This document discusses multi-agent systems and their applications. It provides examples of multi-agent systems for spacecraft control, manufacturing scheduling, and more. Key points:
- Multi-agent systems consist of interacting intelligent agents that can cooperate, coordinate, and negotiate to achieve goals. They offer benefits like robustness, scalability, and reusability.
- Challenges include defining global goals from local actions and incentivizing cooperation. Games like the prisoner's dilemma model social dilemmas around cooperation versus defection.
- The document outlines architectures like the blackboard model and BDI (belief-desire-intention) model. It also provides a manufacturing example using the JADE platform.
3. Intelligent Agent
Observe Environment through sensor
Act for achieving goals ( in economics, it’s rational )
Smith's role is to police and maintain the
Matrix by eliminating potential threats to the
stability of the system
http://matrix.wikia.com/wiki/Agent_Smith
Purpose.
4. Limit of Agent
• Before “s”
• Interconnection and Distribution are in CS
• But I&D isn’t for cooperate, ( agreement | compete )
with systems have different interests
• and… also single processor’s limit
• If we have 10^10 processors…
Let's coordinate our plans for Sunday evening.
The man cooperated with the police
5. Multi Agent
consists of a number of agents, which interact with one-another
- Oxford Univ
successfully interact, they will
require the ability to
cooperate,
coordinate,
and negotiate
with each other
6. For Multi Agent
Design Agent
Autonomous action for
successfully carry out task
Design Society
Interacting with other agent
( Specially, different goals )
Social Dilemmas
Lots of Selfish, whole group loses
7. Multi Agent Learning Design
https://www.youtube.com/watch?v=yE62Zwhmzi8 : DeepMind Role
Advantage
Robustness
Scalability
Reusability of constituents
Challenges
Global goals from local actions
Incentive assignment
Learning while others are learning
Traffic
Economy
Markets
Workplace
Sports
Family
Compete
Cooperate
Coordinate
Communicate
Negotiate
Predict action
8. Matrix Game Social Dilemmas (MGSD)
Cooperate Betrayal
Cooperate R, R S, T
Betrayal T, S P, P
R : Reward for cooperate
S : Betrayal damage
T : Temptation of Betrayal
P : Penalty of Betrayal
Limit : Temporally
9. Cooperation in MGSD
Direct reciprocity : Tit-for-tat
Indirect reciprocity : reputation
Norm enforcement
Learning …
Limitation of MGSD
Real World is Sequential Decision!
Cooperate vs defection => Not just binary, it’s complex
10. Sequential Social Dilemmas
Understanding Agent Cooperation
https://deepmind.com/blog/understanding-agent-cooperation/
Reward가 없지만… laser로 other agent를 잠깐 없앰
먹을게 (초록) 많으면 공존하면서 reward를 얻고,
적으면 서로 공격하기 시작함
13. Multi Agent “REAL” Example
Spacecraft Control
Required – decide action for unexpected event “quickly”
Deep Space 1 [ NASA + Caltech ]
Launched on 1998.
> Launched on 1990s…
Multi-agent Frameworks for Space Applications
15. Why Multi Agent System for Manufacturing?
현재 Manufacturing은 high mix, low volume이 되어감
=>
Complex Algorithm을 풀 때 MAS는 좋은 접근 방법 중 하나!
https://github.com/mskiitd/MAS-GUI
MAS for Manufacturing framework
16. MAS Architecture : BlackBoard Arch.
Blackboard Global database for sharing different info
- input data, partial solutions, final solution
Knowledge Source Independent module for solving problem
(Agent)
Control Online decision for choosing knowledge source to
execute
Communication between agents in decoupled way
Easy to add new agent
17. MAS Architecture : BlackBoard Arch.
Communication between agents in decoupled way
Easy to add new agent
18. Belief Desire Intention Arch.
Belief == Current State ( Env + other Agent )
Desire == Obj of agent
Intention == Action
19. Step 1 : Gatekeeper
Check feasibility
( Machine Type, Due Date )
state
action
state
action
Step 2 : Take Bids
Agent Calculate by each knowledge
Step 3 : Execution
Agent Calculate by each knowledge
20. Agents sequence
Bid == Objective Performance
For Local Scheduler Agent, Machine Agent
- Due Date Satisfaction
- MakeSpan
- …