Multi Agent Systems

1
Multi Agent Systems
LabIS http//labis.vsb.cz/
2
Content

• Introduction to the Theory of Multi-Agent Systems
• A Brief Summary of the Project Logic and
  Artificial Intelligence for Multi-Agent Systems
  (Logic MAS)
• The Project in more details
• Logic and Communication (TL Language, Ontology)
• Geographical Data and Infrastructure
• Process Management (has been presented by
  Prof. Ivo Vondrák)
• Further research
• Discussion

3
Multi Agent System

• Distributed application
• runs on many computers over the network
  (Internet).
• Composed of agents
• there can be hundreds or thousands of agents
  which cooperate to achieve their goals.
• No centralized control
• agents are (less or more) intelligent and
  autonomous
• generally, there is no central dispatcher

4
Agent Characterisation

• (Jacques Ferber)
• Agent is a physical or virtual entity
• a) which is capable of acting in an environment
• b) which can communicate directly with other
  agents
• c) which is capable of perceiving it's
  environment (to a limited extent)
• d) which can offer services and possesses skills
• e) may be able to reproduce itself
• f) tends towards satisfying it's objectives
• g) may be able to learn by experience
• h) etc., more...

5
Object vs. Agent
Object Agent
Encapsulates methods and attributes Encapsulates behaviour
Communicates by method calling Communicates by asynchronous messages in a higher language
Performs a predefined task when called Performs a task if it wants to
Synchronization of threads is necessary Agents are not dependent on each other
Less autonomous More autonomous
Predefined behaviour Dynamic, intelligent and autonomous behaviour
without intelligence n 0.
Intelligence of a degree n gt 0.
6
Agents can perceive an environment

• Agents can carry out actions which modify agent's
  environment.
• Agents can perceive the environment and make
  decisions based on it.

7
Agents communicate with each other

• Outer languages
• define an envelope format for languages and are
  not concerned with the content
• (KQML, FIPA-ACL)
• Inner languages
• are concerned only with the content. The content
  of a message can be specified in any language
  that agents can understand
• (Prolog, JESS, XML, ...)

Hello
1030
What's the time?
fine, thanks
How are you?
8
Agents learn by experience

• Artificial intelligence
• Expert systems
• Logic and intelligent decisions in difficult
  situations
• Autonomy

?
?
?
9
Knowledge representation

• Ontology
• is widely used for describing some parts of the
  real world. Can be represented in different
  logical languages (e.g. W3C OWL) which can be
  easily interpreted by computers and humans.
  Today, there are many ontology databases and
  libraries which are focused on different topics
  (mobile devices, cars, vines, weather and many
  others).

What is the weather?
OK, I will consider it ...
It is raining, rather take an umbrella
10
Agents features

• Agents can clone themselves
• They are able to move from one computer to
  another one.

Clone
Change PC
11
Multi Agent World
Agents
Knowledge Desires Plans
Communication
Environment
Perception
Actions
Internal representation of environment
12
Some Application Areas

• Robotics
• Traffic simulation and navigation
• Ubiquitous computing
• Air traffic control
• Network management
• Distributed data mining
• Manufacturing systems
• And many others...

Emergency and critical situations
13
Standardization

• FIPA (Foundation for Intelligent Physical Agents)
  
• Standards for Intelligent Multi-agent systems.
• Cooperation with universities and companies.
• Conferences, seminars, workshops on multi-agent
  systems
• Standards for communication languages and
  protocols.

14
Content

• Introduction to the Theory of Multi-Agent Systems
• Logic and Artificial Intelligence for Multi-Agent
  Systems (Logic MAS)
• The Project in more details
• Logic and Communication (TL Language, Ontology)
• Geographical Data and Infrastructure
• Process Management (has been presented by
  Prof. Ivo Vondrák)
• Further research
• Discussion

15
LOGIC MAS The pilot project conducted at the
Research Laboratory of Intelligent Systems
(LabIS http//labis.vsb.cz/ )

• The main goal
• Research of information technologies needed for
  coordination of autonomous intelligent agents in
  extraordinary or emergency situations

16
Logic Multi-Agent World

• See the next presentantion
• VSE-MAS-B

17
Communication Language based on TIL

• FIPA SL
• TL Language
• TL vs. TIL
• Ontologies
• Examples
• Implementation
• Example

18
Communication Language based on TIL

• A crucial point in a multi-agent environment
  communication
• Message
• a unit of communication
• can be of an arbitrary form, but
• has to be structured in terms of attributes
• sender, receiver,. . .
• performative (communicative act) inform,
  query, request,. . .
• content semantics of the message, encoded in
  a content language
• ontology vocabulary of concepts used in the
  content

19
FIPA SL

• FIPA (The Foundation for Intelligent Physical
  Agents)
• creates widely accepted standards for MAS
• one of them is the content language FIPA SL
  (Semantic
• Language)
• FIPA SL is based on FOL (First-order logic)
• gt
• widely known, well elaborated logic, but
• mathematical logic stenography of mathematics
• weak expressive power, over inferring / under
  inferring
• - an extension is needed in order to make it
  possible to communicate in a standardised natural
  language
• the type of a message (query, request,. . . )
• the distinction between empirical and analytical
  concepts
• to properly analyse modalities possible,
  necessary
• to properly analyse attitudes believing,
  knowing, thinking, wanting, seeking, etc.

20
FIPA SL

• SL Syntax well defined (grammar)
• SL Semantics (meaning) poor
• the standard work just contains Section Notes on
  Semantics (really just notes)
• derived from FOL semantics? not applicable
  because of numerous extensions beyond FOL
• intuitive is it?
• The poor semantics may lead to serious
  misunderstandings between agents.

21
TL Why?

• We propose a new content language TL
• TL is a FIPA compliant content language
• TL is based on Transparent Intensional Logic
  (TIL)
• gt rich bi-dimensional hierarchy of types
• great expressive power
• fine-grained rigorous semantics
• Makes it possible to analyse natural language
  expressions in a fine-grained way (and formal
  languages too, e.q. SL)
• a universal semantic tool (specification of
  semantics, combining different languages
  together,. . . )
• Appropriate for
• (complex) knowledge representation
• human-level communication

22
TL vs. TIL

• The TIL language of constructions is not apt for
  computerised processing, because
• the TIL language
• is not standardised
• uses non-ASCII characters (not found on the
  keyboard)
• We need a standardised notation for types as well
  as for encoding constructions
• The epistemic base needs extending a type of
  sequences is useful (and also a construction of
  it)
• There is no way of fixing ontologies in TIL (any
  entity of any order can be trivialised) we need
  to specify the chosen conceptual systems to work
  with

23
TL vs. TIL

• the TL type base is an extension of TIL epistemic
  base

TL name TIL equivalent Description
Bool ? Truth values
Indi ? Individuals
Date ? Times
PSW ? Possible worlds
String
Integer
Float ?
Action Type of action
24
TL vs. TIL

• Compound (molecular) types
• collections of partial functions (as in TIL)
• (Float,Indi? Bool) (???)
• finite sequences of objects of particular types
  Indi
• Constructions
• variable x or xType
• trivialization A oA
• application (F A B C) F A B C
• abstraction \xTypeA ?x A
• sequence (SequenceType A B C)
• shortcut application C_at_w,t Cwt

25
Grammar of TL (fragment)

• TLExpression Construction
• Macro.
• Macro SequenceMacro.
• Construction Trivialization
• Variable
• Application
• Abstraction.
• Trivialization "" Concept
• "" PrimitiveValue
• "" Construction.
• Concept UpperCaseAtom
• "" "(" UpperCaseAtom ArgList ")".
• ...

26
TL and Ontologies

• An ontology is a vocabulary of domain-specific
  concepts (FIPA)
• All concepts used by a content language must be
  defined in an ontology
• Any concept in an ontology can be used by a
  content language
• TL types must be specified for concepts to be
  used in TL
• Ontologies are usually frame-like structures
  (frames with slots)
• The latest trend is Semantic Web OWL
  (Description Logics).
• It is not neglected by FIPA
• but not supported (as yet)

27
Example

• Every prime number is odd. (not true)
• Ontology (part)
• TIL analysis
• o? ?xo? oPrime x oOdd x
• TL notation
• (All \xInteger(Impl (Prime x) (Odd x)))

Name TIL type TL type
Prime (??) (Integer-gtBool)
Odd (??) (Integer-gtBool)
All, ? (?(??)) ((Integer-gtBool)-gtBool)
Impl, ? (???) (Bool,Bool-gtBool)
28
Example

• The highest mountain is in Asia.
• TIL analysis
• ?w?toInwt ?w?toHighestwt oMountainwtwt oAsia
• TL notation
• \w\t(In_at_w,t ((\w\t(Highest_at_w,t Mountain_at_w,t)
  w) t) Asia)

Name TIL type TL type
Highest (?(??))?? ((Indi-gtBool)-gtIndi)_at_
Mountain (??)?? (Indi-gtBool)_at_
In (???)?? (Indi,Indi-gtBool)_at_
Asia ? Indi
29
Implementation of TL

• TL is implemented in the MAS framework JADE
  (work in progress)
• JADE
• open-source, extensible development framework in
  Java
• strictly FIPA compliant
• Ontologies are frame-like, edited by Protege.
• Working on integration of OWL ontologies into TL.
• Any agent can simply choose TL as the content
  language for his messages.

30
Example

• The scenario contains
• large car parking, railway station
• Agents
• Driver would like to park close to the railway
  station
• Dispatcher of the car parking managing available
  pull-ins
• Sketch of their dialogue
• Driver I want you to park me somewhere not far
  from the railway station.
• Dispatcher OK, I can park you at this pull-in
  (concrete position).
• Driver Good.

31
Ontology for the Example
Concept TL type
TheDriver Indi
TheDispatcher Indi
TheTrainStation Indi
Pull-in Indi
Near (Indi,Indi-gtBool)_at_
Arrange (Indi,Bool_at_-gtAction)
Park (Indi,Indi-gtBool)_at_
32
Communication

• Driver
• call for proposal
• \xIndi(Arrange TheDispatcher
• \w\t(And
• (Park_at_w,t TheDriver x)
• (Near_at_w,t TheTrainStation x)))
• A call for proposal is a communicative act which
• takes one argument
• returns the action to be proposed
• such that the responding agent will fill in the
  parameter x the parking place

33
Communication

• Dispatcher
• propose
• (\xIndi(Arrange TheDispatcher
• \w\t(And
• (Park_at_w,t TheDriver x)
• (Near_at_w,t TheTrainStation x)))
• (Pull-in
• (Position
• "ltgmlPolygongt
• ltgmlouterBoundaryIsgt
• ltgmlLinearRinggt
• ltgmlcoordinatesgt
• 0,0 100,0 100,100 0,100 0,0
• lt/gmlcoordinatesgt
• lt/gmlLinearRinggt
• lt/gmlouterBoundaryIsgt
• lt/gmlPolygongt")))

34
Communication

• Driver
• Accept proposal
• (same content as the last one)

1 Call for proposal
Dispatcher
2 Propose
Car Agent
3 Accept proposal
35
Future work on TL communication

• Integration of TL with Semantic Web
• namely the Ontology Web Language (OWL)
• specification of ontologies for TL based
  communication
• Methodology of agents communication using the TL
  language
• Future Research
• Specification of an accurate inference machine
  based on the expressive language such as TL
• Implementation of the inference machine

36
Content

• Introduction to the Theory of Multi-Agent Systems
• Logic and Artificial Intelligence for
  Multi-Agent Systems (Logic MAS)
• The Project in more details
• Logic and Communication (TL Language, Ontology)
• Geographical Data and Infrastructure
• Process Management (has been presented by
  Prof. Ivo Vondrák)
• Further research
• Discussion

37
Connecting GIS and Multi-agent systems
38
Relations between GIS and MAS

• Identity situated agent corresponds to a
  feature stored in the geo-database
• Causal agents can affect other features or
  their attributes (behaviour of agents is affected
  by state of GIS)
• Temporal synchronization between MAS and GIS
• Topological provides information about
  topological relations between features

39
Environment perceiving
40
Environment built in a vector layer
Creation of infrastructure agents
Geo Database
Vector geodata export to geodatabase
41
Agent Based Environment
42
Agent types

• Infrastructure agents
• streets, crossroads, other infrastructure objects
• Mobile agents
• vehicles, humans
• Database agent
• providing access to geographic databases
• Visualization agent
• performing visualization of static and dynamic
  agents

43
Implementation environment

• JADE
• Java Agent DEvelomement framework framework
  that simplifies agent creation and deployment
• PostGIS
• Open source spatial database, used for data
  storage and perception of the environment
• GRASS, JUMP, 3D visualization module
• Used for visualization purposes only

44
Implementation environment
User Interface (JUMP, GRASS, 3D)
Database Agent
ACL
ACL
Geo Database
Other Agents
ACL
45
Application screenshots
46
Application video
47
Content

• Introduction to the Theory of Multi-Agent Systems
• Logic and Artificial Intelligence for
  Multi-Agent Systems (Logic MAS)
• The Project in more details
• Logic and Communication (TL Language, Ontology)
• Geographical Data and Infrastructure
• Process Management
• Discussion

48
Process management in MAS

• MAS software process is similar to the standard
  information system development process
• There are some differences, however
• functions, properties and skills of MAS are very
  close to the real world situations
• emphasis is placed on social behavior of system
  components and the whole system
• emphasis on the autonomy of components
• emphasis on communication level between
  components
• components of such systems are Agents
• Methodology used during the development process -
  UML

49
MAS Development Process

• Tools that support this development process
• standard software process and methodology
• UML tools with some extensions Behavior
  Activity Diagram
• etc.

50
Basic definitions

• Agent term describes a software abstraction, an
  idea, or a concept, similar to OOP terms such as
  methods, functions, and objects. The concept of
  an agent provides a convenient and powerful way
  to describe a complex software entity that is
  capable of acting with a certain degree of
  autonomy in order to accomplish tasks on behalf
  of its user. But unlike objects, which are
  defined in terms of methods and attributes, an
  agent is defined in terms of its behavior.
• Behavior refers to the actions or reactions of an
  object or organism, usually in relation to the
  environment. The complexity of the behavior of an
  organism is related to the complexity of its
  nervous system (internal structure, knowledge and
  skills).
• Realization is the element that provides an
  description for implementation of an element that
  specifies behavior, by other words, the
  Realization is one of possible implementation of
  Process this implementation is specified by one
  Behavior Activity Diagram

51
Basic definitions

• Process is a naturally occurring or designed
  sequence of operations or events, possibly taking
  up time, space, expertise or other resource,
  which produces some outcome. A process may be
  identified by the changes it creates in the
  properties of one or more objects under its
  influence. The Processes form the Behavior of an
  Agent.
• Activity is something done as an action. It is
  atomic action within the Process.
• Process or Activity Nodes are the graphical
  representation of afore mentioned terms.

52
Agents Behaviors

• What is the Behavior of an Agent?
• We have to specify
• everything that an Agent can do,
• including Agents abilities and potency of
• solving the problems he/she/it can meet, and of
• responding to changes in the environment
• There are several types of behavior templates
  corresponding to Agents classification
• simple reactive behavior
• simple proactive behavior
• hybrid behavior (combined reactive and proactive
  behavior)
• The Behavior is defined by Processes within the
  Agents life

53
Agents Processes

• Process is a tool for describing Agents
  elementary Behaviors
• Each description of a Process consists of the
  following
• input objects
• output objects (in agreement with the goals of
  the process)
• a set of possible realizations (executing the
  function, mapping Inputs ? Outputs)
• Its execution depending on a current situation
• Each Process can be realized in several ways
  possible realizations of the process

54
Internal Structure of Agents

• Processes can be defined as
• internal within an Agent
• external in the global repository
• (Internal) Processes within a given Agent
• executable only by the given Agent
• they can extend or restrict globally defined
  processes too
• (External) Processes saved in the global
  repository
• executable by all the Agents within the MAS
• possibility of learning and changing the Behavior

55
Internal Structure of Agents
Process may have different control flows defined
by Realizations
Process defined in the global repository
Only one possibility to connect the Agents
message passing
56
Activity Diagram Extension
57
Activity Diagram Extension
58
Realizations of Process

• Realization describes an algorithm of
  transforming Inputs into Outputs depending on a
  current situation (Agents reactions)
• Each Realization has to be in accordance with the
  parent process definition (objectives, input and
  output objects)
• Each Realization is one Behavior Activity
  Diagram based on the UML Activity Diagram
  (from the visual point of view)
• For these purposes, the Multi-Agent Technology
  calls for an extension of the standard UML
  approach

59
Metamodel of MAS Model
60
Intelligence Behavior

• Intelligence contained in the activities
• weather forecast, etc.
• Intelligence of the control flows routing
• suitable car to a given cargo assignment, etc.
• Intelligent selection of Process Realizations
• reconfiguration approach

61
Behavior Reconfiguration

• Agents Behavior can evolve during its life
• as a reaction to a current situation in the
  moment of emergency (when executing a process).
• This reconfiguration is based on the current
  state (objects, environment, etc.) of an Agent
  and/or the whole MAS, and it works with sets of
  particular Realizations of Processes.

62
Behavior Reconfiguration
63
Behavior Reconfiguration

• Each Process node within the Behavior Activity
  Diagram can be a reconfiguration point the
  position of Behavior that can be changed
• Basic reconfiguration algorithm
• selection of applicable Realizations from a given
  Process set
• selection of the most suitable Realization
• execution of the selected Realization
• Tools for support of the reconfiguration
• simple branching based on conditions
• logical tools
• Artificial Intelligence, Expert Systems

64
Selection of an applicable Realization

• Process approach- Realizations are combinedwith
  a given Process withinthe modeling phase- easy
  selection- low intelligence and flexibility
• Logical approach- real-time selection of
  Realizations
• - errors can occur- describing from the
  logical point of view may be difficult- high
  intelligence and flexibility

65
Selection of a Realization

• Modified reconfiguration algorithm
• specification of the set of process Realizations
  based on the process approach
• defining a set of applicable Realizations for a
  given Process based on the process approach
• selection of the most suitable Realization from a
  defined set based on the logical approach
• execution of the selected realization
• Terms usable in this algorithm
• Input and Output Objects (step 1 and 2)
• values of Objects and Scores (step 3)
• FCA, TIL, etc.

66
Selection of a Realization

• Process and Logical approach together

67
Behavior Reconfiguration

• Which are the assets of this approach?
• possibility of an independent life of each Agent
• intelligence built in an Agent Behavior
• easy distribution of knowledge on process
  management among the Agents
• easy way to distribute knowledge on system
  adaptability and evolution among the Agents

68
AgentStudio
69
Future Architecture of AgentStudio
70
Future Work

• development of the tool for Agents behavior
  modeling (.NET)
• behavior-model extensions (message
  specifications, interfaces, etc.)
• automatic generation of Agents based on their
  internal Behavior
• involvement of Logic in Agents behavior
  management (reconfiguration, branching,
  controlling, etc.)
• distribution of the Behavior specifications and
  learning process
• development of application methods of the whole
  MAS management
• knowledge management and distribution
• system states monitoring

71

• Introduction to the Theory of Multi-Agent Systems
• Logic and Artificial Intelligence for
  Multi-Agent Systems (Logic MAS)
• The Project in more details
• Logic and Communication (TL Language, Ontology)
• Geographical Data and Infrastructure
• Process Management
• Discussion