Reusable agent behaviours

1
Reusable agent behaviours

• Julien Subercaze

2
Summary

• Introduction to agents
• Problem of agents behaviours
• State of the art
• Using Web Semantic to model behaviours
• Agent Architecture
• Information and Knowledge issues
• Conclusion

3
Agents

• Some definitions for agent
• Autonomous entity with ability to interact with
  its environment. 9
• a computer system that is situated in some
  environment, and that is capable of autonomous
  action in this environment in order to meet its
  design objectives. 1
• agents are able to act without any external
  control they have their own control over their
  behavior and internal states in any possible
  environment 3
• Common points
• Environment
• Autonomy

4
Agents standards

• FIPA Foundation for Intelligents Physical
  Agents
• IEEE Computer Society standards organization
• Founded in 1996
• Provides standardsfor heterogeneous and
  interacting agents and agent-based systems
• Implementations respecting FIPA standards
• JADE http//www.jade.tilab.com
• SeMoA http//semoa.sourceforge.net
• Pastiche http//code.google.com/p/pastiche/
• Jack http//www.agent-software.com/

5
Agent architecture

• FIPA provides lots a specification concerning
  protocols, system architecture, but not for the
  agents themselves.
• Mainly, depending on the purpose differents
  architectures are used
• Context agents
• Mobile agents
• Gaming agents
• Decision agents
• Fuzzy agents
• .

6
Architecture for intelligent agents

• Intelligents agents
• Software agents
• Ability to take decisions according to its
  knowledge and the environment.
• Follow a goal
• -gt Belief Desire Intention Model
• Agent oriented abstraction 5

7
BDI Model
8
Open Problems

• Main Problem in heterogenous MAS is the
  interoperabilty.
• Reusability of components.
• Web Semantic Standards provides a solution for
  the knowledge modelling of the agent
• Represent the agents knowledge using ontologies
  6
• Architecture based on Semantic Model AgentOWL
  4

9
Knowledge representations

• Current MAS are dedicated to one type of
  knowledge
• KIF (Knowledge Interchange Format 6)
• Topic Maps 6
• OWL/RDF
• Uncertain knowledge (Ontobayes , extension of
  OWL 2)
• How to manage multiple knowledge representation ?

10
Storing different knowledge

• Behaviours can be represented using any
  representation
• Each representation requires an adapted reasoning
  mechanism
• Examples
• Rules Set requires Rule Engine
• Ontobayes requires Bayesian Engine

11
Agent model

• Guidelines
• Flexibility
• Reusability

12
Knowledge Base
We use a JCR repository to manage the different
knowledge representations. The knowledge Base
provides all the methods to manage the agent
knowledge
13
Behaviour execution I

• The agent load his core behaviour
• To complete his goal he will decide the best
  knowledge representation
• The related knowledge handler and reasoner will
  then be loaded

14
Behaviour execution II

• Behaviour can be seen as a finite automata or as
  a rule set.
• In the implementation, we used SWRL (Semantic Web
  Rule language) to implement the core behaviour,
  and Pellet as reasoner engine. 7
• Ontobayes decision support can be expressed using
  OWL-DL.
• Agent have all the same actions list, but
  different implementations.
• Execution of a rule (using Pellet) leads to the
  execution (or not ) of actions (rdfsequence) and
  gives the next rule to execute.
• As the agent have same actions, and KB
  management, the behaviours are exchangeable.

15
Knowledge exchange

• Using Virtual Knowledge Communities for the
  exchange between software agents
• Prototype of Hammond,Maret,Braure,Subercaze 7

Wake me up prototype 8
16
Knowledge and Information I

• These two terms are frequently used as synonyms
• But we need precise definitions
• Information constitute messages which a
  knowing mind may assimilate, understand,
  comprehend and incorporate into its own knowledge
  structures
• Knowledge knowledge involves the mental
  processes of comprehension, understanding and
  learning that go on in the mind and only in the
  mind, however much they involve interaction with
  the world outside the mind,and interaction with
  others.

17
Knowledge Acquisition Model - I

• One way to acquire knowledge from received
  information is to experience it
• Ground idea
• 1. Alice -gt Bob Information
• 2. Bob stores the information
• 3. Bob experiences the information
• 4. Bob acquires knowledge through the
  experience and store it

18
Knowledge Acquisition Model - II

• How to extract knowledge from the experience
• for software agents ? (Part 4)
• Use of machine learning, to discover hidden
  features from the feedback of the environment
• Inductive learning using statistical data
  features
• Deductive learning using pre existing knowledge

19
Future prototype ?

• Software agents video game bots
• In Quake 3 teams of bots can fight on a arena

20
Implementation Framework
21
Similar work

• ACASA (University of Pennsylvania)
• Project on Human Behavior Models (HBM)
• http//www.seas.upenn.edu/7Ebarryg/HBMR.html
• Use of a video game for the prototype Unreal
  Tournament

22
Future of the project

• Framework implementation
• Extend AgentOWL for behaviours reusability in
  top of JADE
• Provide a GUI for agent development
• Extend to other representation than OWL-DL
  SWRL and Ontobayes, for example implement Deontic
  Logic in SWRL
• Open issues
• SWRL development No working plugin for protégé
• High level architecture without related tools
  leads to high development time.

23
Bibliography

• 1 G. Weiss. 1999. Multiagent systems a modern
  introduction to distributed artificial
  intelligence, MIT Press.
• 2 Y. Yang, J.Calmet Decision Making in
  Ontology-Based Uncertainty Model. Proc. of 21st
  European Conference on Operation Research,
  Reykjavik, Iceland, 2006.
• 3 FIPA Abstract Architecture Specification
  http//www.fipa.org/specs/fipa00001/SC00001L.pdf
• 4 Michal Laclavik, Marian Babik, Zoltan Balogh,
  Emil Gatial, Ladislav Hluchý Semantic Knowledge
  Model and Architecture for Agents in Discrete
  Environments. ECAI 2006 727-728
• 5 J. Calmet, P. Maret, R. Endsuleit
  Agent-Oriented Abstraction,RACSAM (Revista  Real
   Academia de Ciencias, Serie "A" de Matemáticas)
   Vol. 98 (1), 2004.
• 6 J.Sowa Knowledge Representation Logical,
  Philosophical, and Computational Foundations .
  Brooks Cole Publishing Co., 1999.
• 7 D. Hodzev Bsc Thesis Needs analysis and
  improvements of an agent-based prototype. INSA
  Lyon, 2008.
• 8 J. Subercaze, P. Maret, N.M. Dang, K. Sasaki
  Context-aware applications using personal
  sensors.. Dans Bodynets, Florence. 2007.
• 9 Wooldridge Introduction to MultiAgent
  Systems. 2002.