An introduction to agents DARTS 2003

1
An introduction to agents DARTS 2003 Session
2A 22nd October 2003

• Joris Maervoet and Stijn Bernaer
• KaHo Sint-Lieven Hogeschool, Ghent

2
Talk outline

• An introduction to agents
• Agent definitions
• Agent applications Helms Deep/NASA
• Agent applications overview
• Agent-oriented design
• Programming agents
• The AMobe project

3
1 Agent definitions
4
Agent definition

• Encapsulated computer system, situated in some
  environment, and capable of flexible autonomous
  action in that environment in order to meet its
  design objectives (Jennings)
• Agent interaction is inevitable to achieve
  individual objectives and to manage
  inter-dependencies (Jennings)

5
Properties
Re-active On external asynchronous stimuli
Autonomous Controls own actions, operates without direct intervention of others
Pro-active Goal directed, takes initiatives
Continuous and persistent Is living, not running
Social Communicates with other agents and humans
Learning Adaptiveness based on experience
Mobile Moves among machines in network
6
Why Agents?
Load reduction Trusted Agent travels between server and client, carries safe protocol
Delay independence Agents perform in real time, not over uncertain networks
Protocols Agents encapsulate protocols, for better adaptability and maintainability
Asynchronous Together with autonomy, agents can act on their own, without necessity for continuous interaction
7
Why Agents?
Adaptive Agents react autonomously on changes in the environment
Heterogeneity Distributed systems are heterogeneous in nature. Agents are optimal for system integration
Robustness Mobile agents can make a distributed system more robust and fault tolerant by deciding autonomously in case of an error
8
Agents and objects, whats the difference?
Object A
Object B
method
Agent A
Agent B
? ACL message
9
2 - Agent applications 2 examples
10
(No Transcript)
11
The battle at Helm's Deep

• Gathering 70,000 people, dressing them and
  choreographing each other is out of question
• Stephen Regelous created Massive (special-effects
  program)
• Massive is able to generate characters with their
  own minds (also called agents)

Source Wired News, Digital actors in Rings can
think
12
The battle at Helm's Deep

• making realistic crowds is making realistic
  individuals

Taken from Pierre Vinet/ New Line Productions
13
The battle at Helm's Deep

• Agents brains (look like complex flow charts)
  define how they see and hear, how fast they run,
  
• Agents movements are based on these of stunt
  actors to allow for example realistic ducks to
  avoid a sword
• Like real people, agents are influenced by their
  environment

14
The battle at Helm's Deep
Taken from Entertainment Weeklys EW.com
15
The battle at Helm's Deep

• Each agent makes subtle responses to its
  surroundings with fuzzy logic rather than yes/no
  decisions
• Placing agents into a simulation, each agent
  makes decisions from its point of view (no crowd
  control)
• Other examples generating flocks of birds,
  duplicating film stars

16
http//technology.arc.nasa.gov
17
NASA

• Traditionally ground systems deal with spacecraft
  planning scheduling, establishment of
  communications and (in some cases) processing
• So the ground system is responsible for managing
  the spacecraft and its activities (remote control)

Source Nasa Ames Research Center, Remote Agent
Project Website
18
Remote Agent

• The first artificial intelligence control system
  to control a spacecraft without human supervision
• Remote Agent successfully demonstrated the
  ability to plan onboard activities and correctly
  diagnose and respond to simulated faults in
  spacecraft components

19
Remote Agent

• Advantages
• Faster reactions to problem situations (getting
  instructions from Earth could take a while)
• Less operation costs and less remote control the
  Remote agent can take care of itself and does not
  need a hundred of specialised people on board or
  on the ground

20
Remote Agent

• Made up of 3 components
• Planner and scheduler (PS) produces flexible
  plans, specifying the basic activities that must
  take place in order to accomplish the mission
  goals
• Smart executive (EXEC) carries out the planned
  activities
• Mode Identification and Recovery (MIR) monitors
  the health of the spacecraft and attempts to
  correct any problems that occur

21
Remote Agent

• These components work together and communicate to
  ensure that the spacecraft accomplishes the goals
  of the mission

Taken from Nasa Ames Research Center, Remote
Agent Project Website
22
3 Agent applications overview

• Source Agentlink Roadmap

23
Agent-based systems - overview

• Assistant agents e.g. TAC, where agents book
  hotels and make travel arrangements
• Multi-agent decision systems e.g. components in
  a network may jointly seek to allocate scarce
  resources of the network
• Multi-agent simulation systems used to model
  some complex real-world domains, e.g. biological
  populations

24
Industrial and Commercial Applications

• Manufacturing
• eCommerce e.g.
• Telecommunications
• Supply chain management
• Entertainment and leisure
• e.g. The Creature Games, SimCity
• the film The Two Towers

25
Simulation applications

• Agent-based simulation characterised by
• Agent-based computing
• Social sciences studies interaction among social
  entities and include social psychology,
  management, policy and some areas of biology
• Computer simulation techniques such as discrete
  event, object-oriented, equation-based simulation

26
Simulation applications

• Advantages
• Forecasting some complex real-world environments
  (economy, society, biology)
• Examples
• Flight Simulators train pilots to respond
  appropriately to unexpected events
• Southwest Airlines increased revenues by 10
  million dollar by using a agent-based simulation
  of cargo routing

27
Application opportunities

• Ambient intelligence providing an environment of
  thousands of embedded and mobile devices
  interacting to support user-centred goals and
  activity
• Bioinformatics
• Grid computing geographically separated
  computers that share applications, data and
  computational resources
• Electronic business

28
4 - Agent-oriented design
29
A new software paradigm

• procedures ? data types ? objects ? agents
• Point of view a further powerful abstraction, a
  new software paradigm
• Methodologies (e.g. AUML) in development
• Agent patterns

30
Pitfalls (Jennings)

• Political you oversell agents, you get dogmatic
  about agents
• Management you dont know why you want to use
  agents, you want generic solutions to one-off
  problems
• Conceptual you believe agent technology is a
  silver bullet, you forget that you are
  developing software, you forget agents are
  multi-threaded software

31
Pitfalls (Jennings)

• Analysis and design you ignore related
  technology, you dont exploit concurrency, you
  start from a tabula rasa
• Agent level you want your own architecture, you
  use too much AI / no AI
• Society level you see agents everywhere, you
  have too few/many agents, you spend all the time
  on an infrastructure, your agents interact too
  freely, your system lacks structure

32
5 Programming agents
33
JADE and LEAP

• Java Agent Development framework (JADE) is an
  agent development environment implemented in
  J2SE.
• Developing FIPA-compliant (Foundation for
  Intelligent Physical Agents) agent applications
  for interoperable intelligent multi-agent
  systems.
• Lightweight Extensible Agent Platform (LEAP)
  replaces JADEs core functionality for devices
  with low memory/CPU capacities.
• JADE-based agentplatform which is standardized
  for mobile devices and compatible with mobile
  Java environments

34
JADE and LEAP

• JADE still in development. (last 3.0)
• LEAP project recently taken over by JADE

Huge number of successful applications
35
JADE and LEAP

• JADEs functionality
• Distributed agentplatform on several JVMs,
  behaviour model, AMS (automatische
  (de)registratie) DF ACC.
• GUI
• Debugging tools
• Intraplatform mobility
• Multiple DF
• ACL messages, several protocols
• Interface for external applications

36
JADE and LEAP
Taken from LEAP User Guide
37
JADE and LEAP

• LEAP on Symbian OS devices PersonalJava.
• LEAP on PocketPC devices install CLDC/MIDP or
  PersonalJava.
• LEAP on Palm OS install CLDC/MIDP.
• LEAP (or just JADE) on Windows NT/98/2000/XP
  install J2SE.

38
6 - The AMobe-project
39
IT research group

• Department of Engineering at KaHo Sint-Lieven
  Hogeschool
• Research areas personnel scheduling, agent
  technology, timetabling, mobile devices,
  optimisation, artificial intelligence
• Funded projects (IWT)

40

• 97 99 OCAPI
• ObjeCt georiënteerde Agenten voor gedistribueerde
  PlannIngssystemen using agents to optimize
  rosters and the routes of mobile nurses
• 99 01 COALA
• COoperating And Learning Agents creating an
  ontology for planning systems, Semantic Web
• 01 03 CoFfTeA
• Component Framework for Timetabling Applications
  component-based framework for timetabling
  applications
• 02 04 AMobe
• Application Development for Mobile Devices
  agent-oriented software-development on mobile
  devices
• 02 04 TITAN (internal project)
• Developing a timetabling application for the
  Engineering department
• 03 05 DINGO
• DIstributed Negotiation - Gedistribueerde
  Onderhandeling

41
AMobe

• Application Development for Mobile Devices
• 2 coordinators, 2 researchers
• How are mobile devices integrated in
  applications?
• 2 case studies (evaluation and iteration)
• Website http//ingenieur.kahosl.be/projecten/AMob
  e

42
AMobe
Permanent presence Mobile positioning
Functionality
uLinux Palm OS Epoc32 Windows CE J2ME
Waba LEAP JIAC IV
2.5G GPRS - EDGE WLAN - bluetooth 3G UMTS
??
Mobile devices
Technology
Platform
price market situation dimensions
Other factors
Architecture
Software patterns
43
Case studies

• Tele Atlas
• geographical database, employees on location
• TI
• Mobile devices assist people with non-congenital
  brain injuries

44
Framework for efficient datatransfer

• First case study gained interest IDEWE and WGK
  have similar cases
• Common agent-construction for efficient
  datatransfer between server and mobile devices
• Why agents? Agents can decide at which moment
  which data will be sent. Agents enable loosely
  coupled software development

45
Server (ev. DB server)
Server (ev. synchro server)
Mobiel toestel
INTERFACE AG.

• SYNCHRO AGENT
• ToestelIDTransactieIDpriorichtingaanvragerDBa
  ctie
• ---------------------------------------------
  ----
• 002 927 form -gt dca002_at_
  get
• 002 928 form lt- uca002_at_
  hereis
• 003 929 form -gt pol9_at_
  hereis
• 003 930 form lt- pol5_at_
  get
• actieve transacties
• prioriteitsdrempel

RECEIVING SLAVE AGENT
RECEIVING SLAVE AGENT
DOWN-LOAD CLIENT AGENT
RECEIVING SLAVE AGENT
TASK AGENTS
SENDING SLAVE AGENT
SENDING SLAVE AGENT
UPLOAD CLIENT AGENT
POLICY AGENT
SENDING SLAVE AGENT
POLICY AGENT
POLICY AGENT
SERVER DB AGENT
CLIENT DB AGENT
DB
DB
INFORMATION AGENTS
46
caesar.clt
Demo in JADE
arno
panamarenko
caesar
gaudi
Start/stop
Watch
Force blixa
Could I send a report?
Fill
Spartacus(x) will be your partner
Create
I am active

• arno.txt
• GMLGMLGML

blixa
spartacus(x)
spartacus(x).txt GMLGMLGML
Watch size
Fill
Please send everything
Copy arno.txt
end
ToSend.txt GMLGMLGML
Server
Mobile device
47
Demo agents over GPRS

• Purchase Sierra Wireless AirCard 750 (PCMCIA cf.
  PenPCs Tele Atlas) Mobistar abonnement voor
  GPRS
• First test JADE agent on portable (with GPRS -
  link) communicates with JADE agent on server
  within our LAN (both Windows XP)