Schemas and Schema-based Architectures

In the last two decades several theoretical and computational models have been inspired, directly or indirectly, by theories of sensorimotor and cognitive development (Bates, 1979; Bickhard, 1998; Piaget, 1954) that describe schematic structures, or ‘schemas’, as crucial in behavior and cognition. The term ‘schema’ was firstly introduced by Bartlett (1932) to mean a map or structure of knowledge stored in long-term memory. Successively Piaget (1954) described schemas in a more operational sense, roughly as mental representations of some physical or mental actions that can be performed on an object or event. He considered schemas as the building blocks of thinking, and the basic structure underlying behavior and cognition (in a process that he described as ‘assimilation and accommodation’). One tenet of schema theory is that schemas are specialized subsystems realizing a tight coupling between perception and action. A schema can be used for recognizing a specific entity (say a dog) or a class of entities (say journalists or swimmers), or for controlling a specific action (say opening a door or skying). Some of these schemas may require parameters to be filled in. There can also be more complex schemas for planning sequences of actions, as well as for more complex cognitive operations such as doing inferences. Central to schema theory is not only what schemas can individually do, but also how they are organized and what they can collectively do. This view has inspired several other researchers in cognitive science, artificial intelligence and cognitive robotics. In these fields several schema-like structures have been proposed, including frames (Minsky, 1988), scripts (Schank and Abelson, 1977), schemas (Arbib, 1992, 1989; Drescher, 1991; Neisser, 1976; Norman and Shallice, 1986; Shapiro and Schmidt, 1982), neural schemas (Mccauley, 2002), semiotic schemas (Roy, 2005), and behaviors (Brooks, 1991; Maes, 1990). Architectures including distributed and competitive functional units are often referred to as ‘behavior-based’ or ‘schema-based’. Several integrated frameworks have been proposed for designing them; among the most popular ones, we can mention the behavior-based approach proposed in (Arkin, 1998), the NSL/ASL in (Weitzenfeld et al., 2000) and the Robot Schema (RS), a formal language for designing robot controllers proposed in (Lyons and Arbib, 1989) which includes perceptual and motor schemas.