Abstract not found

The paper discusses an architecture for intelligent agents based on the use of A-Prolog - a language of logic programs under the answer set semantics. A-Prolog is used to represent the agent's knowledge about the domain and to formulate the agent's reasoning tasks. We outline how these tasks can be reduced to answering questions about properties of simple logic programs and demonstrate the methodology of constructing these programs. Keywords: Intelligent agents, logic programming and nonmonotonic reasoning. 1 INTRODUCTION This paper is a report on the attempt by the authors to better understand the design of software components of intelligent agents capable of reasoning, planning and acting in a changing environment. The class of such agents includes, but is not limited to, intelligent mobile robots, softbots, immobots, intelligent information systems, expert systems, and decision-making systems. The ability to design intelligent agents (IA) is crucial for such diverse tasks as ...

The Situation Calculus is a logic of time and change in which there is a distinguished initial situation and all other situations arise from the different sequences of actions that might be performed starting in the initial one. Within this framework, it is difficult to incorporate the notion of an occurrence, since all situations after the initial one are hypothetical. These occurrences are important, for instance, when one wants to represent narratives. There have been proposals to incorporate the notion of an action occurrence in the language of the Situation Calculus, namely Miller and Shanahan’s work on narratives [22] and Pinto and Reiter’s work on actual lines of situations [27, 29]. Both approaches have in common the idea of incorporating a linear sequence of situations into the tree described by theories written in the Situation Calculus language. Unfortunately, several advantages of the Situation Calculus are lost when reasoning with a narrative line or with an actual line of occurrences. In this paper we propose a different approach to dealing with action occurrences and narratives, which can be seen as a generalization of narrative lines to narrative trees. In this approach we exploit the fact that, in the discrete Situation Calculus [13], each situation has a unique history. Then, occurrences are interpreted as constraints on valid histories. We argue that this new approach subsumes the linear approaches of Miller and Shanahan’s, and Pinto and Reiter’s. In this framework, we are able to represent various kinds of occurrences; namely, conditional, preventable and non-preventable occurrences. Other types of occurrences, not discussed in this article, can also be accommodated. 1

Some of the recent work on representing action makes use of high-level action languages. In this paper we show that an action language can be represented as the sum of two distinct parts: an "action description language" and an "action query language." A set of propositions in an action description language describes the effects of actions on states. Mathematically, it defines a transition system of the kind familiar from the theory of finite automata. An action query language serves for expressing properties of paths in a given transition system. We define the general concepts of a transition system, of an action description language and of an action query language, give a series of examples of languages of both kinds, and show how to combine a description language and a query language into one. This construction makes it possible to design the two components of an action language independently, which leads to the simplification and clarification of the theory of actions. 1 Introducti...

In this paper we suggest an architecture for a software agent which operates a physical device and is capable of making observations and of testing and repairing the device components. We present novel de nitions of the notions of symptom, candidate diagnosis, and diagnosis which are based on the theory of action language AL. The new de nitions allow one to give a simple account of the agent's behavior in which many of the agent's tasks are reduced to computing stable models of logic programs. 1

this paper we address the problem of diagnosing faulty behavior of a dynamic system. Similar to the work in (Baral, McIlraith & Son 2000) our approach is based on the semantics of action languages (Gelfond & Lifschitz 1998). Unlike (Baral, McIlraith & Son 2000) we do not use action language L. Instead we assume that the set of all possible trajectories of the system is described by a system description of an action description language AL d (Baral & Gelfond 2000) (possibly augmented by logic programming rules of A-Prolog). We believe that this language is sufficiently powerful for a wide range of applications. The use of AL d allows us to give a definition of diagnosis which, we believe, is simpler than related definitions of (Baral, McIlraith & Son 2000). Simplicity of the definition and the recent discoveries of the close relationship between A-Prolog and reasoning about effects of actions allow us to develop a collection of simple algorithms for computing diagnoses. The algorithms are based on the ideas from answer set programming (Marek & Truszczynski 1999; Niemela 1999; Lifschitz 1999) and are implemented on top of SMODELS (Niemela & Simons 1996) - a system for computing stable models of logic programs. According to our definition all diagnoses are created equal. In practice however, some diagnoses are more equal than others. Selection of the "best" diagnosis is often based on some heuristic information about the plausibility of different types of faults. By way of example we show how our choice of A-Prolog as a knowledge representation language allows us to incorporate this heuristic information in the corresponding diagnostic systems. The heuristics improve the quality of diagnoses as well as efficiency of computation

This paper presents some preliminary work on causal reasoning about actions studying the causes of derived formulas in a given transition, in terms of subsets of the performed actions. We present a