An Architecture For Intentional Agents With Reactive Behavior

Alex Gain, Graham Haug

Abstract


This paper is centered on the design and implementation of intelligent agents. Recently, an architecture has been proposed which relies on the notion of intention. In this work, intentions are related to agent beliefs, goals, and action and are employed to solve a key problem at the center of intelligent behavior: selecting the next action for the agent to perform. However, this existing architecture results in agents which are driven only by their goals but cannot respond to domain triggers in a timely manner. For example, if an agent populates a room and the room is observed to be on fire, the agent’s next action should be to leave the room regardless of its current goal and plan. The goal of our research is to design and implement an architecture for intentional agents that can react to environmental triggers regardless of the current goals and plans and allow for the agent to resume goal-driven behavior afterwards. Our design is based on well-established knowledge representation paradigms: Answer Set Prolog (ASP) and Action Language (AL). These languages have been very effective for describing possible agent actions and details of an agent’s environment, both of which contribute to the agent’s knowledge base. In order to expand the functionality of agents to allow for reactive behavior, we first modify Action Language (AL) to allow statements for triggered actions that describe how agents should respond to outside changes. We explicitly define the syntax and semantics for this newly modified AL. We extend the Theory of Intentions (TI) to allow these triggered actions to be treated as intended actions. We design and write ASP programs which will automatically implement any necessary reasoning tasks in the architecture. With these extensions, the architecture for intentional agents is expanded to allow agents featuring reactive behavior capability. Lastly, we discuss ideas for future improvements and expansions. Our research demonstrates that we can design intention-driven agents that can react to dynamic environments by utilizing Action Language and Answer Set Prolog.


Keywords


Artificial Intelligence, Knowledge Representation, Declarative Programming

Full Text: PDF

Refbacks

  • There are currently no refbacks.