Incorporating Planning into BDI Systems

Many architectures of autonomous agent have been proposed throughout AI research. The most common architectures, BDI, are procedural in that they do no planning, seriously curtailing an agent’s ability to cope with unforeseen events. In this paper, we explore the relationship between propositional planning systems and the process of means-ends reasoning used by BDI agents and define a mapping from BDI mental states to propositional planning problems and from propositional plans back to mental states. In order to test the viability of such a mapping, we have implemented it in an extension of a BDI agent model through the use of Graphplan as the propositional planning algorithm. The implemented prototype was applied to model a case study of an agent controlled production cell.