An abstract architecture for rational agents

Abstract

A number of architectures have been proposed for systems that are situated in dynamic environments and have to reason, plan, and act under stringent time constraints. In this paper, we consider architectures based on the notion of a rational agent, in which the system is viewed as having attitudes of belief, desire, and intention, which in turn influence its decision making and determine its behavior. A basic interpreter for situated systems is first described. This interpreter is then specialized to form an abstract belief-desire-intention (BDI) architecture, and it is shown how this interpreter satisfies some of the theoretical properties of proposed BDI formalisms. The abstract interpreter is developed further to produce different types of rational agents, based on variations in their commitment strategies. Finally, the design of a practical system is considered, taking into account the complexity limits on an agent's deductive and planning capabilities