Components of Learning in a Reactive Environment

Abstract

The goal of the C—MU/UCI World Modelers Project [Carbonell (this volume)] is to develop an integrated model of learning in a complex, reactive environment — in particular, a simulated physical world that contains three—dimensional objects. Within this framework, we are designing cognitive architectures within which to cast our theories of learning, and in this paper we outline one such architecture. The paper is organized around four types of learning that we believe are central to dealing with reactive environments — the formation of object concepts, the acquisition of procedures, the construction of cognitive maps, and the discovery of problem solving heuristics. In each case, we discuss issues of representation and performance in addition to problems of learning.