous agent environments. However, autonomous systems require not only large knowledge bases and knowledge sharing; they also require efficient run-time performance. In agents optimized for performance, control structures and domain knowledge are often intertwined, resulting in fast execution but knowledge bases that are brittle and scale poorly. Our hypothesis is that combining ontology representations and tools with agents optimized for performance will capitalize on the strengths of the individual approaches and reduce their weaknesses. Our strategy is to use automatic translators that convert ontological representations to agent representations, hand-coded agent knowledge for ontological inference, and explanation-based learning to cache ontological inferences. The paper outlines the rationale for this approach and design decisions and trade offs encountered. We also discuss criteria for evaluating success and understanding the consequences of design decisions on agent performance and knowledge base manageability.