a framework for agent collaboration in multi-agent systems submitted by: mohamed gamaleldin atwany...

A Framework for Agent Collaboration in Multi-Agent Systems

Submitted by:

Mohamed Gamaleldin Atwany

Supervised by:

Abdel-Aziz Khamis, Phd. Magdy Aboul-Ela, Phd.Dept. of Computer and Dept. of Computer and Information Sciences, Information Systems, Cairo University Sadat Academy for

Management Sciences

A thesis submitted to the Department of Computer Science, Institute of Statistical Studies and Research, Cairo University, in partial fulfillment of

the requirements for the degree of Master in Computer Science

July 2002

A Framework for Agent A Framework for Agent Collaboration in Multi-Collaboration in Multi-

Agent SystemsAgent Systems

[email protected]@acm.org

http://www.geocities.com/matwany/http://www.geocities.com/matwany/macf.pptmacf.ppt

The Agenda

1. Introduction to Agents and Multi-Agent Systems

2. Multi-Agent Collaboration

3. Proposed Multi-Agent Collaboration Framework

4. Proposed Framework Implementation

5. The Case Study: e-Trade Agent Team

6. Summary and Conclusion

Introduction to Agents and Multi-Agent Systems

Defining Agents

have partial representation of

the environment

perceive and act upon its

environment

may be able to reproduce itself

can communicate directly with

other agents

possess skills and can offer

services

possess resources of its own

driven by a set of tendencies

autonomous behavior

Possess behavioral flexibility

and rationality

An agent is a virtual or physical computational entity that


Types of Agents Cognitive Agents

Intentional (Rational) agentsHave explicit goals motivating their actions

Module-based agentsReflexive cognitive agents

Reactive agents Drive-based agents

Directed by motivation mechanisms Agents

Respond to stimuli from the environment, behavior guided by the local state of the world in which they are immersed


Defining Intelligent Agents Able to pursue its goals and executes its actions such that it

optimizes some given performance measure Operates flexibly and rationally in a variety of environmental

circumstances, given the information they have and their perceptual and effectual capabilities

Has explicit goals motivating its action


OO Paradigm vs. Agent Paradigm Object is the basic unit Entity state definition is

unconstrainedType of messages are unconstrained

Abstraction level is lower

Agent is the basic unit Entity state defined via

Belief, commitments, goals Types of messages include

request, inform, query Abstraction level is higher

and hence, it is more suited to the development of open systems


Defining Multi - Agent SystemsA multi-agent system is a system composed of number of interacting agents

and characterized by being comprised of the following elements

An environment

A set of passive environment objects that agents can perceive, create, destroy and modify

A number of agents representing system’s active entities

A number of relations that link objects and agents to each other

A number of operations that enables agents to perceive, produce, consume, transform and

manipulate environment objects

Laws of the universe


Key Issues in Multi - Agent Systems Communication

Interaction

Coordination interactions

Cooperation interactions

Negotiation interactions

Organization interactions


Key Issues in Multi-Agent Systems

Communication A threefold problem involving knowledge of interaction protocol,

communication language and transport protocol

Forms the basis for interaction and social organization Speech Acts Theory

views natural human language as actions (a suggestion, a commitment, or a reply) classified to types (Assertive acts, Directive acts, …etc.)

KQML (content, communication, and message layers)

Conversations Defined as a series of communications among different agents that follows a

protocol and with some purpose A layered conversational model (protocol, conversation, and policy layers)



Interaction An interaction situation is an assembly of behaviors

resulting from the grouping of agents acting in order to

attain their objectives, paying attention to the resources

available to them and to their individual skills

Occurs between two or more agents brought into a

dynamic relationship through a set of reciprocal actions



Coordination Refers to either

a state of an agent community where agents’ actions fit well with each other or

to the process of achieving a state of coordination within an agent community

Agents coordinate their actions for four main reasons Agents require information and results other agents’ supply Limited resources have to be shared to optimize carried actions and

try avoid possible conflicts Enables cost reduction by eliminating pointless actions and avoiding

redundant actions Agents might have separate interdependent objectives that they need

to achieve while profiting from goal interdependencies



Cooperation Defined as coordination among non-

antagonistic agents where participants succeed or fail together

A cooperative situation is validated if eitherAdding a new agent could result in an increase in

performance levels of the groupAgent actions serve to avoid or to solve potential

or actual conflicts.



Negotiation Defined as

Interaction between agents based on communication for the purpose of coming to an agreement, or

A process by which a joint decision is reached by two or more agents, each trying to reach an individual goal or objective, or

Coordination among competitive or simply self-interested agents or,

As a distributed communication-based search through a space of possible solutions.



Negotiation Is much related to distributed conflict resolution and decision-

making Requires agents to use a common language Supports cooperation and coordination between agents The Process:

Agents make proposals Proposals are commented (refined, criticized, or refuted) by other

agents other agents then communicate their possibly conflicting positions, Agents then trying to move towards agreement by making

compromises or searching for alternatives



Organization Defined as an arrangement of relationships between

components or individuals which produce a unit, or system, endowed with qualities not apprehended at the level of the components or individuals.

An organization links, in an inter-relational manner, diverse elements or events or individuals, which thenceforth become the components of a whole.

An organization ensures a relatively high degree of interdependence and reliability, thus providing the system with the possibility of lasting for a certain length of time, despite chance disruptions


Applications of Multi-Agent Systems

Problem Solving Multi-Agent Simulation The Construction of synthetic worlds Collective robotics Kinetic program design


Collaboration in Multi-Agent Systems Defined as forms of high-level cooperation that requires the

(development of) mutual understanding and a shared view of the task being solved by several interacting entities

Collaboration occur within a team of agents cooperating to achieve some collective goal.

As a team of cooperating agents, participating agents succeed or fail together.

Sharing a mental state within a team of agents enables reasoning about their beliefs, commitments, and intentions and hence, reason about the success or failure of collaboration.


Collaboration in Multi-Agent Systems

Multi-Agent Collaboration Theories The Theory of Joint Intentions

defines logic of rational action that is intended to be used as a specification of agent design

The basic argument is that a joint activity is one that is performed by individuals sharing certain specific mental properties which affect and are affected by properties of the participants

The Shared Plans Theory several deficiencies noted in Pollack’s mental state of plans Defines the concept of a shared plan Describes the entire web of a team’s intentions and beliefs when

engaged in teamwork The Theory of Cooperative Problem Solving Process

presents a model of cooperative problem solving (CPS) characterizes agents’ mental states leading them to solicit, and

take part in, cooperative action



Multi-Agent Collaboration Frameworks GRATE

a general framework that enables the construction of multi-agent systems for the domain of industrial process control

Applications could be built very rapidly because much of the general domain behavior is already defined

STEAM enables a team of agents to act coherently in a way that overcomes

the uncertainties of complex, dynamic environments in which team members often encounter differing, incomplete and possibly inconsistent views of the world and mental state of other agents

The Issue of Interoperability The frameworks does not support interoperability

Open systems Readiness Heterogeneous agents, no pre-specified interaction protocols, no pre-

specified organization



The Development of a Shared Mental State

The shared mental state consists of the following set of shared knowledge structures:

a dependency graph of achievement goals a dependency graph of commitments to achieve these goals a dependency graph of actions believed to achieve these

goals a dependency graph of commitments to these actions a dependency graph of intentions of actions agents are

committed to achieve a dependency graph of mutual beliefs about goal relevance

and achievement status, status of commitments, status of intentions, and status of actions

Proposed Framework

Proposed Framework for Multi-Agent Collaboration

Scope Creating, sharing, and maintaining a shared mental state

within a team of agents

Objectives Framework based on a formal model of teamwork Support different phases of cooperative problem solving Transparent to existing interaction protocols and agent

organizations Transparent to development environments Transparent to agent architectures

Proposed Framework

The Methodology

Is based on the observation that behavior can be analyzed without any knowledge of the implementation details

The proposed framework should be based on two teamwork models

The proposed framework should adopt a layered conversational model

Proposed Framework

Overal Object ModelConceptual Framework

Framework Implementation

MAS Systems

State Model

Conversational Model

Framework Implementation Derivation Mechanism

BRL

ACL

Message Encoding & Decoding Facil ity

XML Message Format

Query Interaction Protocol

Conversational Pattern

Conversational Policy

Cooperation Mechanism

Negotiation Mechanism

Team Organization

Coordination Mechanism

Communication Mechanism

Formal Teamwork Model

Ontology

Extention Mechanism

Proposed Framework

Components

Define a pattern of interaction for information exchange that agents should follow

Define unambiguous rules for reasoning about agent and team behavior

Maintain a clear separation between the generic specification defined by the framework and possible implementations of that framework

Proposed Framework

Components

The State Model

_

Planning

_

Pre-Planning

_

Post-Planning

_

Execution

_

End State

_

individual commitment to subaction

_

joint commitment to subaction

_

team cannot achieve P

_

team

_

formation

_

conditions

_

are matched

_

no mutual

_

belief that

_

team T

_

can

_

achieve P

_

joint

_

commitment to

_

main action

_

drop joint

_

commitment to

_

main action

_

intend action

_

drop intention

_

to main joint

_

action

_

drop intention to action

_

jointly intend

_

main action

_

mutual belief that the goal is

_

achieved, unachievable, or

_

irrelevant

_

State State

_

intiator agent

_

recognizes problem

Proposed Framework

Components

The Conversational Model Contents

A Query Interaction Protocol A Set of Collaborative Conversational Patterns A Collaborative Conversational Policy

Proposed Framework

ComponentsThe Conversational Model

Conversational PatternsForm Team Conversational Pattern

This conversational pattern defines the preconditions, post-conditions,

reasoning, and mental states used to form a team.

Valid Team States

Pre-Planning

Pre-Conditions

the team formation facilitator agent has a designated goal to achieve

through team action.

Step 1: Reasoning Performed by the Team Formation Facilitator Agent

a. belief that a candidate team is able to achieve designated goal.

b. attempt to solicit assistance in order to form team to achieve designated

goal.

Step 2: Messages Sent by the Team Formation Facilitator Agent

c. announce to candidate team agents its attempt to solicit assistance in

order to achieve designated goal.

Step 3: Reasoning Performed by Other Candidate Agents

d. maintain belief about team formation facilitator announced attempt.

Proposed Framework

ComponentsThe Conversational Model

Conversational PolicyThe proposed agent conversation policy consists of the

following components: Domain and problem specific rules to be defined by

the agent developer. Teamwork rules explicitly defined by the proposed

framework state model through the definition of possible team states and the rules for reasoning about team states.

Teamwork rules defined by the Cooperating Problem Solving Theory

Teamwork rules defined by the Joint Intentions Theory

Proposed Framework

Integration into MAS Architectures

Agent X

InternalRepresentation A

Agent X

InternalRepresentation B

Agent X

InternalRepresentation B

Facilitator for Internal

Representation A

Facilitator for Internal

Representation B

All Agents use Framework

Implementation E2

Agent XInternal

Representation AFramework

Implementation E1

Agent YInternal

Representation BFramework

Implementation E1

Agent ZInternal

Representation BFramework

Implementation E2

Facilitator for Implementation

E1 to E2

Facilitator for Implementation

E2 to E1

Scenario #1:All agents use the same framework implementationFacilitators translate between internal representations and the framework BRL

Some agents use framework implementation E1, while others use framework implementation E2.Facilitators translate messages between framework implementations E1 and E2

Proposed Framework Implementation

The Components A behavior representation language for modeling agent

mental behavior (BRL) Ontology Modal and temporal operators grammar The extension mechanism

An agent communication language (ACL) Speech acts mechanisms

A message interchange format XML based message format Mapping to BRL elements

A set of facilitators and components XML message encoding/decoding facility

Proposed Framework ImplementationComponents

Proposed Ontology – Object Model


Proposed Ontology - Frames

Frame Field Name Description

belief id unique identification Belief

agent id unique identification for agent or team that owns this belief

Belief Message clause id unique identification for the message clause, see next

section

goal id unique identification Goal

agent id unique identification for agent or team that owns this belief

commitment id unique identification

agent id unique identification for agent or agent group that owns this

commitment

Commitment

goal id or action

id

unique identification for goal (action) that an agent is

committed to achieve (execute)


Modal and Temporal Operators

Modal clause Description

achievable Expresses the status of achievability of a given goal

irrelevant Expresses the status of relevance of a given goal

Exist Expresses the status of existence of a designated knowledge

element.

Temporal clause Description

Added Expresses the status of addition of a new designated knowledge

element.

Dropped Expresses the status of dropping a designated belief

Attempted Expresses the past status of attempt in the past of a designated

action as a result of an agent intention

attempting Expresses the current status of attempt in execution of a

designated action as a result of an agent intention

achieved Expresses the status of achievement of a given goal


Speech Acts and The Inquire Mechanism

Initiator Participant A Participant B

A. inquire

B.1 inform

B.2 inform


A Structured Message format based on XML

<team_message team_id="trade_team" message_id="attempt_to_solicit_assistance_1">

<meta_info content_type="framework-implementation-horn-clause"/>

<delivery_info sender_id="buyer">

<recipient_list>

<recipient agent_id="merchant"/>

<recipient agent_id="delivery"/>

</recipient_list>

<reply_to_list>

<recipient agent_id="merchant"/>

<recipient agent_id="delivery"/>

<recipient agent_id="buyer"/>

</reply_to_list>

</delivery_info>

<message_content>

<message_clause truth_value="true"

Proposed Framework Implementation

The ComponentsMessage encoding/decoding facility object model

Proposed Case Study

Teamwork in e-Trade

The Problem

Trade as an Organization of Trade Agents

Trade as an Interaction of Trade Agents

Trade as a Task Environment

Trade as a Cooperative System

Trade as a Coordinated System

Collaboration Within a Trade Team

Proposed Case Study

Teamwork in e-TradeMapping the Purchase Process to CPS Model Phases

Purchase Process Phase CPS Model Phase Description

Decision to buy

merchandise

Recognition The buyer recognize the

potential for performing

purchase

Pre-team Buyer negotiates with a

group of merchants to

select the best offer, and

then, construct a trade

team in order to execute

the transaction

Negotiation phase

Planning Buyer and merchant agree

on all purchase attributes

Exchange phase

Settlement phase

execution Trade team members

execute purchase

transaction steps in a

timely coordinated manner

Proposed Case Study

Teamwork in e-Trade A Knowledge-Level Model for Reasoning

about Collaboration Consisting of a set of mental elements

categorized into beliefs, goals, commitments, and intentions and their dependencies

Specifying a number of inference rules that allow reasoning about teamwork state

Enable exchange of beliefs, goals, intentions, and commitments

Proposed Case Study

Teamwork in e-TradeTrade Team Goal Hierarchy

perform paymentG1.1

receive paymentG1.2

receive merchandise

G1.4

settle buyer part of transaction

G1.6

settle merchant part of transaction

G1.5

deliver merchandise

G1.3

goal dependencyperform trade

Goal G1

Proposed Case Study

Teamwork in e-TradeA Teamwork Knowledge-Level Model

Agent Goal Attributes Goal Identification Agent Identification Goal Type Goal Addition Trigger Goal Drop Trigger

Proposed Case Study


Agent Actions Attributes Actions Identification Agent Identification Parent Action Actions Type Actions Dependency

Proposed Case Study


Agent Commitment Attributes Commitment Identification Agent Identification Commitment Type Commitment Addition Trigger Commitment Drop Trigger

Proposed Case Study


Agent Intention Attributes Intention Identification Agent Identification Intention Type Intention Preconditions Intention Post ConditionsA

Proposed Case Study


Agent Belief Attributes Belief Identification Agent Identification Belief Addition Trigger Belief Drop Trigger

Proposed Case Study

Teamwork in e-Trade An Approach for the Specification and

Development of MAS Collaborative Behavior

Develop MAS

Verify MAS

Develop Input Scenarios

Verify prototype

Develop prototype

Proposed Case Study

Teamwork in e-Trade

Collaborative Analysis Facility Based on proposed framework and proposed

framework implementation Encode agent strategy with the prototype All possible collaborative scenarios are

encoded into program input Validate generated behavior

Proposed Case Study

Teamwork in e-Trade Case Study Results

Agent interaction and communication is crucial for maintaining a shared and consistent view of the trade problem

A common view of the goals, actions, commitments, and intentions help agents reason on teamwork activities and state

The use of conversational model helped agents reason about teamwork activities and state

The implementation enabled agents to express collaborative mental behavior, using a set of agent interaction mechanisms, and transmitted using a common message format

By reviewing generated output, the MAS developer is able to verify team and individual collaborative behavior

Summary and Conclusion Multi-agent systems are complex systems consisting of a

number of agents, each of which by itself might represent an organization consisting of one or more agents

within a MAS, agents interact in order to achieve their individual and collective goals in an act defined as collaboration

The lack of support for interoperability and open systems in existing environments

Proposed framework provides transparency to current development environment, interaction protocols, and agent organizations

Summary and Conclusion Separating framework from possible implementation enables

the evolution of implementations that match environment needs

The proposed implementation provides an ACL, a BRL, a message format, and a message encoding/decoding facility

The agent paradigm suitable for the developing of open systems as found in the domain of e-Trade

The proposed framework and the proposed framework implementation enabled the development of a MAS in the domain of e-Trade

The proposed iterative approach eases the process of specification and development of MAS collaborative behavior

QuestionsQuestions

Thank YouThank You

a framework for agent collaboration in multi-agent systems submitted by: mohamed gamaleldin atwany...

Documents

agents actions

agent community agents

number of agents

grouping of agents

different agents

multiagent systems key

redundant actions agents

lower agent