Abstract

This paper is concerned with providing a common framework for both the logical specification and execution of agents. While numerous high-level agent theories have been proposed in order to model agents, such as theories of intention, these often have little formal connection to practical agentbased systems. On the other hand, many of the agent-based programming languages used for implementing 'real * agents lack firm logical semantics. Our approach is to define a logical framework in which agents can be specified, and then show how such specifications can be directly executed in order to implement the agent's behaviour. We here extend this approach to capture an important aspect of practical agents, namely their resource-bounded nature. We present a logic in which resource-boundedness can be specified, and then consider how specifications within this logic can be directly executed. The mechanism we use to capture finite resources is to replace the standard modal logic previously used to represent an agent's beliefs, with a multi-context representation of belief, thus providing tight control over the agent's reasoning capabilities where necessary. This logical framework provides the basis for the specification and execution of agents comprising dynamic (temporal) activity, deliberation concerning goals, and resource-bounded reasoning.

Citations