PRACTIONIST: A NEW FRAMEWORK FOR BDI AGENTS V. Morreale a S. Bonura a G. Francaviglia a M. Cossentino b S. Gaglio cb a R&D Laboratory - Engineering Ingegneria Informatica S.p.A. b ICAR-Italian National Research Council c DINFO-University of Palermo Abstract In this paper, we present PRACTIONIST (PRACTIcal reasONIng sySTem), a new frame- work built on the Bratman’s theory of practical reasoning to support the development of BDI agents in Java (using JADE) with a Prolog belief base. We aims at reducing the gap between the expressive power of the BDI model and the difficulty of efficiently implementing its features. In PRACTIONIST we adopt a goal-oriented approach and stress the separation between the deliberation process and the means-ends reasoning, with the abstraction of goal used to formally define both desires and intentions during the deliberation phase. Moreover, PRACTIONIST agents are able to reason about their beliefs and other agent’s beliefs, since beliefs are not simple grounded literals or data structures but modal logic formulas. 1 Introduction One of the most interesting and popular agent theories is the Belief-Desire-Intention (BDI) model. According to the philosophical viewpoint, the internal states and the decision process of a BDI agent are modeled in terms of mental states, such as beliefs, desires and intentions, which respectively represent the information, motivational, and deliberative attitudes of the agent. The BDI model derives from the philosophical tradition of human practical reasoning, which was first developed by Bratman [2]. It states that humans decide, moment by moment, which actions to perform in order to pursue their goals. Practical reasoning involves two processes: (1) deliberation, to decide what states of affairs to achieve; and (2) means-ends reasoning, to decide how to achieve these states of affairs. Bratman’s theory stresses the role of intentions in human reasoning, as it states that intentions are important because they affect the selection of next actions to be executed. In the context of rational agents, the BDI model appears very attractive for several reasons. Firstly, the abstractions used in the model are really intuitive: the reader will agree that it is easy to understand the distinction between the process of deciding what to do and the one involving how to; similarly, the notions of belief, desire and intention are very familiar in human reasoning, so they could be easily used in BDI agent design. Moreover, this model provides a clear functional decomposition, which indicates what sort of subsystems might be required to build an agent. Nevertheless, the main issue in developing BDI agents is to figure out how to efficiently implement the above-mentioned processes [13]. The best-known BDI agent architecture is the Procedural Reasoning System (PRS), developed by Georgeff and Lansky [7]. Some concrete implementations of PRS have been developed, such as dMARS [6], developed at the Australian AI Institute, the UM-PRS implemented in C++ at the University of Michigan [9], and a Java version of PRS called JAM [8]. Finally, it is worth mentioning JACK [3], which is a commercially available programming language that extends the Java language with BDI features. It should be remarked that in the PRS and all above-mentioned descendant implementations, beliefs are treated as a collection of arbitrary data structures, desires are usually treated as events,