Download Consensus Protocol: Multi-agent Systems

Consensus: Multi-agent
Systems (Part1)
Quantitative Analysis: How to make a
decision?
Thank you for all referred pictures and information.
Agenda

Introduction


Definitions
Questions

Reaching Agreements

Auction


Task allocation
Auction algorithm
2
Multiagent Systems, a Definition

A multiagent system is one that consists of a
number of agents, which interact with oneanother

Swarm of Robots


Agents will be acting on behalf of users with
different goals and motivations


Exchange information
Heterogeneous or Homogeneous
To successfully interact, they will require the
ability to cooperate, coordinate, and negotiate
with each other, much as people do
3
Multiagent Systems, a Definition

Why we apply multi-agent systems to solve
the problem?

A single agent cannot perform parallel tasks
alone.

Multi-agent can accomplish given tasks more
quickly.
4
Swarm Intelligence

Application of Swarm Principles: Swarm of
Robotics
http://www.domesro.com/2008/08/swarm-robotics-for-domestic-use.html

http://www.youtube.com/watch?feature=playe
r_embedded&v=rYIkgG1nX4E#!
5
Multiagent Systems (MAS)

Questions In Multiagent Systems:

How can cooperation emerge in societies of selfinterested agents?

What kinds of languages/protocols can agents
use to communicate?

How can self-interested agents recognize conflict,
and how can they reach agreement?

How can autonomous agents coordinate their
activities so as to cooperatively achieve goals?
6
Multiagent Systems (MAS)

How to make a group decision among them?
or How to achieve the group mission?

Find the optimal decision of group

Resolve conflicts among individuals

Maximize the overall performance of group
7
Multiagent Systems is Interdisciplinary

The field of Multiagent Systems is influenced and
inspired by many other fields such as:

Economics


Game Theory




Strategy for decision making
Conflict and cooperation between decision-makers
Logic
Social Sciences



Profit, Bargain
Leader, follower
Trust
This has analogies with artificial intelligence itself
8
Objections to MAS

Isn’t it all just Distributed/Concurrent Systems?
There is much to learn from this community,
but:

Agents are assumed to be autonomous, capable of
making independent decision


they need mechanisms to synchronize and coordinate
their activities at run time
Agents are self-interested, so their interactions
are “economic” encounters
9
Objections to MAS

Isn’t it all just AI?


We don’t need to solve all the problems of artificial
intelligence in order to build really useful agents
Classical AI ignored social aspects of agency.

These are important parts of intelligent activity in
real-world settings
10
Social Ability

The real world is a multi-agent environment:


Some goals can only be achieved with the
cooperation of others
Similarly for many computer environments:
witness the Internet

Social ability in agents is the ability to interact with
other agents via some kind of agent-communication
language, and perhaps cooperate with others
11
Other Properties

mobility:


veracity:


agents do not have conflicting goals, and that every agent will
therefore always try to do what is asked of it (helps)
rationality:


an agent will not knowingly communicate false information
(only true information)
benevolence:


the ability of an agent to move around an electronic network
agent will act in order to achieve its goals, and will not act in
such a way as to prevent its goals being achieved
learning/adaption:

agents improve performance over time
12
Agents and Objects

Main differences:

agents are autonomous:


agents are smart:


agents embody stronger notion of autonomy than objects, and in
particular, they decide for themselves whether or not to perform
an action on request from another agent
capable of flexible (reactive, pro-active, social) behavior, and the
standard object model has nothing to say about such types of
behavior
agents are active:

a multi-agent system is inherently multi-threaded, in that each
agent is assumed to have at least one thread of active control
13
Reaching Agreements

How do agents reaching agreements
when they are self interested?

There is potential for mutually beneficial
agreement on matters of common interest

The capabilities of negotiation and
argumentation are central to the ability of an
agent to reach such agreements
14
Definitions: Negotiation and Argumentation

Negotiation (Compromise)

Dialogue between two or more parties






intended to reach an understanding
resolve point of difference
gain advantage in outcome of dialogue
to produce an agreement upon courses of action
to bargain for individual or collective advantage
“tries to gain an advantage for themselves”
Argumentation

how conclusions can be reached through logical reasoning

Including debate and negotiation which are concerned with
reaching mutually acceptable conclusions
http://en.wikipedia.org/wiki/Negotiation
http://en.wikipedia.org/wiki/Argumentation_theory
15
Mechanisms, Protocols, and Strategies

Negotiation is governed by a particular mechanism,
or protocol

The mechanism defines the “rules of encounter” between
agents

Mechanism design is designing mechanisms so that
they have certain desirable properties

Given a particular protocol, how can a particular
strategy be designed that individual agents can
use?
16
Mechanism Design

Desirable properties of mechanisms:







Convergence/guaranteed success
Maximizing social welfare
Pareto efficiency
Individual rationality
Stability
Simplicity
Distribution
17
Auctions

An auction takes place between an agent
known as the auctioneer and a collection of
agents known as the bidders

The goal of the auction is for the auctioneer
to allocate the good to one of the bidders


Resource allocation
The auctioneer desires to maximize the price;
bidders desire to minimize price
18
Auction Parameters

Goods can have




Winner determination may be



first price
second price
Bids may be



private value
public/common value
correlated value
open cry
sealed bid
Bidding may be



one shot
ascending
descending
19
English Auctions

Most commonly known type of auction:



first price
open cry
Ascending

Dominant strategy is for agent to successively bid a
small amount more than the current highest bid until
it reaches their valuation, then withdraw

Susceptible to:


winner’s curse
shills
20
Dutch Auctions

Dutch auctions are examples of open-cry
descending auctions:

auctioneer starts by offering good at artificially
high value

auctioneer lowers offer price until some agent
makes a bid equal to the current offer price

the good is then allocated to the agent that
made the offer
21
First-Price Sealed-Bid Auctions

First-price sealed-bid auctions are one-shot
auctions:





there is a single round
bidders submit a sealed bid for the good
good is allocated to agent that made highest bid
winner pays price of highest bid
Best strategy is to bid less than true valuation
22
Vickrey Auctions

Vickrey auctions are:


second-price
sealed-bid

Good is awarded to the agent that made the
highest bid; at the price of the second highest bid

Bidding to your true valuation is dominant
strategy in Vickrey auctions

Vickrey auctions susceptible to antisocial behavior
23
Lies and Collusion

The various auction protocols are susceptible to lying
on the part of the auctioneer, and collusion among
bidders, to varying degrees

All four auctions (English, Dutch, First-Price Sealed
Bid, Vickrey) can be manipulated by bidder collusion

A dishonest auctioneer can exploit the Vickrey
auction by lying about the 2nd-highest bid

Shills can be introduced to inflate bidding prices in
English auctions
24
Applying to Algorithms

Node is represented an
agent

Edge indicates the
corresponding agents that
have to coordinate their
actions

1
4
2
3
Only interconnected agents
have to coordinate their
actions at any particular
instance
25
Task Allocation

Task Allocation Method in term of multi-agent system is given into two
meanings: for achieve the common goal involve one task or more than
one tasks.

Task Allocation problem:

The goal of task allocation is, given a list of n tasks and n agents, to find a conflictfree matching of tasks to agents that maximizes some global reward.

Behaviors of Task allocation


Agent stay focus on a single task until the task is over


Opportunism
Agent can switch tasks if another task is found with greater interesting or priority


Commitment
Coordination
Coordination is linked to communication, the ability of agents to communicate
about who should service which task



Individualism
Agent have no awareness of each other.
Communication is used to prevent multiple agents from trying to accomplish the
same task
26
Methods of Task Allocation
Methods of Task allocation
Centralized Methods
Pros
•
•
•
Cons
Cheaper and easier to
build the structure.
Fit to manage tasks for
each agent, then ease to
work.
Reduce conflict of actions.
•
•
•
A single point of failure.
Limited Bandwidth.
Congestion of
transportation.
Conflict of assignment.
Collecting information of
each sub-decision making
through the center.
Decentralized Methods
•
•
No single point of failure
Each of agent has
capability to coordinate
their actions by
themselves.
•
•
Distributed Methods
•
local information
exchanging among
neighbors
Support Dynamic network
topology
Support Large-scale
network
No global information
•
•
27
Auction Algorithm

The auction algorithm is an iterative method to find a best prices
and an assignment that maximizes the net benefit, for solving the
classical assignment problem

Task assignment






m agents and n tasks, matching on one-to-one
Benefit cij (cost function) for matching agent i to task j
Assigning agents to tasks so as to maximize the total benefit
Agents place bids on tasks, and the highest bid wins assignment
A central system acting as the auctioneer to receive and evaluate
each bid
Once all of bids have been collected, a winner is selected based on a
predefined scoring metric (Bid Price)
28
Auction Algorithm
29
Auction Algorithm
30
Negotiation

Auctions are only concerned with the allocation of goods: richer
techniques for reaching agreements are required

Negotiation is the process of reaching agreements on matters of
common interest

Any negotiation setting will have four components:





negotiation set: possible proposals that agents can make
protocol
strategies, one for each agent, which are private
rule that determines when a deal has been struck and what the
agreement deal is
Negotiation usually proceeds in a series of rounds, with every agent
making a proposal at every round
31
Negotiation in Task-Oriented Domains

Imagine that you have three children, each of whom needs to be delivered
to a different school each morning.

Your neighbor has four children, and also needs to take them to school.

Delivery of each child can be modeled as an indivisible task.

You and your neighbor can discuss the situation, and come to an agreement that it is better
for both of you (for example, by carrying the other’s child to a shared destination, saving him
the trip).

There is no concern about being able to achieve your task by yourself.



The worst that can happen is that you and your neighbor won’t come to an agreement about
setting up a car pool, in which case you are no worse off than if you were alone.
You can only benefit (or do no worse) from your neighbor’s tasks. Assume, though, that one
of my children and one of my neighbors’ children both go to the same school (that is, the
cost of carrying out these two deliveries, or two tasks, is the same as the cost of carrying
out one of them).
It obviously makes sense for both children to be taken together, and only my neighbor or I
will need to make the trip to carry out both tasks.
--- Rules of Encounter, Rosenschein and Zlotkin, 1994
32
Researches: Machines Controlling and
Sharing Resources
 Electrical
grids (load balancing)
 Telecommunications
 PDA’s
(schedulers)
 Shared
 Traffic
networks (routing)
databases (intelligent access)
control (coordination)
33
References

Micheal Wooldridge, “An Itroduction to Multiagent Systems,” John Wiley&Sons, May 2009.

S. Sodee, M. Komkhao and P. Meesad: Consensus Decision Making on Scale-free Buyer
Network. Intl. J. Computer Science pp. 1554-1559, 2011.

S. Sodsee, M. Komkhao, Z. Li, W.K.S. Tang, W.A. Halang and L. Pan: Discrete-Time
Consensus in a Scale-Free Buyer Network. In: Intelligent Decision Making Systems, K.
Vanhoof, D. Ruan, T. Li and G. Weets (Eds.), pp. 445–452, Singapore: World Scientific 2010.

S. Sodsee, M. Komkhao, Z. Li, W.A. Halang and P. Meesad: Leader-following Discrete-time
Consensus Protocol in a Buyer-Seller Network. Proc. Intl. Conf. Chaotic Modeling and
Simulation, Greece, 2010.

T. Labella, M. Dorigo, and J. Deneubourg, “Self-Organized Task Allocation in a Group of
Robots”, Proceedings of the 7th International Symposium on Distributed Autonomous Robotic
Systems (DARS04). Toulouse, France, June 23-25, 2004.

B.B. Biswal and B.B. Choudhury, “Cooperative task planning of multi-robot, systems”, 24th
international Symposiam on Automation & Robotic in Constructions (ISARC), 2007.
34