Extending UML for Modeling and Design of MultiAgent Systems

1
Extending UML for Modeling and Design of
Multi-Agent Systems

• Krishna Kavi University of North Texas
• David Kung University of Texas at Arlington
• Hitesh Bhambhani University of Texas at
  Arlington
• Gaurav Pancholi University of Texas at
  Arlington
• Marie Kanikarla University of Texas at Arlington

2
Presentation Topics

• Introduction
• Agent Versus Object
• BDI Formalism
• UML for MAS
• New Modeling Constructs
• New Diagrams
• Example Application
• Related Work
• Conclusions and Future Work

3
Introduction

• Agents are being used in an increasingly wide
  variety of applications from simple e-mail filter
  programs, to complex mission control and safety
  critical systems.
• In this paper we propose a framework and
  extensions to UML to support MAS development.
• The main motivation for developing the framework
  is to provide a MAS modeling language to help
  application engineers to focus their effort on
  agent-oriented modeling rather than having to
  define and construct an agent model from the
  scratch.

4
Agent Versus Objects

• Agent-oriented programming is the next evolution
  of OO programming.
• Objects are passive components whereas agents are
  autonomous.
• Objects are reactive whereas agents are proactive
  and situation-aware.
• Beliefs are similar to attributes
• Plans are similar to methods
• Goals have no object oriented equivalent

5
BDI Formalism

• The BDI architecture associates with agents,
  beliefs, desires or goals to achieve , and
  intentions or plans to act upon to achieve its
  desires.
• In formal terms, one can utilize logic to
  describe these components and reason about MAS
• In practical terms, beliefs can be viewed as the
  state of the world. Desires may be associated
  with a value so that they can be prioritized.
  Intentions reflect the actions that must be
  exercised to achieve the goal values.

6
Conceptual Model

• .

Environment
Beliefs
Beliefs
FIPA
Goals
Plans
Plans
Goals
KQML
Environment
Blackboard
Plans
Beliefs
Goals
7
Extending UML for MAS

• UML is excellent for object-oriented
  (OO)systems, But it lacks the capability to model
  and specify multi-agent systems.
• So extending UML such that agent system
  developers have a "common" language for
  visualizing, specifying, constructing and
  documenting the artifacts of an agent based
  system.
• The modeling constructs that are introduced are
  Agent, Belief, Goal, Plan, FIPA Performative,
  KQML Performative, and Blackboard.

8
Extending UML for MAS (contd..)

• Agent is the basic modeling construct
• Application specific BDI agent types are
  implicitly defined as subclasses of Agent and
  hence, they inherit the model-defined structural
  and behavioral features and relationships.
• The framework here enforces the BDI model and
  also provides the flexibility for re-using an
  existing design and implementation.

9
New Modeling Constructs

• Agent
• Situated performs actions according to input
  received from its environment and these action
  might change the environment.
• Autonomous - operates without the direct
  intervention of humans or other agents.
• Reactive - must take timely actions in response
  to changes in the environment.
• Proactive - not only react, but also exhibit
  goal-oriented behavior

10
New Modeling Constructs (contd..)

• Belief
• Beliefs are the agents observations and/or
  sensing of the environment. Or, it is the
  description of the state of the world (state as
  viewed by an agent)
• Every belief has a list of goals that may be
  affected by the changes to the belief, and is
  updated by the sensors or other agents.
• When a belief instance is changed , the affected
  Goal instances are informed.

11
New Modeling Constructs (contd..)

• Goal
• Goals specify the states of the world that are
  desirable
• Every goal is associated with a utility
  value,which indicates how achievable the goal is,
  and how valuable the goal is to the overall
  system. ( i.e. goals can be prioritized.)
• Changes to the goals utility values result in
  pre-empting some plans and initiating new plans.
• Goals can be proactive or reactive. Proactive
  goals reflect the desires of an agent, and
  Reactive goals reflect how an agent can be
  situated in an environment.

12
New Modeling Constructs (contd..)

• Plan
• Plans are sequences of actions generated by the
  agent to change the state of its environment.
• Plans can be implemented by the command pattern.
  The generic command class may implement the
  thread and the command subclass each implements
  an action of the agent.
• Execution of plans affects the environment which
  in turn changes the beliefs.

13
New Modeling Constructs (contd..)

• Blackboard
• Black board is shared communication mechanism.
• This is a concrete class, to permit the use of
  shared blackboards. Agent can define polymorphic
  methods for reading, reading and removing,
  writing, appending to the blackboard.
• FIPA- ACL / KQML
• These are Communication Languages. The beliefs
  which may be shared and modified by other agents
  is achieved by these Languages.

14
New Diagrams

• Agent Goal Diagram(AGD)
• AGD can also illustrate roles of an agent
• Depicts the goals of an agent and their
  relationships to the environment
• Use Case Goal Diagram (UCGD)
• UCCD combines existing UCD and UGD
• Shows relationships between use cases and goals
• Provides a high level guidance to Agent Sequence
  Diagram(ASD)
• Agent Domain Model (ADM)
• Conventional System Domain Model is extended to
  include Agents

15
New Diagrams(contd..)

• Represents domain knowledge internal to an agent
• Agent Sequence Diagram(ASD)
• Depicts interactions among the beliefs, goals,
  plans and other objects of an agent
• Agent Design Diagram(ADD)
• Introduced to document the design of an agent
• Facilitates reuse of an agents design
• Agent Activity Diagram(AAD) and Agent state-chart
  Diagram(ASCD)
• Introduced to model the internal activity and
  info flows

16
Example Application

• Intelligent Elevator System(IES)
• In addition to common features, IES must optimize
  the service and minimize the total movements of
  all cars
• Optimized service can be accomplished by
  minimizing the response time to requests
• Distributed decision making among the elevator
  agents
• Follow the Unified Process to design the system

17
Example Application (contd..)

• Step 1 Identify use cases and goals from
  requirements
• Use Cases
• Go to floor
• Request Elevator
• Open Door
• Close Door
• Goals
• Minimize Response Time
• Minimize Movement
• Step 2 Refine use case and goal diagrams
• New use cases added and existing ones revisited

18
Example Application (contd..)

• Step 3 Refine system domain model and agent
  domain models
• System domain model constructed or refined
• For elevator example, it consists of objects
  representing various parts of an elevator
• Agent domain model captures application dependant
  beliefs, goals and plans
• Step 4 Specify system and agent sequence
  diagrams
• Each use case has a system level sequence diagram
• Agent sequence diagrams to show the intrinsic
  interactions within an agent

19
Use Case Goal Diagram
Elevator System
? Elevator Car Agent
Elevator Subsystem
Goto Floor
Minimize Turn Around Time
Passenger

Request Elev
Minimize Movement
Open Door
Close Door
20
Agent Domain Model
Request
1..
Requests
ElevCarState
floorNo
ElevatorCarAgent
direction
status
floorNo
direction
Create (g Goals, b Beliefs)
affects
SelectGoal () Goal
load
Has-plan
Has-plan
ServeRequest
ServeRequest
Minimize
MinimizeTurn
Plan2
Plan1
Movement
AroundTime
type reactive
execute()
execute()
typereactive
del
egate to
delegate to
ltltThreadgtgt
ltltThreadgtgt
ServReq
ServReq
Plan2Cmd
Plan1Cmd
execute( )
execute( )
21
Agent Sequence Diagram
ElevatorCarAgent
MinimizeTurnaroundTime
Minimize Movement
rRequests
xread()
update(x)
beliefChanged(self)
beliefChanged(self)
asynchronous
sgetElevCarState()
sgetElevCarState()
eval ()
eval ()
goalEvaluated(self)
goalEvaluated(self)
vgetUtilValue()
ugetUtilValue()
ugtv pursue()
out(x)
vgtu pursue()
create(r,s)
create(r,s)
execute()
exclusive-or
execute()
22
Related Work
23
Conclusions and Future Work

• A framework and the necessary extensions to UML
  to address model and design of MAS
• Agents are peers in decision making process
• Our approach allows flexibility of using FIPA,
  KQML or any other agent communication language
• Various diagrams are introduced
• Framework utilizes interfaces and abstract
  classes to provide flexibility in implementation
• In process to complete the implementation of the
  IES

24
Conclusions and Future work

• Plan to apply framework to modeling and design of
  intelligent agents for MavHome smart home project
• Design and implementation of a CASE environment