Interleaving Learning, Problem Solving, and Execution in the ICARUS Architecture

Abstract

In this paper, we review Icarus, a cognitive architecture that utilizeshierarchical skills and conceptsfor reactive execution in physical environments. In addition, we presenttwo extensions tothe framework. The rst involves the incorporation of means-ends analysis,which lets the systemcompose known skills to solve novel problems. The second involvesthe storage of new skills thatare based on successful means-ends traces. We report experimentalstudies of these mechanismson three distinct domains. Our results suggest that the two methodsinteract to acquire usefulskill hierarchies that generalize well and that reduce the eort requiredto handle new tasks. Weconclude with a discussion of related work on learning and prospectsfor additional research.