An architecture for intentional agents



Journal Title

Journal ISSN

Volume Title



The goal of this research is to investigate the use of Answer Set Prolog (ASP) based languages and action languages in the design and implementation of intelligent agents. In particular we are interested in better understanding an agent's intentions and how they are related to the agent's beliefs, goals, and actions. There has been substantial work on intentions as part of a solution to a problem that is at the center of intelligent behavior: the problem of selecting the action to perform next. We believe that an agent architecture that includes intentions will allow us to improve the current methodology of ASP-based agent design. In this dissertation we will define an architecture for the design and implementation of agents whose behavior is determined by their intentions.

We divide this task into two parts. The first is describing a model of an intentional agent and its environment. The second is to incorporate such a model into an architecture, to formally describe reasoning tasks and behavior of intentional agents, and to create a prototype implementation.

The domain model is a transition diagram whose nodes represent the mental state of the agent and the physical state of the environment.
The former is described by a theory of intentions TI which is independent from the description of the environment. Both are described using action language AL The agent's reasoning tasks include explaining unexpected observations (diagnosis) and determining which of his actions are intended at the present moment. Intuitively, an intentional agent only attempts to perform those actions that are intended and does so without delay. We present a prototype implementation of the architecture based on a refinement of the architecture in which the reasoning tasks of the agent are reduced to computing answer sets of programs constructed from the agent's knowledge.



Intelligent agents, Intentions