Externalisation and Internalization: A New Perspective on Agent Modularisation in Multi-Agent System Programming

Abstract

Abstract-Agent modularisation is a main issue in agent and multi-agent system programming. Existing solutions typically propose some kinds of constructs -such as capabilities -to group and encapsulate in well-defined modules inside the agent different kinds of agent features, that depend on the architecture or model adopted-examples are goals, beliefs, intentions, skills. In this paper we introduce a further perspective, which can be considered complimentary to existing approaches, which accounts for externalizing some of such functionalities into the computational environment where agents are (logically) situated. In this perspective, agent modules are realised as suitably designed artifacts that agents can dynamically exploit as external tools to enhance their action repertoire and -more generally -their capability to execute tasks. Then, to let agent (and agent programmers) exploit such capabilities abstracting from the low-level mechanics of artifact management and use, we exploit the dual notion of internalization, which consists in dynamically consulting and automatically embedding high-level usage protocols described in artifact manuals as agent plans. The idea is discussed providing some practical examples of use, based on CArtAgO as technology for programming artifacts and Jason agent platform to program the agents.

Environment programming in multi-agent systems: an artifact-based perspective

“…The artifact will eventually generate a todo signal at the specified time, so as to allow the agent to react with a proper plan and perform the task. First investigations about this perspective are reported in [40].…”

“…Such a feature has been conceived in particular for open systems composed by intelligent agents that dynamically decide which artifacts to use according to their goals and dynamically discover how to use them. Actually, the notion of manual can be extended from artifacts to workspaces [40]: in that case manuals may contain the description of usage protocols that can involve multiple kinds and instances of artifacts.…”

Long-Lived Agents on the Web: Continuous Acquisition of Behaviors in Hypermedia Environments

“…MAS researchers have already examined the specification of operating instructions for artifacts in an attempt to design maintainable and extensible goal-directed autonomous consumers of resources (e.g. [11]). Following the methodology in [11], we enable autonomous agents to dynamically read and interpret high-level operating instructions that encapsulate low-level operations of artifacts.…”

“…[11]). Following the methodology in [11], we enable autonomous agents to dynamically read and interpret high-level operating instructions that encapsulate low-level operations of artifacts. Such information is explicitly described in artifact manuals and is meant to provide the agent with the operating instructions that are necessary to adequately exploit artifact functionalities.…”

Modularization in Belief-Desire-Intention agent programming and artifact-based environments

“…The agent oriented programming (AOP) paradigm ( Shoham, 1993 ) copes naturally with such requirements, reducing user intervention ( Maamar & Moulin, 1997 ) and promoting famed methods for facing complexity: decomposition, abstraction and hierarchy ( Jennings, 1999 ; Cuesta, Gomez & Gonzalez, 2008 ; Federico Bergenti & Zambonelli, 2004 ). Additionally, the agents and artifacts (A&A) meta-model ( Weyns, Omicini & Odell, 2007 ; Ricci, Piunti & Viroli, 2011 ) promotes a modular approach for programming the environment where agents are situated, providing a suitable model to externalize ( Ricci, Piunti & Viroli, 2009 ) the functionality required by agents to act and perceive. Unfortunately, current AOP proposals usually address this complexity taking the multi-agent system (MAS) as a whole, disregarding the individual agent level, despite an agent is intrinsically a complex system as well.…”