SURVEY OF MULTIAGENT SYSTEM

NAME: MUHAMMAD USMAN ALI

Reg. No.: L1S17MSEE0019
 Survey of Multi Agent Systems.
Muhammad Usman Ali1
1
Electrical Engineering Department,
University of Central Punjab,
Lahore, Pakistan.
1
[email protected]

Abstract—Multi Agent Systems (MASs) is a mean used to solve the difficult problems by dividing them into
small parts. It’s been in the attention of students from different field of lines like Electrical, Civil and Computer
science etc. Each task assigned to independent entity, which are called agents. That agent had to decide a action
which is best for solving the task by using multiple inputs. Multiple applications in the world had been founded
MASs which are smart grids, more complex problems and different kind of networks in computers. On one side
MASs had a large number of applicability still on the other side it had to face many challenges. This paper is
expected to be a detailed note on MASs for researchers of this field.

Keywords— MAS (Multi Agent Systems), MAS applications.

I. INTRODUCTION
In a past couple of years DAI: Distributed - Artificial
Intelligence had gain an attention of large number of
researchers due its very complex problems [1]. Its
algorithms had been divided into three parts: Parallel
Artificial Intelligence (PAI), Distributed Problem
Solving (DPS) & Multi Agent Systems (MASs). PAI
includes construction of parallel algorithms their
languages and their architectures to enhance the
efficiency of its algorithms. DPS includes the division
of a main task into the sub tasks in which each is
assigned to a node with the sets of neighboring nodes
which are called the computing entities. These identities
limits the flexibilities by having their resources and
shared knowledge. The main focus in this paper is on
MAS which contains the independent entity called
agent. Agents communicates along with their neighbors
agents to learn the new actions and strategies. These
agents used their knowledge to solve the assigned task Figure 1. Paper Overview.
by performing the different kind of actions on
environments. This capability of MAS makes it a
suitable and very important tool to solve the complex
II. AGENT INTRODUCTION
problems of different fields include civil & electrical
engineering etc. In order to develop MAS a wide range Russell et al. [8] define the agent as: “A flexible
of addressing is required of difficult challenges like autonomous entity capable of perceiving the
coordination among agents [2], learning [1] & security environment through the sensors connected to it.''. It is
[1]. defined by a different way in [3] as: “An encapsulated
Agents can be implemented on many disciplines like computational system that is situated in some
Electrical & Civil engineering etc. So a detailed field environment and this is capable of flexible, autonomous
study is required of the concepts of MAS. action in that environment in order to meet its design
objective.'' These are the different definitions of 4. Entity:
different researchers we defined it as: Every agent can execute an action that can cause the change
Agent: An entity which is placed in an environment and in environment. This agent can perform two types of action it
senses different parameters that are used to make a can be discrete or continuous. I case of continuous agent can
decision based on the goal of the entity. The entity perform the infinite actions. Where as in discrete agent can
perform only a finite number of actions.
performs the necessary action on the environment based
The main objective of each agent is to solve the task that is
on this decision. assigned to it. To achieve this goal first of all agent sense all
In the above definition four words are used whose the parameters from the environment. By using this data
description are as follows: agent can analyze the environment better. Agent can also use
1. Entity: the information from its neighbor agents. The figure 2
It belongs to the agent type. Agent can be different describes the structure of an agent.
things it can be a software, its can any electrical
parameter it can be a robot etc.
2. Environment:
The location of the agent is environment. Now this
environment can be anything it can be a network is case
of computer systems. The different features of
environment can affect the complexity of agent based
system [1]:
 Accessibility:
The accuracy/perfectness by which the agents from the
environment sense the data. If the environment is
accessible it means the data sensed by agent is up to date
and accurate. Vice versa is same, if the environment is Figure 2: Agent Structure
inaccessible then the data sensed will not be an accurate
or up to date data. Following are the features that gives agent the
 Determinism: capability to solve the most complex problems [4], [5].
Determinism states to the expectedness of the outcome  Sociability:
of an action. If the environment is deterministic it As information request agents can share their
means the outcome will be predictable. Similarly vice knowledge in order to enhance their performance in
versa if the environment is non deterministic the result reaching the goal.
of an action will not be predictable.  Autonomy:
 Dynamism: Independently every agent can make its own decisions
Dynamism bring up to the variations that happen in the and can take the suitable action.
environment that are self-determining of the actions  Proactivity:
taken by the agent. Environment can be static if changes Using the history of agent it can predict the possible
in environment only occurs as the significance of the act actions of future.
of agents. And in the remaining cases it will be Agent can take its actions independently, the real
dynamic. When the environment variates, the formerly advantage is that the agents can share the information
sensed information may not be longer precise. with one another and can take its future actions. These
 Continuity: agents together can solve the complex tasks and
It refers to the discreetness and continuity of the problems. This is called Multi Agent Systems (MAS).
environment of the agent. In contrast two this MAS is
categorized in two: discrete & continuous. If the
environment is continuous it then through a continuous III. MULTI AGENT SYSTEMS (MAS)
function it affects the state of an agent. If the The main features of MAS include that it had a very low
environment is discrete, it forces the agent to go in to a lost, they are more reliable and they are more flexible.
set of encoded states. These properties make it a very effective solution to
3. Parameters: solve the complex tasks. In this a complex task is sub
Agent sense different type of data from the environment divided into small tasks. And to each task an agent will
which are called parameters. For example if a robot is be assigned [13].
moving then its parameters are its velocity and position.
MAS agents usually have an incomplete information reduces the load on the mediator agent as compared to
the main thing is that agents communicate with each the facilitator employment.
other. It reduces the communication overhead in return
it confirms the scalability but the overheads remain low A. MAS Features:
as the quantity of agents increase. For locating an agent Following seven important features of MAS had been
in order to minimize the outlined overhead, mainly for outlined and discuss the different categorizations that
a larger scale MAS, the concept of middle agent is used. arise while considering each feature. Table 1 shows a
Middle agents can be categorized as: summary of these features.

 Facilitator:
The implementation of a facilitator is shown is figure 3
(a). This facilitator plays a role of a mediocre between
the agent which is sending the request known as
requester and the agent that is giving the service known
as requestee. The facilitator courses the request to the
suitable agent. The reaction is sent back to the facilitator
which communicates it to the requester. As is obvious
that facilitator had become a bottleneck and a potential
single point of failure [6]. To reimburse the effects of a
central facilitator, numerous cooperative facilitators are
employed to answer to the requests [7]. This method
needs the facilitators to communicate to remain
coordinated and balance the load.

Table 1: MAS features.

B. Distinguishing MAS with Alike systems:

Wooldridge [1], Zhao et al. [9], and Sadeghi et al. [10]

compared MAS with expert systems and object-
oriented programming language, two concepts that also
involve decision-making and knowledge sharing. An
expert system detects the environment and learns the
knowledge, then makes a decision on Wooldridge [1],
Zhao et al. [11], and Sadeghi et al. [12] compared the
MFA to expert systems and an object-oriented
programming language, two concepts that also involve
decision-making and knowledge sharing. An expert
system detects environment and learn knowledge, then
make a decision. Solving a task Unlike MAS, where
each agent can communicate with any other agent, an
expert system can communicate and exchange data with
predefined entities. Although both the MAS and the
Figure 3: Implementation of (a) Facilitator (b) expert systems use a decision-making function, the
Mediator. input is different, which affects the original decision. In
 Mediator: an expert system, the decision is based on the collected
This employment varies from the above in that the environmental data and on the knowledge of the expert
requester and requester agents can communicate system, while an agent also uses his goal. According to
directly with each other as shown in Figure 3b. This the decision,
 Table 2: A study on the differences of MAS with
expert systems and object-oriented programming.

Figure 4: MAS Summary application

An expert system advises a controller to perform an technologies. This difficulty is removed by the help of
action. The controller can reject the decision of the these agents. MAS is further divided into the four
expert system, since it is a separate system that also uses different categories:
other inputs. However, an agent acts directly on the 1) Cloud Computing [12].
environment after making a decision. 2) Social Networking [13].
3) Security [14].
4) Routing [15].
IV. MAS APPLICATION
In this section, we present a taxonomy of MAS B. Agents in Robotics
applications based on a vast discipline: i) Computer
networks (VAT section), ii) Robotics (section IV-B), Hardware agents mention to the corporal hardware that
iii) Modeling (section IVC), iv) Cities and built builds up the robot, whereas the agents of software are
environments section IV-D), and v) smart grids (section the decision makers, path planners, task managers and
IV-E). A summary of these applications is described in communicating agents. The hardware agents uses the
Figure 4. different type of sensors like cameras, to seizure
A. Cloud Computing pictures of the environment. The groups of agents had
The difficulty in the networks of computers been made into the teams. Now these teams works on
meaningfully grows because of the appearance of latest their experience. Strengthening knowledge is used
sideways with the Probabilistic Neural Networks (PNN) Figure6: Blackboard communication.
to rise the correctness of final decision made by an
agent.
VII. MODELLING & SIMULATION FOR AGENT
C. Agents for modelling complex systems BASED SYSTEMS
It is expensive and suffers important processing
overhead because of the need for powerful modeling This section consist of multiple modeling and
platforms and huge difficulty. ABM: Agent Based assessment methods which are used to examine the
Modeling [17] a low-cost and inferior source resolution performance metrics, which differently dependent on
for modeling complex systems. the MAS application and goal, of the calculated agent-
based system associated to the state-of-the-art [20]. In
the following we outline three fundamental evaluation
V. MAS CHALLENGES methods:
MAS is facing many challenges including localization,  JADE: Java agent development framework in
task allocation, fault detection, learning, security, one of the most widely used simulation
organization and coordination control. Coordination methods. It is Java based and paybacks from
control includes consensus, controllability, third-party libraries. It is a standard based on
synchronization, connectivity and formation [18]. written FIPA. It also supports the simulating
distributed system. It has a graphical interface
of for designing MAS. It hides the complexity
VI. AGENT COMMUNICATION
of MAS from the designers. It is also an open
There are three approaches that used in agent source. It can also be linked to the simulation
communication: software MATLAB.
 Speech act: An agent can perform as a speaker  GAMA: It is also a platform for simulation for
(S) that can produce utterance to change the the systems which are agent based systems. In
views of the hearer (H). any kind of application it can simulate the
 Message passing: Agents directly message to MAS. It also supports the GAML, which is a
each other in this method as shown in figure 5. high level language which can be used to
The agents use point to point which can be simulate the MAS.
called as broadcast communication to talk to  MATLAB: When it is related to complex
other agents. mathematical evaluation it is used to study the
MAS performance. MATLAB can also be
linked with JADE for further analysis of MAS.
 Mathematical Analysis: This mathematical
analysis is also used to evaluate MAS.

Figure 5: Broadcast communication.

 Blackboard: the agents shared the data with a

common blackboard as shown in figure 6. All
the agents store the data in a common
blackboard which other agents can easily
access. Each agent can access data of multiple
agents from blackboard [19]. Table 2: Summary of MAS simulator.
The methods are specially designed to evaluate the
MAS. Because of the wide application the evaluation
methods can be engaged to assess the performance of
the agent based systems.
 [11] A. P. Garcia, J. Oliver, and D. Gosch, ``An intelligent
agent-based distributed architecture for smart-grid integrated
network management,'' in Proc. IEEE 35th Conf. Local
VIII. SUMMARY Comput. Netw. (LCN), Oct. 2010, pp. 1013_1018.
In this survey, a high-level inclusive discussion is [12] S. D. J. McArthur et al., ``Multi-agent systems for power
engineering applications_Part I: Concepts, approaches, and
proposed concerning miscellaneous features of MAS
technical challenges,'' IEEE Trans. Power Syst., vol. 22, no.
which benefits the newcomers to grip and clear the 4, pp. 1743_1752, Nov. 2007.
basic concepts of MAS, study the latest applications in [13] H. Rezaee and F. Abdollahi, ``Average consensus over
numerous disciplines, the challenges in developing high-order multiagent systems,'' IEEE Trans. Autom. Control,
MAS, and the different methods to analyses the vol. 60, no. 11, pp. 3047_3052, Nov. 2015.
performance of MAS. Firstly MAS, agents and their [14] L. Ma, H. Min, S. Wang, Y. Liu, and S. Liao, ``An
key features are defined. Then the main challenges and overview of research in distributed attitude coordination
applications were discussed while presenting references control,'' IEEE/CAA J. Autom. Sinica,
for the studies. Then, communications among the vol. 2, no. 2, pp. 121_133, Apr. 2015.
agents was discussed and the paper is concluded by a [15] J. Qi, R. Vazquez, and M. Krstic, ``Multi-agent
deployment in 3-D via PDE control,'' IEEE Trans. Autom.
conversation on evaluation methods to analyze the
Control, vol. 60, no. 4, pp. 891_906, Apr. 2015.
effectiveness of an MAS agent-based system. [16] R. Merris, ``Laplacian matrices of graphs: A survey,''
Linear Algebra
REFERENCES Appl., vols. 197_198, pp. 143_176, Jan./Feb. 1994.
[17] C. Godsil and G. F. Royle, Algebraic Graph Theory, vol.
207. Springer,
[1] M. Wooldridge, An Introduction to Multiagent Systems. 2013
New York, NY, [18] H. F. Ahmad, ``Multi-agent systems: Overview of a new
USA: Wiley, 2009. paradigm for distributed systems,'' in Proc. 7th IEEE Int.
[2] A. H. Bond and L. Gasser, Readings in Distributed Symp. High Assurance Syst. Eng., Oct. 2002, pp. 101_107.
Arti_cial Intelligence. San Mateo, CA, USA: Morgan [19] Q. Liu, L. Gao, and P. Lou, ``Resource management
Kaufmann, 2014. based on multiagent technology for cloud manufacturing,'' in
[3] S. Shamshirband, N. B. Anuar, M. L. M. Kiah, and A. Proc. Int. Conf. Electron., Commun. Control (ICECC), Sep.
Patel, ``An appraisal and design of a multi-agent system 2011, pp. 2821_2824.
based cooperative wireless intrusion detection computational [20] F. M. Al-Shrouf, ``Facilitator agent design pattern of
intelligence technique,'' Eng. Appl. Artif. Intell., vol. 26, no. procurement business systems,'' in Proc. 32nd Annu. IEEE
9, pp. 2105_2127, 2013. Int. Comput. Softw. Appl. (COMPSAC), Aug. 2008, pp.
[4] A.-M. Zou, K. D. Kumar, and Z.-G. Hou, ``Distributed 505_510.
consensus control for multi-agent systems using terminal
sliding mode and Chebyshev neural networks,'' Int. J. Robust
Nonlinear Control, vol. 23, no. 3, pp. 334_357, Feb. 2013.
[5] M. H. Bowling, ``Convergence and no-regret in
multiagent learning,'' in Proc. NIPS, 2004, pp. 209_216.
[6] A. Zidan et al., ``Fault detection, isolation, and service
restoration in distribution systems: State-of-the-art and future
trends,'' IEEE Trans. Smart Grid, vol. 8, no. 5, pp. 2170_2185,
Sep. 2016.
[7] P. Balaji and D. Srinivasan, ``An introduction to multi-
agent systems,'' in Innovations in Multi-Agent Systems and
Applications. Berlin, Germany: Springer, 2010, pp. 1_27.
[8] S. Russell and P. Norvig, Arti_cial Intelligence: A
Modern Approach,
vol. 25. Egnlewood Cliffs, NJ, USA: Prentice-Hall, 1995, p.
27.
[9] L. C. Jain and D. Srinivasan, Innovations in Multi-Agent
Systems and Application. Springer, 2010.
[10] D. Ye, M. Zhang, and A. V. Vasilakos, ``A survey of
self-organization mechanisms in multiagent systems,'' IEEE
Trans. Syst., Man, Cybern., Syst., vol. 47, no. 3, pp. 441_461,
Mar. 2017.