Industrial and Practical Applications of DAI

1
Industrial and Practical Applications of DAI
Ori Liel

• Agents in action!

2
In this lecture

• Characteristics which make problems fit for an
  agent-based solution.
• A broad overview of practical DAI applications.
• An elaboration on one specific application the
  RAPPID system
• The Industrial life-cycle and how it bears on
  agent systems

3
Characteristics which make problems fit for an
agent-based solution
4
A note about OOD design

• Advantages of Object-Oriented design
• More thoroughly explored (theoretically)
• Better practical support
• Therefore one must also ask what is the drawback
  of an Object Oriented solution.

5
Characteristics Modularity

• A modular problem is one that can easily be
  divided into sub-problems.
• Assign an agent to each problem
• Regular Object-Oriented design is also
  suitable.

6
Characteristics Decentralization

• A decentralized problem can be decomposed into
  stand-alone processes.
• No main thread of control
• Little inter-process communication
• Advantages of decentralization
• Parallelism
• Eliminates wasteful method calls
• Suits modern industrial approaches push
  decision making down as much as you can
• OOD requires multi-threading

7
Characteristics Changeability

• A problem which may change often
• Agent solution is modular decentralized

8
Characteristics Open environments

• When a problem is situated in an environment
  which is
• Dynamic
• Agents are reactive and monitor their environment
• Elements of uncertainty
• No full knowledge of environment
• Non-deterministic
• Agents taught to plan with partial info, and
  backtrack if necessary
• Finally, agents as a natural metaphor

9
An overview of practical DAS applications
10
Applications

• Simulations
• What could we want to simulate?
• An economic system
• A biological system (ant-farm, beehive)
• Traffic flow
• What makes these problems suitable for
  agent-technology?
• Decentralized
• Changeable
• Dynamic environments

11
Applications contd

• Some available simulation tools
• Echo is a simulation tool developed to
  investigate mechanisms which regulate diversity
  and information-processing in systems comprised
  of many interacting adaptive agents, or complex
  adaptive systems (CAS). Echo agents interact via
  combat, trade and mating and develop strategies
  to ensure survival in resource-limited
  environments. http//www.santafe.edu/projects/echo
  /echo.htmlintro
• Swarm is a software package for multi-agent
  simulation of complex systems, originally
  developed at the Santa Fe Institute. Swarm is
  intended to be a useful tool for researchers in a
  variety of disciplines. The basic architecture of
  Swarm is the simulation of collections of
  concurrently interacting agents with this
  architecture, we can implement a large variety of
  agent based models. http//www.swarm.org/intro.htm
  l
• SugarScape Unlike many other research-oriented
  computer models, Sugarscape uses the bottom-up
  approach known as agent-based modeling.
  http//www.brookings.edu/dybdocroot/sugarscape/

12
Applications contd

• E-Commerce
• What are the needs?
• Comparison shopping
• Auctions
• What makes these problems suitable for
  agent-technology?
• The dynamic environment of the internet
• Agents as natural metaphors
• Auctions require communication

13
Applications contd

• E-commerce Wooldriges examples revisited
• Auctions
• Auction bot (recently deceased)
• Kasbah transformed to travel resource
• Comparison shopping
• Jango web site for sale
• Bargain Finder alive and kicking

14
Applications contd

• Manufacturing
• What are the needs?
• Support for product-design process
• Planning and scheduling
• Real-time control
• What makes these problems suitable for
  agent-technology?
• Virtually all of the things we mentioned before

15
Applications contd

• Manufacturing contd
• Product design
• RAPPID elaborated later!
• Planning and scheduling
• AARIA
• Daewoos system
• Mature applied at Daewoo Motors integrated
  automobile production facility in Korea
• Real-time control
• ARCHON
• Mission to integrate pre-existing expert systems

16
Applications contd

• Computer games (separate lecture)
• Air traffic control
• OASIS in Australia
• Medicine
• Patient monitoring

17
The RAPPID system

• Responsible Agents for Product-Process Integrated
  Design

18
The RAPPID system

• Imagine youre designing a product
• Product has many components
• Each component has many characteristics
• Weight
• Space
• Power consumption
• Each component is designed by a different team,
  and teams could be scattered across the globe
• Goal
• Achieving an optimized design in a reasonable
  time

19
RAPPID What does it aim to fix?

• Problems with current design approaches
• Problem often, disputes about resources are
  settled through politics rather than logic
• Solution resources are commodities in a
  marketplace.
• Problem A small inaccuracy in Chief Engineers
  vision can be fatal.
• Solution design as constraint satisfaction
  problem.

20
RAPPID Solution to 1st problem

• The Marketplace
• Designers buying power determined at offset.
• Characteristics assigned initial price
• supply--demand principles set prices for the
  different characteristics.
• Isnt it nice how you can suddenly trade
  electrical power more space?

21
RAPPID - Solution to 1st contd

• Agent architecture
• Components are interface agents for designers.
• Monitor the market for designer
• Make suggestions or even transactions.
• Characteristics are agents
• Perhaps a case of seeing agents everywhere?

22
RAPPID illustration
23
RAPPID Solution to second problem

• The basic idea chief engineer sees which
  characteristics are slacking
• He buys them out and uses the money to insert
  more of other characteristics to the simulation

• Converging the
• design space

24
The Industrial life-cycle and how it bears on
agent systems

• When agents leave the lab

25
Industrial life-cycle - overview

• Requirement definition
• Positioning
• Specification
• Design
• Implementation
• Comissioning
• Operation
• Decomissioning

26
Industrial life-cycle Requirement definition

• Talk about needs
• Market analysis reveals that we are losing sales
  to competitors who are offering sport utility
  vehicles (SUVs), a niche in which we currently
  have no product offering
• Agent Systems
• Remember there are more problems that can be
  automated

27
Industrial life-cycle - Positioning

• define the projects relationship to other
  projects in the enterprise.
• Our current product divisions are luxury auto,
  economy auto, minivan, and light truck. The
  minivan and light truck divisions seem the best
  candidates to host the new SUV offering

28
Industrial life-cycle - Specification

• What the project will do, but not how.
• We benchmark the performance characteristics of
  our competitors SUV offerings
• The result is a list of the features and
  characteristics of the new vehicle.

29
Industrial life-cycle - design

• Decide how to achieve the functions in the
  specification
• Product-engineering deigns a new vehicle,
  Process-engineering designs the factory that will
  make it
• Agent-systems
• Design considerations
• What in a system becomes an agent?
• What types of agents to use?
• How to divide labor between agents?

30
Industrial life-cycle (design) contd

• Completing design testing
• Role-Playing
• Figure out the rules that should guide each agent
• Maybe discover need for more/less agents
• Formal analysis
• Logical analysis
• Simulation

31
Industrial life-cycle - implementation

• Constructing the system
• Purchasing negotiation contracts for the
  equipment needed to construct the new vehicle. A
  plant is selected to house the new line old
  equipment is removed, new equipment is installed

32
Industrial life-cycle implementation (in agent
systems)

• Hardware considerations
• General-purpose computer or specially designed
  parallel architecture?
• Software considerations
• Use existing tools or create your own?
• In science
• suitable tools may not exist
• Constructing tools part of the mission
• In industry
• Making new tools consumes time
• Designers may not be agent experts

33
Summary

• We talked about characteristics which make a
  problem suitable for an agent-based solution
• We touched over some of the areas in the industry
  in which DAI is used
• We examined the RAPPID system
• We talked about the industrial life-cycle of an
  agent system