Exception Handling in Goal-Oriented Multi-Agent Systems

1
Exception Handling in Goal-Oriented Multi-Agent
Systems
Ibrahim Çakirlar, Erdem Eser Ekinci and Oguz
Dikenelli Ege University Computer Engineering
Department Izmir, Turkiye
2
2
Overview

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

3
3
Introduction

• Exception handling is based on the exception
  concept that is defined in software systems such
  as an error occurred during execution flow of a
  program.
• Traditional technologies like Object Oriented
  Programming(OOP) try to satisfy reliability and
  robustness properties via exception handling.

4
4
Introduction

• Classical exception handling techniques can not
  be applied to MASs for the reason of some
  properties
• cooperative,
• autonomy,
• open,
• distributed.
• In this research we design a specialized
  exception handling approach for MASs.

5
5

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Other Exception Handling Approaches
• Case Study Electronic Barter Application
• Conclusion

6
6
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems

• In many cases and researches exception handling
  is considered as a later step of development
  process.
• Lamsweerde and Letier also claim using goals to
  model exception handlers at the requirement
  analysis phase of goal-oriented MAS development.
• Axel van Lamsweerde and Emmanuel Letier.
  Handling obstacles in goal-oriented requirements
  engineering. Software Engineering, 26(10)
  9781005, 2000

7
7
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems
8
8
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems

• Goal-oriented multi-agent system executable
  entities system goal, agent goal, plan.
• In MASs, execution flow designed with the terms
  of cooperation and standalone.

9
9
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems

• We propose a layered exception hierarchy like
  Platon.
• Exception levels
• Multi-Agent Level
• Agent Level
• Plan Level
• Eric Platon, Nicolas Sabouret, and Shinichi
  Honiden. An architecture for exception
  management in multi-agent systems. International
  Journal on Agent-Oriented Software Engineering,
  2008

10
10
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems

• We categorized exception levels due to hierarchy
  of executable entities.

11
11
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems

• In our principle, conveniently with the
  simplicity principle of Goodenough, claim that
  each exception should be handled with the
  executable entities that are same level with the
  exception occurred in.
• If it can not be handled in the same level, it
  hampers the execution flow of one upper level.
  So,it should be handled in the upper level.
• John B. Goodenough. Exception handling design
  issues. SIGPLAN Not., 10(7)41 -45, 1975

12
12
Exception Handling Mechanism for Goal-oriented
Multi-agent Systems
13
13

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

14
14
Abstract Exception Handling Architecture of MASs

• Three phase of agent cyclic execution model
  perception, reasoning and action.
• As other operations exception handling spreads
  over the same execution model.
• Perception phase ? Reasoning Phase ? Action Phase

15
15
Abstract Exception Handling Architecture of MASs

• Perception phase Perception of exceptional/
  unexpected situation.
• Reasoning phase Decide exception cause,
  occurance level and handling level
• Action phase Apply decided exception handling
  policy

16
16

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

17
17
Our Implementation

• Our abstract exception handling architecture is
  implemented in SEAGENT Multi-Agent Development
  Framework.
• The primitive properties of SEAGENT are semantic
  web enability, goal orientation and HTN based
  plan execution.

18
18
Our Implementation - Decision making Cycle
All exceptions are occur in this step.
Plan Level is handled in this step. But for other
levels the cycle goes on.
19
19
Our Implementation Execution Sample
Goal Model
Plan Model
Dynamically added when an exception occurs
20
20

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

21
Case Study Electronic Barter Application
Customer Agent1
Barter Management Agent
Notify for organizing Product Transport
Cargo Agent
Barter Proposals
Transport information is notified.
Sends evaluation result
Notifiy Customer Agent for evaluating the Barter
Requests
matching
Transport Information
Customer Agent2
22
22
Case Study Electronic Barter Application Goal
Model of exchangeBarterProducts System Goal
23
23
Plan Level Exception Handling Definition
HTN Plan for evaluateBarterProposal agent goal
Plan Level Exception Definition For HTN Tasks
Exception Handling Task for UnavaibleDatabase Exce
ption
24
Agent Level Exception Handling
Customer Agent1
Barter Management Agent
Notifiy Customer Agent for evaluating the Barter
Requests
Barter Proposals
When an exception occurs the regular execution
flow is changed.
Customer agent cancels the proposal
matching
The proposal of crashed agent is cancelled
Participant agents are notified
Customer Agent2
25
25
Agent Level Exception Handling Goal Model
Exceptional agent goal definition
26
26
Multi-agent Level Exception Handling
inverseOf
inverseOf definition between system goals
27
27

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

28
28
Other Exception Handling Approaches

• We categorize other exception handling approaches
  in two main category.
• Organizational View
• Agent View

29
29
Other Exception Handling Approaches

• Organizational View These approaches bring some
  organizational entities to MAS and these entities
  monitor agents by listening their internal events
  and messages to detect exceptional situations.
• Centralized
• Decentralized

30
30
Other Exception Handling Approaches

• Centralized organizational view define an
  exception handling entity in the architecture of
  MAS for handling exceptions.
• Tripathi and Miller's guardian agent
• Klein and Dellarocas's Exception Handling
  Service(EHS).
• Decentralized organizational view spread
  responsibility of handling exceptions through
  more than one organizational entity.
• Haegg's sentinel agents.

31
31
Other Exception Handling Approaches

• Agent View These approches listen messages and
  events of the agent, but they propose a mechanism
  to handle the exceptions inside the agent.
• SaGE framework.

32
32
Comparison of Exception Handling Approaches

• Organizational views definitions are weak,
  they need so much information and resource
  (agents/services) for handling exceptions.
• Decentralized views need an agent for each agent.
• Organizational views do not consider agent
  internal knowledge.
• Organizational views do not support agent
  autonomy.
• Agent views focus on agent internal exceptions,
  do not consider organizational exceptions.

33
33

• Introduction
• Exception Handling Mechanism for Goal Oriented
  MASs
• Abstact Exception Handling Architecture for MASs
• Our Implementation
• Case Study Electronic Barter Application
• Other Exception Handling Approaches
• Conclusion

34
34
Conclusion

• We propose an exception handling approach for
  goal-oriented MASs.
• Our approach is classified in agent view but MAS
  Level Exception make it applicable in
  organizational view.
• And we propose life cycle that supports autonomy
  of agents.
• Its obeserved with the case study that our
  approach support the robustness and reliability
  of MASs.

35
35
Thanks, any questions?