international conference on cognitive modeling 2010 brahms tutorial

ICCM 2010

Introduction to Brahms

Multi-Agent Activity Modeling with the

Brahms Environment

Maarten Sierhuis, Ph.D.

Area Manager

Knowledge, Language & Interaction

Palo Alto Research Center (PARC®)

[email protected]

NASA Ames Associate

[email protected]

ICCM 2010 2

NASA ARCBrahms Team

• Co-PI - William J. Clancey, Ph.D.

• Co-PI - Maarten Sierhuis, Ph.D.

• Software Architect - Ron van Hoof

• Software Developer - Mike Scott

• Brahms Modeler - Chin Seah

ICCM 2010

Download This Tutorial From

http://public.me.com/msierhuis/ICCM2010/Brahms ICCM 2010 Tutorial (final).ppt

3

ICCM 2010

Schedule• 1:00pm – 2:00pm

– Work Practice Modeling– Brahms Overview + Demo

• 2:00pm – 2:30pm– Agents, Groups, Beliefs and Facts– Classes and Objects

• 2:30pm – 3:00pm– Break

• 3:00pm – 3:30pm– Thoughtframes– Geography

• 3:30pm – 4:30pm– Activities and Workframes– Composite Activities– Agent Communication

4

ICCM 2010

Human-Centered Design

• A different engineering approach

• Start with understanding people’s practice– Both the way people collaborate, as well as the culture of an

organization is encompassed in the communities of practice of an organization—the work practices of the people (Wenger, 1998)

• What is Practice? – People’s collaboration, “off-task” behaviors, multi-tasking,

interrupted and resumed activities, informal interaction, knowledge and use of space (geography)

– It’s the way people work

• Human-Centered Design Methodology– Understanding or designing a work system by modeling and

simulating people’s work practice

• Design information systems in the context of work

– It’s all about understanding the context

5

ICCM 2010

• Hermeneutics and Work Practice Hermeneutics and Work Practice (Winograd & Flores)(Winograd & Flores)– Interpreting the world in order to actInterpreting the world in order to act

• Situated Action ModelsSituated Action Models (Suchman, Lave) (Suchman, Lave)– Moment-by-moment analysis of interaction between people, Moment-by-moment analysis of interaction between people,

tools and environment in a particular situation (or setting).tools and environment in a particular situation (or setting).

• Activity Theory Activity Theory (Vygotsky, Leont’ev, Nardi)(Vygotsky, Leont’ev, Nardi)– An activity creates and encompasses the context through its An activity creates and encompasses the context through its

enactment of actions and operations of the people engaged, and enactment of actions and operations of the people engaged, and the artifacts used.the artifacts used.

• DistributedDistributed Cognition Cognition (Hutchins)(Hutchins)– The study of representation of knowledge both inside people’s The study of representation of knowledge both inside people’s

heads and in the artifacts and systems they use, i.e. the total heads and in the artifacts and systems they use, i.e. the total cognitive system (e.g. the pilots and the cockpit).cognitive system (e.g. the pilots and the cockpit).

Understanding ContextUnderstanding Context

6

ICCM 2010

What is practice?

Sierhuis’ Definition:Sierhuis’ Definition:

• situatedsituated activitiesactivities of a group of people of a group of people • synchronously or asynchronouslysynchronously or asynchronously• collaborating and communicatingcollaborating and communicating• gaining experience gaining experience

Practice is “doing-in-action” (Suchman, ’87)“Collective learning results in practices.” (Wenger, ‘97)

What is it that results? Can it be described? Can it be modeled?

7

ICCM 2010

Work Practice Modeling

• Groups & AgentsGroups & Agents– work as activitieswork as activities

– beliefs trigger workbeliefs trigger work

– bounded rationality is socially bounded rationality is socially and culturally definedand culturally defined

• Collaboration between Collaboration between AgentsAgents– agents react to and interact agents react to and interact

with other agentswith other agents

– same time/same placesame time/same place

– same time/different placesame time/different place

– different time/same placedifferent time/same place

– different time/different placedifferent time/different place

8

ICCM 2010

WPM cont’d

• Tools & ArtifactsTools & Artifacts– tools used in activitiestools used in activities– artifacts created in activitiesartifacts created in activities

• Environment/GeographyEnvironment/Geography– agents have a location agents have a location – artifacts have a locationartifacts have a location– detecting real-world factsdetecting real-world facts

• CommunicationCommunication– is situated is situated – the means of communication the means of communication

depends on the situation (e.g. depends on the situation (e.g. voice loop, f2f communication, voice loop, f2f communication, telephone, faxing, e-mail)telephone, faxing, e-mail)

– impacts efficiency of workimpacts efficiency of work

9

ICCM 2010

“Day in the life” (DITL) Model

• Description of a day in the life of a …– Person– Group of people

• Abstraction of what people do during the day …– Activities– Interactions– Locations– Use of systems and artifacts

10

ICCM 2010 11

Brahms

• How people work together

• How people get to participate in a collaborative activity

• Role of the environment / use of space

• How communication happens

• The tools and artifacts used

• Individual motives, history and culture

• Not just problem-solving activities

• Brahms is a agent-based modeling and simulation environment

• Simulation as a work system design/analysis tool

• Agents represent social and collaborative Humans

• Agents are situated, deliberative, cognitive and reactive

• Understanding how people really work• Developing multi-agent systems

Theoretical Practical

ICCM 2010

From Simulation to From Simulation to ImplementationImplementation

12

ICCM 2010

2

4

6

8

10

TRL

1992 1996 2000 2004 2008

NYNEX/IRL

JPL

JSC

APOLLO

VICTORIA

ISS Mission Control

BRAHMS – HISTORY OF APPLICATIONS

ISS

SurfaceHab

MobileAgents

OnboardISS

201113

ICCM 2010 14

Brahms Overview

The Language

ICCM 2010

BrahmsBrahms

Work is like a symphony,Well rehearsed, but always different

15

ICCM 2010

HistoryHistory

Social Systems of Work

Work Practice

Communities of Practice

Work Process

Modeling

Expert Systems

1992 - 1997

1998 - Now

Agents

Simulation

EthnographyCommonKADS

16

ICCM 2010

Types of Agent Languages

Goal

17

ICCM 2010

Brahms Agent EnvironmentBrahms Agent Environment

• ComposerComposer for building models. for building models.(or Integrated Development Environment).(or Integrated Development Environment).

• CompilerCompiler for compiling models. for compiling models.

• Virtual MachineVirtual Machine for simulating models. for simulating models.(or Simulation Engine).(or Simulation Engine).

• Agent ViewerAgent Viewer for viewing simulations. for viewing simulations.

18

ICCM 2010 19

Where to download Brahms?

ICCM 2010

Brahms Use License

• Brahms Research License– Free use for research purposes

– Download Brahms from AiS website

– Receive license file to be installed on your machine

• Brahms Commercial License– Used for any commercial application of Brahms

– Contact Maarten Sierhuis

20

ICCM 2010 21

Brahms Download

• “Brahms Agent Environment”.– Go to http://www.agentisolutions.com/download/index.htm

• Requires MySQL 5.x. database to be installed. – Go to http://www.mysql.com– Note: Read AgentViewer_Readme.html to set the appropriate

privileges for the anonymous database user.

• Requires a Brahms License file.– Sent to you when signed up

• Install Brahms Eclipse Plugin from AiS

ICCM 2010 22

Brahms InstallationCheck out Brahms Files

Files created by the Brahms installation are in:

C:\Program Files\ Brahms\AgentEnvironment

ICCM 2010 23

Language Documentation

ICCM 2010 24

ICCM 2010 25

ICCM 2010 26

Brahms Tutorial

ICCM 2010

Brahms LanguageBrahms Language

• Agent Oriented / BDIAgent Oriented / BDI– Agents are first-class citizensAgents are first-class citizens– Agents are belief basedAgents are belief based– Agents are intention based:Agents are intention based:

• Beliefs become intentions that trigger reasoning- and/or situation-action rulesBeliefs become intentions that trigger reasoning- and/or situation-action rules

– Agents can communicateAgents can communicate• Organizational ModelingOrganizational Modeling

– Agents can be modeled within a hierarchical member-of inheritance structureAgents can be modeled within a hierarchical member-of inheritance structure

• Object-basedObject-based– Objects can represent physical artifacts, data and concepts to reason withObjects can represent physical artifacts, data and concepts to reason with– Integration of Java objects as data objects, Java activities and Java agentsIntegration of Java objects as data objects, Java activities and Java agents

• Geography-basedGeography-based– Areas can be conceptual representations of locationsAreas can be conceptual representations of locations– Areas can be located within other areas, creating a hierarchical environment Areas can be located within other areas, creating a hierarchical environment

modelmodel– Agents and objects can be located within an areaAgents and objects can be located within an area

27

ICCM 2010

Anatomy of a Brahms ModelAnatomy of a Brahms Model

• Groups and Agents– Groups contains agents which are located in aGroups contains agents which are located in a

• Geography– and these situated agents haveand these situated agents have

• Beliefs and Facts– that lead them to engage inthat lead them to engage in

• Activities– that take a certain amount of time which are that take a certain amount of time which are

specified within an agent’sspecified within an agent’s• Workframes

– that are triggered by agent’s beliefs. Activities that are triggered by agent’s beliefs. Activities can change an agent’s beliefs or affectcan change an agent’s beliefs or affect

• Classes and Objects– which could then change facts in the world. which could then change facts in the world.

Facts can be detected by agents and Facts can be detected by agents and change an agent’s beliefs which might triggers an agent’schange an agent’s beliefs which might triggers an agent’s

• Thoughtframes– that can further change an agent’s beliefs and that can further change an agent’s beliefs and

trigger an agent’s workframe.trigger an agent’s workframe.

Group = Student, Agent = Alex

Geography = Berkeley, CA

Belief = Alex is hungry

Activity = Eating

Workframe = When hungry go eat

Object = Money, Debit card, ATM

Thoughtframe = If no money go to the ATM machine

28

ICCM 2010 29

Anatomy of a Brahms ModelLanguage Concepts

Groups

Agents

Classes

Objects

AreaDefs

Areas

Paths

Conceptual Classes

Conceptual Objects

Attributes (OA-V)

Relations (ORO)

Beliefs

Facts

Activities

Workframes

Preconditions

Consequences

Detectables

Thoughtframes

agent-based

object-based

geo-based

object flows

mental state/world state

activity-based/subsumption

rule-based

reactive

inferences

ICCM 2010

Time-based Situation-Action

Production Rules represent Qualitative Relationships!!

Situation-action RulesSituation-action Rules

30

World State

F1, F2, F4 …..

E4

Detect F4

ICCM 2010 31

Brahms: agent-oriented languageSymbolic Discrete Event BDI System

Production and Situation-action Rules

WF1 C1 and C2 and C3 => Detect F4, Activity1(t), B5, B7

WF2 C4 => Activity2(t), B6

WF3 C5 and C6 and C7 => Detect F8, Activity1(t), F9

F1, F2, F4 …..

Workframe Rule Memory

Belief Memory

Agent

World State

B1, B2, B3, ………..

Next time event

State at next time event

TF1 C1 and C5 => B2

TF2 C8 => B9

TF3 C9 and C6 and C8 => B10

Thoughtframe Rule Memory

ICCM 2010 32

Brahms Agent

Engine

Beliefs(atomic

formulas)

Desires(workframes

thoughtframes belief matching)

Plans(workframes

Thoughtframesactivities)

Intentions(workframes

thoughtframesinstantiations)

input

actionoutput

Beliefs-Desires-Intentions

ICCM 2010

Event Scheduler

T:0 belief XT:1 fact Y….….T:N belief Z

Multiagent LanguageMultiagent Language

BRAHMSVirtual

Machine

World State

fact F1.

fact Fn

fact Y Java Activity

class Activity2 extends AbstractExternalActivity{ …. }

class Agent5 extends AbstractExternalAgent

{ …. }

Java Agent 5

Brahms Geography

Java VM

33

ICCM 2010 34

Brahms: Distributed MAS

AgentDirectoryService

ICCM 2010 35

Brahms Agent Engine

ICCM 2010

Architecture

Brahms Virtual Machine

VMController

EventNotifier

Logger

Compiler

CompiledConcepts

SourceCode

StaticModel

DynamicModel

WorldState

ModelDB

Logs

ConfigAgent /EngineScheduler

AnalysisTools

JAPI

Interactive Development Environment

BrahmsComposer

(Eclipse Plugin)

Brahms: the environment

36

ICCM 2010 37

Model Development Cycle

1. Build Model in Composer

2. Compile Model

3. Fix compiler errors

4. Run Simulation

5. Parse History File MySQL DB

6. Display Results in AgentViewer

1

2

ICCM 2010 38

Composer: Eclipse Plugin

ICCM 2010 39

Model File

ICCM 2010

Composer: Eclipse Plugin

40

ICCM 2010 41

Compile Model

ICCM 2010 42

Model in UML (Poseidon)

Compiler generates.XMI file

ICCM 2010 43

Simulate

ICCM 2010 44

Parse History File

ICCM 2010 45

History File Format

ICCM 2010 46

Brahms AgentViewer

ICCM 2010

DEMO

47

ICCM 2010




• 2:30pm – 3:00pm– Break



48

ICCM 2010

Agents, Groups, Beliefs and Facts

49

ICCM 2010

What is a Brahms agent?

• Agents model human behavior.• Agents could be autonomous intelligent systems• Attributes of an agent:

– autonomy, – social ability, – reactivity, – pro-activeness, – mobility– bounded rationality.

50

ICCM 2010 51

Brahms Agentagent agent-name { GRP.group-membership }

{

{ display : ID.literal-string ; }

{ cost : ID.number ; }

{ time_unit : ID.number ; }

{ location : ARE.area-name ; }

{ icon : ID.literal-string ; }

{ GRP.attributes }

{ GRP.relations }

{ GRP.initial-beliefs }

{ GRP.initial-facts }

{ GRP.activities }

{ GRP.workframes }

{ GRP.thoughtframes }

}

external agent agent-name ;

See Brahms Language Reference

http://www.agentisolutions.com/documentation/language/ls_agt_stx.htm

You can refer to an agent using an agent reference:

• agent-name• the keyword current

This refers to the current agent,Like this in Java

• frame variable• activity parameter

ICCM 2010

What is a Brahms group?

• A Brahms group describes the abstract properties and behaviors of a group of agents

• Types of groups:– Functional

– Organizational

– Social

– Community of Practice

• Groups can be members of multiple groups

• Agents can be members of multiple groups

52

ICCM 2010 53

Brahms Groupgroup group-name { group-membership }{{ display : ID.literal-string ; }{ cost : ID.number ; }{ time_unit : ID.number ; }{ icon : ID.literal-string ; }{ attributes }{ relations }{ initial-beliefs }{ initial-facts }{ activities }{ workframes }{ thoughtframes }}

group-membership ::= memberof group-name [ , group-name ]*


http://www.agentisolutions.com/documentation/language/ls_grp_stx.htm

ICCM 2010 54

Groups

group BaseGroup { … }

group Student memberof BaseGroup { … }

agent Alex memberof Student, BrahmsModeler { … }

BaseGroup

Student

Alex

BrahmsModeler

ICCM 2010

NASA Mission ControlOrg Chart

Divisions

Branches

Groups

Roles

Agents

55

ICCM 2010

56

Apollo 12 Groups and Agents

Roles

Functions

Agents

ICCM 2010 57

Brahms Groups, Agents and Attributes: What are Brahms attributes?

• Attributes represent a property of a group/class or agent/object.

• Attributes have values.

• Scope of an attribute:– Private – cannot be inherited.– Protected – access only for members of group.– Public – access by any group or agent.

• Attribute values are assigned or changed by asserting new beliefs or facts.

ICCM 2010 58

Typed Language

• Value Types

• Collection Types

• Java Types

• Meta Types

• Value Assignments


http://www.agentisolutions.com/documentation/language/ls_att_stx.htm

ICCM 2010 59

Value Types

• Value Types are attribute types that assign a specific value[ int | long | double | symbol | string | boolean ]

• All types can have the value unknown

group Student {attributes:

public boolean male; [unknown | true | false] public int howHungry; [unknown | { + | - } unsigned]private double preferredCashOut; [unknown | [ integer.unsigned ]private long perceivedtime; [unknown | { + | - } unsigned { l | L }public symbol colorHair; [unknown | [ letter ][ letter | digit | ‘-’ ]* ]protected string name; [unknown | " [ letter | digit | ‘-’ | ‘:’ | ‘;’ | ‘.’ ] ” ]

}

ICCM 2010 60

• The only Collection Type is the map attribute type[ map ]

• Allows for the assignment of multiple values to the attribute where each value is addressable using an index or key.

• The attribute values are index/value or key/value pairs. The index or key can be any positive integer or string value.

• All types can have the value unknown


public map myMap;}

Collection Types

myMap(1) = 10myMap(2) = 20LmyMap(3) = currentmyMap(4) = truemyMap(5) = “a string”myMap(6) = asymbolmyMap(7) = unknown

myMap(“string1”) = 10myMap(“string2”) = 20LmyMap(“string3”) = currentmyMap(“string4”) = truemyMap(“string5”) = “a string”myMap(“string6”) = asymbolmyMap(“string7”) = unknown

ICCM 2010 61

• The only Java Type[ java ]

• Allows direct references from Brahms to Java objects. • The compiler must be able to load the Java class for the type, the compiler uses

this method to ensure that the Java type is valid. The Java classes for the types used in the Brahms language must therefore be in the Java classpath.

• The Java type name must be resolvable to a fully qualified name using the Java import statements (jimport).

jimport gov.nasa.arc.brahms.modat.kfx.KfxFile;group Student {

attributes:public java(KfxFile) javaKfxFile;

}

Java Types

gov.nasa.arc.brahms.modat.kfx.KfxFile.java:

public class KfxFile implements Serializable {

}

ICCM 2010 62

• Meta Types allow for binding of concepts that are considered to be a subtype of the meta types

• Meta Types allow Brahms User Defined types as values of attributes, variables and parameters.


public Group myGroup;}

Meta Types

Group ::= any groupAgent ::= any agentClass ::= any classObject ::= any objectActiveClass := any group or classActiveInstance ::= any agent or objectActiveConcept ::= any active class and active instanceConcept ::= any active concept, conceptual concept and geography concept

Example: Maarten.myGroup = Student

ICCM 2010 63

Brahms Facts & Beliefs:What is a Brahms belief?

• Represents an agent’s interpretation of a fact in the world.– “South Hall is 65 degrees but Alex believes its 80

degrees.”• Represents an agent’s conception of the world (s)he lives in.

– “I am a student at University of California, Berkeley.”• Beliefs are “local” to an agent.• Agents can reason about their beliefs.• Agents can communicate their beliefs.

ICCM 2010 64

Brahms Facts & Beliefs:What is a Brahms fact?

• Represent some physical state of the world. – Alex is male is true.

• Facts are globally true in the world.

• Agents do not reason with or act directly on facts.

• Agents can detect facts in the world (noticing or sensing).

ICCM 2010 65

Belief & Fact Grammar

( [ value-expression | relational-expression ] )

value-expression ::= obj-attr equality-operator value | obj-attr equality-operator sgl-object-ref

equality-operator ::= = | !=value ::= ID.literal-string | ID.number | PAC.param-name |

unknown

Examples: (Alex.male = true) (Alex.male != false) (Alex.hasCash = cash-obj-1)

relational-expression ::= tuple-object-ref REL.relation-name sgl-object-ref { is ID.truth-value }

Examples: (Alex isMarriedTo Kim) (Alex isMarriedTo Kim is false)

ICCM 2010 66

Belief Creation & Changes

• Initial beliefs can be defined in agent or group• Initial beliefs are inherited via group membership• Beliefs can be created/changed using a conclude

statement• Beliefs can be communicated by other agents or

objects• Beliefs can be created by detecting facts• Beliefs can be automatically created by the agent’s

engine

ICCM 2010 67

Initial Beliefsgroup Student {

attributes:public boolean male;

initial_beliefs:(current.male = unknown);

}

agent Alex memberof Student {}

agent Kim memberof Student {initial_beliefs:

(current.male = false);}

ICCM 2010 68

Beliefs versus Object Attribute values

Object Orientation

object Alex instanceof class student {

Public Boolean male = true;

}

object Kim instanceof class student {

Public Boolean male = false;

}

Belief-basedagent Alex memberof Student {

attributes:public boolean male;

initial_beliefs:(current.male = true);(Kim.male = false);

}

agent Kim memberof Student {attributes:

public boolean male;intial_beliefs:

(Alex.male = false);(current.male =

false);}

ICCM 2010 69

Conclude a belief

• The conclude statement can be used to create a new, or change an existing belief

conclude ( ( resultcomparison ) { , belief-certainty } { , fact-certainty } ) ;

resultcomparison::= [ result-val-comp | PRE.rel-comp ]result-val-comp::= BEL.obj-attr BEL.equality-operator PRE.expression

|BEL.obj-attr BEL.equality-operator ID.literal-symbol |BEL.obj-attr BEL.equality-operator ID.literal-string |BEL.obj-attr BEL.equality-operator BEL.sgl-object-ref |BEL.tuple-object-ref BEL.equality-operator BEL.sgl-object-ref

belief-certainty::=bc : ID.unsignedfact-certainty::=fc : ID.unsigned

conclude((current.male = true), bc:100, fc:0);

ICCM 2010 70

Relationship between Brahms Facts and Beliefs

ICCM 2010 71

Brahms Facts & Beliefs

WorldFact: (It is 10:40AM)

Fact: (It is Thursday Nov. 19, 1969)

Fact: (the door of the SEQBay is closed)

Fact: (AlBean is located in the SEQBayArea)

Fact: (PeteConrad is located in the SEQBayArea)

Agent Al Bean

belief: (the door of the SEQBay is closed)

belief: (AlBean is located in the SEQBayArea)

Agent Pete Conrad

belief: (AlBean is located in the SEQBayArea)

belief: (PeteConrad is located in the SEQBayArea)

belief: (PeteConrad is ready to offload the ALSEP)

belief: (PeteConrad is located in the SEQBayArea

detect

open

open

belief: (the door of the SEQBay is closed)

belief: (PeteConrad is ready to offload the ALSEP)

communicate

ICCM 2010

Classes and Objects

72

ICCM 2010

Brahms Objects• Objects are data and real world artifacts.• Objects could be inanimate objects or computational objects.• Why objects and agents?

– Most agent languages only represent agents.– Brahms incorporates our theory of work practice, and from a social

and practice perspective people do differentiate between intentional agents (i.e. humans) and artifacts

• Examples:– Fax machines– Database– Instruments– Rock samples– Photo Cameras– Space Suits– ATM

73

ICCM 2010

Brahms Class

• Describes the properties/behavior of abstract object classes

• Types of classes:– Physical artifacts– Computers– Data

• Multiple Inheritance

74

ICCM 2010

More on Brahms Objects

• Objects do not need a location to exist and act in the world.– For example, Alex’s bank account.

• Objects, unlike agents, can react to both facts and beliefs.– For example, data in a computer are beliefs but might not

necessarily be facts.

• Facts about objects defines the context• Agent beliefs about objects

– Defines how people get to do what they do …

75

ICCM 2010 76

BREAK!!!

ICCM 2010




• 2:30pm – 3:00pm– Break



77

ICCM 2010

Thoughtframes

78

ICCM 2010

What are some of Alex’s thoughts?

79

ICCM 2010

Example Thoughtframes

• When student is studying Do– Read While You Study

• While Reading, continously– Assess How Hungry You Are– Based on norms about how much you want to spend

for lunch and how hungry you are, select how much cash you need

– Based on needed cash and norms decide if you need to get money and how much

80

ICCM 2010

Need Cash To Eat?

• IF amount of cash < preferred cash in pocket THEN need cash

• IF amount of cash >= preferred cash in pocket THEN don’t need cash

81

ICCM 2010

How much cash needed?

• IF need cash AND very hungry THEN take out $15 (remember it’s a student!)

• IF need cash AND not very hungry THEN take out $10

• IF need cash AND not hungry at all THEN take out $5

82

ICCM 2010

How hungry am I?

• IF my hunger level > 20 THEN my hungriness is high

• IF my hunger level <= 10 THEN my hungriness is low

• IF 10 >= my hunger level <= 20 THEN my hungriness is medium

83

ICCM 2010

How long before I am hungry?

• IF an hour has gone by THEN my hunger level goes up by 3

84

ICCM 2010

Thoughtframes

• Thoughtframes are production rules

• Conclude new beliefs

• Do NOT take time

• Allows representation of agent’s reasoning (inference).

• Can be located inside composite activities– Activities constrain when an agent uses specific reasoning

• Fire immediately whenever preconditions match beliefs.

• Allows for implementation of problem solving activities– While “in” an activity, the agent reasons using its thoughtframes

– Conclusions of new beliefs in TFs can execute new TFs, WFs and activities

85

ICCM 2010

Thoughtframes Syntax

thoughtframe thoughtframe-name {display: literal-string;repeat: truth-value;priority: unsigned;variables: variable declarationwhen(precondition-declaration)do {

thoughtframe-body-element}

}

thoughtframe-name ::= name thoughtframe-body-decl ::= do { [ thoughtframe-body-element ; ]* }thoughtframe-body-element ::= CON.consequence

86

ICCM 2010

Preconditions

thoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when (knownval(current.needCash = true) and knownval(current.hungryness = high))

do conclude((current.preferredCashOut = 15), bc:100, fc:0);

}

}

precondition ::= [ known | unknown ] ( novalcomparison ) |

[ knownval | not ] ( evalcomparison )


http://www.agentisolutions.com/documentation/language/ls_tfr_stx.htm

87

ICCM 2010

Knownval Preconditionthoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when (knownval(current.needCash = true) and knownval(current.hungryness = high))do {

conclude((current.preferredCashOut = 15), bc:100, fc:0);}

}

• Assume agent Alex has the following beliefs:– (Alex.needCash = true)– (Alex.hungryness = high)

• Knownval matches on Exists(belief b) [ (ObjOrAgt.attr = value) ] in agent’s belief-set

• Both knownval preconditions are TRUE• You are allowed to leave the knownval keyword out

88

ICCM 2010

Not Preconditionthoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when (not(current.needCash = true) and knownval(current.hungryness = high))

do {conclude((current.preferredCashOut = 15), bc:100, fc:0);

}}

• Assume agent Alex has the following beliefs:– (Alex.needCash = false)– (Alex.hungryness = high)

• Not matches on Forall(belief b) [ not(ObjOrAgt.attr = value) ]in agent’s belief-set

• Knownval precondition is TRUE• Not Precondition is TRUE, because not(Alex.needCash = true) equal to TRUE

89

ICCM 2010

Known Preconditionthoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when (known(current.needCash) and knownval(current.hungryness = high))


}}


• Known matches on Exists(belief b) [ (ObjOrAgt.attr = <any-value>) ] in agent’s belief-set

• Knownval precondition is TRUE• Known Precondition is TRUE, because (Alex.needCash = true)

90

ICCM 2010

Unknown Preconditionthoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when (unknown(current.needCash) and knownval(current.hungryness = high))


}}


• Unknown matches on Forall(belief b) [ not(ObjOrAgt.attr = <any-value>) ] in agent’s belief-set

• Knownval precondition is TRUE• Unknown Precondition is FALSE, because (Alex.needCash = true)

91

ICCM 2010

Where to define thoughtframes?

• Within a group or agent• At the “top-level”

group Student {attributes:relations:initial_beliefs:initial_facts:activities:workframes:thoughtframes: thoughtframe tf_HowMuchMoneyToGet_HungryEQhigh {

when ( (current.needCash = true) and (current.hungryness = high) )


}//do }//tf_HowMuchMoneyToGet_HungryEQhigh

}//group Student

92

ICCM 2010

Where to define thoughtframes?

• Inside a composite activity within a group or agent• Model problem-solving as an activity

group Student {attributes:relations:initial_beliefs:initial_facts:activities:

composite_activity SolveCashOutProblem( ) {activities:workframes:thoughtframes:

thoughtframe tf_PreferredCashOut {when ((Alex_Cash.amount < 10.00)) do {

conclude( (Alex_Cash.amount = Alex_Cash.amount + 10.00), bc:100, fc:0);}//do

}//tf_PreferredCashOut}//composite_activity SolveCashOutProblem

}//group Student

93

ICCM 2010

What are Brahms Variables?

• Variables can be used in a Workframe or Thoughtframe to write more generic Workframes and Thoughtframes.

• Before a variable can be used it has to be declared.

• The scope of the variable is bound to the frame it is declared in.

94

ICCM 2010

Using Variables in Thoughtframe

thoughtframe tf_PreferredCashOut {when( (Alex_Cash.amount < 10.00)) do { conclude((Alex_Cash.amount = Alex_Cash.amount + 10.00));}//do

} //tf_PreferredCashOut

Without Variable With Variable

thoughtframe tf_PreferredCashOut {variables:

forone(Cash) cashobj;when((current.myCash = cashobj)

and(cashobj.amount < 10.00))

do { conclude( (cashobj.amount = cashobj.amount + 10.00));}//do

} //tf_PreferredCashOut

95

ICCM 2010

Brahms Variable Types

Did you notice “forone(cash) myCash” declaration statement?

• Variables supports three quantifiers: • Foreach

– binds to each belief-instance that can be bound to the variable, a separate frame instantiation is created.

• Forone– binds to only one belief-instance, and only one frame is created. A forone variable binds

to the first belief-instance found and ignores other possible matches.

• Collectall– binds to more than one belief-instance. The variable is bound to all matching belief-

instances, but only one frame is created.

See Brahms Language Referencehttp://www.agentisolutions.com/documentation/language/ls_var_stx.htm

96

ICCM 2010

Forone Variable

• Let’s assume three Cash objects: Cash-1, Cash-2, Cash-3

• Forone binds to one of the Cash objects, e.g. Cash-3

• Thoughtframe fires once!

• Belief about Cash-3.amount is changed


forone(Cash) cashobj;when(knownval(current.myCash = cashobj)

andknownval(cashobj.amount < 10.00))

do { conclude( (cashobj.amount = cashobj.amount + 10.00), bc:100, fc:100);}//do

}//tf_PreferredCashOut

97

ICCM 2010

Foreach Variable


• Foreach binds to every Cash object, Cash-1, Cash-2, and Cash-3

• Thoughtframe fires three times!• Once for each variable binding to one

of the Cash objects• Beliefs about Cash-1.amount,

Cash-2.amount and Cash-3.amount are changed

• All at the same time, because thoughtframes take zero time!!


foreach(Cash) cashobj;when( (current.myCash = cashobj)

and (cashobj.amount < 10.00))



98

ICCM 201099

Collectall Variable


• Collect creates a list with every Cash object, Cash-1, Cash-2, and Cash-3 in it

• Thoughtframe fires one time!• Loops over the entire list • Beliefs about Cash-1.amount,

Cash-2.amount and Cash-3.amount are changed

• All at the same time, because the thoughtframe fires only once!!


collectall(Cash) cashobj;when( (current.myCash = cashobj)

and(cashobj.amount < 10.00))



ICCM 2010100

Need Cash To Eat?

• IF amount of cash < $12 and want preferred cash in pocket THEN need cash

• IF amount of cash > $11 THEN don’t need cash

thoughtframe tf_cash_Needed { variables: forone(Cash) cs; when(knownval(current.needCash = false) and knownval(current.calculatedCash = false) and knownval(current.howHungry > 20.00) and knownval(current hasCash cs) and unknown(current.chosenDiner) and knownval(cs.amount < 12.00) and knownval(current.preferredCashOut > 0.00)) do { conclude((current.calculatedCash = true), bc:100); conclude((current.needCash = true), bc: 100); }}

thoughtframe tf_cash_Not_Needed { variables: forone(Cash) cs; when(knownval(current.needCash = false) and knownval(current.howHungry > 20.00) and knownval(current.calculatedCash = false) and knownval(current hasCash cs) and unknown(current.chosenDiner) and knownval(cs.amount > 11.00)) do{ conclude((current. calculatedCash = true), bc:100); conclude((current.needCash = false), bc:100); }}

ICCM 2010101

Which Diner?

• IF not found a diner yet,

• AND amount of cash > $15 THEN goto Blake’s diner.

thoughtframe tf_chooseBlakes { variables: forone(Cash) cs; when(knownval(current hasCash cs) and knownval(cs.amount > 15.00) and knownval(current.checkedDiner = false)) do { conclude((current.chosenDiner = Blakes_Diner), bc:100); conclude((current.checkedDiner = true), bc:100); }}

ICCM 2010102

Forward Reasoning

Student.preferredCashOut

Init: T=0: (Alex_Agent.preferredCashOut = 8.00);

Student.preceivedTime

Cash.amountT=24700: (Alex_Cash.amount = 13.00);

T=25200: (Alex_Agent.perceivedtime = 8);

Campanile_Clock.timeT=28800: (Campanile_Clock.time = 9);

tf_feelHungry

T=28800: (Alex_Agent.howHungry = 21.00);T=28800: (Alex_Agent.perceivedtime = 9);

Student.howHungrytf_cash_Neededtf_cash_Not_Needed

T=28800: (Alex_Agent.needCash = false);

Student.needCash Student.checkedDiner

tf_ChooseRaleightf_ChooseBlakes

Student.chosenDiner

T=28800: (Alex_Agent.chosenDiner = Raleigh_Diner)T=28800: (Alex_Agent.checkedDiner = true);;

T=28800: (Alex_Agent.checkedDiner = false);

ICCM 2010

Geography

103

ICCM 2010

Brahms Geography• Agents and objects can be located (initial location).• Agents and objects can move to/from locations.• When agents/objects come into a location, the Brahms engine

automatically creates a location fact (agent.location = <current-area>).

• Agents know where they are and notice others:– When agents come into a location, the Brahms engine automatically gives the

agent a belief about its new location (same as fact), and– … gives the agent a location belief for all other agents and objects currently in

that location.– When an agent/object leaves a location, the location fact and beliefs are retracted

(from all agents that are in that location the moment the agent/object leaves.)

• Agents and objects can carry (containment relation) other agent/objects.– Contained objects are NOT noticed until they are put into the area.

104

ICCM 2010

Geography Objects

• Geography Model is separate from Agent and Object Model

• Conceptual Geography Model– Areas are a special type of geography object

– Areas have attributes and relations

– Areas can define initial facts

– Areas are instances of an Area Definition

– Area Definition is a special geography class type

• Facts about areas represent state of a location– E.g. temperature

• Agent location attribute is inherited from Brahms BaseGroup. For objects from BaseClass

105

ICCM 2010

BaseAreaDef

• BaseAreaDef is the Brahms base class for areas

• Every area is an instance of BaseAreaDef

• Default special relations in BaseAreaDefpackage brahms.base;/** * areadef BaseAreaDef * * This areadef serves as the base for every area definition in a * brahms model and provides conceptual classes with a minimum work set. * * library brahms.base */areadef BaseAreaDef {

relations:public BaseAreaDef isSubAreaOf;public BaseAreaDef hasSubArea;

} // BaseAreaDef

106

ICCM 2010

• Default Brahms AreaDefs: – World, City, Building

Brahms Geography:Area Definitions

areadef City {display: City;

}//areadef City

areadef areadef-name {extends areadef-name [ , areadef-name ]*} }{{ display : ID.literal-string ; }{ icon : ID.literal-string ; }{ GRP.attributes }{ GRP.relations }{ GRP.initial-facts }}

107

ICCM 2010

Brahms Geography:Areas

• Instance of AreaDef => Parent• Part of other Area => Super-Areas • Inverse: Has Parts => Sub-Areas• Inhabitants => Located Agents and Objects• Has Paths => movement

area area-name instanceof ADF.areadef-name { partof area-name } {{ display : ID.literal-string ; }{ icon : ID.literal-string ; }{ GRP.attributes }{ GRP.relations }{ GRP.initial-facts }}

108

ICCM 2010

Paths between areas

• Areas are not necessarily a 2-dim. Grid• Areas can sub-areas, sub-areas can have sub-sub-

areas, etc

109

ICCM 2010110

package projects.atm;

areadef University extends BaseAreaDef { }areadef UniversityHall extends Building { }areadef BankBranch extends Building { }areadef Restaurant extends Building { }

// ATM Worldarea AtmGeography instanceof World { }

// Berkeleyarea Berkeley instanceof City partof AtmGeography { }

// inside Berkeleyarea UCB instanceof University partof Berkeley { }area SouthHall instanceof UniversityHall partof UCB { }area Telegraph_Av_113 instanceof BankBranch partof Berkeley { }area SpraulHall instanceof UniversityHall partof UCB { }area Bancroft_Av_77 instanceof BankBranch partof Berkeley { }area Telegraph_Av_2405 instanceof Restaurant partof Berkeley { }area Telegraph_Av_2134 instanceof Restaurant partof Berkeley { }

AtmGeography

Berkeley

Telegraph_Av_X UCB

SouthHall SpraulHall

Bancroft_Av_X

Part-Of

Alex_Agent Kim_Agent

Inhabitants

agent Alex_Agent memberof Student {location: SouthHall;

agent Kim_Agent memberof Student {location: SouthHall;

Tutorial Scenario Geography

ICCM 2010111

Sub-Area Facts

ICCM 2010112

Where in the world is Alex?


}

ICCM 2010113

Where in the world is Alex?


}

ICCM 2010114

Moving• Agents and objects can move

– Use move(to_location) activity in a workframe– Can specify duration in clock-ticks– Default zero duration, unless

• Define a Path object between two areas– Defines duration to move from area1 to area2– Bi-directional path

• Engine retracts and creates location facts and beliefs– Can specify (sub-)area arrival and departure detection

• Engine calculates shortest path between areas• Contained objects and agents move with the agent

ICCM 2010115

Paths

Path from South Hall to Blakes Diner

Path from South Hall to Raleigh Restaurant

path path-name{ { display : ID.literal-string ; }area1 : ARE.area-name ;area2 : ARE.area-name ;{ distance : ID.unsigned ; }}

ICCM 2010116

Paths to Restaurants

//paths to and from banks from spraul //and south hallspath SpH_to_from_WF {

area1: SpraulHall;area2: Bancroft_Av_77;distance: 200;

}

AtmGeography

BerkeleyTelegraph_Av_113

UCBSouthall

SpraulHall

Telegraph_Av_2134

Telegraph_Av_2405

Bancroft_AV-_77200

20080

100

360

240

400

280240

320

Student

Restaurant

Atm

ICCM 2010117

Agent/Object Movement

ICCM 2010




• 2:30pm – 3:00pm– Break



118

ICCM 2010 119

Activities and Workframes

ICCM 2010 120

Beyond Task/Goal AnalysisDescribing & Explaining Behavior

Watching Soccer with 50,000 fans

Communities of Practice

Being Dutch at the World Cup

Resting

Teamwork

1. All human activity is purposeful, but not every goal is a problem to be solved and not every activity performs a task (e.g., resting).

2. “Off-task” activities (waiting, talking, playing) have composite structure, norm-based variations, and purposes relevant to work functions.

3. Some goals are states to sustain, not requiring search or pathfinding (e.g., brainstorming).

4. Must reformulate how Deliberation relates to: Activity Motives and Attentive Coordination

ICCM 2010

Goals vs. Activities

Goal Stack

Problem-Solving Approach Activity Approach

6. Add 3 and 45. Solve x-4 = 34. Solve equation on screen3. Get experiment credit2. Pass introductory psychology1. Get college degree

Act: Taking an introductory psychology classDuration: 1 semester

Act: Going to collegeDuration: 4 years

Act: Doing an experimentDuration: 2 hours

Act: Solving equation x-4 = 3Duration: 2 mins

Act: Adding 2 and 4Duration: 1 sec

121

ICCM 2010 122

Activities

Task

Activity

Goal

What are Activities? • socially constructed engagements, • situated in the real world, • taking time, effort and application of knowledge, • defined beginning and end, • NOT necessarily need goals in the sense of problem-solving tasks.• can be interrupted.

When in an activity people might articulate a task they are working on, and the goal they want to accomplish (Clancey ’97)

Tasks and Goals are constructed within an Activity (Clancey ’97) (Kuutti ’96)

ICCM 2010 123

What are some of Alex’s activities?

• Studying in South Hall.

• Walking to a restaurant.

• Getting money from a bank’s cash machine.

• Ordering food from a restaurant.

Characterization:• has a duration• is situated in the real world• can be interrupted / resumed, but stay active• can be decomposed and/or subsumed

ICCM 2010 124

Brahms Activities• Primitive activities

– Lowest level, user-defined, but not further specified.

– Parameters are time, and resources

• Predefined activities– Primitive activities with predefined semantics (communicate, move, etc.)

• Composite activities– User-defined detailed activities

– Decomposed in sub-activities

– Describes what an agent does while “in” the activity

• Java activities– User-defined primitive activities that are implemented in a Java class

– Uses the Brahms API.

ICCM 2010 125

Activity Definition• Declaration and reference

– All activities have to be declared in the activities section of either a group, agent, class, object, or composite-activity. The declared activities can then be referenced in the workframes defined for the group, agent, class or object.

• Parameters– It is possible to define input parameters for primitive activities. These input parameters can be

used to make activities more generic. In the reference the values for the input parameters have to be passed.

• Priority– Activities can be assigned a priority. The priorities of activities in a workframe are used to

define the priority of a workframe. The workframe will get the priority of the activity with the highest priority defined in the workframe.

• Duration– Activities in general have a duration. The duration of the activity can be defined to be a fixed

amount of time. The random attribute has to be set to false and the max-duration attribute has to be set to the maximum duration in seconds. The duration of the activity can also be defined to be a random amount of time. To define a random amount of time the random attribute has to be set to true, the min-duration attribute has to be set to the minimum duration of the activity in seconds and the max-duration attribute has to be set to the maximum duration of the activity in seconds.

ICCM 2010 126

Activity Definition• Resources

– Artifacts (objects) can be defined as being a resource or not by setting the resource attribute to either true or false. In general artifacts that are worked on by agents are not considered to be a resource in an activity (a form, a fax). Artifacts that are used by an agent in an activity are considered to be resources ( a fax machine, a telephone).It is possible to associate artifacts with activities for statistical purposes and for the purpose of generating object flows by defining them in the list of resources for an activity. Artifacts which are defined as resources are also called resource objects. Resource objects associated with activities will only collect statistics and will not be used for the object flow generation. Artifacts which are defined not to be a resource and which are associated with an activity are also called touch objects. Touch objects should be associated with one or more conceptual object(s) for object flow purposes and statistical purposes.

• Defaults– display = <activity-name>– priority = 0– random = false– min_duration = 0– max_duration = 0– resources = none

ICCM 2010 127

Primitive Activitiesprimitive-activity ::= primitive_activity activity-name ( { param-decl [ , param-decl ]* } ) {

{ display : ID.literal-string ; }{ priority : [ ID.unsigned | param-name ] ; }{ random : [ ID.truth-value | param-name ] ; }{ min_duration : [ ID.unsigned | param-name ] ; }{ max_duration : [ ID.unsigned | param-name ] ; }resources : [ param-name | OBJ.object-name ]

[ , [ param-name | OBJ.object-name ]*;}

primitive_activity Study (Books course_book) {

display : “Study for a Course” ;{ priority : 10 ;{ random : true ;{ min_duration : 1800 ; /* 30 mins */{ max_duration : 7200 ; } /* 2 hours */resources : course_book;

}

ICCM 2010 128

Predefined Primitive Activities (1)

• Move– Moves an agent/object

from one area to another area.

• Agent/object is moved

• Contained objects/agents are moved

• Location beliefs/facts are retracted/created

• Location facts are created

• Create Agent/Object/Area– Creates new agents/

objects/areas dynamically• Agent can be member of

multiple groups

• Objects can be an instance of a class

• Can bind new agent/object to a return parameter

• Can give new agent/object a name and location

ICCM 2010 129


• Communicate– Communicates agent’s beliefs

from/to receiver agent(s)• Agent needs to have the beliefs

to communicate• Can specify type of

communication (phone | fax | email | face2face | terminal | pager | none)

• Can specify to which agents/objects is being communicate

• Can specify when to communicate (start | end)

• Broadcast– Communicates agent’s beliefs

from/to all agent(s) in specific areas

• Can specify areas and subareas• Agent needs to have the beliefs

to communicate• Can specify type of

communication (phone | fax | email | face2face | terminal | pager | none)

• Can specify to which agents/objects is being communicated

• Can specify when to communicate (start | end)

ICCM 2010 130


• Get– Allows an agent or object to pick up or transfer one or more

objects and/or agents from an area, other agent or object, and carry it with it while performing activities.

• Put– Allows an agent or object to put down (drop) or transfer one or

more objects and/or agents, referred to as items, carried by the agent or object performing the activity.

• Gesture– The gestures as indicated by the gesture activity are visualized in

the virtual reality environment provided that environment supports the specified gestures.

ICCM 2010 131

Move Activity

move PAC.activity-name ( { PAC.param-decl [ , PAC.param-decl ]* } ) { { display : ID.literal-string ; }{ priority : [ ID.unsigned | PAC.param-name ] ; }{ random : [ ID.truth-value | PAC.param-name ] ; }{ min_duration : [ ID.unsigned | PAC.param-name ] ; }{ max_duration : [ ID.unsigned | PAC.param-name ] ; }{ PAC.resources }location : [ ARE.area-name | PAC.param-name ] ;{ detectDepartureIn : [ ARE.area-name | PAC.param-name ] [ , [ ARE.area-name | PAC.param-name ]]* ; }{ detectDepartureInSubAreas : [ ID.truth-value | PAC.param-name ] ; }{ detectArrivalIn : [ ARE.area-name | PAC.param-name ] [ , [ ARE.area-name | PAC.param-name ] ]* ; }{ detectArrivalInSubAreas : [ ID.truth-value | PAC.param-name ] ; } }

move moveToRestaurant( ) { location: Telegraph_Av_2405;}

move moveToSouthHall() { location: SouthHall;}

move moveToLocationForCash(Building loc) { location: loc;}

move moveToLocation(Building loc) { location: loc;}

ICCM 2010 132

Java Activities

A Brahms Java Activity is a primitive activity but its actual behavior is specified in Java code.

– Java code may cause an action to happen outside the Brahms model completely (e.g. pop-up a dialog that say’s “hello world”)

– Java code can generate the output values and assign them to unbound variables in Brahms

– Java code can generate new model objects within the Brahms model– Java code can generate beliefs/facts into objects/agents/areas within the

Brahms model– Java code can integrate Brahms beliefs to external systems

ICCM 2010 133

Brahms Java Activity Syntax

java-activity ::= java PAC.activity-name ( { PAC.param-decl [ , PAC.param-decl ]* } ) { { display : ID.literal-string ; }{ priority : [ ID.unsigned | PAC.param-name ] ; }{ random : [ ID.truth-value | PAC.param-name ] ; }{ min_duration : [ ID.unsigned | PAC.param-name ] ; }{ max_duration : [ ID.unsigned | PAC.param-name ] ; }{ PAC.resources }class : [ ID.literal-string | PAC.param-name ] ; { when : [ start | end | PAC.param-name ] ; }}

java sayHelloWorld() {max_duration: 0;class: "gov.nasa.arc.brahms.atm.HelloWorld";when: start;

} // sayHelloWorld

ICCM 2010 134

“Hello World” Java Activity

java sayHelloWorld() {max_duration: 0;class: "gov.nasa.arc.brahms.jact";when: start;

} // sayHelloWorld

public class HelloWorld extends AbstractExternalActivity { public HelloWorld() { } // HelloWorldActivity /** * Implement the activity */ public void doActivity() throws ExternalException { System.out.println("************************************************"); System.out.println(”*My Brahms Agent is saying: 'Hello Agent World’*"); System.out.println"*************************************************"); } // doActivity} // HelloWorld

ICCM 2010 135

Brahms - FACET Integration

“Agentify” FACET– Wrap FACET as a

Brahms Agent

– Design Agent Communicative Acts

– Create Java Interface to FACET

ICCM 2010 136

Activities and Workframes

• Activities describe what people do …

• Workframes describe when people do what they do …

• Thus, they describe when activities are performed …

ICCM 2010 137

Brahms Workframes• Workframes are situation-action rules:

– Activities are associated with a conditional statement or constraint, – Workframes are different from production rules, in that they take time.– If the conditions of a rule are believed, then the associated activities are

performed.• We call these preconditions

– Precondition match on the beliefs held by the agent

• Workframes can be associated with groups/agents and classes/object.– A workframe defines when an activity (or activities) of an agent/object may

be performed.

• Having two or more agents with different workframes, performing the same activity, can represent individual differences.

• Conclusions are facts or beliefs or both that may be asserted when a workframe is executed.

ICCM 2010 138

Workframe Syntaxworkframe workframe-name{ { display : ID.literal-string ; } { type : factframe | dataframe ; } { repeat : ID.truth-value ; } { priority : ID.unsigned ; } { variables : [ VAR.variable ]* } { detectables : [ DET.detectable ]*} { when ( { [ PRE.precondition ] [ and PRE.precondition ]* } ) | do { [ PAC.activity-ref | CON.consequence ]* } }}

activities: primitive_activity eat( ) { priority: 0; max_duration: 400; }

workframe wf_eat { repeat: true; variables: forone(Cash) cs; forone(Diner) dn; when(knownval(current hasCash cs) and knownval(current.location = dn.location)) do { eat(); conclude((current.howHungry = current.howHungry - 3.00), bc:100, fc:0); conclude((cs.amount = cs.amount - dn.foodcost), bc:100, fc:100); conclude((current.readyToLeaveRestaurant = true), bc:100, fc:0); }}

ICCM 2010 139

AgentViewer Screenshot of Eat Worfkrame and Activity

ICCM 2010 140

Frame Repeat Attribute

• Repeat– A frame can be performed one or more times depending on the value of the “repeat” attribute. – A frame can only be performed once if the repeat attribute is set to false. – A frame can be performed repeatedly if the repeat attribute is set to true.– In case the repeat attribute is set to false, the frame can only be performed once for the specific

binding of the variables at run-time.

workframe wf_eat { repeat: true; <= can repeat for the same Cash and Diner Objects priority: 1; variables: forone(Cash) cs; forone(Diner) dn; when(knownval(current hasCash cs) and knownval(current.location = dn.location)) <= when the agent’s location is the diner location do { eat(); conclude((current.howHungry = current.howHungry - 3.00), bc:100, fc:0); conclude((cs.amount = cs.amount - dn.foodcost), bc:100, fc:100); conclude((current.readyToLeaveRestaurant = true), bc:100, fc:0); }}

ICCM 2010 141

AgentViewer Screenshot of RepeatingEat Worfkrame and Activity

ICCM 2010 142

Priority Attribute

• Priority– The workframe priority can be set in one of two ways.

– The priority can be set by setting the value for the priority attribute or,

– The priority can be deduced based on the priorities of the activities defined within the workframe, the workframe will get the priority of the activity with the highest priority.

– If no priority is specified the priority will be deduced from the activities, otherwise the specified priority is used.

Use of Activity Prioritiesactivities: primitive_activity eat( ) { priority: 1; max_duration: 400; }

primitive_activitity do_work( ) { priority: 0; max_duration: 400; }

workframes: workframe wf_Eat { repeat: true; when(knownval(current.howHungry > 20.00) and knownval(current.location = Telegraph_Av_2405)) do { eat(); conclude((current.howHungry = current.howHungry - 5.00), bc:100, fc:0); conclude((Alex_Cash.amount = Alex_Cash.amount - Raleigh_Diner.foodcost), bc:100, fc:100); conclude((current.readyToLeaveRestaurant = true), bc:100, fc:0); } }

workframe wf_Do_Work { repeat: true; do { do_work(); } }

T1=1 : do_work( )T2=400 : do_work( )…Tn=n : eat( )Tn+1=n+400: do_work( )…

Example 1

T1=1 : do_work( )T2=300 : eat ( )T3=n+700: do_work( )T4=n+800: do_work( )

interruptresume

Example 2

143

ICCM 2010 144

Interrupting Workframe and Activity using Priorities

ICCM 2010 145

Brahms Detectables(for reactive behavior)

• Associated with workframes and activities• Active while a workframe/activity is active• Used for:

– Agents noticing states of the world, and being able to act upon those

• 3-steps: (i) detect fact, (ii) notice (fact becomes belief), (iii) conditionally act on belief

– Control the execution of workframes and activities• Example: do act A until you notice fact F

• Type: continue | impasse | abort | complete | end_activity

ICCM 2010 146

Detectable Grammar

workframe workframe-name{ { display : ID.literal-string ; } { type : factframe | dataframe ; } { repeat : ID.truth-value ; } { priority : ID.unsigned ; } { variables : [ VAR.variable ]* } { detectables : [ DET.detectable ]*} { when ( { [ PRE.precondition ] [ and PRE.precondition ]* } ) | do { [ PAC.activity-ref | CON.consequence ]* } }}

detectable detectable-name { { when ( [ whenever | ID.unsigned ] ) } detect ( ([ result-val-comp | PRE.rel-comp ]) { , detect-certainty } ) { then continue | impasse | abort | complete | end_activity} ;}

Defaults* when = whenever* dc = 100* action = continue

composite-activity PAC.activity-name ( { PAC.param-decl [ , PAC.param-decl ]* } ) { { display : ID.literal-string ; }{ priority : [ ID.unsigned | PAC.param-name ] ; }{ end_condition : [ detectable | nowork ] ; }{ WFR.detectable-decl }{ GRP.activities }{ GRP.workframes }{ GRP.thoughtframes }}

ICCM 2010 147

Detectables in Workframes

workframe wf_waitAtmAsksPin {repeat: true; variables:

forone(BankCard) bkc4; forone(Cash) cs4; forone(Atm) at4;forone(Bank) bk4;

detectables:detectable atmAsksPin {

when(whenever)detect((at4.pinAsked = true), dc:100)

then complete;}

when(knownval(current.waitAtmAsksPin = true) and knownval(current hasBankCard bkc4) and

knownval(at4 contains bkc4) and knownval(current.pinCommunicated = false)) do {

waitAtmReply();conclude((current.waitAtmAsksPin = false), bc:100, fc:100);

}}

ICCM 2010 148

Wait for Reply Detectablecomposite_activity WaitAndProcessReply (ReplyAgent agt ) {

end_condition: detectable;detectables:

detectable dt_wait_for_reply {when (whenever)

detect((agt.replied = yes))then end_activity;

}activities:

composite_activity ProcessReply(ReplyAgent agt) { … }

workframes:workframe wf_Replied_n {

when (knownval(agt.answer = some_answer))do {

ProcessReply(agt);conclude((agt.replied = yes), bc:100, fc:0);

}}

}

ICCM 2010 149

Composite Activities

ICCM 2010 150

Composite Activities

• Decompose activities into sub-activities and the workframes that can execute them.

• Defines a workframe-activity hierarchy• Execution is different than traditional rule

hierarchies:– Subsumption hierarchy– While “in” an activity the higher-level activity

is still active.

ICCM 2010 151

Composite Activity Syntax

composite-activity PAC.activity-name ( { PAC.param-decl [ , PAC.param-decl ]* } ) { { display : ID.literal-string ; }{ priority : [ ID.unsigned | PAC.param-name ] ; }{ end_condition : [ detectable | nowork ] ; }{ WFR.detectable-decl }{ GRP.activities }{ GRP.workframes }{ GRP.thoughtframes }}

composite_activity study() { activities: primitive_activity reading ( ) { max_duration: 1500; } //end primact

workframes: workframe wf_readingWhileStudying { do { reading(); } //end do } //end wf

thoughtframes: thoughtframe tf_needCashToEat { variables: forone(Cash) cs; when (knownval(cs.amount < current.preferredCashOut)) do { conclude((current.needCash = true)); } //end do } //end tf…} //end composite activity study

ICCM 2010 152

Workframe-Activity Hierarchy

ICCM 2010 153

Activity Subsumption

ACTIVITY 4 ACTIVITY 3 ACTIVITY 1

(other activities)

WF 1

(other activities)

ACTIVITY 2

WF 2

ACTION X

(other workframes)

WF 4 (impasse)

WF 5

ACTIVITY 6

(other activities)

ICCM 2010 154

Model Parallel Activities?• Example: Talking on cell phone while driving

composite_activity Driving( … ) {end_condition: detectable;activities:

composite_activity Talking_on_Cell_Phone( … ) {…}…

workframe Talking_while_Driving {when(knownval(mycell.isRinging = true) {do { Talking_on_Cell_Phone(…); }

}}

ICCM 2010 155

Workframe/Activity States

highest priority

precondition true precondition false

not highest priority

impasse detectable

end workframe oriabort | complete | end_activity detectable

highest priority

repeat = true

impasse resolved

Workframe Interruptiongroup PrimitiveActivityPerformer {

attributes:public boolean execute_PAC_1;

activities:primitive_activity PAC_1(int pri) {

display: "PAC 1";priority: pri;max_duration: 900;

}

primitive_activity PAC_2(int pri, int dur) {display: "PAC 2";priority: pri;max_duration: dur;

}

workframes:workframe wf_PAC_1 {

repeat: true;

when (knownval(current.execute_PAC_1 = true)) do {

PAC_1(1);conclude((current.execute_PAC_1 = false));

}}

workframe wf_PAC_2 {repeat: true;

do {PAC_2(0, 1800);conclude((current.execute_PAC_1 = true), bc:25);PAC_2(0, 600);

}}

}

Workframe-Activity Hierarchy

Wf_PAC_1

PAC_1(1)

Wf_PAC_2

PAC_2(0, 1800) PAC_2(0, 600)

Tn+1 = Tn + 1800

Tn+2 = Tn+1 + 900

156

ICCM 2010 157

Brahms External Agent

A Brahms External Agents are Brahms agents but its actual behavior is specified in Java code.

– Defined in your Brahms program as an external agent with a name– Java code implements the behavior of the agent– Java code to communicate (receive/send beliefs) from other Brahms

agents– Java code can access Brahms model– Java code can generate facts for objects/agents/areas within the Brahms

model– Java code can “agentify” external systems

ICCM 2010 158

Speech Dialog Agent

ICCM 2010 159

Agent Communication

ICCM 2010

Two Ways of Agent Communication

• Communicating individual beliefs– simple, but unstructured– mostly used in simulation

• Communicating via speech acts– more complicated, but more structured– good for defining standard conversation

protocols– used in MAS using FIPA standard

160

ICCM 2010 161

Brahms Communications

• Activities that transfer beliefs to/from one agent to one or several other agents, or to/from an (information carrier) object. Examples are:– Face to face conversations.– Reading or writing a document.– Data entered into computers.

• An agent/object has to have the belief before it can communicate (i.e. tell) the belief to another agent/object.

• Recipient agent/object will overwrite original beliefs with communicated beliefs.

ICCM 2010 162

What does Alex need to communicate?

Banking Institution

PIN Information

Amount of Money

PIN Verified

Withdrawal Amount

Approved

ICCM 2010 163

Communicate Activity Grammar

communicate PAC.activity-name ( { PAC.param-decl [ , PAC.param-decl ]* } ) { { display : ID.literal-string ; }{ priority : [ ID.unsigned | PAC.param-name ] ; }{ random : [ ID.truth-value | PAC.param-name ] ; }{ min_duration : [ ID.unsigned | PAC.param-name ] ; }{ max_duration : [ ID.unsigned | PAC.param-name ] ; }{ PAC.resources }{ type : [ phone | fax | email | face2face | terminal |pager | none | PAC.param-name ] ; }with : [ [ AGT.agent-name |OBJ.object-name |PAC.param-name ] [ , [ AGT.agent-name |OBJ.object-name | PAC.param-name ] ]* ;about : TDF.transfer-definition [ , TDF.transfer-definition ]* ;{ when : [ start | end | PAC.param-name ] ; }}

transfer-definition ::= transfer-action ( communicative-act | DET.resultcomparison )transfer-action ::= send | receivecommunicative-act ::= OBJ.object-name | PAC.param-name

ICCM 2010 164

Transfer Definition

• The direction of communication– send: agent send to will always receive– receive: agent receiving from does not “know” it is

communicating; needs to have the belief

• Transfer of beliefs happens either at the start or at the end of the activity– when: start | end;

ICCM 2010 165

Transfer Definition• Defines the kind of belief the agent/object is

communicating• Attribute value is not taken into account …

– use ‘?’ as value• send(current.preferredCashOut = ?)

– don’t use a value• send(current.preferredCashOut)

• Relations objects/agents– send(current hasCash ?);– send(current hasCash);– send(current hasCash <Cash>);

• Can pass <cash object> as parameter• send(current hasCash <cash object>);

ICCM 2010 166

Communication Activities

Group Student:

communicate communicatePIN(Atm at3) { max_duration: 1; with: at3; about: send(current.pinCommunicated), send(current.believedPin); when: end;}

communicate sendRequestAmount(Atm at3) { max_duration: 1; with: at3; about: send(current.amountCommunicated), send(current.preferredCashOut); when: end;}

Class Atm:

communicate askBankVerification_1(Bank ba3) { max_duration: 1; with: ba3; about: send(current.currentAccountCode), send(current.currentAccountPin), send(current.currentAmountRequested), send(current.askedVerification); when: end;}Class Bank:

communicate replyAtm(Atm at3, Account bka) { max_duration: 1; with: at3; about: send(current.correctPin), send(current.enoughBalance), send(bka.balance); when: end;}

ICCM 2010 167

Workframe with Communicate Activity

workframe wf_communicatePIN {repeat: true;variables:

forone(BankCard) bkc3; forone(Atm) at3;forone(Bank) ba3;forone(Building) bd3;

when(knownval(current hasBankCard bkc3) andnot(current contains bkc3) andknownval(current.chosenBank = ba3) andknownval(at3 ownedbyBank ba3) andknownval(current.pinCommunicated = false) andknownval(current.location = at3.location) andknownval(current.pinRemembered = true) andknownval(current.waitAtmAsksPin = false))

do {processCommunicatePin();communicatePIN(at3);conclude((current.waitAtmAsksAmount = true), bc:100, fc:100);conclude((current.pinCommunicated = true), bc:100, fc:100);

}}

ICCM 2010 168

Alex Communicates with ATM

ICCM 2010 169

Brahms Libraries

• Libraries are Brahms Groups/Classes with domain-independent common capabilities

• Brahms comes with a number of libraries

– Also, we’re always extending Brahms capabilities …

– Libraries are a useful way to add capabilities without the need to add to the language

• Most libraries provide generic Activities developed as Java activities

• User can create their own libraries– Add Libraries to

../AgentEnvironment/Models/brahms– Use import statement, just like with

Java libraries

• Current Libraries:– Calendar

• Time and Date library

– Communication • FIPA compliant Communicative

Acts Library for agent communication

– Input/Output• File IO library for

copying/deleting files

– Java Utilities• Utilities for handling Java objects

for Brahms Agents and Objects

– System Utilities• Some simple Java print activities

and string manipulation

ICCM 2010 170

Communication Library

• A Communicator is able to communicate with other agents through communicative acts

• The Communicator specifies a set of activities that can be used by communicators to create, read, manipulate, retract, and send communicative acts

• Defines class CommunicativeAct

ICCM 2010 171

CommunicativeAct

• The CommunicativeAct models a communication event between two actors

• Defines a message that is based on the Communicative Act standard defined by FIPA (Foundation of Intelligent Physical Agents)

• Specifies an Envelope with the address information (from, to, date, ...) and transport hints

• Specifies a Payload for the message content and content properties

• Envelope and payload are maps

• http://www.fipa.org/specs/fipa00037/index.html

ICCM 2010 172

Message Envelope• This map is used to contain address-related information for a communicative act. The key-value

conventions are:"from” - ActiveInstance (required)

"to” - ActiveInstance (required)

"date” - string (required, formatted as yyyyMMddThhmmssSSSZ (UTC date/time)

"comments” - string (optional)

"acl-representation" - string (optional)

"encrypted” - boolean (optional)

"payload-length" - string (optional)

"payload-encoding" - string (optional)

"intended-receiver" - ActiveInstance (optional)

"received" - ReceivedObject (optional)

Example:

(<communicative_act>.envelope(“from”) = <sending_agent> )

(<communicative_act>.envelope(“to”) = <receiving_agent> )

ICCM 2010 173

Message PayloadThis map is used to contain content of a communicative act. The key-value conventions are:"performative” - symbol (required - INFORM, REQUEST, SUBSCRIBE, etc.)"sender” - ActiveInstance (required)"receiver" - ActiveInstance (required)"reply-to" - ActiveInstance (optional)"action" - symbol (optional)"subject" - symbol (optional)"content" - any instance type (optional)"language" - string (optional)"encoding" - string (optional)"ontology" - string (optional)"protocol" - string (optional)"conversation-id" - string (optional)"reply-with” - string (optional)"in-reply-to” - string (optional)"reply-by" - string (optional, date/time formatted as yyyyMMddThhmmssSSSZ (UTC

date/time))Example: (<communicative_act>.payload(“sender”) = <sending_agent> )

ICCM 2010 174

Example Sending ComAct

workframe wf_ConfirmGetAvailableTime {

variables:

forone(ActiveInstance) sender;

forone(string) convid;

forone(CommunicativeAct) reply;

forone(Activity) act;

when ( (comact.payload("performative") = REQUEST) and

(comact.payload("action") = getAvailableTime) and

(comact.payload("sender") = sender) and

(comact.payload("purpose") = act) and

(comact.payload("conversation-id") = convid))

do {

createCommunicativeAct(current, sender, INFORM, convid, getAvailableTime, classtype, current, reply);

setPayloadProperty(reply, "content", payloadobj1);

setPayloadProperty(reply, "purpose", act);

sendCommunicativeAct(reply, sender);

}//do

}//wf_ConfirmGetAvailableTime

ICCM 2010 175

Receiving ComAct

ICCM 2010

Brahms Hosting Environment

176

ICCM 2010

Distributed MAS Communication Framework

177

ICCM 2010

Robotic Recon Field TestJune 20, 2009

Blackpoint Lava Flow, AZ

178

Flight Control Team Science Operations Team

ICCM 2010

Conversation Protocol

179

ICCM 2010 180

More Information …

• E-mail: [email protected]

[email protected]

• Brahms URL: http://www.agentisolutions.com

• Brahms Papers in Publications on: http://homepage.mac.com/msierhuis http://Bill.Clancey.name

international conference on cognitive modeling 2010 brahms tutorial

Technology

pm activities

activities artifacts

work system

pm agents

efficiency of work

way people

work practicehermeneutics

people wenger