Integrating classical and reactive planning within an architecture for autonomous agents

Abstract

In this paper we describe an architecture for agents that have to act in dynamic environments. The architecture, which has been partially implemented in Common Lisp, consists of a number of layers, each of which occupies a different position on the flexibility versus computational tractability trade-off. By integrating a high-level declarative planner with a reactive planner using situated-action rules, the architecture combines the flexibility of the former with the efficiency of the latter. Moreover, the lower levels of the architecture can be automatically configured on the basis of previous activities of the higher levels, thus avoiding the need for providing a complete set of situated-action rules that is inherent in a purely reactive planning approach.

We would like to thank George Kiss and Nigel Shadbolt for valuable discussions. We would also like to thank a number of anonymous referees who make useful comments on an earlier draft of this paper. The first author is financially supported by an SERC postgraduate studentship.