Agent Technologies

Based on tutorials and presentations: M. Wooldridge, G. Becerra, M.

Calisti, A. Tveit, S. Green, L. Hurst, B. Nangle, P. Cunningham, F. Somers,
R. Evans, P. Storms, D. Kotz, Picco, Murphy, Gray, Suri, Tschudin,
Bradshaw, Rus, Harrison, Karjoth, Ranganathan, Suri, Singh, Huhns,
Laamanen, D.B. Lange, M. Oshima, V. Terziyan and others

1
I am grateful to the anonymous photographers and artists,
whose photos and pictures (or their fragments) posted on
the Internet, have been used in the presentation.

2
 Mobility and Self-Management, Abilities to
Communicate, Cooperate, and Negotiate with others
- are among the basic abilities of an Intelligent Agent

3
 Multi-Agent System (MAS)
A network of problem solvers that work
together to solve problems that are beyond
their individual capabilities:
- solve problems that may be too large for a
centralised single agent,
- provide better speed and reliability,
- tolerate uncertain data and knowledge.

4
 Two extremes of MAS
Cooperative Distributed Problem Solving or Cooperative
Multi-Agent Systems (CMAS)
Agents that had all been designed by a single designer.
Agents could be counted on to act for the greater good of the
system, since they could all be programmed that way by the
designer, who was only concerned with increasing the
general system’s performance and not the performance of
individual agents.

Self-Interested Multi-Agent Systems (SMAS)

Individually motivated agents, who had been designed by independent
designers. Hence such agents are considered self-interested, competitive or
non-cooperative and may exhibit antagonistic behaviour.
E.g., a set of personal meeting-scheduling agents where each agent tries to
schedule a meeting at the best time for its particular owner.

5
 Agents in a multi-agent system
must:
be self-managed to automatically create suitable
configuration for each new context;
move to be able to find each other and access
environmental resources (mobility);
communicate amongst themselves;
coordinate their activities;
negotiate once they find themselves in conflict.

9
To achieve these agents must be
intelligent
Our next lecture

http://www.cs.jyu.fi/ai/vagan/Agent_Intelligence.pptx
10
… and even emotionally intelligent!

http://www.cs.jyu.fi/ai/EBI.ppt 11
 Emotional (Ethical / Moral)
vs. Rational (Safe) Choice

Which option must be chosen by an autonomous driver ?

VS VS

12
http://www.technologyreview.com/view/542626/why-self-driving-cars-must-be-programmed-to-kill/
 (Multi) Agent Technologies

• Self-Management (next lecture)

Mobility
• Communication
• Coordination
• Negotiation

13
 Mobile Agents

“A mobile agent is a program that can migrate

from machine to machine in a heterogeneous
network. The program knows why and chooses
when and where to migrate. It can suspend its
execution at an arbitrary point, transport
itself to another machine and resume
execution.”

14
 Mobile agent
vs.
virus
The agent can be categorized into two types; one is known as an intelligent
agent and the second is malicious agent, called computer virus. These types of
agents have some common characteristics; both are intelligent by nature and
both have the properties of reproduction and transfer from one host to
another host in the network, called mobility or transferability. The
fundamental difference between these agents is that the mobile intelligent
agent uses deliberate transport infrastructure while the malicious agent
hijacks the resources. The intelligent agents are used in a distributed
environment because they are not cumbersome for the network traffic; they
overcome network latency, operate in heterogeneous environment and possess
fault-tolerant behavior, on the other hand the malicious agents are designed to
be inefficient and wasteful for computer network resources and beyond.
https://core.ac.uk/download/pdf/25901857.pdf

15
 …humans cannot but agents can (?)

https://www.youtube.com/watch?v=yfePpMTbFYY
https://www.youtube.com/watch?v=VA9-R5X6Fkw
16
Mobile Agent : Conceptual Diagram

Mobile Computing Laboratory,

Dept of IECS, FCU

17
 Mobile Agents

Machine 1

...
1. Send agent
2. Send child agents /
collect partial results

3. Return merged
and filtered results Dynamically selected
proxy site Machine n

18
Humans and Agents: “Mind” vs. “Body”
Typically humans are characterized as having both a mind (nonphysical) and
body/brain (physical). Agent as virtual copy of human must have also a kind
of “mind” and a kind of “body”. Philosophy still concerns the questions:
whether the mind and body are separate or the same thing; if separate then
who controls whom and how do they interact… The models of humans’ and
agents’ “mobility” depend on the vision chosen for the “body-mind”
dilemma.
Materialism is the belief that nothing exists apart from the material world
(i.e., the brain is a physical matter like all the body and the mind is just the
function of the brain).
Subjective Idealism believes that physical objects and events are reducible
to mental objects, properties, events. Ultimately, only mental objects (i.e. the
Mind mind) exist and that what we think of as our body is merely the perception of
mind.
Dualism is the view that the mind and body both exist as separate entities.
Body Cognitive psychology has placed a new emphasis on this debate. They have
taken the computer analogy of Artificial Intelligence and applied it to this
debate as a new version of dualism. They argue that the brain can be
Belongings compared to computer hardware that is "wired" or connected to the human
body and the mind therefore is like software, allowing a variety of different
Agent Mobility: software programs: to run. This can account for the different reactions people
What is moving: mind, have to the same stimulus.
McLeod, S. A. (2007). Mind Body Debate. Retrieved from:
body, “luggage” or all? www.simplypsychology.org/mindbodydebate.html

19
“Preparing for the trip”

20
 “Full” mobility (before)

Home Host
Distributed Agent Platform

21
 “Full” mobility (during)

Home Host
Distributed Agent Platform

22
 “Full” mobility (after)

Home Host
Distributed Agent Platform

23
“Partial” or “Mind” mobility (before)

Home Host
Distributed Agent Platform

24
“Partial” or “Mind” mobility (during)

Home Host
Distributed Agent Platform

25
“Partial” or “Mind” mobility (after)

Home Host
Distributed Agent Platform

26
“Telepresence” or mobility imitation
remote control by communication

Home Host
Distributed Agent Platform

27
“Surrogates” and Telepresence

2009

Telepresence

http://spectrum.ieee.org/robotics/humanoids/the-reality-of-robot-surrogate
s
28
Why Mobile Agents?

29
 Reason 1: Bandwidth conservation
Text documents,
numerical data, etc.

Dataset

Server Client/Proxy

Dataset

Client/Proxy
Server

30
 Reason 2: Reduce latency
Latency, a synonym for delay, is an
expression of how much time it takes
for a packet of data to get from one
designated point to another

1. Observe 2. Move to better

high average location
latency to
clients

If many distributed computational processes in the network need coordination,

then move coordination entity (controller) closer to the computational processes.
31
 Reason 3: Acting Autonomously and
Asynchronously
1. Mobile client
2. Mobile client moves to
performing task task
the network to perform
task autonomously

expensive cheap

task

task

task

Mobile devices often rely on expensive or fragile network connections. Tasks requiring a
continuously open connection between a mobile device and a fixed network are probably
not economically or technically feasible. To solve this problem, tasks can be embedded
into mobile agents, which can then be dispatched into the network. After being
dispatched, the agents become independent of the process that created them and can
operate asynchronously and autonomously. The mobile device can reconnect at a later
time to collect the agent. 32
 Reason 4: They are robust and fault tolerant

Mobile agents ability to react dynamically to unfavorable situations and events

makes it easier to build robust and fault-tolerant distributed systems. If a host
is being shut down, all agents executing on that machine are warned and
given time to dispatch and continue their operation on another host in the
network.

33
 Reason 5: Load balancing

Jobs/Load

Load balancing is distributing processing and

Jobs/Load migrate in a heterogeneous communications activity evenly across a computer
network so that no single device is overwhelmed.
network of machines
34
 Agent Mobility in Heterogeneous
Environments

http://ceur-ws.org/Vol-752/52-58mitrovicetal.pdf

35
 ABSTRACT
Mobile agents are becoming pre-eminent by not only outperforming in comparison with the conventional techniques such
as RMI, RPC etc. but also by surpassing their loopholes. They promise to solve many major issues of high network
bandwidth consumption during communication, bottleneck problem of centralized system, even can act as intrusion
detection agents, and may also be used as monitoring of various nodes in multifarious domains like e-commerce services,
for load balancing in cluster, health care monitoring systems, air traffic control systems, and many more. In this paper, the
agent server required to allow mobile agents on any machine in network are compared for homogenous and
heterogeneous nodes. The homogeneity and heterogeneity of nodes is defined at the hardware level and type of OS
installation. Basically, a mobile agent is moving the code to data rather data to code. Agent and agent server are two
different parts, in which agent is a computational, operational and communicative entity while the agent server takes care
of fundamentals execution and security features. To all intents and purposes, these agent servers help mobile agents to
interact and engage with the underlying system acting as an execution environment for them. Agent servers, also called as
agency or agent runtime environment, may differ for different platforms and this contrast lies in the software architectural
components which they contribute being a middle layer in between the mobile agents and system platform. This paper
focuses on architectural dissimilitude between agencies of heterogeneous and homogeneous distributed systems.

36
 (Multi) Agent Technologies

 Self-Management
 Mobility
Communication
 Coordination
 Negotiation

37
 Agent-to-agent communication

Indirect communication (shared memory) Message passing

 information available for all  direct exchange
 no direct communication  common language
 simple architecture  conversation - sequences of
messages
Agent
Agent Agent
Agent

Blackboard
Blackboard Agent
Agent Agent
Agent AA Message Agent
Agent BB
Agent
Agent
(Sender)
(Sender) (Receiver)
(Receiver)

Agent
Agent Agent
Agent

38
 Communication is not just for fun;
it is very pragmatic in agents’ world

39
 Speech Act Theory
High level framework for human
communication
“Language as Action” (J.L. Austin)

• Speakers do not just exchange true or false sentences;

• Speakers perform speech acts:
• Requests, suggestions, promises, warnings, etc.

40
Watch this video to get the point of
the Speech Act Theory
https://www.youtube.com/watch?v=3-son3EJTrU

Language as a Window into Human Nature

41
Watch also this one on Semantics and
Pragmatics of Speech Act Theory
Semantics and Pragmatics - Speech Acts, An Overview
https://www.youtube.com/watch?v=rs6O77SkIOo

42
… and this one: Literal vs Conventional
Meaning of Communication
Philosophy: Conventional Implicature
https://www.youtube.com/watch?v=YD82l_bUhLc

43
 SA Theory (cont)
• Locution:
• Taking into account context and reference, i.e., who is the speaker and the hearer,
which is the object, etc.
• Illocution
• The act of transporting intentions, i.e., the speaker wants the hearer to do something or
to think something as a consequence
• Perlocutions
• Actions that occur as a result of the illocution
• Example: „Open the window!“
• Locution: Monique is the speaker, Steve is the hearer, the window is the object
• Illocution: Monique wants Steve to open the window
• Perlocution: Steve opens the window

44
 Semantics
• What do we mean by `semantics´?
• Providing semantics is the process of ascribing meaning. (Y. Labrou)

To describe the state of an agent before sending a

particular message and after having received it is a useful
way to ascribe meaning to communication primitives

• What do we mean by agents´states?

• Which language to use for describing agents´states?

45
 Plan-Based Semantics
• Here is their semantics for request:
request(s, h, f)
• pre:
• s believe h can do f
(you don’t ask someone to do something unless you think they can do it)
• s believe h believe h can do f
(you don’t ask someone unless they believe they can do it)
• s believe s want f
(you don’t ask someone unless you want it!)
• post:
• h believe s believe s want f
(the effect is to make them aware of your desire)

46
 FIPA ACL
• The standard FIPA Agent Communication Language is FIPA ACL
• FIPA ACL is based on speech acts
• To send and process messages corresponds to perform actions, i.e.,
communicative acts (CAs)
• CAs are described a formal semantics based on modal logic

http://dl.acm.org/citation.cfm?doid=1293731.1293735

http://www.fipa.org/ 47
 INFORM
FIPA ACL

• Basic structure :
• Performative (aka “function”)
22 performatives in FIPA
• Housekeeping (aka “variables”) sender John
e.g., sender, receiver, subject, etc. receiver Mary
language English
• Content (aka “data”) in-reply-to Query #7
the actual content of the message

YES ! I love you !!!

48
FIPA-ACL message elements
Element Category of Elements
performative Type of communicative
acts
sender Participant in
communication
receiver Participant in
communication
reply-to Participant in
communication
content Content of message
language Description of Content
encoding Description of Content
ontology Description of Content
protocol Control of conversation
conversation-id Control of conversation

reply-with Control of conversation

in-reply-to Control of conversation
reply-by Control of conversation

49
 FIPA-ACL message elements
• query
• request Element Category of Elements
• agree
:performative Type of communicative
• inform acts
sender Participant in
• failure
communication
• refuse receiver Participant in
communication
• subscribe reply-to Participant in
• cfp communication
content Content of message
• propose
language Description of Content
• propagate
• encoding Description of Content
proxy
• not-understood ontology Description of Content
protocol Control of conversation
conversation-id Control of conversation

reply-with Control of conversation

in-reply-to Control of conversation
reply-by Control of conversation

50
 FIPA CAs Examples
• query
Communicative acts
• The door is open?
• Open the door (for me) • request

Performatives
• OK! I’ll open the door • agree
• The door is open • inform
• I am unable to open the door • failure
• I will not open the door • refuse
• Say when the door becomes open
• subscribe
• Anyone want to open the door?
• cfp
• I can open the door for you…at a price
• propose
• Inform all others that the door is open
• propagate
• Inform your mother that the door is
open • proxy
• Door? What’s that? Don’t understand... • not-understood

Full list of performatives is here:

http://www.fipa.org/specs/fipa00037/SC00037J.html#_Toc26729686 51
 Semantics of CAs
• Feasibility Precondition:
• The conditions (i.e., one or more propositions) which need to be
true before an agent can plan to execute an action
• Rational Effect:
• The rational effect of an action is a representation of the effect
that an agent can expect to occur as a result of the action being
performed.

Effect 52
 Agents´Interactions
• Ongoing conversations between agents have typical patterns.
Certain message sequences are expected, and, at any point in the
conversation, other messages are expected to follow. These typical
patterns of message exchange are called interaction protocols (IP)
• By their nature, agents can engage in multiple dialogues, perhaps
with different agents (also humans), simultaneously. The term
conversation is used to denote any particular instance of such a
dialogue. Thus, the agent may be concurrently engaged in multiple
conversations, with different agents, within different protocols.

53
 FIPA standard IPs
• fipa-query: the receiver agent is requested to perform some kind of
inform action
• fipa-request: the receiver is requested to perform some action.
• fipa-contract-net: an agent (manager) solicits proposals from other
agents (contractors) by specifying the task and the conditions placed
by the manager upon the execution of the task. The contractor‘s
proposal includes the preconditions that the contractor is setting out
for the task (e.g., price, time, etc.)
(request :sender A
• ..some more :receiver B
:content (some act)
:protocol fipa-request
)

54
 The Contract Net Protocol
An important generic protocol

• Manager announces tasks via a

(possibly selective) multicast
• Agents evaluate the announcement.
Some submit bids
• Manager awards a contract to the most
appropriate agent
• Manager and contractor communicate
privately as necessary

55
Check how communication protocols are
used in this “game”; this is interesting …
http://taga.sourceforge.net/ TAGA game simulates a global
travel market in the Agentcities
environment. A customer from City
A want to have a recreational tour in
resort R, so he/she needs a round-
trip flight ticket and corresponding
hotel accommodation when he/she is
in resort R. Moreover, the customer
will be happy if his/her preferences
are satisfied, e.g. living in good
hotel, enjoy a concert or go to a
famous restaurant. In TAGA, all
such service providers can sell their
services on the Web and thus form a
Web Travel market. Travel agents
will help the customer to buy the
travel package from the Web travel
market. All the TAGA agents are
FIPA compliant and FIPA Interaction
Protocols are used in agent
communication. A marketing
mechanism is defined as an big
interaction protocol which embeds
some sub-interaction protocols.

56
 Watson Conversation Service:
Interaction protocols for human-agent communication

57
 (IBM) Watson Virtual Agent:
an example of communication agent (cognitive conversational self-
service engine) designed to manage certain interaction protocols
Watson Virtual Agent understands common customer service queries
from day one because it has been pre-trained on frequently occurring
customer intents, or the true purpose of the many different ways that a
customer can ask a question. For example, intents to do with store https://www.ibm.com/watson/developercloud/conversation.html
information, payment requests and account management. Watson has
been trained to answer these questions and you can quickly tailor
Watson’s responses with no complicated coding or scripting required. [
http://www.ibm.com/marketplace/cloud/cognitive-customer-engagement/us/en-us]

Watson Conversation combines a number of cognitive

techniques to help you build and train a bot - defining
intents and entities and crafting dialog to simulate
conversation. The system can then be further refined with
supplementary technologies to make the system more
human-like or to give it a higher chance of returning the
right answer. Watson Conversation allows you to deploy a
range of bots via many channels, from simple, narrowly
focused Bots to much more sophisticated, full-blown
virtual agents across mobile devices, messaging platforms
like Slack, or even through a physical robot.

58
Enjoy the performance of artificial agent in virtual
banking conversation with a human on this video

59
A lot can be captured from you following
the samples of your communications

60
 Conclusions
• Agent-to-agent communication is fundamental to realize the
potential of agent paradigm
• For agents´ interoperability standards are essential
• Agent Communication Language for defining which communicative acts to
use
• Content languages to express the knowledge interchange format
• Ontologies to agree upon a common vocabulary and meanings to describe a
subject domain
• Predefined Interaction Protocols in order to define which sequence
of messages is expected
• The semantics approach can be oriented either to ascribe meaning to
a single CA or to an IP (more pragmatic way of working)

61
 (Multi) Agent Technologies

• Self-Management
• Mobility
• Communication
 Coordination
• Negotiation

62
 Why is coordination needed?

• Dependencies between agents actions

• Handle global constraints

• E.g. Time, Money, or computational resources

• No individual have sufficient competence or

resources to solve the entire problem alone

63
 Not the same but all important
• Collaboration is working together towards a shared objective. Co-creation
is collaboration towards joint production of a mutually valued outcome.
• Coordination is sharing information and resources so that each party can
accomplish their part in support of a mutual objective.
• Cooperation assumes exchange of relevant information and resources in
support of each other’s (possibly selfish) goals, rather than a shared objective.

64
 Terminology of Coordination - I
• Coordination
• An agents reasoning about its own and the behavior of other
agents in order to ensure that the community acts in a coherent
manner

• Coherence
• Agents actions can be performed and that they do not conflict
with one another

65
 Terminology of Coordination - II
• Convention
• Means of monitoring commitments in changing circumstances

• Commitment
• Agents promises to undertake a specified sequence of actions
or/and to reach a specific state

66
 Example of Commitment
promise (Awho, Bwhom, Pwhat) - a commissive act
A

B
promise (A, B, P1), where P1 =x(1  x  3), y (¬ Dirt(x,y));
promise (B, A, P2), where P2 =x(4  x  6), y (¬ Dirt(x,y)).
67
 Example of Convention
dropCommitment (Ccommit, Qif) - a conventional act

B
dropCommitment (C1, Q1), where Q1 = x(1  x  3),  y (Monster(x,y));

C1 = promise (A, B, P1), where P1 =x(1  x  3), y (¬ Dirt(x,y)).

68
Coordination as a distributed
Goal-Search Problem

Agent 1 Agent 2

O verall G oal 1 O verall G oal 2

AND OR
S ubgoal 1 S ubgoal 2 S ubgoal 3

S ub..S ubgoal n

71
Coordination as a distributed Goal-
Search Problem
A distributed goal
search tree involving
Agent1 and Agent2.
The dotted arrows
indicate
interdependencies
between goals and data
in different agents,
solid arrows
dependencies within
an agent. The
superscripts associated
with goals and data
indicate the agent
which contains them.

72
 Read more on coordination in:

“Commitments and Conventions: The Foundation

of Coordination in Multi-Agent Systems”

by NICK R. JENNINGS
http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=B8559C8CAD3
6F066029B6DA5D9FF25DF?doi=10.1.1.65.5594&rep=rep1&type=pdf

73
Collaboration and Coordination
for reaching a community goal
a

b
sin(x)
ln(x)

Community goal:
x+y Input: a, b
Output: ln [sin(a) + sin(b)]

74
 Interesting challenge:
Agents + Robots + Humans Coordination

Read article here: http://www.ihmc.us/users/mjohnson/papers/070924-Coordination%20in%20Human-Robot%20Teamwork.pdf

75
 Coordination and game theory
 Collaborative MAS case (trying to make everyone as “happy” as
possible):
 Pareto efficiency, or Pareto optimality, is a state of coordination process involving two or more
collaborative agents, in which it is impossible to make any one individual better off without
making at least one individual worse off.
 Self-Interested MAS case (trying to make oneself “happier” than
others):
 Nash equilibrium is a is a state of coordination process involving two or more self-interested
agents, in which each agent is assumed to know the equilibrium strategies of the other agents,
and no agent has anything to gain by changing only their own strategy.
 See also a “Prisoner's dilemma”: http://en.wikipedia.org/wiki/Prisoner's_dilemma

76
 “Prisoner's dilemma"
 "prisoner's dilemma" (Poundstone, 1992) : Two criminals are
arrested and imprisoned. Each prisoner is in solitary confinement
with no means of communicating with the other. The prosecutors
lack sufficient evidence to convict the pair on the principal charge.
They hope to get both sentenced to a year in prison on a lesser
charge. Simultaneously, the prosecutors offer each prisoner a
bargain. Each prisoner is given the opportunity either to: betray the
other by testifying that the other committed the crime, or to
cooperate with the other by remaining silent. The offer is:
 If A and B each betray the other, each of them serves 5 years in prison
 If A betrays B but B remains silent, A will be set free and B will serve 20 years in
prison (and vice versa)
 If A and B both remain silent, both of them will only serve 1 year in prison (on the
lesser charge)

77
 t han
rd rs
“Prisoner's dilemma" a t e r r e w
d
a
p r i s o n e
t w o
r e
g: Two es t e r
foare he
 "prisoner's dilemma" (Poundstone, 1992) a n t e rcriminals m e T
e r s l f - i t c o e r .
arrested and imprisoned. Each ff
oprisoner l is e
s in solitary e o u th ward
confinement
o
e r i o n a ib l ch al r e
a r t n r a t o s s e a ld
with no means of communicating p l y with y p
the other. r a y The prosecutors
i d u o u
a p u re o n l b e t
n d i v e y w
n s
lack sufficient y i n g
evidence a l l
to convict t h e the pair
t o on the
g i principal t h charge. m a
t r a e m , i n g e m u i n h e n h u d
Theyshope b e to get t h
both a
sentenced
e n r t
to ha year u r s
in prison y w on a l i t y
lesser, a n
a u e w ith er, m is fo that p betra r e a
n t h is
o f
e c
Bcharge. i n g
tSimultaneously, o t h e s
r prosecutors
the is to eeach
offer t I n
. prisoner o i
r a del s
p e r a t h e so n s u l t e r s i l n a v i o
o
co betEach y
ra nprisoner p r i r e o n
is opportunity s e h m
bargain. a l t isisgiven pthe
h r k e pt iveither e b to:pbetray l e the
tio rt that he committed t im
will by
other l y r a
testifying a
of the o f t
other y b o t h
o o p e r a
the crime,
b y s or to
u r e g p b o t h t h e s c c t e d
p
cooperate i n
stwithethe d s other byif remaining d
ar silent. e i
d Theiaoffer ] is:
e r e a r d o w p r d
 IfinAt and aBlleach y l betray r e wthea other, b i a s
each
t of themis serves k5iyears
a t i p e in prison
g i c t t e r i c w h . [ W
 If lAobetraysbBe but B remains t e m silent, i te A willctbe i onset free and B will serve 20 years in
a s y s s p a
get(andlavice
prison
y a versa) e s d e
e s t e d
s p a m t e r
 If A and di lar g
B both remain
l
silent,
f - i n both of them will only serve 1 year in prison (on the
lesser charge)m i " s e
si onal
i
" r at 78
 (Multi) Agent Technologies

• Self-Management
• Mobility
• Communication
• Coordination
 Negotiation

80
 Intra-agent negotiation
model
• Buyer (service requestor):
– The main goal is to get the good (or service) s maximizing the utility
expressed as:
U(s) = Value – Price
An offer is considered unacceptable if U(s) < 0
An offer can be considered acceptable when U(s) > 0
• Seller (service provider):
– The main goal is to maximize the profit expressed as:
(s) = Price – Cost

81
 I VALUE cleaned room
as much as 100 Euro.
Therefore I have to buy
service from a cleaning
Buyer’s logic:
agent for a PRICE
cheaper than that … U(Cleaned_Room) =
= 100 – PRICE > 0

Service
consumer
agent

82
 To clean the room I need to
spend resources which
Seller’s logic: COST me as much as 45
Euro. Therefore I have to
П(Cleaned_Room) = sell my service for a
PRICE bigger than that …
= PRICE - 45 > 0

Service
provider
agent
“Window” for negotiations: 45 < PRICE < 100 83
 Negotiation Protocols
• A specification of the possible valid actions (how the interaction can
take place)
• Possible participants
• Buyers, sellers, mediators, etc.
• Negotiation states
• Offering, accepting proposal, negotiation closed, etc.
• Events that determine state transitions
• No more offers, timeout expiration, etc.
• Valid actions of the participants
• Accept, reject, counter-propose, etc.

84
 Agreements
• Negotiation to agree about what?

• Price
• Quality
• Quantity, volume, number, etc.
• Payment modality More
Morethanthanone
oneissue
issuecan
can
• Delivery process be
benegotiated
negotiated
• Delivery date atatthe
thesame
sametime.
time.
• Maintenance
• .....

85
 What is an auction?
„Auction are popular, distributed and autonomy preserving ways of
allocating items among agents.“ [Sandholm 1999]
„An auction is a method of allocating scarce goods, a method that is
based upon competition“. [Agorics, 1996]
• Auctions provide an efficient way to
• Quickly find partner with highest valuations
• Run one-to-many interaction
• Determine fair prices
• Allocate resources

An auction is a mechanism to find prices that allow commitments

86
 Auction Principles
• An auction protocol establishes the rules governing an auction:
• how bids and offers can be done
• The order in wich prices are quoted
• how to determine the winner
• Auction Protocols:
• FPSB (First-Price Sealed Bid)
• Vickrey or SPSB (Second-Price Sealed Bid)
• English
• Dutch
• An auction strategy is used to determine the bid to be offered

87
 FPSB, Vickrey
• In FPSB and Vickrey (or SPSB) auctions the main three phases are:
• Bidding phase: every participant can submit only one secret bid per
auction.
• Winner determination phase: the auctionner evaluates the highest (or the
lowest one when appropriate) bid. The bidder who offered the highest (or
the lowest one when appropriate) bid is the winner.
• Auction ending: the winner and the seller are notified about the results and
about each other identity.
• In FPSB auctions the winner pays the highest (or the lowest one
when appropriate) bid.
• In Vickrey (SPSB) auctions the winner pays the second highest (or
the second lowest one when appropriate) bid.
• Advantages: one-round bid, no information revealed, the buyers do
not need to be present.
FPSB: First-Price Sealed-Bid SPSB: Second-Price Sealed-Bid
88
 50
FPSB example

Service
consumer agent 60

70

80

The “winner” will get

50 Euro for his job 89
 50
SPSB (Vickrey)
example
Service
60
consumer agent

70

80

The “winner” will get

60 Euro for his job 90
 90
FPSB example
(other way around)

80

Service
provider agent

70

The “winner” will pay

90 Euro for the service
91
 90
SPSB (Vickrey)
example
(other way around)

80

Service
provider agent

70

The “winner” will pay

80 Euro for the service
92
 English
• In an English auction there are three main phases:
– Bidding phase: every participant submits bids. Every bidder can submit several
bids up to a pre-fixed maximum amount. A new bid has to be greater (lower when
appropriate) than the highest current bid in the market.
– Winner determination phase: the auctionner evaluates the highest (lowest when
appropriate) bid. The bidder who offered the highest (or lowest when appropriate)
bid is the winner.
– Auction ending: the winner and the seller are notified about the results and about
each other identity.
– All auction participants have to be informed about the current highest (lowest
when appropriate) bid
– Delays in delivering bids can change auctions results: buyers should be present
– Advantage: no counterspeculation, but long process
– Art auctions, etc.

93
English auction example
1
2
3 4
5

94
 Dutch
• In a Dutch auction the main phases are:
– Bidding phase: every participant can submit one bid.
– Winner determination phase: the auctionner continuosly lowers (increases when
appropriate) the price until a bidder accepts the price.
– Auction ending: the seller is notified about the results and about each other
identity.

– All auction participants have to be informed about the current

price
– Delays in delivering bids can change auctions results
– Advantage: efficient, reveals only information about the winner
– Dutch flower market, fish markets

95
 More Auctions
http://www.swdsi.org/swdsi06/proceedings06/Papers/MIS15.pdf

96
 Bored and tired
from the abstractions?
You, probably, wonder:
“How on earth all these
will help me during my everyday work in a company?”
PAAMS 2019 : 17th International answers
Conference on Practical Applications the question:
of Agents and Multi-Agent Systems

How abstract agent approaches, methods, and tools help

system designers to accomplish the mapping between
available agent technology and application needs?
018
2
S
https://www.paams.net/ AM
PA
97
Topics: PAAMS 2019 : 17th International
Conference on Practical Applications
of Agents and Multi-Agent Systems

https://www.paams.net/
98
 ABSTRACT: Deep neural networks have achieved impressive successes in fields ranging from object recognition to complex games such as Go. Navigation, however,
remains a substantial challenge for artificial agents, with deep neural networks trained by reinforcement learning failing to rival the proficiency of mammalian spatial
behaviour, which is underpinned by grid cells in the entorhinal cortex. Grid cells are thought to provide a multi-scale periodic representation that functions as a metric for
coding space and is critical for integrating self-motion (path integration) and planning direct trajectories to goals (vector-based navigation). Here we set out to leverage the
computational functions of grid cells to develop a deep reinforcement learning agent with mammal-like navigational abilities. We first trained a recurrent network to
perform path integration, leading to the emergence of representations resembling grid cells, as well as other entorhinal cell types. We then showed that this representation
provided an effective basis for an agent to locate goals in challenging, unfamiliar, and changeable environments - optimizing the primary objective of navigation through
deep reinforcement learning. The performance of agents endowed with grid-like representations surpassed that of an expert human and comparison agents, with the metric
quantities necessary for vector-based navigation derived from grid-like units within the network. Furthermore, grid-like representations enabled agents to conduct shortcut
behaviours reminiscent of those performed by mammals. Our findings show that emergent grid-like representations furnish agents with a Euclidean spatial metric and
associated vector operations, providing a foundation for proficient navigation. As such, our results support neuroscientific theories that see grid cells as critical for vector-
based navigation, demonstrating that the latter can be combined with path-based strategies to support navigation in challenging environments.

Check this recent article related to your assignment 99…

 https://
en.wikipedia.org/wiki/Simultaneous_localiz
ation_and_mapping
http://ieeexplore.ieee.org/document/1638022/
http://ieeexplore.ieee.org/document/1678144/

In robotic mapping and navigation, simultaneous

localization and mapping (SLAM) is the computational
problem of constructing or updating a map of an unknown
environment while simultaneously keeping track of an
agent's location within it. Published approaches are employed
in self-driving cars, unmanned aerial vehicles, autonomous
underwater vehicles, planetary rovers, newer domestic robots
and even inside the human body. [Wikipedia]

Check SLAM approach related to your assignment 100

…
 https://
en.wikipedia.org/wiki/Simultaneous_localiz
ation_and_mapping
http://ieeexplore.ieee.org/document/1638022/
http://ieeexplore.ieee.org/document/1678144/

In robotic mapping and navigation, simultaneous

localization and mapping (SLAM) is the computational

… However, within your assignment,

problem of constructing or updating a map of an unknown
environment while simultaneously keeping track of an
agent's location within it. Published approaches are employed

I expect you being capable to suggest

in self-driving cars, unmanned aerial vehicles, autonomous
underwater vehicles, planetary rovers, newer domestic robots
and even inside the human body. [Wikipedia]

(“invent”) something of your own,

which will be simpler and more
elegant than SLAM …

Check SLAM approach related to your assignment 101

…
 Some reports on our industrial experience with agent technology:
Khriyenko, O., & Terziyan, V. (Eds.). (2010).
UBIWARE (Smart Semantic Middleware for Ubiquitous Computing) Final Project Report. Technical Report (50 pp.).
UBIWARE Project: University of Jyvaskyla.

(Brief presentation of the UBIWARE Final Report is here).

(UBIWARE platform documentation and UBIWARE download package).

Terziyan, V. (Ed.). (2007). Smart Resource (Proactive Self-Maintained Resources in Semantic Web) Final Project Report.
Technical Report (98 pp.). SmartResource Project: University of Jyvaskyla.
(Presentation of the SmartResource Final Report is here).
(Download SmartResource platform here).

All reports, presentations and downloads from SmartResource and UBIWARE projects 102