Architecting Agents Using Core Competencies

1
Architecting Agents Using Core Competencies
D. N. Lam K. Suzanne Barber
barber, dnlam_at_mail.utexas.edu http//www.lips.utex
as.edu The Laboratory for Intelligent Processes
and Systems The University of Texas at Austin
Design Agents with Ease, Confidence, and
Discipline
PC Context Example application
Communication transfer of messages between agents Communication transfer of messages between agents Communication transfer of messages between agents
  Sensing to tell other agents that a sensor signal will be transmitted
  Modeling to send beliefs and other types of information
  Planning to collaborate on generating plans involving several agents
  Acting to declare to other agents that a task is completed for synchronization
Organization determination of agent relationships Organization determination of agent relationships Organization determination of agent relationships
  Modeling to know who has authority over which agents
  Planning to know who makes the final decisions on choosing a plan
Coordination management of dependencies and conflicts Coordination management of dependencies and conflicts Coordination management of dependencies and conflicts
  Sensing to prevent interference when emitting signals to sense the environment
  Modeling to maintain a consistent models with other agents
  Planning to ensure that each agents plan does not conflict with other agents plans
  Acting to respond to changes in the environment during cooperative actuation
Example domain
  Functional Requirement Task(s)
1 sense radar interference and radar positions S sensor sense -gt data
1 sense radar interference and radar positions S data interpret -gt variables
2 send/receive messages from other agents Comm message send_message -gt
2 send/receive messages from other agents Comm receive_message -gt message
3 change radar frequency A command act -gt actuator
4 maintain beliefs about radar and positions of ships M variables update_beliefs -gt beliefs
5 create autonomy-based organization beliefs negotiate_on_organization -gt organization
6 decide on autonomy level Org beliefs, organization select_autonomy_level -gt autonomy level
7 decide on best planning strategy given current autonomy level P autonomy level select_planning_strategy -gt planning strategy
8 create plan to reduce frequency interference P beliefs, planning strategy create_plan -gt plan
9 create schedule to change frequency P plan create_schedule -gt schedule
10 detect conflicts in plans with other agents Coord plan, beliefs detect_plan_conflict -gt conflict
11 suggest solution for conflicts Coord conflict, plan suggest_conflict_solution -gt solution
Visualizer Dependencies among tasks, resources,
and interactions are visualized as dependency
graphs. Tasks associated with the same core
competency, automatically cluster together. Nodes
can be added and removed for model
accuracy. Tasks are grouped into agent modules
based on clustering or other metrics.
Task List Editor define agent tasks associated
with Core Competencies
Evaluator
2
Architecting Agents Using Core Competencies
Design Agents with Ease, Confidence, and
Discipline
D. N. Lam K. Suzanne Barber

• dnlam, barber_at_mail.utexas.edu
• http//www.lips.utexas.edu
• The Laboratory for Intelligent
• Processes and Systems
• The University of Texas at Austin

3
(No Transcript)
4
Agent Competencies
Core Competencies (CCs) establish the essential
agent functionalities, while Pluggable
Competencies (PCs) describe the additional
required functionalities for agents interacting
in a multi-agent system.
5
Usage of Pluggable Competenciesin the context of
each CC
6
Architecting Process
7
Example Domain
Given a set of domain-specific functional
requirements produced by domain analysis, the
architect identifies and extracts a task list
that reflects the domain-independent functional
requirements. Each task is associated with a
Core (or Pluggable) Competency. Task syntax
C input taskname -gt output
8
Task List Editor
The architect can explore possible agent tasks in
the Repository and select predefined tasks
and/or
modify and add to the task list directly using
the task syntax.
9
Visualizer
From the task list, the functional composition of
the agent is visualized as a dependency graph.
10
Missing or unconnected tasks are corrected to
produce a more accurate functional model of the
agent.
Tasks associated with the same core competency
automatically form clusters.
11
Evaluator
To assist in determining the encapsulation of
tasks into agent components, a dependency matrix
(displays the dependencies for each task and
component) and an evaluation table (displays the
coupling and cohesion values for the tasks and
components).
12
Architecture Specifier
The architecture specification consists of (1)
the components that were formed, (2) the tasks
that each component encapsulates and (3) inputs,
outputs, and interactions that the component may
have as a result of the encapsulated tasks. The
Architecture Specifier uses the task list and
component encapsulations to produce an XML file.