John E. Laird, Douglas J. Pearson, and Scott B. Huffman
Most work on adaptive agents have a simple, single layer architecture. However, most agent architectures support three levels of knowledge and control: a reflex level for reactive responses, a deliberate level for goaldriven behavior, and a reflective layer for deliberate planning and problem decomposition. In this paper we explore agents implemented in Soar that behave and learn at the deliberate and reflective levels. These levels enhance not only behavior, but also adaptation. The agents use a combination of analytic and empirical learning, drawing from a variety of sources of knowledge to adapt to their environment.