Introduction

    The problems addressed in this research are the lack of interoperability among agent systems, the lack of common design and evaluation methodologies, and the lack of a common definition for an agent or multi-agent systems.   The goal of this research is to formalize the capabilities that should be expected of agents as well as the inter-dependencies between these capabilities in order to establish a foundation on which all agents can be designed and built.  To encourage the rapid development of interoperable, compatible, and flexible agents, tools based on formalized design and evaluation methodologies will be produced to assist designers construct agent architectures and evaluate possible strategy implementations (i.e. design options for implementing agent capabilities).

Core Competencies

    An agent is characterized according to its functionalities, which can be grouped under six core competencies: sensing, modeling, organizing, planning, acting, and interoperating. In fact, all agents, including reactive agents, must implement each of these core competencies explicitly or implicitly. A reactive agent has an implicit role within an organization to sense raw data, to model the data by associating it to a value, to select a plan (rule), and to act out that selected plan. Interoperation with other agents is implicitly achieved by interacting with the environment.  Unfortunately, core competencies alone do not ensure agency but instead provide a measure of agency. For example, reactive agents are one type of agent, but more sophistication can be design to provide the capabilities 1) to merge multiple sensor inputs; 2) to create declarative, behavioral, and intentional models; 3) to change or adapt roles in an agent organization; 4) to plan efficient schedules with limited distributed resources; 5) to act in cooperation with other entities; and 6) to negotiate and coordinate with other agents. The science for selecting the required core competencies, understanding the inherent dependencies between those core competencies as well as understanding and evaluating techniques for implementation remains a research issue.

DACAT

    With the foundational concept of core competencies established, a toolkit called DACAT is being developed to facilitate the design and evaluation of various agent designs.  The first tool, called TLReader (Task List Reader), is used to help the designer build an agent architecture based on   core competencies from a list of tasks, which were decomposed from the desired functionalities. Each task has several attributes, including input, output, interaction, and the core competency to which the task is related. The list of tasks is translated into XML, which the TLReader deciphers and visualizes as nodes and connections. Nodes represent tasks, resources, and interactions, and the directed connections represent the nodes’ dependencies with other tasks, resources, and interactions.

    The TLReader tool also helps the designer organize the agent architecture on the basis of core competencies. Tasks that are related to the same core competency clump together, while tasks that are not related drift graphically apart. As a result, these clumps of functionality can be grouped together as a core competency module within the agent. By using the core competency modules as the agent architecture, the dependencies among module should be minimized by appropriate designer selections. As a result, plug-and-play of strategies is simplified because all dependencies are more easily identified. Recognizing these dependency among tasks, resources, and interaction gives the designer a better understanding of the inner workings of the agent.

    The core competencies-based organization is an intuitive abstraction of the functionality of an agent. Existing agents can be evaluated using core competencies by defining all the tasks the agent is capable of and using the TLReader to visualize the core competency-based architecture. Thus, evaluation of any agent is facilitated and comparisons among agents with different architectures are possible. At this stage, evaluation consists of an analysis of the dependencies within an agent and of the flexibility of the agent’s architecture.

Conclusion

The benefits of establishing the concept of core competencies include 1) sharing a common characterization of agents and having a measurement of agency, 2) a basis from which to derive methodologies for agent design and evaluation, and 3) the ability to modify a component of an agent with complete knowledge of its affects on other components of the agent. By providing tools for the creation and modification of agents, rapid development and interoperability among agents is facilitated.