1. Ongoing Research on Software Engineering for Multi-Agent Systems
Carlos José Pereira de Lucena
Software Engineering Lab (LES) – PUC-Rio
October 2004 – SBES’04

2. Presentation Overview
Software Agents: Motivation
Perspectives on Software Engineering for MASs
MAS-SE at LES
MAS Modeling (two approaches)
Developing Method based on Aspects
Technology Evaluation
Technology Transfer
Future research directions at LES
Conclusions
Software Engineering Lab (LES) – PUC-Rio

3. Software systems will be
The Scenario
With the advances in network-based applications, the introduction of autonomous components in software systems has become usual
Software systems will be
Everywhere
Always connected
Always active to perform something on the user’s behalf
SELMAS 2001, 2002, 2003 e 2004
ICSE 2005
Software Engineering Lab (LES) – PUC-Rio

4. New Characteristics of Software Systems1
Situatedness
Components execute in the context of an environment, and they can influence and be influenced by it
Openness
Systems have decentralized management and can dynamically change their structure
Locality in control
Components represent autonomous and proactive loci of control
Locality in interactions
Components interact with each other accordingly to local patterns
Futuro de Computer Science e Software Engineering
Computabilidade
Peter Wegner => interação + forte que algoritmos
1 F. Zambonelli, H.V.D. Parunak. Towards a Paradigm Change in Computer Science and Software Engineering: A
Synthesis. Knowledge Engineering Review, 2004.
Software Engineering Lab (LES) – PUC-Rio

5. Software Agents
The complexity introduced by this new kind of software systems goes beyond the capabilities of traditional software engineering abstractions, such as objects and components
Software Agent
An agent is an interactive computer system that is situated in some environment, and that is capable of autonomous action in this environment in order to meet its design objectives2
Outras definições:
Object with attitude
Object that say no and go
2 M. Wooldridge, N.R. Jennings. Intelligent Agents: Theory and Practice. Knowledge Engineering Review 10(2),
1995, pp
Software Engineering Lab (LES) – PUC-Rio

6. O que são Sistemas Multi-Agentes?
No ponto de vista de IA
Um sistema multi-agente é uma sociedade de agentes (agentes IA) individuais, que interagem por meio de troca de conhecimento e de negociação para alcançar um objetivo local ou global
No ponto de vista de ES
Um sistema multi-agente é um sistema de software composto por diversos locos de controle (agentes ES) encapsulados e independentes, que interagem no contexto específico de uma aplicação
Software Engineering Lab (LES) – PUC-Rio

7. Agency Characteristics
Characteristics of multi-agent systems include
Autonomy
Capable of acting without direct external intervention
Interactivity
Communicates with the environment and other agents
Adaptation
Capable of responding to other agents and/or its environment to some degree
Pro-activity
Goal-oriented - it does not simply react to the environment
OMG, White paper
Software Engineering Lab (LES) – PUC-Rio

8. Agency Characteristics (cont’d)
Characteristics of multi-agent systems include
Learning
Modify its behavior based on its experience
Rationality
Able to choose an action based on internal goals
Mobility
Able to transport itself from one environment to another
Software Engineering Lab (LES) – PUC-Rio

9. Evolution of SE Paradigms
Assembly Languages
Functional Abstraction
Structured Programming
Object-Orientation
Components
Design Patterns
...
Software Agents
Abstractions determined by the machine architecture
Time
Abstractions determined by the problem domain
Agents 1950
Software engineering X Knowledge Engineering
Cognitvo X Reativo
Software Engineering Lab (LES) – PUC-Rio

10. Today’s Technological Environment
The advances in Internet technologies increased the demand for the development of distributed, heterogeneous and complex systems
The evolution of the Web towards a Semantic Web is creating contents that can be directly manipulated by processes
Software agents embody distribution and heterogeneity and, thus, they are indicated as the new abstraction for the engineering of complex distributed systems
HTML -> XML -> RDF -> OWL
Software Engineering Lab (LES) – PUC-Rio

11. The Multi-Agent Paradigm
The Multi-Agent paradigm offers
A decomposition model
An abstraction for software development based on the problem domain
A model of system development as a composition of organizational units
A decentralization model to reduce the unit coupling
Antropomórfica
Generalização na forma de comunicação de agente
Software Engineering Lab (LES) – PUC-Rio

12. Experimental Development since 1998
AulaNet Learning Management System
2BuyNet e-Business (B2C)
VBroker e-Business (B2C)
VMarket e-Business (C2C)
iDeal e-Business (B2B)
CommercePipe e-Business (C2B)
PortalWare Knowledge Management
SkillO Skill Management
WebClipper Web Crawling
Software Engineering Lab (LES) – PUC-Rio

13. The State of the Art of SE-MAS
Software Engineering for Multi-agent Systems at a very early stage
Several proposals, little agreement
AOSE/AAMAS, SELMAS/ICSE
JAAMAS
Ad-hoc or proprietary development
Lack of an assessment of the impacts of the agent technology on software development
There is a need to develop a proper Software Engineering for the Agent Paradigm
AOSE/AAMAS
AAMAS 3 anos em 2004
AOSE 5 anos em 2004
JAAMAS desde 1998 – 6 anos em 2004
Métricas
Software Engineering Lab (LES) – PUC-Rio

14. Agent-oriented requirements analysis and specification
Research Topics
Agent-oriented requirements analysis and specification
Best practice in agent-oriented development
Engineering large-scale agent systems
Experiences with field-tested agent systems
Formal methods for agent-oriented systems, including specification and verification logics
Software Engineering Lab (LES) – PUC-Rio

15. Research Topics (cont’d)
Implications of agent-oriented software on organizational and social structures within and between companies (e.g. changes in roles, responsibilities, transparency, business processes and decision schemes)
Integration of agent-oriented software into existing business processes and implications on business process re-engineering
Market and other economic models in agent systems engineering
Software Engineering Lab (LES) – PUC-Rio

16. Research Topics (cont’d)
Methodologies for agent-oriented analysis and design
Model checking for agent-oriented systems
Model-driven architecture (MDA) for MAS
Practical coordination and cooperation frameworks for agent systems
Refinement and synthesis techniques for agent-based specifications
Relationship of AOSE to other SE paradigms (e.g., OO)
Software Engineering Lab (LES) – PUC-Rio

17. Research Topics (cont’d)
Re-use approaches for agent-oriented software, including design patterns, frameworks, components, and architectures
Software development environments and CASE tools for AOSE
Standard APIs for agent programming
Standardizations for AOSE
UML and agent systems
Verification and validation techniques for agent-based systems
LES cobre quase tudo
OMG – RFI (Request For Information)
Software Engineering Lab (LES) – PUC-Rio

18. MAS Conceptual Framework MAS Modeling
MAS-SE at LES
MAS Conceptual Framework
TAO
MAS Modeling
ANote
MAS-ML
Developing Methods and Frameworks
Aspect-oriented agent patterns
Software Engineering Lab (LES) – PUC-Rio

19. MAS Conceptual Frameworks
Describe the structure of the abstractions of MAS
Examples
KAoS, d'Inverno and Luck, Yu and Schmidt
Issues
The description of the interplay between these abstractions
The description of the other MAS abstractions surrounding the agent, such as environment and organization
The use of the abstractions to guide the creation of models
Software Engineering Lab (LES) – PUC-Rio

20. The TAO Conceptual Meta-model
Defines a core set of MAS abstractions
The entities that may be described in MAS, their properties and the relationships associated with them (structural aspects of MAS)
The creation and destructions of entities and also defines domain-independent behavior (dynamic aspects of MAS)
Defines a core set of MAS relationships
TAO groups together the abstractions that are frequently described in the literature for MAS
Software Engineering Lab (LES) – PUC-Rio

21. The TAO3 Abstractions Relationships
Agent, object, environment, organization and role
Relationships
Inhabit, ownership, play, control, dependency, association, aggregation/composition and specialization
3 V. Silva et al. Taming Agents and Objects in Software Engineering. In: Garcia et al. (eds.) Software Engineering
for Large-Scale Multi-Agent Systems, LNCS 2603, 2003, pp
Software Engineering Lab (LES) – PUC-Rio

22. MAS Modeling languages and methodologies
MAS modeling languages that do not extend UML
Gaia, MaSE
ANote (Choren et al, SELMAS 2004), CAMLE (Shan and Zhu, SELMAS 2004)
MAS modeling languages that extend UML
AUML, AORML
MAS-ML (Silva et al, LNCS 2940, 2004)
Methodologies that use or extend MAS modeling languages
MESSAGE, Prometheus
Prometheus -> AUML
Message -> AUML
Software Engineering Lab (LES) – PUC-Rio

23. MAS Modeling Languages that Do Not Extend UML
Issues
Do not have a conceptual framework
Overuse or lack information on diagrams (scattered design of the system)
Miss the definition of the non-agent entities (environmental resources)
Do not explicitly show the agency properties such as adaptation or autonomy
Software Engineering Lab (LES) – PUC-Rio

24. MAS Modeling Languages that Extend UML
Issues
Do not explicitly define the extensions applied to the UML meta-model
Do not define guidelines to implement the systems modeled using the language
Do not model the same of the structural aspects and dynamic aspects frequently described in MAS
Heavy
Software Engineering Lab (LES) – PUC-Rio

25. Lightweight, agent-oriented notation language Focus
The ANote4
Lightweight, agent-oriented notation language
Focus
Specification of systems with distributed computing, agents and knowledge representation (ontology)
Characteristics
Agent-centric / Agency features
Goal-oriented
Design views
4 R. Choren, C. Lucena. Modeling Multi-agent Systems with ANote. Journal of Software and Systems Modeling, 2004.
Software Engineering Lab (LES) – PUC-Rio

26. ANote Conceptual Meta-model
Software Engineering Lab (LES) – PUC-Rio

27. Each concept defines a perspective or view
Views
Each concept defines a perspective or view
ANote defines seven views each of which has its own representation (diagram)
Software Engineering Lab (LES) – PUC-Rio

28. ANote Goal View
Provides an initial identification of a tree of goals that outline the functions performed by the agents
They are specified from elaborations, process descriptions and system problem (functional decomposition)
Software Engineering Lab (LES) – PUC-Rio

29. ANote Agent View
Specifies the agent types of the multi-agent application solution and their relationships (system structural base)
They are specified from the logical subdivision of the system’s functional goals
Software Engineering Lab (LES) – PUC-Rio

30. UML Class Diagram used as notation
ANote Ontology View
Identifies the non-agent components of the system, specified with an ontology
They are specified from the system environment resources, their relationships and functions (object-oriented approach)
UML Class Diagram used as notation
Software Engineering Lab (LES) – PUC-Rio

31. Captures agent behavior in specific contexts
ANote Scenario View
Captures agent behavior in specific contexts
Describe how goals can be achieved by an ordering of actions and events
It specifies the normal course of action and the alternative paths of execution (emergent context)
Shows the agents, interactions and plans involved in achieving a goal in a context (enterprise modeling)
Scenario table used as notation
Software Engineering Lab (LES) – PUC-Rio

32. ANote Planning (Action) View
Specifies the execution states, or actions, an agent has to perform to compute an action plan (derived from a scenario that illustrated a goal)
It allows the modeling of the agents’ workflows
Software Engineering Lab (LES) – PUC-Rio

33. ANote Interaction (Message) View
Shows the structural organization of agents that send and receive messages while executing an action plan
It allows the modeling of the interaction protocols (conversation graphs)
Software Engineering Lab (LES) – PUC-Rio

34. ANote Organization View
Specifies the multi-agent system implementation unit that offers services (set of goals), accessed by an interface (set of message protocols)
There is no basic approach to split the agents into logical organizations (it is not a property of the abstraction, but a design decision)
Software Engineering Lab (LES) – PUC-Rio

35. ANote Diagrams in a Nutshell
Software Engineering Lab (LES) – PUC-Rio

36. ANote Diagrams in a Nutshell
Software Engineering Lab (LES) – PUC-Rio

37. Web-based Insurance System Basic Railcar Control System
Some Usage Examples
Agent Marketplace
Web-based Insurance System
Basic Railcar Control System
Trading Agent Competition (TAC)
Software Engineering Lab (LES) – PUC-Rio

38. Ongoing Research on ANote
Tool support
Graphical tool
Internal XML representation
Eclipse Platform plug-in
Translation (from model to code) support
MAS Framework (TAC)
Transformation from ANote models XML representations to MAS Framework
Software Engineering Lab (LES) – PUC-Rio

39. Extended UML notation language Focus
The MAS-ML5
Extended UML notation language
Focus
Specification of the structural and dynamic aspects of MAS through a conservative extension of UML
Characteristics
Agent/organization-centric (based on TAO meta-model)
UML extension
Structural and dynamic aspects
5 V.T. Silva, C.J.P. Lucena. From a Conceptual Framework to a Multi-Agent System Modeling Language. Journal of Autonomous Agents and Multi-Agent Systems 9(1/2), 2004, pp
Software Engineering Lab (LES) – PUC-Rio

40. MOF Architecture MOF ER meta-meta-model meta-meta-model UML TAO
Meta-meta-model layer
instantiation
UML
meta-model
TAO
meta-model
Meta-model layer
unification
MAS-ML
meta-model
Domain model layer
MAS-ML
models
Instance layer
Software Engineering Lab (LES) – PUC-Rio

41. Meta-classes of UML meta-model
The MAS-ML Meta-model
Duty
Right
Features
Classifier
Structural Feature
Behavioral Feature
Belief
Goal
Axiom
Operation
Property
Class
AgentClass
ObjectRoleClass
OrganizationClass
AgentPlan
Meta-classes of UML meta-model
New meta-classes
New stereotypes
Legend
AgentRoleClass
AgentAction
1..* *
sender
receiver 1
Element
AgentProtocol
AgentMessage
Constraint
postcondition
precondition
0..1
EnvironmentClass
Active
Passive
Software Engineering Lab (LES) – PUC-Rio

42. The MAS-ML Meta-model Class AgentClass ObjectRoleClass
OrganizationClass
AgentRoleClass
EnvironmentClass
define
play 1
1..*
0..*
sub-org
play in
inhabit
sub-
org
Relationship
Directed Relationship
Classifier
Element
Association
Dependency
Ownership
Inhabit
Control
Play
Generalization
1..*
Meta-classes of UML meta-model
New meta-classes
Legend
Software Engineering Lab (LES) – PUC-Rio

43. MAS-ML Class Diagram
Represents the relationships between the classes and other MAS entities
Virtual Marketplace
association
Organization A
AgentA
Item
Book
Organization A
AgentB
specialization
Imported Books
Second-hand Books
Legend:
inhabit
Object / Environment
Organization
Agent
Software Engineering Lab (LES) – PUC-Rio

44. MAS-ML Organization Diagram
Models the organizations and the relationships between the organizations and other MAS entities
Software Engineering Lab (LES) – PUC-Rio

45. Models the roles and the relationships between the roles
MAS-ML Role Diagram
Models the roles and the relationships between the roles
MAS-ML usa diagramas estáticos diferentes que dão enfoque em diferentes visões da modelagem
Em AORML, por exemplo, os diagramas ficam sobrecarregados de informação.
Problema com o uso de diferentes diagramas contendo um sub-conjunto de informações em comum: consistência
Software Engineering Lab (LES) – PUC-Rio

46. MAS-ML Sequence Diagram
Represents the interactions between the MAS instances and the intra-actions defined by each instance
New elements to represent interactions
The concept of message was extended
New stereotypes associated with messages were created
Extensions to model protocols were defined
New elements to represent intra-actions
Extensions to model the execution of plans and actions were defined
Software Engineering Lab (LES) – PUC-Rio

47. MAS-ML Sequence Diagram
New pathnames and icons
Software Engineering Lab (LES) – PUC-Rio

48. plan organization method invocation environment agent action
Bob/buyerOfBooks/Bookfinder/Place-A :
UserAgent/Buyer/
GeneralStore/VirtualMarketplace
Place-A :
VirutalMarketplace
BrandNewBookstore/marketOfUsedBooks/Bookfinder/Place-A :
Second-handBookstore/MarketOfUsedGoods/
GeneralStore/VirtualMarketplace
environment
plan
Entering
organization
Search for organization
organization
getOrganizations()
Vector
method
invocation
Check_organization_goals
Managing entrance
Request (OrgGoals)
Get_organization_goals
Inform (OrgGoals)
Check_organization_roles
Request (OrgRoles)
agent
Get_organization_roles
Inform (OrgRoles)
action
Bob/buyerOfSecond-handBooks/Bookfinder/Place-A :
UserAgent/BuyerOfSecond-handBooks/
GeneralStore/VirtualMarketplace
Select_role_to_play
<<role_commitment>>
message stereotype
Inform (RoleToPlay)
Entering
organization
Register_itself_in_organization
Inform (Buyer_of_second_hand_books)
Search_for_announcement
Request (Announcement)
...
Software Engineering Lab (LES) – PUC-Rio

49. Simple Matching System Agent Marketplace
Some Usage Examples
Simple Matching System
Agent Marketplace
Paper Submission and Reviewing System
Simple Supply-chain Management System
Software Engineering Lab (LES) – PUC-Rio

50. Ongoing Research on MAS-ML
Preliminary approach to code generation
Tool support
Graphical tool
Create the MAS-ML DTD based on the extended MAS-ML grammar
Transform XMI based on MAS-ML DTD into XMI based on UML DTD
MAS-ML Formalization
MAS-ML Analyzer
Description Logic
Ontology
Software Engineering Lab (LES) – PUC-Rio

51. The Aspect-Oriented Agents6
Agents with aspect-orientation
Focus
Specification of an architecture and of a pattern language for agents using the aspect-oriented approach
Characteristics
Aspect-oriented
Architectural Method (kernel and agenthood properties)
Pattern language
6 Garcia, A. F.; Lucena, C. J. P.; Cowan, D. D.; Agents in Object-Oriented Software Engineering. In Software: Practice & Experience, Elsevier, vol. 34, Issue 5 , pp , May 2004.
Software Engineering Lab (LES) – PUC-Rio

52. The Architectural Problem
The agent properties are not orthogonal
Crosscut each other and the agent’s basic functionality
Interaction
Collaboration
Mobility
Learning
Roles
Adaptation
Autonomy
Basic Functionality
The problem is that although separation of concerns is critical to MAS engineers, it is often difficult to achieve it in realistic systems.
The agent properties are not orthogonal – they crosscut each other and the agent’s basic functionality.
For example, this is the architecture of a simple agent; it describes the high-level structure of an agent. Each architectural component modularizes an agent property or agent concern.
Note that the adaptation component, which modularizes the adaptation concern, crosscuts the basic’s agent functionality, the autonomy property, and the interaction property…
architectural component
Software Engineering Lab (LES) – PUC-Rio

53. The Architectural Problem (cont’d)
Multiple agent types  heterogeneous architectures
A Reactive Information Agent
A Cognitive User Agent
Interaction
Interaction
Need for a more flexible
architectural approach
Adaptation
Autonomy
Adaptation
Autonomy
Basic Functionality
Basic Functionality
The architectural definition of the system’s agents is not easy since a multi-agent system can include multiple agent types and each agent type has heterogeneous architectures.
The incorporation of some properties into the agent architecture and their relationships depend on the agent type…
Interaction, adaptation, and autonomy are often considered as basic concerns in the development of agent architectures, while properties like mobility, learning, roles and collaboration are optional concerns. These additional properties also affect or crosscut multiple architectural components of an agent.
For example, a reactive information agent can include… A cognitive user agent encompasses different components….
Moreover, note that the agent properties are composed in different forms… In this reactive agent, adaptation crosscuts… In this cognitive agent, adaptation crosscuts…
So, the main architectural problem is that the agent architecture should support the flexible incorporation and removal of agent properties, and a flexible composition of the architectural agent components in order to allow the construction of heterogeneous agent types.
Mobility
Learning
Collaboration
Roles
Roles
Software Engineering Lab (LES) – PUC-Rio

54. The Design Problem Basic Functionality Collaboration Mobility
Adaptation
Autonomy
Interaction
Learning
Basic Functionality
Agent
goals
...
efectors
adapters
threads
move()
returnHome()
itinerary
processInformation()
LCs
startRole()
itinerary remoteEnvironments
roles
processInf ()
sendMsg()
receiveMsg()
addAdapter()
removeAdapter()
notifyAdapters()
makeDecision()
startThread()
addGoal()
executePlan() …
Role
agents
protocol
Collaboration
getName()
addAgent() …
Plan
goal
preCond()
posCond() …
public Result searchKW(String KW) {
adaptKnowledge(KW);
processInformation(KW);
...
<< searching-specific code >>
if (result = null) {
this.move(KW) }
InterfaceAgent
InformationAgent
UserAgent
user DB
userName
receiveInstruction()
monitor()
...
The problems are not related to the architectural definition of the agents. It is also difficult to deal with the agent properties at the detailed design level.
The transition of an architectural specification to a detailed object-oriented design is not straightforward. It is difficult to preserve the architectural separation of the agent concerns through the OO design and implementation.
The object-oriented abstractions, such as classes, and the composition mechanisms are not able to modularize the agent properties. As the agent complexity increases, these properties tend to crosscut several classes that implement the agent’s basic functionality.
For example, as you can see, in this example, these classes, on the left side, implement the basic functionality of three agent types. The methods implement the basic services of the agent types. These classes, on the right side, implement the collaboration concern.
The introduction of the agent properties into the system leads to the problem of crosscutting concerns. Each agent concern cannot be modularized in a single class.
For example, the learning concern crosscut the agent’s basic functionality and the collaboration concerns; even we try to modularize the learning strategies into separate classes, new learning-specific methods and attributes need to be introduced to the agent’s basic classes to connect them with the learning components.
In addition, the problem of crosscutting concerns is not limited to the class-level modularity, the introduction of the agent properties to the system also impacts the method-level modularity. The implementation of the agent properties also crosscuts several methods, which implement the agent’s basic functionality and other agent properties.
For example, the information agent has the basic function of searching for information based on some keywords. Note that the implementation of the search algorithm is intermingled with calls to the components that implement the agent properties, such as adaptation, learning, and mobility.
The agent’s basic functionality should not be mixed with the implementation of the agent properties because these properties are not part of the agent’s basic functionality.
The presence of crosscutting concerns makes it difficult to reuse and maintain the basic services of the agent and the implementation of the agent properties.
getUser()
checkPreference() …
Caller
Answerer
search(Keyword)
search(Keywords) …
caller
answerer
receiveRequest()
sendResponse()
...
Plan1
Plan2
sendRequest()
receiveResponse()
...
action1()
action2()
action3()
...
action1()
action2()
...
Software Engineering Lab (LES) – PUC-Rio

55. The Proposed Solution software architecture Architectural Method
detailed design
Pattern
Language
This research work proposes an aspect-oriented approach for developing multi-agent systems.
The hypothesis is that the use of aspects through the different development phases supports improved separation of MAS concerns and, as a consequence, leads to more reusable and maintainable agent-based systems.
The approach is composed of:
an architectural method for guiding the development of heterogeneous agent architectures
a pattern language to document aspect-oriented solutions for the detailed design of multi-agent systems
an aspect-oriented framework that follows the architectural solution and the design patterns
Implementation
Framework
implementation
Software Engineering Lab (LES) – PUC-Rio

56. The Architectural Method
Step 1: define the agent kernel
Kernel
Belief
Goal
Plan
Agent
name
goals
plans …
new()
addBelief()
setGoal()
executePlan()
...
- In the first step, the core of the agent is defined. Its kernel is decomposed in terms of goals, beliefs, plans and so forth.
Garcia, A., Kulesza, U., Lucena, C. Aspectizing Multi-Agent Systems: From Architecture to Implementation. “Software Engineering for Multi-Agent Systems III”, Springer LNCS, State-of-the-Art Survey, December (prelo)
Software Engineering Lab (LES) – PUC-Rio

57. The Architectural Method (cont’d)
Step 2: Define the agenthood properties
Goal
Creation
Autonomy
Execution
Autonomy
Message
Reception
Kernel
Message
Sending
Belief
Goal
Interaction
Plan
Adaptation
Agent
(2nd Step: Propriedades Mínimas)
The architecture is composed of a central component and aspectual components.
In the 2nd step, the agenthood properties are modularized as architectural aspects.
Each component crosscuts or affects the kernel and other aspectual components.
For example, the adaptation aspect crosscuts not only the Kernel, but also the other architectural aspects, such as the Interaction aspectual component.
Note that the aspect-oriented architecture is flexible enough to support different architectural compositions.
name
goals
plans
new()
addBelief()
setGoal()
executePlan()
...
Knowledge
Adaptation
Legend:
aspectual component
crosscutting interface
Software Engineering Lab (LES) – PUC-Rio
normal interface

58. The Architectural Method (cont’d)
Step 3: define the agent types
Kernel
InterfaceAgent
receiveInstruction()
monitor()
memorize()
...
InformationAgent
search(Keyword)
search(Keywords)
search(Sentences) …
UserAgent
getUser()
checkPreference()
Agent
name
beliefs
goals
plans
addBelief()
addGoal()
executePlan()
Software Engineering Lab (LES) – PUC-Rio

59. The Architectural Method (cont’d)
Step 4: define the additional properties
Kernel
Information
Gathering
Learning
InterfaceAgent
receiveInstruction()
monitor()
memorize()
...
InformationAgent
search(Keyword)
search(Keywords)
search(Sentences) …
UserAgent
getUser()
checkPreference()
Collaboration
Protocol
(4rd Step: Propriedades Opcionais)
Legend:
aspectual component
crosscutting interface
Software Engineering Lab (LES) – PUC-Rio
normal interface

60. Additional Properties
The Pattern Language
Agenthood
Additional Properties
Environment
Adaptation
Mobility
incoming message
absent knowledge
new message
new goal
new
event
outgoing
message
Interaction
Kernel
Learning
absent knowledge
new event
new message
agent creation
collaboration
joining
Autonomy
Roles
Legend:
Crosscuts
Software Engineering Lab (LES) – PUC-Rio

61. The Adaptation Pattern
The Adaptation concern
Observes events
Triggers the adaptation strategy
Adapts the agent knowledge or behavior
Problem: how to separate the adaptation concern from the other agency concerns?
Software Engineering Lab (LES) – PUC-Rio

62. The Adaptation Pattern (cont’d)
Observable
addAdapter()
removeAdapter()
notifyAdapters()
Role
Agent
Beliefs
Observer
collaboratingAgents
collaborationProtocol
adapters
addAdapter()
removeAdapter()
notifyAdapters()
addAgent() …
goals
plans
adapters
addAdapter()
removeAdapter()
notifyAdapters()
addAgent()
removeAgent()
executePlan()
...
name
addAdapter()
removeAdapter()
notifyAdapters()
Collaboration
Adapter
adapt()
PrimitiveBeliefs
CompositeBeliefs …
update()
Chair
Reviewer
Kernel
papers
submissionDeadline
reviewDeadline
addAgent()
removeAgent() …
chair
toReviewPapers
reviewDeadline
fillForm()
updateForm()
getChairName() …
MyPaper
new()
coauthors
keywords
publisher
status
Legend:
Adaptation Concern
Software Engineering Lab (LES) – PUC-Rio

63. The Adaptation Pattern (cont’d)
Solution
Behavior
Adapation
adapters
adaptBelief()
findPlan()
adaptSpecificBelief()
findSpecificPlan()
Agent
Belief
knowledge
elements
observed
Aspect
newGoal_()
failedPlan_()
planFinal_()
changedBelief_() newMsg_()
Knowledge
Adaptation
executePlan()
Plan
setGoal()
Adapter
adapt()
Legend:
_beforeAdvice
afterAdvice_
_aroundAdvice_
Software Engineering Lab (LES) – PUC-Rio

64. Method Evaluation 4 qualitative studies and 1 quantitative study
3 prototypes in different application domains
Portalware URL:
ExpertCommittee URL:
Aspect-oriented metrics suite
Cohesion, coupling, size, separation of concerns
1. Sant'Anna, C., Garcia, A., Kulesza, U., Lucena, C., Staa, A. Design Patterns as Aspects: A Quantitative Assessment. Proceedings of the SBES’04, Brasília, Brazil, October 2004.
2. Garcia, A. et al. Separation of Concerns in Multi-Agent Systems: An Empirical Study. In: "Software Engineering for Multi-Agent Systems II", Springer-Verlag, LNCS 2940, State-of-the-Art Survey, January 2004, pp
3. Sant’Anna, C., Garcia, A., Chavez, C., Lucena, C., Staa, A. On the Reuse and Maintenance of Aspect-Oriented Software: An Assessment Framework. Proceedings of the SBES’03, Manaus, Brazil, October 2003.
4. Garcia, A., Chavez, C., Torres, V., Lucena, C. Engineering Multi-Agent Systems with Patterns and Aspects. Journal of the Brazilian Computer Society, SBC, Number 1, Volume 8, July 2002, pp
Software Engineering Lab (LES) – PUC-Rio

65. Ongoing Research on Aspect-Oriented Agents
The generative approach
More experimentation
Refinement of patterns (ex: learning and mobility)
Software Engineering Lab (LES) – PUC-Rio

66. Technology Evaluation
Trading Agent Competition
The Trading Agent Competition (TAC) is an international forum designed to promote and encourage high quality research into the trading agent problem.
The scenarios
TAC Classic - a "travel agent" scenario based on complex procurement on multiple simultaneous auctions
TAC SCM - a PC manufacturer scenario based on sourcing of components, manufacturing of PC's and sales to customers
The TAC 2004 is the fifth event (Classic).
The TAC research community:
USA, Canada, Mexico, Brazil, France, UK, Sweden, Netherlands, Germany, Turkey, Israel, Macau, China, Japan, Australia, etc.
Software Engineering Lab (LES) – PUC-Rio

67. TAC Classic
Software Engineering Lab (LES) – PUC-Rio

68. TAC – Trading Agent Competition
Final Scores for TAC CLASSIC (2004).
Position
Agent
Affiliation
Average
Score
Games
Played 1
whitebear04
Cornell University, USA. 35 2
Walverine
University of Michigan, USA 3
LearnAgents
PUC-Rio, Brazil. 4
SICS02
SICS, Sweden. 5
NNN
Hebrew University, Israel. 6
UMTac-04
University of Utrecht, Netherlands. 7
The Chinese University of Hong Kong, China. 8
RoxyBot
Brown University, USA.
Software Engineering Lab (LES) – PUC-Rio

69. LearnAgents – A Multi-Agent System for the TAC Classic
Flight, Hotel,
and Ticket Sensors
Flight and Hotel
Price Predictor
Allocator Master +
Allocator Slaves
Corporate
Knowledge
Base
Monitor
Agent
Flight, Hotel,
and Ticket Negotiators
Ordering Agents
(Flight & Hotels, Tickets)
Software Engineering Lab (LES) – PUC-Rio

70. LearnAgents – A Multi-Agent System for the TAC Classic
CPU 1
Flight, Hotel, Ticket Sensors (3)
Update Prices in the Knowledge Base
Filter the events from the environment
Flight and Hotel Price Predictor (2)
Flight: Maximum Likelihood
Hotel: Moving Average
Allocator Master (1)
Calculates 50 allocations per minute
Integer Programming
Ordering Agent (2)
Decide which allocation to use, and decides the next strategy.
Flight, Hotel and Ticket Negotiators (28)
Negotiates based on the strategy defined by the Ordering Agent
Hotel: MiniMax, Neural Network, Temporal Difference Learning, LMS Learning
CPU 2
Monitor Agent (1)
Stores important data from each agent
Monitor the Performance of each Agent and the Agency
CPU 3
Allocator Slave (2)
Calculates Ticket Allocation
Calculates High Price
Integer Programming
Software Engineering Lab (LES) – PUC-Rio

71. Monitor the Performance of each Agent
Monitor the Performance of the Agency
Parameters of the Agents
Tuning
Software Engineering Lab (LES) – PUC-Rio

72. Some Future Directions at LES
The generative approach
Law governing in Multi-agent systems
Architectures
Component-based
ASF Framework (MAS-ML-based)
Tool support
Case study development
Software Engineering Lab (LES) – PUC-Rio

73. Technology Transfer: MindsatWork Ltda.
Agent-Based WebClipper (instance of a Multi-Agent based Framework)
A Multi-Agent Based Architecture for the Development of Control and Acquisition Systems (Petrobras, Biogenie)
Software Engineering Lab (LES) – PUC-Rio

74. Technology Transfer: Modulo Security Solutions S.A.
A Multi-Agent System for Monitoring News about Information Security
An Agent-Based Remote Analysis of the Vulnerability of a Computer System.
Software Engineering Lab (LES) – PUC-Rio

75. Development and adoption of agent technology
Conclusions
Software Everywhere
Development and adoption of agent technology
We are in the infancy of SELMAS
Software Engineering Lab (LES) – PUC-Rio

76. References    TAO Conceptual Framework ANote
V.T. Silva, A. Brandão, C. Chavez, C. Lucena, P.S.C. Alencar. Taming Agents and Objects in Software Engineering. In: A. Garcia, et al. (eds.) Software Engineering for Large-Scale Multi-Agent Systems. Lecture Notes in Computer Science, LNCS 2603, 2004, pp
ANote
R. Choren, C. Lucena. Modeling Multi-agent Systems with ANote. Journal on Software and Systems Modeling, 2004.
R. Choren, C. Lucena. Agent-Oriented Modeling Using ANote. Proceedings of the 3rd International Workshop on Software Engineering for Large-Scale Multi-Agent Systems, SELMAS, 2004, pp   
Software Engineering Lab (LES) – PUC-Rio

77. References (cont’d)     MAS-ML
V.T. Silva, C.J.P. Lucena. From a Conceptual Framework to a Multi-Agent System Modeling Language. Journal of Autonomous Agents and Multi-Agent Systems, 9(1/2), 2004, pp
V. Silva, R. Choren, C. Lucena. Using the MAS-ML to Model a Multi-Agent System. In: C. Lucena et al. (eds.) Software Engineering for Multi-Agent Systems II. Lecture Notes in Computer Science, LNCS 2940, 2004, pp
The Aspect-Oriented Architectural Method
A. Garcia, C. Lucena, D. Cowan. Agents in Object-Oriented Software Engineering. Software: Practice and Experience, Elsevier, 2004, pp
A. Garcia, et al. An Aspect-Based Approach for Developing Multi-Agent Object-Oriented Systems. Proceedings of the 15th Brazilian Symposium on Software Engineering, SBES, 2001, pp    
Software Engineering Lab (LES) – PUC-Rio

78. References (cont’d)    The Aspect-Oriented Agent Architecture
A. Garcia, U. Kulesza, C. Lucena. Separation of Concerns in Open Multi-Agent Systems: An Architectural Approach. Proceedings of the 3rd International Workshop on Software Engineering for Large-Scale Multi-Agent Systems, SELMAS, 2004, pp
Case Studies
K. Silva, R. Choren, C. Lucena. Applying A Note to modeling Multi Agent Systems - A Case Study in E-Business. Proceedings of the IASTED International Conference on Software Engineering, SE, 2004, pp
F. Melo, et al. Deploying Agents with the CORBA Component Model. In: W. Emmerich, A.L. Wolf (eds.) Component Deployment. Lecture Notes in Computer Science, LNCS 3083, 2004, pp   
Software Engineering Lab (LES) – PUC-Rio

79. References (cont’d)    Case Studies (cont’d)
V. Silva, R. Choren, C. Lucena. A UML Based Approach for Modeling and implementing Multi-Agent Systems. Proceedings of the 3rd International Conference on Autonomous Agents and Multi-Agent Systems, AAMAS, 2004.
A. Garcia, et al. Engineering Multi-Agent Systems with Aspects and Patterns. Journal of the Brazilian Computer Society, Special Issue on Software Engineering and Databases, 8(1), 2002, pp
A. Garcia, et al. Separation of Concerns in Multi-Agent Systems: An Empirical Study. In: C. Lucena, et al. (eds.) Software Engineering for Multi-Agent Systems II. Lecture Notes in Computer Science, LNCS 2940, 2004, pp   
Software Engineering Lab (LES) – PUC-Rio

80. References (cont’d) TAC  
Jose Alberto R. P. Sardinha, Ruy L. Milidiú, Carlos J. P. Lucena, Patrick M. Paranhos, Pedro Cunha. LearnAgents - A multi-agent system for the TAC Classic. Poster Session at AAMAS' Trading Agent Competition, New York, NY, July 2004.
Jose Alberto R. P. Sardinha, Alessandro F. Garcia, Carlos J. P. Lucena, Ruy L. Milidiú. On the Incorporation of Learning in Open Multi-Agent Systems: A Systematic Approach. Sixth International Bi-Conference Workshop on Agent Oriented Information Systems(AOIS-2004) at CAiSE 2004, Riga, Latvia, Jun (to appear)  
Software Engineering Lab (LES) – PUC-Rio

81. Prof. Carlos José Pereira de Lucena lucena@inf.puc-rio.br
Thank You
Prof. Carlos José Pereira de Lucena
Software Engineering Lab (LES)
Software Engineering Lab (LES) – PUC-Rio