Direct Execution of Agent Specifications

Abstract

Although agent-based systems are beginning to be used in critical applications, it is clear that more precise, and logically well-founded, development techniques will be required for agent-based applications in the future. Thus, our aim is to provide a formal framework supporting the principled development of agent-based systems, and comprising: logics in which the high-level behaviours (of both agent and system) can be concisely specified; techniques for refinement and verification of such specifications; and a programming language providing concepts close to the specification notation used. While the specification and verification aspects are covered elsewhere, this talk concerns the programming of agent-based systems based upon the direct execution of specifications. The core formalism we use is temporal logic, which is both simple and intuitive, yet powerful. This enables us to specify the basic dynamic attributes of both the agent and the multi-agent environment. Thus, the basic specification of an agent’s behaviour is given as a temporal formula, then transformed into a simple normal form, called Separated Normal Form (SNF), which provides a simple and intuitive description of what is true at the beginning of execution, what must be true during any execution step, and what constraints exist on future execution states [3]. Given an agent specified using SNF, we use a forward-chaining process on a set of SNF clauses in order to construct a model for the specification, and thus animate the agent specification [2]. Further, temporal specifications of the above form can be extended with aspects of deliberation [4] and with a dimension of (bounded) belief [6].