The Bond Agent System (1)

1
The Bond Agent System(1)

• EEL 5937 Multi Agent Systems
• Lecture 8, Jan 30, 2003
• Lotzi Bölöni

2
Origins

• Bond 1 1995-96, TCL based agent system
• Purdue University
• Structural biology applications
• Bond 2 1996-2001
• Purdue University
• Java based, general purpose distributed object
  system and agent framework
• Handwritten communication frameworks, persistency
  models etc.
• Introduced the notion of agent surgery
• Bond 3 2002-
• General purpose agent system
• Testbed for agent research
• Integration of best of breed technologies

3
Architecture
4
Builds on Protégé-2000 ontology framework

• Uses the ontology framework as a knowledgebase
  library.
• Every agent has a knowledgebase which is relying
  on a set of ontologies which can be pre-edited in
  Protégé-2000.
• The Protégé forms are used as data entry widgets
  in Bond

5
Builds on Jade agent framework

• Bond takes from the Jade agent framework
• The communication framework
• The message formats
• The directory service
• The agent containers
• Bond builds on
• The Jade behavior architecture.
• Things which Bond is not using / replaces
• The Jade Java-class based ontology
  implementations.
• Generally speaking, everything written in the
  Jade user manuals will still work in Bond.

6
What does Bond add to the underlying components

• Multiplane state machine organization
• Blueprint agent description language
• Mutability, agent surgery
• Strategies metadata extended behaviors
• Collection of strategies (strategy data base)

7
Blueprint. Describing the structure of a Bond
agent.
8
Blueprint

• Python based language. Describes the structure of
  the agent.
• Describes the state machine of the agent.
• Describes the strategies associated with the
  states.
• Describes the ontologies to be used.
• WARNING It is a work in progress. New features
  will be added during the semester.

9
Blueprint example

• createDefaultAgentUI()
• includeKnowledgeBase("Sample")
• create a plane which goes round and round
• createPlane("RoundAndRound")
• s bond.agent.strategy.strategydatabase.
  WaitAndTransitionStrategy(agent, 5000, SUCCESS)
• addFirstState(s, "Round1")
• s bond.agent.strategy.strategydatabase.
  WaitAndTransitionStrategy(agent, 5000, SUCCESS)
• addState(s, "Round2")
• s bond.agent.strategy.strategydatabase.
  WaitAndTransitionStrategy(agent, 5000, SUCCESS)
• addState(s, "Round3")
• addTransition('Round1', 'Round2', SUCCESS)
• addTransition('Round2', 'Round3', SUCCESS)
• addTransition('Round3', 'Round1', SUCCESS)

10
Result.

• -- the agent presented here.

11
Ontologies in Bond.
12
Ontology hierarchies in Bond.

• The Bond core ontology contains descriptions of
  the basic entities involved in building an agent
• Contains classes for agent description, beliefs,
  desires, intentions, actions
• Simple ontologies for time and space
• Domain specific ontologies contain the classes
  involved in the description of a specific domain
• E.g. computational grid ontology
• You will need to develop your own ontologies for
  your projects.
• Agent specific ontologies contain the initial
  knowledgebase of an agent
• Typically, they do not define new classes, but
  contain instances for the initial values.

13
Specifying ontologies in Blueprint

• The BondCore ontology is loaded by default
• For the domain specific and agent specific
  ontologies, you need to specify them.
• The ontologies must be in the kb subdirectory.
• includeKnowledgeBase(GridComputing")
• includeKnowledgeBase(Security")
• includeKnowledgeBase(AgentSpecificOntology")

14
Strategies in Bond
15
Behaviors

• Behaviors in Jade are the basic active units.
• Java classes with a couple of functions
• void action()
• Contains the main operation of the behavior
• It is called repeatedly until the behavior
  terminates
• Boolean done()
• It should return true, if the strategy has
  terminated
• Void onStart()
• Initialization performed when the strategy is
  first executed.
• Int onEnd()
• Executed when the strategy is terminated
• The return value of this strategy will be used to
  perform the transition on the finite state
  machine.

16
Strategy Behavior Metadata

• Behaviors are the hard-coded functionality.
  Basically, hand-written code.
• Strategies add meta-data to the behaviors
• Describe their functionality
• Their resource requirements
• Performance
• Pre- and post-conditions
• Runtime introspection
• The goals are
• Automatic assembly of agents
• Automatic reconfiguration
• Reasoning about agents.
• WARNING This is work in progress. The meta-data
  attached to strategies will change in the course
  of the semester.

17
Example strategy

• Wait and transition.
• Waits for the specified number of milliseconds,
  then performs the transition specified.

18

• package bond.agent.strategy.strategydatabase
• import bond.agent.strategy.Strategy
• import java.util.logging.Logger
• import bond.agent.BondAgent
• public class WaitAndTransitionStrategy extends
  Strategy
• private static final Logger LOGGER
  Logger.getLogger("bond.agent.strategy.strategydata
  base")
• private long theDelay
• private int transition
• private long startTime
• public WaitAndTransitionStrategy(BondAgent
  theAgent, long theDelay, int transition)
• super(theAgent)
• this.theDelay theDelay
• this.transition transition

19

• public void onStart()
• startTime System.currentTimeMillis()
• super.onStart()
• public boolean done()
• long currentTime System.currentTimeMillis()
  
• if (currentTime - startTime gt theDelay)
• return true
• return false
• public int onEnd()
• startTime 0L
• return transition

20

• public void action()
• if (startTime 0L)
• onStart()
• block(500) // block for 1/2 second