This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

Bio-Inspired Routing Algorithms Survey

for Vehicular Ad-hoc Networks
Salim Bitam, Abdelhamid Mellouk, SMIEEE, and Sherali Zeadally, SMIEEE

 time detour routes computation, in addition to non-safety

Abstract—Vehicular Ad-hoc NETworks (VANETs) play a key applications such as nearby information services (e.g., finding
role in the design and development of Intelligent Transportation the closest fuel station, restaurant or hotel), automated toll
Systems (ITS) which aim to improve road safety and payment, and infotainment services (e.g., which provide access
transportation productivity. VANETs cover Vehicle-to-Vehicle Internet-based services) [1].
(V2V) and Vehicle-to-Roadside (V2R) communications. One of
the most important challenges of this type of network is the Initially, ITS applications supported by two kinds of fixed
timely and reliable dissemination of messages among vehicular infrastructure-based vehicular networks which are Vehicular
nodes which enable drivers to take appropriate decisions to Wireless Local Area Networks (V-WLAN) and Vehicular
improve road safety. In the past decade, many routing protocols Cellular Network (VCN). A WLAN connection is established
for VANETs that can support reliability and safety requirements
with V-WLAN and a 3G/4G LTE (Long-Term Evolution)
have been proposed. These protocols suffer from several
limitations including complexity, lack of scalability to large scale
connection is established with VCN. V-WLAN is formed by a
networks, routing overheads, etc. To address these limitations, set of access points fixed at traffic intersections whereas VCN
various bio-inspired approaches have been proposed to route is based on a set of fixed cellular gateways such as 3G/4G LTE
packets among vehicular nodes in an optimized manner. We base stations, supplying wireless connections over land areas
survey recent proposed bio-inspired routing algorithms for the known as cells.
VANET environment. In particular, we identify the key features,
strengths, and weaknesses of these algorithms and compare them Despite many advances in fixed infrastructure-based
by using various criteria. Moreover, we propose a unified formal vehicular networks (i.e. V-WLAN and VCN) when
model of the bio-inspired multi-modular approaches applied to disseminating data packets in broadcasting mode, through the
VANET routing. We highlight main future research directions in use of low-delay and high bandwidth fixed units (i.e. access
this area. point or cellular gateway), these operated infrastructure
networks are narrowly deployed. This is because of
Index Terms— Vehicular Ad-hoc Network, Bio-inspired infrastructures’ high costs, and geographic limitations,
Algorithm, Routing Optimization, Reinforcement Learning. especially in rural areas, islands, mountains, desert, and so on.
Thus the network transmission range is very restricted to cover
I. INTRODUCTION digital service areas widely deployed as a vehicular network. In

I n the last few years, we have witnessed an increasing interest

in the transportation sector and the research community to
improve road safety by providing timely and accurate
addition, because the vehicles drive at high speed, the
communications provided by these operated infrastructure-
based networks suffer from frequent handovers (i.e. unreliable
information to vehicular drivers and transport authorities. One network connectivity) and run at a great risk of breaking up
way to improve road safety is by sharing road traffic radio links [2].
information through wireless and mobile networks with little or Moreover, infrastructure-based vehicular networks suffer
no infrastructure provided by the well-known Intelligent from another important drawback which is the cost of
Transportation System (ITS). deployment of multiple fixed units and devices at the road
ITS applications include a variety of safety applications edge. More specifically, deployment of V-WLAN or VCN
such as vehicle safety, prevention of collisions, traffic requires the use of expensive communication transceivers in
monitoring, control of traffic flows, blind crossing, and real- addition to data transfer costs which are incurred for the usage
of the operator’s infrastructure [3] [4].

This paper has been submitted on January 2014 for consideration for These aforementioned limitations of vehicular network
publication in IEEE Communication Surveys and Tutorials. require all vehicles to engage in a unicast and multi-hop
S. Bitam is with the Laboratory LESIA, Department of Computer Science, routing process with each other and with stationary and fixed
University of Biskra, 07000 Biskra, ALGERIA (e-mail: salim.bitam@
laposte.net).
gateways. As result, a new type of vehicular networks known
A. Mellouk is with the Image, Signal and Intelligent Systems Laboratory- as vehicular ad-hoc network (VANET) has been conceived to
LISSI, Department of Networks and Telecoms, IUT C/V, University of Paris- provide a reliable and ubiquitous connectivity on the road to
Est Creteil (UPEC), 94400 Vitry sur Seine, Paris - FRANCE (e-mail:
[email protected]).
traffic users that enhances the vehicular infrastructure networks
S. Zeadally is with the College of Communication and Information, and enables ITS applications in terms of network reliability,
University of Kentucky, Lexington, 40506, Kentucky, USA (e-mail: and network coverage [5], where only on-board devices could
[email protected]).

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

be deployed, and the communication cost is negligible [4]. messages should be guaranteed. In addition, multimedia
information can also play a role in traffic management and road
A Vehicular Ad-hoc NETwork (VANET) is considered as a
safety. For instance, a timely video stream that depicts a
specific kind of Mobile Ad-hoc NETwork (MANET) which
particular road situation (such as a major traffic congestion,
consists of a set of mobile nodes (Vehicles) and fixed nodes
fire, or natural disaster) ahead on a highway can enable drivers
known as Road-Side Units (RSUs). A VANET provides digital
to make appropriate decisions on whether to proceed or turn
data communication between vehicles through Inter-Vehicle
back [9]. Furthermore, there are also various types of comfort
Communication (IVC) and, between vehicles and RSUs
applications for VANETs which improve passengers’ traveling
through Vehicle-to-Roadside Communication (VRC). Due to
experience and provide informative services such as fuel
their restricted range of motion in terms of directions and
station locations, weather information, emergency and
speeds, VANET vehicles move according to an organized
breakdown services, interactive communication, Internet
mobility model with some differences between highways,
access, etc.
urban or rural areas. Moreover, a vehicle is also equipped with
some sort of radio interface or On-Board Unit (OBU) that It appears here that we need to develop and deploy efficient
enables short-range wireless IVCs and/or VRCs along with a routing algorithms which allow packets to be routed from the
Global Positioning System (GPS) integrated into vehicles to sender to the receiver via a set of intermediate nodes with a
facilitate location-based services. high level of Quality of Service (QoS) [10], [11] to ensure the
timely and reliable delivery of information and multimedia
It is widely accepted that ITS safety applications are the
data, and ITS applications [12]. QoS, in this context, means
most sensitive services in VANETs because of the significant
sending data between nodes with a minimum end-to-end delay
impact they can have on human lives.
and routing overheads. In addition, QoS also implies ensuring
It is worth mentioning that there are two types of ITS safety the successful delivery of the maximum number of transmitted
applications that can be distinguished in terms of their messages (i.e., with a minimum number of dropped packets)
transmission modes. The former is performed by relying on where the network bandwidth is optimally used. There are two
safety messages for broadcast or geocast mode, often used in main transmission modes in VANET routing: the first one is
single hop, and sent by fixed access points. We can cite called unicast in which data packets are sent to one
Freeway Management, Crash Prevention and Safety, Road destination. The second mode, known as multicast, aims to
Weather Management as examples of ITS safety applications transmit data packets to multiple destinations at once.
offered by fixed infrastructure vehicular networks [6]. The
Over the last couple of years, several routing protocols [13],
second type of ITS safety applications is provided by network
[14], [15], [16], [17] have been proposed in the literature to
vehicles which transmit safety messages to other vehicles over
support road safety, multimedia, and comfort applications in
multiple hops using a unicast mode. Collision Notification,
VANETs. These routing protocols were designed to discover
Collision Avoidance, Driver Assistance are examples of ITS
and maintain routes in order to disseminate information
safety applications provided by vehicles which form a
between any two VANET nodes (vehicles and/or RSUs) where
VANET and require routing solutions [6].
the sender is the source node, and the receiver the destination
ITS safety applications for VANET rely on the aggregation node. VANET routing protocols can be broadly classified into
of VANET information using safety messages which are sent two categories: topology-based routing protocols and
by each vehicle in the vehicular environment. These safety geography-based ones.
messages carry information such as the state of the other
Topology-based routing protocols determine routes based on
vehicles including their location, speed, acceleration, brake
topological links between nodes along the source-destination
state, etc.
path [13], [14]. Common characteristics shared among
The timely delivery and processing of such vehicular topology-based routing protocols include the accuracy of the
information can help vehicle drivers avoid accidents or to found destination, the reduced length of the discovered paths,
react appropriately [7]. Following an emergency brake or absence of loops, and a low rate of dropped messages.
weather conditions, traffic accidents or collisions can also be However, they also suffer from various drawbacks such as the
prevented by receiving emergency or warning messages instability of the found routes, high end-to-end delays and
generated and sent by VANET nodes; either particular routing overheads.
vehicles (such as police cars, ambulances, etc.) or by RSUs at
In contrast, geography-based protocols route data packets
the appropriate time, through a unicast and multi-hop routing.
without any knowledge of the underlying network topology.
We note here that the infrastructure-based vehicular networks
Geography-based protocols make use of the geographic
(i.e. V-WLAN or VCN) are not well-suited for certain direct
positions of the source and destination nodes to discover
vehicle-to-vehicle or vehicle-to-infrastructure communications
routes and to make packet forwarding decisions [15], [16].
due to limited presence of access point or cellular gateway in
This category of protocols is known for its robustness of the
the whole road traffic environment [8].
routes found as well as the reduced number of dropped
To ensure the timely reception of these kinds of safety messages during congestion. However, geography-based
messages, fast, reliable and efficient dissemination of protocols often incur considerable end-to-end delays. They

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

The evolutionary algorithms category consists of

computational techniques inspired by natural evolution such as
inheritance, mutation, selection and crossover which aim to
find the optimal route between the transmitter and the receiver.
In this category, two sub-classes can be cited; Sequential
Genetic Algorithms which aim to evaluate only one objective
such as the Intersection-based Geographical Routing Protocol
(IGRP) [31] which tries to maximize the probability of
connection from vehicles to the Internet. The second sub-class is
the Parallel Genetic Algorithms which have as a goal to
Figure 1. Taxonomy of Bio-inspired Routing Algorithms for VANET optimize multi-objective transmission criteria like the power
also use non-real-time information (e.g., using static roadmap consumption of wireless communication devices of vehicles, and
the time required to perform the automatic configuration of the
data) about the positions of vehicles when establishing routes
Optimized Link State Routing (OLSR) protocol applied to
which often leads to inaccurate destination node found.
VANETs [32].
All the routing protocols discussed above belong to the
The second category consists of all VANET routing
classical computational field where pure theoretical
algorithms inspired by swarm intelligence such as Ants, Bees,
(mathematical) ideas (such as FIFO paradigm or Hamiltonian
Birds, etc. Three sub-classes can be cited in this category; Ant
systems) were considered to obtain provably optimal routing
Colony Optimization (ACO), Particle Swarm Optimization
solutions. However, such solutions are often inadequate for
(PSO), and Bees Colony Optimization algorithms (BCO).
large-scale vehicular networks because of their high
Concerning ACO, several VANET routing protocols have been
complexity and low performance [18], [19], [20], [21].
inspired by the behavior of ants when they search the food
Recently, a new area, called bio-inspired computing, has
source, and then they communicate this discovery by a
emerged which looks at biology as a source of inspiration, and
chemical substance called pheromone. For example, [33] was
mimics the laws and dynamics of natural species [15]. Bio-
proposed as a VANET routing protocol known as Mobility-
inspired computing has received a lot of interest lately because
aware Ant colony optimization Routing DYMO (MAR-DYMO),
of its high performance results and outstanding capabilities to which used the ACO principle to find optimal routes between
solve theoretical and experimental problems, as well as its source node and the destination, with predictions on the mobility
high suitability for various applications [22], [23], [24], [25], of vehicles in terms of position and speed.
[26].
The way birds flock has been an inspiration source of PSO-
One of the fields where bio-inspired techniques have been based VANET routing protocols, like parallel Particle Swarm
widely applied is computer networking. In particular, Optimization (pPSO) proposed in [34]. pPSO was conceived to
biologically inspired methods have been used to address the improve the QoS parameters of the Ad hoc On-demand Distance
routing problem in order to ensure the timely and reliable Vector routing (AODV) protocol [35] such as packet delivery
transmission of safety messages for VANETs in an optimal ratio, normalized routing load, and end-to-end delays.
fashion.
In the swarm optimization category, some BCO-based VANET
There are several advantages of bio-inspired approaches for routing protocols have considered the most important behaviors
data packet routing in VANETs. Bio-inspired methods are in Bees’ life such as food source searching and marriage, in order
more efficient for large-scale vehicular networks due to the to design these approaches. We can cite Hybrid Bee swarm
similarity between the manner of finding VANET routes and Routing (HyBR) [36] as a unicast routing protocol proposed for
species behavior to satisfy their natural needs like the VANETs. HyBR applied a topology-based routing inspired by the
discovery of food source paths for Ants [25] or Bees [27]. bees’ communication when the network is dense, whereas a
geography-based scheme inspired by bees’ marriage is performed
Another strong advantage of bio-inspired methods is the low
when the network density is low.
complexity to perform computational problems [28].
Adaptability and self-organization are considered as promising Finally, we survey some bio-inspired algorithms which were
aspects provided by the bio-inspired methods to cope with inspired by other biological phenomena such as human behavior
frequent changes of VANET topology, and adjustment of used to propose the Fuzzy-Assisted Social-based routing (FAST)
routes can be achieved without or with minimal human protocol [37] which leverages a friendship mechanism to make
intervention [29]. Moreover, robustness could be improved by secure routing decisions from the source to the destination of real-
bio-inspired networking approaches to maintain an acceptable time vehicular traffic. Moreover, the natural immune system
routing performance despite network disruptions [30]. was considered as an idea to conceive a routing protocol called
Datataxis [38]. This VANET routing idea was proposed to
In this survey, we present a taxonomy of bio-inspired effectively cover a vehicular urban sensing environment, and to
routing algorithms dedicated for VANETs. We classify bio- diffuse the collected data which can be further used to improve
inspired VANET routing algorithms into three main categories: vehicular traffic.
Evolutionary algorithms, Swarm intelligence algorithms, and
other biologically inspired algorithms, as shown in figure 1. It is worth noting that each category consists of two types of

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

protocols: it can contain either mono or multi-objective routing packet with certain constraints such as a reduced end-to-end
processes that try to find best route with one or multiple criteria delay and few dropped packets. On the other hand, due to the
such as end-to-end delay, hop count, transmission cost or high-speed mobility of vehicles, there are frequent changes in
others. The second type is the unicast or multicast transmission the network topology, and selected paths could rapidly change
mode which can differentiate the different VANET routing and then, cause a delayed transfer of data, as well possible data
protocols. packet losses. This problem has prompted researchers to find
robust routing protocols that discover efficient and long
Bio-inspired approaches have showed their strengths and
effectiveness in various application scenarios. They have a strong
lifetime paths.
potential to address the challenges associated with VANET B. Mobility models used for VANET routing
routing such as vehicular network scalability, self-organized
To test and validate research studies in VANET routing, it
control of VANET’s dynamics, complexity of messages
is hard to conduct real experiments in vehicular areas due to
exchanges, and VANET robustness as will be highlighted below.
high risks, dangers and elevated costs of such tests. Therefore,
Based on the Reinforcement Learning (RL) paradigm, we to study VANET routing, modeling and simulating of
suggest a unified formal model of the bio-inspired multi- vehicular network with a high level of accuracy and reality are
modular approaches applied to VANET routing. This model necessary. To address these challenges, several research
aims at finding multiple solutions represented by a global activities have been proposed to simulate realistic VANET
objective function. The purpose of this contribution is to deal mobility models. We classify VANET mobility models
with VANET routing problems such as routes discovery and proposed in the VANET literature into four categories based
maintaining, parameter tuning, routing optimization etc., on the type of vehicular movement: random-based mobility
considered as a part of some global function to be built. This model, geographic Map-based mobility model, group-based
model performs an iterative process which is sufficient for mobility model, and prediction-based mobility model.
routing, especially when we have no a priori knowledge about
the global function to be built. Moreover, RL is more suitable 1) Random-based Mobility Models
for bio-inspired methods compared to traditional exact In this category, the Movement (location, velocity and
methods, where RL is focused on on-line performance, acceleration) of vehicles changes randomly over time. More
without any sub-optimal actions explicitly introduced. specifically, a vehicle selects a random destination and a speed
The rest of the paper is organized as follows. Section II between 0 and a maximum speed, and then it pauses for a
highlights a background of VANET routing and existing fixed number of seconds. Recently proposed random-based
standards. The basic bio-inspired concepts and operations mobility models include Random Waypoint Mobility Model
applied to VANET routing are explained in section III. The [39], and Random Walk Mobility Model [40]. Despite its
motivations for using bio-inspired techniques in VANET simplicity, random-based mobility models cannot be applied
routing are presented in section IV. In sections V, VI and VII, to certain vehicular scenarios with geographic restrictions and
we present each set of routing algorithms based on our obstacles such as buildings, streets, curves, etc. Moreover,
classification; Evolutionary algorithms for VANET routing are vehicles move in random fashion and do not displace with an
reviewed in section V, swarm intelligence optimization for increased or decreased velocity over the time as with the real
VANET routing is presented in section VI and the section VII movement of vehicles.
presents other bio-inspired routing studies. Based on all these
different approaches, we propose in section VIII a unified 2) Geographic Map-based Mobility Models
formal model for bio-inspired multi-modular approaches To overcome the limitations of random-based mobility
applied to VANET routing. Section IX presents some future models concerning geographic restrictions presented in
trends. Finally, section X concludes the paper. vehicular environment, several mobility models have been
proposed aiming to allow vehicles to move over constrained
II. VANET ROUTING BACKGROUND AND EXISTING and limited urban or rural maps. Vehicular maps consist of a
STANDARDS set of roads, streets, curves, obstacles, buildings, etc. which
This section presents a background on VANET routing and restrict vehicles displacements. Various models have been
VANET mobility models as well existing standards ratified for introduced, namely Manhattan Grid Mobility Model [41], City
VANET routing. Section Mobility Model [42], and Freeway mobility model
[43]. These types of models imply geographic restrictions as
A. Definition of VANET routing in reality and is very simple simulate. However some traffic
Routing is considered as one of the most important processes parameters such as network density, differentiating between
in VANET research that allows vehicular applications to vehicular scenarios in rush hours and off-pick hours may be
provide their digital services. Routing is the process of considered or predicted to improve these types of models.
selecting the best paths between the sender (i.e. source node)
and the receiver (i.e. designation node) through a set of
VANET nodes. To improve vehicular safety and to satisfy the
needs of traffic users, VANET routing should forward a data
4

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

3) Group-based Mobility Models For DSRC, 75 MHz of spectrum at 5.9 GHz was allocated
to be used by DSRC operations in an approximate range of
Group-based mobility models focus on using the
1000 meters [48]. The DSRC spectrum band is organized into
collaboration among network nodes which have the same goal
7 channels of 10 MHz wide each, allowing parallel
such as traveling for the same destination. Therefore, the
transmissions of different applications. These channels are
network is divided into several groups, each with its own
reserved for both public and private data communications,
mobility behavior. Group-based mobility models include:
with higher priority given to public communications such as
Reference Point Group Mobility Model [44], Virtual Track
safety messages. One channel is restricted for safety
Mobility Model [45]. Despite the expected good outcomes
communications only (channel 178 called Control CHannel
(such as the reduced routing overhead) of this category, there
CCH), while the two channels at the ends of the spectrum
are some limitations with group-based models such as nodes
band are reserved for special purposes, namely critical safety
in the same geographic group do not always move in the same
of life and high power public safety. Moreover, all the
direction. Also, this category does not consider the network
remaining channels are Service CHannels (SCH) which can be
density.
used for either safety or non-safety applications.
4) Prediction-based Mobility Models Concerning energy, DSRC standard limited the peak
In prediction-based mobility models, the vehicle keeps transmit output power over the frequency band of operations
information about its previous location and velocity which to no more than 750 mW (milliWatt) (28.8 dBm decibel
will be correlated in order to predict its future movement. milliWatt), and the maximum Effective Isotropically Radiated
Hence, the changes in speed and direction of vehicles are Power (EIRP) to no more than 30 W (44.8 dBm), as the
smooth and realistic. Moreover, network density may be amount of power emitted by a DSRC isotropic antenna. DSRC
determined using prediction rules according to the vehicular used Orthogonal Frequency Division Multiplexing (OFDM) as
traffic flow. Examples in this category include the Gauss- the method of encoding digital data (digital modulation).
Markov mobility model [46], and the Markov-History based Furthermore, since transmissions are performed via the
Mobility Model [47]. Despite the practical realistic aspect of wireless interface, DSRC works in high vehicle speed mobility
prediction-based mobility models, when modeling mobile conditions (up to 120 mph (miles per hour)), and in extreme
nodes movement and network density, these models hide some weather conditions (e.g. rain, fog, snow, etc.) with a short time
VANET topology details such as obstacles, traffic lights, etc. delay.
which cannot be omitted. 2) Standards for Wireless Access in Vehicular Environments
(WAVE)
C. VANET Routing Standards
A VANET routing standard is a normative specification of As a part of IEEE 1609 series of Standards [49], IEEE
a wider array of technologies applicable to routing in vehicular Standard for Wireless Access in Vehicular Environments
networks. The standard allows telecommunication and (WAVE) was proposed to identify wireless access for
automotive industry operators to produce routing devices with vehicular wireless and mobile networks. WAVE was specified
uniform features in order to be used with certain portability. as a set of rules aiming to ensure safety, health, or
Also, standards are designed to support a variety of environmental protection, or ensure against interference with
applications based on vehicular communication which or from other devices or networks.
guarantee interconnectivity and interoperability among Initially, the physical layer and the Medium Access
different devices. This subsection presents several important Control of vehicular wireless connections were managed by
standards proposed for VANET routing. existing IEEE 802.11, 802.11a and 802.11p standards
respectively proposed for wireless networks (Wi-Fi).
1) Dedicated Short Range Communication (DSRC) Therefore, data is modulated using Orthogonal Frequency
Proposed initially by the Federal Communications Division Multiplexing (OFDM) and forwarded with data rates
Commission (FCC) of the US Department Of Transportation of up to 54 MBits/s within a radio frequency band of 5 GHz
(DOT) in 1999, Dedicated Short Range Communications range giving rise to transmission range of 100 meters. In
(DSRC) is a two-way, short-to-medium-range wireless WAVE, network upper layers are supported by IEEE 1609
communications capability that permits very high data standards. IEEE 1609.4 resides above the IEEE 802.11p and is
transmission critical in communications-based active safety considered as a MAC extension standard applied when DSRC
and non-safety applications in vehicular wireless networks. performs in a multi-channel environment. It helps a device
DSRC was developed to support vehicle-to-vehicle and with several radios to effectively switch among those
vehicle-to-roadside communications, where vehicles and channels, as well as to remain tuned to a single channel all the
RoadSide Units (RSUs) are assumed to be equipped with time [50].
OnBoard Units (OBUs) and wireless transmission devices,
respectively. The network services in the network and transport layers
are supported by IEEE 1609.3 standard. This standard defined
a new network layer protocol known as WAVE Short Message

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

Protocol (WSMP) aimed to avoid the packet overhead the number of iterations must not exceed a maximum
associated with Internet Protocol (IP) using WAVE Short number to deal with the non- stagnation state.
Messages (WSMs) and WAVE Service Advertisements
2) Swarm Intelligence Optimization (SIO)
(WSAs). The security protocols are defined in IEEE 1609.2 by
the application of standard mechanisms for authenticating and Considered as bio-inspired algorithms, SIO includes
encrypting messages, especially WSMs and WSAs. computational techniques that are used to solve optimization
Concerning authentication, the sender’s authority to transmit problems by imitating the collective behavior of ant, bird and
messages is verified through a digital signature based on the fish or other biological species which leads to the emergence
Elliptic Curve Digital Signature Algorithm (ECDSA) [51]. of intelligent global behavior of these species. The result is a
self-organized behavior with simple rules used by the species
III. OVERVIEW OF BASIC BIO-INSPIRED CONCEPTS USED FOR used to fulfill their needs (i.e., food searching, reproduction,
VANET ROUTING defense, etc.) without any centralized administrative authority
We briefly present the basic concepts and operations often in the entire system [53].
used by different bio-inspired approaches, either for routing B. Basic Bio-inspired Algorithms
protocols or for optimization algorithms.
In this section, the main bio-inspired algorithms applied to
A. Fundamental Concepts and Operations the computational area and their principles are described
1) Basic Parameters of an Optimization Problem: (figure 2).
 Optimization problem definition: in the computational 1) Genetic Algorithm (GA)
area, the optimization problem is defined as the selection
of a best solution (optimal solution) among others with Genetic algorithm is a bio-inspired algorithm proposed in
respect to some criteria [52]. 1970s [54]. The algorithm imitates the process of natural
 Search Space (SSp): this includes all feasible solutions to evolution. GA is an optimization approach based on a
solve an optimization problem. SSp can be represented by population of individuals (called chromosomes) which
a graph in which each feasible solution can be marked by generate another population during an iteration. The initial
its value for the problem. population is set by using a random initialization. Afterward,
 Objective function (Fitness function): is a computational all chromosomes (parents) are probably selected two by two in
function used to evaluate SSp solutions to find some order to participate in a crossover operation, followed by a
extreme (minimum or maximum) solution by considering mutation operation. Both of these operations result in a set of
the optimal one based on one or multiple objectives. chromosomes (offsprings) which are compared with the
parents so that the best chromosomes can be used for the next
 Population and individual encoding: a population is a set
generation. The processes of selection, crossover, mutation,
of individuals where each individual represents a SSp
and generation are rerun until a stopping criterion is satisfied.
solution. Each individual is encoded using a data structure
We briefly describe the selection, crossover, and mutation
based on Boolean, Strings, Trees etc. Some individual
operations below.
encoding types include: Binary encoding which consists of
a set of binary variables; Discrete encoding is based on a a) Selection Strategy
discrete vector of integer values; natural encoding in The selection strategy determines the chromosomes which
which an individual is represented by a vector of integer will be chosen to be the parents of the offsprings of the next
values. Encoding plays a major role in the efficiency and generation. Parents are selected according to their fitness using
effectiveness of the resolution algorithm which helps to various strategies. For example, Roulette Wheel Selection
reduce the computational complexity [52]. (RWS) [55] is based on the assignment of a selection
 Initialization: is the way of assigning a set of initial values probability to each chromosome that is proportional to its
from SSp to form the initial population. There are several fitness. These probabilities are represented on a roulette-wheel
ways to choose first solutions (initial population). For
example, the random initialization in which the first SSp
individuals are selected at random. With uniform
initialization, SSp individuals are equally likely to be
chosen (every individual of n individuals of SSp has equal
probability “1/n” to be selected).
 Stopping criterion: is defined as a measured value which
controls the number of times of execution of the resolution
algorithm. This criterion can be static in which a fixed
number of iterations is set before starting the resolution
algorithm. The stopping criterion can also be dynamic
when the iterations are carried out and stopped only if the
fitness function does not improve after a certain number of
times (also known as the stagnation state). We note that
Figure 2. Main Bio-inspired Algorithms
6

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

that has a circular shape and contains one indicator. Once the 3) Particle Swarm Optimization (PSO)
roulette-wheel is wheeled; it will stop with the indicator
The PSO algorithm [26] is similar to the way birds flock in
showing one of the represented probabilities and its
real life. Using swarm optimization, PSO considers a particle
corresponding chromosome is selected.
defined by its position (i.e., quality of solution) and its
b) Crossover Operation velocity (i.e., its ability to move toward another solution
Crossover is a binary operation that is applied to two which may be better). This algorithm ensures that each
chromosomes in order to generate two new chromosomes. particle chooses among the best previously encountered
Crossover simulates the reproduction of individuals, where the positions or moves toward a new position using the particle’s
representation of each chromosome is partitioned and gives velocity if this new one is the best. PSO performs these steps
rise to new chromosomes each of which consists of one part of until it reaches a stopping criterion.
one parent and another part of the second parent. Crossover 4) Bees Life Algorithm (BLA)
operation aims to generate new diversified solutions in SSp. It
Several computational problems have been solved using
is the choice of far solutions (offsprings) in SSp from the point
new approaches inspired by Bees’ behaviors. The general
of view of fitness compared to the original solutions (parents).
approach is called Bees Life Algorithm in which the two
There are several crossover strategies such as one-point
major behaviors of bees have been used: food foraging and
crossover, where one cut-off point is randomly chosen in the
reproduction.
representation of both chromosomes to be divided into two
parts. The first part of the first parent is added to the last part Food foraging behavior is observed when bees search new
of the second one to form the first offspring chromosome, and nest sites or during the food source foraging. To do this, some
vice versa. Moreover, a two-point crossover has also been bees (called scouts) navigate and explore the region to find a
proposed where two cut-off points are applied [56]. food source. If found, they come at the dance floor in the
c) Mutation Operation
beehive to share this discovery with their nest mates via the
language of dance which can be round or waggle related to the
Mutation is a unary operation that is applied to one distance of discovery. Some bees (called foragers) are
chromosome generated after crossover operation. Mutation recruited to exploit this discovery.
aims to perform a small change in the representation of a
chromosome to reach a new solution near to the original one Reproductive behavior is guaranteed by the queen which
in SSp from the point of view of fitness. There are various mates with several drones in a mating-flight. After three days,
mutation strategies and one popular one is the one-point the queen lays its eggs. The unfertilized egg will give rise to a
mutation [57], where one cut-off point is randomly chosen in drone, while, the fertilized egg gives rise to worker or queen
the representation of the offspring to be changed with a depending on the food quality given to the larvae [60].
random value. This strategy generates a new chromosome 5) Fuzzy Logic of Human Brain (FL)
which is near to the first one in the SSp. Another mutation
strategy called Uniform Mutation, in which a random value is Fuzzy logic (also called probabilistic logic) corresponds to
selected between the user-specified upper and lower bounds approximate reasoning rather than fixed and exact. FL [61] is
which replaces a value of the chosen cut-off point of the inspired from the human brain which imitates the
chromosome [58]. interpretation of uncertain sensory information [62]. Fuzzy
logic is based on the use of a set of variables which may have
2) Ant Colony Optimization (ACO) a truth value that ranges in degree between 0 and 1, against
ACO is a SIO algorithm which is inspired by the food source binary logic where variables may take only 0 or 1 values. The
searching behavior of ants and was proposed in 1992 [59]. In partial truth may represent more realistic systems, especially
their natural environment, ants exploit their areas randomly: the continuous systems which require an interval of values to
once they have found food, ants return to the colony while be explained.
laying down pheromone (a chemical substance) trails. This 6) Escherichia Coli Bacteria Behavior
path is likely to be used by other ants which follow the trail
with more pheromone being laid down and the quality of the An Escherichia coli (E. coli) bacterium is an ideal organism
path is reinforced against other paths that are less visited. for chemotaxis modeling because of the rich experimental
Nevertheless, over time, the pheromone trail starts to information collected over years of extensive research. Similar
evaporate thereby reducing its attractive strength. to many other bacteria, Escherichia coli bacteria are
Consequently, a short path, by comparison, gets marched over considered as an illness cause, affecting species immune
more frequently, and thus the pheromone density becomes system. These bacteria can move from a set of rigid flagella
higher on shorter paths than longer ones. The idea of ACO is that enables the bacteria to swim in regular or irregular
similar to this behavior by representing SSp solutions with directions. More specifically, two types of movements are
paths traveled by ants. performed. The first type is performed when the flagella turns
clockwise, then the bacteria tumble and do not move in any
particular direction (random movement in any direction). In
contrast, the second type is accomplished when the flagella
7

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

turn counter-clockwise causing the bacteria to swim in a To address the data acquisition problem in WSNs, the authors
specific direction. The sensors of an E. coli are the receptor of [67] proposed the use of mobile sinks (particular nodes that
proteins that are stimulated by external substances. Based on are responsible to receive and process the network data to help
the level of nutrients (or attractants) a bacterium moves in make decisions relevant to the network context: forest, farm
different search modes. etc.). In [67], the authors proposed a routing protocol called
SIMPLE inspired by Swarm Intelligence, to maximize the
When the bacteria are placed in a substance with a nutrient
network lifetime. Since SIMPLE is based on swarm
with homogenous concentration, they exhibit a search
intelligence, it does not require any global information of the
behavior but with increased run length of swimming and
entire network to improve the network’s lifetime. A cooperative
decreased tumble time. However, when the bacteria detect a behavior between nodes (also known as collective behavior) is
change in the concentration level of the nutrition, they swim planned to achieve global optimal performance. SIMPLE was
along the gradient of concentration towards the most nutrition shown to yield better scalability and robustness against node
rich area, and spend less time tumbling [38]. This behavior has failures when compared with Ad hoc On-demand Distance
been applied to solve several computational problems by Vector routing (AODV) [35], Dynamic Source Routing (DSR)
representing SSp solutions with the bacteria area where the [68], and a traditional routing algorithm called Max-min
bacteria change place from one solution to another one [63]. Remaining Energy Protocol (MREP) [69].

IV. MOTIVATIONS FOR USING BIO-INSPIRED APPROACHES IN B. Computational Complexity

VANET ROUTING Another strong motivation for applying bio-inspired
methods to solve computer networking problems is because of
Bio-inspired networking approaches make use of computa-
the computational complexity associated with networks in
tional methods to solve networking problems. The basic
terms of performance costs (run time, and/or number of
principle used by these approaches is the imitation of natural
resources used to solve some specific problem). Traditional
behaviors of living creatures (such as humans, insects, animals)
networking solutions that work reasonably well for small-scale
when they try to find solutions to their natural needs (such as
networks suffer from high complexity for large-scale networks
searching for food, reproduction, defense, traveling etc.).
[18]. However, bio-inspired approaches yield optimal
Over the last decade, we have witnessed the emergence of solutions with low complexity [28]. In [26], the authors used a
bio-inspired techniques that have been deployed to solve biologically derived algorithm Particle Swarm Optimization
various networking problems in areas such as routing, (PSO) to optimize multicast routing for MANETs. PSO was
congestion control, security, etc. The main motivation behind successfully applied in [70] to solve the Multicast Routing
the deployment of bio-inspired networking techniques stems Problem (MPR) in which the best multicast routes are
from the strong similarity between communication scenarios required. The results obtained in [70] demonstrate reduced
in networking and the natural communication of species. Next, complexity in terms of shorter times of finding the best
we review some of the most relevant computer networking multicast routes compared to GA algorithm.
characteristics that make them particularly suitable for bio-
C. Adaptability and Self-Organization
inspired solutions.
With the significant development of Next Generation
A. Network Scalability Networks (NGNs) (i.e., MANETs, WSNs, VANETs, etc.),
Scalability is the ability of a network to efficiently handle a additional network characteristics have emerged which did not
large number of nodes. Scalability is considered as one of the exist in previous traditional wired networks. Many of these
most challenges of networking research because if it is not characteristics (distributed and decentralized administration,
well supported, the performance of the network can quickly heterogeneity of nodes, frequent changes of topology
degrade [64]. Scalability is quite well supported by traditional information, seamless communications infrastructure etc. [71])
networking methods if the network density is low (i.e., small stem from node mobility which causes network nodes to
number of nodes). However these traditional methods are less frequently change their information states (e.g., routing tables,
efficient for large-scale networks [19], [20], [21]. In contrast, connectivity, sending packets etc.) to adapt to various
by their nature, bio-inspired solutions are scalable (best path networking conditions.
toward food with ants [25], best new nest-mates for bees [17],
Moreover, when parameters such as network bandwidth,
best organization of travels for birds [60] etc.) even if the
power consumption of routers and network nodes change,
number of individuals is very high (i.e., high density
NGN devices should allow self-configuration, self-operation,
networks). As a result, computer networking scientists have
self-organization, and self-optimization to ensure sustained
been motivated to explore new solutions for networking
performance without requiring any manual intervention from
problems using the same bio-inspired principles which
human operators to modify existing parameters. We note that
include: local cooperation between individuals (between
many current solutions of vehicular networks require human
neighbors), local communication to reach global information
intervention when a node of the network fails since the
about the state of the network [60]. Scalability, in terms of
network topology is formed by fixed access points or
large number of nodes in Wireless Sensor Networks (WSNs)
stationary gateways. To route packets for example, it is not
which collect data, remains a significant challenge [65], [66].
8

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

possible to use a network topology based only on fixed nodes

which can fail and lead to the use of wrong routes. In such
situations, we need routing solutions that are adaptive and
self-organization policies based on a dynamic topology [72].
Bio-inspired techniques enable the deployment of adaptive
and self-organizing solutions and maintain better up-to-date
network state information than traditional techniques [64]. Figure 3. Evolutionary Algorithms Sub-classes
Several popular bio-inspired methods have been recently
proposed to solve various network problems in an adaptive V. EVOLUTIONARY ALGORITHMS FOR VANET ROUTING
and self-organizing manner. For example, to address the In this section, evolutionary algorithms applied for
routing problem in MANETs, the bio-inspired AntHocNet VANET routing are presented. We can distinguish two sub-
algorithm [73], [29], based on Ants’ behavior when searching classes: Sequential Genetic Algorithms and Parallel Genetic
their food, was proposed. AntHocNet algorithm uses two Algorithms, as shown figure 3.
processes to collect routing information. One process is
similar to the exploration phase of searched areas in which A. Sequential Genetic Algorithms for VANET Routing
each ant runs through an end-to-end path (between the source The sequential genetic algorithm has been applied to
node and the destination node) repeatedly with the aim of VANET routing in some works aiming to evaluate one or
using the most traveled path. Pheromone diffusion is the more objectives as presented in the following sub-sections.
second process where ants update routing tables of visited
1) Mono-objective Sequential Genetic Algorithms
nodes with routing information which can be used later by
other nodes to find other paths, similar to the behavior of Ants. For instance, the Adaptive Message Routing (AMR)
AntHocNet exhibited a significant improvement in terms of protocol [83] uses genetic algorithm to search a QoS route
packet delivery ratio, routing overhead and average end-to-end with minimum end-to-end delay as the single objective. In this
delay over conventional, well-known routing protocols such as study, the authors required maintaining a threshold for the
AODV and Optimized Link-State Routing (OLSR) [32]. connectivity probability, as well as for the hop count
considered as two constraints. Here, VANET was modeled as
D. Network Robustness a grid consisting of a set of fixed cells that with RSU at its
Network robustness is the ability to maintain an acceptable center. While a vehicle moves in a particular cell, its location
performance despite network disruptions. Robustness ensures information is aggregated and reported to the appropriate
RSU. In addition to the intermediate and adjacent street
a resilient network against link/node failures or removal of
intersections, the aggregated information enables the RSU to
nodes or targeted attacks [74]. Recently, robustness has been
construct a more stable and optimal route for vehicles in terms
extensively addressed by many networking research studies of QoS requirements, as shown in figure 4. AMR starts when a
[75], [76] especially by using results inspired from biological vehicle enters a new cell where this vehicle sends a signaling
systems [77], [78]. Improved results with bio-inspired message to the cell’s RSU through its neighboring vehicles.
networking approaches stem from the collaborative reaction of When the RSU receives the signaling message it considers the
network nodes as in natural systems when faced with internal location information of this new vehicle to compute the
and external perturbations [30], [79]. For example, the human optimal route among those available in its routing database.
immune system consists of a set of elements responsible for
the recognition and defense against pathogens (i.e., viruses, The optimal route is calculated using the geographic
bacteria, parasites, etc.) to ensure the robustness of the human position’s information of each selected intersection as well as
the street IDentifier (ID) by taking into consideration the QoS
body’s health. The immune system’s behavior has been
requirements. More specifically, AMR uses a genetic
applied to intrusion detection in networks in [80] and showed
better results compared to traditional Discrete Tree Transform
(DTT) [81] and Singular Value Decomposition (SVD) [82]
methods (which do not use biological based approaches)
because of their inability to detect and recognize some
intrusions [80]. The above discussions illustrate the strong
potential and benefits of bio-inspired networking approaches
over traditional networking solutions.
Before starting the presentation of each set of the VANET
bio-inspired routing, let us remind that our proposal is based
on a taxonomy which is divided in 3 categories : Evolutionary,
Swarm Intelligence and all others. The following sections are
dedicated for each of them.
Figure 4. AMR Road Map Model

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

algorithm with end-to-end delay as objective as well as the iterations). The simulations have shown that the GA-based
connectivity probability and hop count as constraints. Each IGRP could find a near-optimal solution in terms of
route of successive intersections between the vehicle and the probability of connectivity compared with the optimal value. It
RSU (i.e. the solution) is represented by natural encodings is worth noting that this proposal is based on the intersection
(strings format), such as routes (A-C-E-F), (A-B-D-F), or (A- nodes which are more appropriate for urban scenarios and can
C-D-F). The initial population which contains a set of be limited in highways or rural environments.
individuals (routes) is generated at random.
In [84], a hybrid geographic and Delay Tolerant Networks
After, the RWS strategy is applied to select two parents (DTN) routing protocol was proposed. This unicast protocol
(routes) to perform the probabilistic one-point crossover. Here, used a genetic algorithm to enhance the DTN routing protocol
a common intersection is searched between parents, and then which may be limited whenever the network is sparse. This
the two parts of routes are swapped to form two offsprings. proposal starts with a greedy forwarding in which the source
The following step is the mutation in which a random node sends a packet to an immediate neighbor that is
individual is randomly chosen from the population using a geographically closer to the destination. If there is no neighbor
uniform mutation. After this step, a random intersection is closer to the destination, perimeter forwarding is then
picked and is replaced with the adjacent intersection. These considered in which each node forwards packets through the
steps are performed until a fixed maximum number of perimeter to its first neighbor counterclockwise from itself
generations is reached. We note that AMR copes with network (this principle is called the right hand rule).
dis-connectivity by allowing vehicles to store and carry
Often, greedy and perimeter forwarding processes are less
messages while moving until it can forward the messages to a
efficient than DTN for VANET where the network is sparse
next valid node.
leading to network partitioning in terms of node density.
Numerical and simulation results show that AMR Therefore, a connection strategy between partitions should be
improved end-to-end delay and connectivity probability when defined which tries to find the switching time from the
compared with GPCR [16]. AMR also exhibited an adaptive perimeter mode to the DTN mode and to select a proper node
behavior when the network density changes. However, the for storing packets and carrying them from one partition to
complexity of finding the optimal routes can be increased another. This study used a genetic algorithm to train the DTN
because AMR assumes that a threshold of GA generation node evaluation system to determine the optimal DTN node.
should be executed even if the optimal solution is found early In other words, the GA is applied to determine how effective
on. Moreover, another possible drawback with AMR is that each parameter is in choosing the DTN node. There are
the applied crossover and mutation can give rise to long routes several parameters considered for the switching decision such
in which duplicated intersections can occur. as the vehicle’s speed, the vehicle’s direction, the distance
between vehicle and destination, and the probability of loss of
Saleet et al. in [31] proposed a routing protocol called the
network connectivity. The GA is initialized by the
Intersection-based Geographical Routing Protocol (IGRP). It
construction of the first generation set of chromosomes
is mostly used for packet forwarding from vehicles to the
(individuals).
Internet through a gateway (through which a packet must
traverse) by an effective selection of road intersections. IGRP Each chromosome is initialized by random values of the
proposed to discover an optimal path between the source and abovementioned parameters. Some chromosomes are selected
the destination in a unicast way, which consists of successive using selection strategies such as Roulette-wheel selection
interconnected road intersections using a genetic algorithm. (RWS) [55], to generate some offsprings using one-point
These road intersections are selected in a way to maximize the crossover and a probabilistic mutation (i.e., a chromosome is
probability of connectivity among the intersections as the mutated with a probability equal to 0.01%). This process is
protocol objective. This work requires satisfying QoS rerun four times. After an experimental study, the authors
requirements such as tolerable delay, bandwidth, and error obtained an improvement in terms of average delivery delays,
rate. and average delivery ratio against the original GeoDTN+Nav
especially, for the fourth generation. This proposed approach
A natural encoding has been used to define each
can be evaluated for rural scenarios to study the case of sparse
intersection in the backbone route. Then, a chromosome
networks. In addition, the vehicle speed (40 km/h) used in the
(individual) expresses one route by its ordered intersections. A
performance evaluation is very low to cope with the loss of
set of chromosomes is randomly selected to generate the initial
network connectivity limitation in VANET.
population. More specifically, each chromosome begins with
the intersection adjacent to the source node and the following 2) Multi-objective Sequential Genetic Algorithms
intersections are chosen at random until the Internet gateway
xChangeMobile [85] is a multi-objective, sequential
is reached. The authors used the roulette wheel selection
genetic algorithm that uses unicast geography-based routing
strategy to select parents and one point crossover to generate
for VANETs. It is an inter-vehicular content exchange system
offsprings. Moreover, a uniform mutation operator is applied
that makes use of two routing protocols, VanetDFCN and
to ensure solutions’ diversity and avoid the stagnation state on
ChunkXChange which are enhanced by genetic algorithms
wrong solution (local optimum).
with the aim of addressing multimedia packets lost due to
This GA algorithm is rerun until a maximum number of obstacles along the road and interrupting the transmission
iterations is reached (in this work the authors used 30 caused by gaps in the multimedia messages. First, the

10

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

xChangeMobile system uses VanetDFCN as a broadcasting evaluating the resulting individual; then the algorithm decides
protocol which is an extension of the Delayed Flooding with whether the new offspring replaces the current one which then
Cumulative Neighborhood (DFCN) [86]. DFCN is a gets transferred (the replaced solution) to the external archive
mechanism for broadcasting packets from one node to its which is an additional population composed of less optimal
neighbors. DFCN has been extended to VanetDFCN by solutions found during the optimization process.
supporting additional parameters to decide if the received
The xChangeMobile system has been evaluated using the
packet needs to be forwarded or not in order to optimize the
Madhoc simulator [87] for a highway scenario. The MOCell
use of intermediate nodes. VanetDFCN requires the use of
algorithm was implemented in the jMetal framework [88]. The
GPS devices for each node (vehicle). VanetDFCN forwards
results obtained have shown that the two objectives (missing
the received packet only if the following three conditions are
chunks and bandwidth) are in conflict, and choosing a solution
satisfied: first, the packet should be received for the first time.
with better bandwidth leads to a decrease in the number of
Second, the received packet does not exceed a certain distance
missing chunks. In addition, the authors found out that around
called the Distance To Live (DTL) which indicates the
half of the obtained solutions have more than 50% of missing
maximum length of a path the message can travel and cannot
chunks in vehicles’ cache memories due to the short
be exceeded. Finally, a time estimate that the sender and the
simulation time (60 seconds). It is worth noting that this work
receiver are in the same range is used to compute the number
can be extended if it is further evaluated for both urban and
of grouped TCP packets called chunks, they could exchange.
rural scenarios.In [89], a Statistical Location-Assisted
It is worth noting that a chunk is a data container in which
Broadcast (SLAB) protocol was proposed as a multi-hop and
TCP packets are stored. A chunk consists of two parts: a
multicast protocol for VANETs. SLAB is based on machine
unique identifier placed in the header of chunk, and a chunk’s
learning techniques such as genetic algorithms and particle
content which is composed of one up to many TCP packets,
swarm optimization to automatically discover decision
where its size is fixed by the proposed xChangeMobile
threshold values, and then to make a decision to re-broadcast.
protocol. Consequently, if the number of chunks is higher than
The threshold values studied are: the period of HELLO
a given threshold, then the packet can be forwarded. Note that
beacons and the number of broadcast packets transmitted per
the time estimate between the sender and the receiver is based
receiving node to decrease routing overheads. These threshold
on their positions (latitude and altitude), speeds, and
values vary with node density, spatial distribution pattern, and
directions.
channel quality as multiple criteria to be evaluated. The
The second protocol used by the xChangeMobile system is authors of SLAB use the threshold values obtained for a
the ChunkXChange protocol which manages data gaps (i.e., broadcasting technique called distance-to-mean method [90],
missing chunks) which may be encountered in chunks. This in order to allow nodes to favor re-broadcasting when a small
protocol uses three functions: (i) making requests for chunks part of their transmission area is covered by other nodes.
that are missing in the local cache memory, (ii) answering to
The distance-to-mean method estimates the coverage by
possible chunk requests received from other nodes in the
network, and (iii) updating the cache memory when a reply for considering the distance from the node to the spatial mean of
a requested chunk is received. In the case of missing chunks the neighbors from whom the message has already been
(detected in the cache memory of the current node), the received. Therefore, the node refrains from retransmitting the
request is sent through VanetDFCN to fill the gaps. Therefore, message only if this distance is small because of the even
when a receiver receives this request and knows the missing distribution of neighbors around the sender node.
chunks, it replies by sending the available requested chunks in Evaluations of SLAB showed high reachability (i.e., the
its local memory through VanetDFCN. average fraction of nodes to receive broadcast messages) and
To optimize the performance of xChangeMobile the efficient bandwidth consumption (i.e., the fraction of re-
authors aim to minimize the number of remaining missing broadcasting nodes and the number of bytes sent per covered
chunks in all the vehicles and the bandwidth usage which is node) for both urban and highway scenarios with varying node
measured as the total number of messages sent in the network. densities. However, the use of HELLO beacons messages by
To achieve these objectives, a multi-objective algorithm based SLAB can flood the network.
on a Genetic Algorithm (MOCell) is used. This algorithm B. Parallel Genetic Algorithms for VANET Routing
starts with the initialization step by creating a population of N
individuals, where each individual represented a particular set Genetic algorithms were also applied for VANET but in a
of parameters of the proposed protocols (VanetDFCN and parallel manner and to evaluate only multi-objective routing.
ChunkXChange). These population individuals are evaluated Here, a parallel genetic algorithm was proposed in [91] to find
using the two previously mentioned objectives (missing the optimal parameter setting (automatic configuration) of the
chunks and bandwidth) through a Pareto front. unicast Optimized Link State Routing (OLSR) [32] protocol
when applied to VANETs. Thus, parallel genetic algorithm
After this evaluation process, crossover and mutation was applied to improve the efficiency of OLSR in VANETs
operations are successively applied to the populations’ by trying to reduce the power consumption of wireless
individuals until the stopping criterion is reached. In other communication devices of vehicles, and to minimize the time
words, for each population, the MOCell algorithm consists of required to perform the automatic configuration. OLSR is a
selecting two parents from its neighborhood, re-combining proactive routing protocol conceived for MANETs. To find
them in order to obtain an offspring, mutating it, and routes, and to cope with nodes’ mobility, OLSR uses an

11

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

TABLE I
EVOLUTIONARY ALGORITHMS FOR VANET ROUTING
(NOTATION. +: SUITABLE, -: LESS SUITABLE, M: MEDIUM, L: LOW, N/A: NOT AVAILABLE)
Evaluated objectives Unicast/ Performance characteristics
Sub- multicast Routing Mobility Packet
Algorithm Scalability Robustness Complexity Routing
category Type Criteria Transmission decider model Delay Delivery Bandwidth
mode overhead
ratio
[83] End-to-end Source
Unicast Urban + ND L + ND ND ND
AMR delay node
Probability
[31] Mono Source
of Unicast Urban + ND L + ND + ND
IGRP -obj node
connectivity
[84] Urban
Delivery Distributed
Sequential Hybrid Unicast & - ND L - ND ND ND
delay decision
Genetic DTN highway
Algorithms [85] Missing TCP
Source
xChange packets & Multicast Highway - ND L - ND ND +
node
Mobile Bandwidth
Multi Reachability
-obj (probability Urban
[89] Distributed
of Unicast & + ND L + ND - ND
SLAB decision
connectivity) highway
& bandwidth
Power
Parallel consumption
[91] Multi Distributed
Genetic & automatic Unicast Urban M ND L M ND + +
GAP -obj decision
Algorithm configuration
time

efficient periodic flooding of control messages by using energy-aware OLSR configuration found by the parallel GA
special nodes called Multi-Point Relays (MPRs). The main compared when the standard OLSR configuration is used.
message types used by an OLSR node include: a) periodic However, this proposed algorithm can suffer from many
HELLO messages to detect neighborhood connections; b) drawbacks because of limitations such as the long end-to-end
Topology Control (TC) messages generated by MPRs, and delays, instability of routes, and routing overheads (because of
broadcasted periodically with the purpose of providing each the proactive property of OLSR) of OLSR in vehicular
node in the network with sufficient link-state information to environment.
allow route calculation.
After a general analysis of GA-based routing protocols
Moreover, TC messages are used to compute next hop proposed for VANETs, we can conclude that this kind of
destinations for all nodes in the network using the shortest hop inspiration dealt with the routing problem with linear
forwarding paths. The authors of [91] have proposed to complexity which is the optimal one compared to complexity
modify OLSR in order to reduce the power consumption and of conventional routing protocols often characterized by a
increase the packet delivery ratio which can decrease due to high complexity of finding or maintaining paths used to
data exchanges of each OLSR node. The modified OLSR disseminate data packets in VANETs. We note that the
starts by encoding different configuration parameters of OLSR proposed GA-based VANET routing lacked the study of
which represent one chromosome of the GA. These connectivity robustness, as well as packet loss when
configuration parameters represent time fields used to decide forwarding packets. In addition, vehicular rural areas were not
if a message will be forwarded. These parameters are: considered to evaluate these proposals. These limitations can
HELLO_INTERVAL, REFRESH_INTERVAL, TC_INTERVAL, be investigated in future GA-based studies. Table 1
WILLINGNESS, NEIGHB_HOLD_TIME, TOP_HOLD_TIME, summarizes the main characteristics of recently proposed GA-
MID_HOLD_TIME, and DUP_HOLD_TIME (more detailed based routing protocols for VANETs.
explanations are given in [32]).
To initialize the first population, a uniform distribution is VI. SWARM INTELLIGENCE FOR VANET ROUTING
proposed to ensure the diversity of the individuals of this Life species have been considered as a source of
population. Next, the recombination (crossover) operation is inspiration for VANET routing. Therefore, three main
applied according to an operator called the classic arithmetic categories arise from swarm intelligence studies and they
recombination operator. In the classic arithmetic crossover, include: ant colonies, particle swarm systems, and bee
two chromosomes (routes) are selected which should possess colonies, as presented in figure 5. In this section, swarm
at least one common node, from which one is randomly intelligence algorithms for VANET routing will be reviewed.
selected [92]. Hence, the offsprings are generated by the
combination of the first part of the first parent with the second A. Ant Colony Optimization for VANET Routing
part of the second parent, and vice-versa.
The mutation operation is performed with the aim of
simultaneously changing the encoded parameters using the
OLSR power-aware problem specifications in order to ensure
that the new individual belongs to valid OLSR configurations.
This proposal has been simulated using a vehicular urban area
in which the GA iteration was rerun 500 times. The results
obtained show significant reductions in the power
consumption of the VANET nodes when using the automatic Figure 5. Swarm Intelligence Algorithms Sub-classes

12

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

Several studies inspired by Ant colonies have been process simulates a random selection of a route).
proposed for VANET routing, either to optimize a single
It is worth noting that each route’s pheromone quantity is
criterion or as multi-objective ant colony optimization. These
periodically evaporated following a mathematical model based
two classes are presented as follows.
on the current pheromone quantity and an evaporation rate
1) Mono-objective Ant Colony Optimization [33]. Moreover, the authors of MAR-DYMO modified
HELLO messages used to ensure neighborhood connections.
One of the first studies of a single objective, unicast, and
The modification led to aperiodic messages in contrast to
multi-path routing protocol for VANETs was proposed in
native DYMO HELLO messages which were periodic. The
[33]. This protocol is the integration of ACO [93] with the
HELLO messages in MAR-DYMO are used to ensure the
Dynamic Manet On-demand (DYMO) routing protocol [94],
knowledge of local connections between a vehicle and its
[95]. This proposal is called Mobility-aware Ant colony
neighbors and are managed by a Kinetic Graph framework
optimization Routing DYMO (MAR-DYMO). DYMO is a
[97] that allows a vehicle to make predictions about their
reactive, multi-hop routing protocol and is the successor of
neighbors such as their position at a given time instant and the
AODV [96] that is used to discover routes using a simple
duration they will remain as neighbors, for instance.
principle based on sending a Route Request (RReq) packet to
all neighbors, which in turn resend it to their neighbors and so MAR-DYMO was simulated using the Network Simulator
on until the destination node is reached or node that knows the ns-2 [98] and was compared with AODV, DYMO, and Ant-
destination is encountered. Once found, a Route Reply (RRep) DYMO [99] using the following metrics: average delivery
packet is sent to the source through the reverse route. ratio, average end-to-end delay and routing overhead. The
results obtained have shown better performance against
MAR-DYMO uses the ACO principle along with
DYMO and Ant-DYMO. MAR-DYMO only showed minor
predictions on the mobility of vehicles in terms of position and
improvements in the average packet delivery ratio at the
speed aiming to find routing paths with an optimal lifetime, as
expense of increased routing overheads. This work may be
a single objective. To achieve this goal, MAR-DYMO updates
extended to include other VANET scenarios such as rural
the routing table of each node with new information:
areas and other MAC protocol such as IEEE 802.11p suited
pheromone. First, the pheromone deposited into each routing
for VANET.
table entry corresponds to the quality of each route between
source and destination. More specifically, pheromone is added Another ACO based routing approach for VANETs was
to a route entry when the RReq (called Forward Ant) traveled proposed in [100]. It is a geography-based unicast routing
along this route. More pheromone is deposited if RRep (called protocol called Trust dependent Ant Colony Routing (TACR).
Backward Ant) crossed the same route. Figure 6 shows the In this work, the authors introduced an algorithm that creates
pheromone table of the node i after receiving and ant from the VANET clusters by considering direction, position, and
node j. In the link (j-i), the pheromone level Δϕij is calculated relative speed of the vehicle to manage the scalability of the
using equation 1, as follows: network in terms of the number of vehicles. For each cluster, a
selection procedure is executed to choose the most appropriate
∆∅ (1) cluster head. In this procedure, priority is given to the roadside
unit to be selected as the head due to its fixed position and
Where PR is the probability of i receiving a message from high processing capabilities. Otherwise, the cluster vehicle
j, tlink is the route lifetime, and tMAX is a value defined as the that presents a high value of trust (calculated with the slowest
maximum route lifetime. speed in real time) is chosen as the cluster head. As a third
The pheromone, along with ACO, is then used to predict contribution, the authors have proposed an evolutionary-based
and select the best route that connects the source and the routing protocol using the Ant Colony technique.
destination based on the pheromone concentrations level of This position-based routing approach is based on trust for
multiple routes. The pheromone concentration on a route the simple VANET highway scenario. The routing process
indicates how good this path is in terms of the route quality starts with the source node whenever it wants to disseminate a
(i.e., estimated lifetime of the route as a single aimed message towards a destination node. To do this, the message is
criterion). Using this substance, a path is selected. If several sent to the cluster head of the destination following an inter-
paths between the same source and the same destination were cluster routing process if the source node and the cluster head
found, probabilistic Roulette-Wheel Selection (RWS) method are out of range of each other. The cluster head then transmits
as we described earlier in this work can be used. Hence, the the message to the destination using an intra-cluster routing
appropriate path is selected to disseminate packets (this process.
The cluster head starts the inter-cluster routing based on
the Ant Colony algorithm by generating (n-1) forward ants
which are sent to (n-1) network cluster heads through multi-
hop paths. When a cluster head receives an ant, it checks that
it never received this ant, and also checks if the destination
node belongs to its cluster. If yes, the cluster head kills the
forward ant and generates a backward ant which will be sent
Figure 6. Ant Walking and Deposited Pheromone in link i-j to the source node as feedback. If a duplicate ant is received or

13

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

there is no destination in the current cluster, the ant will be considered as the best one to be used to disseminate further
killed. With intra-cluster routing, the cluster head is data packets.
responsible for routing the received message (from the source
Simulation results have showed that this approach can lead
node) to the appropriate destination in a unicast manner using
its member list table. to a fast converged topology in terms of packet delivery ratio,
routing overhead, and end-to-end delay performance metrics
The TACR protocol’s simulation results were compared when compared to Ad hoc On-demand Distance Vector
with AODV and (MAR-DYMO) protocols using routing routing (AODV) [35] and Dynamic Source Routing (DSR)
overhead as the only performance metric. The results obtained [68] protocols. However, predefining the number of iterations
show that TACR presented a large improvement in terms of can lead either to no topology convergence if the iteration
routing overhead against AODV with six times lower number is small, or spending too much time unnecessarily to
overhead than AODV but against MAR-DYMO, TACR is discover the topology when the number of iterations is high.
only slightly better, especially when the number of vehicles
does not exceed 75 vehicles. It would be interesting to conduct Another proposed bio-inspired protocol, called Multicast
further performance tests on TACR for other performance with Ant Colony Optimization for VANETs (MAV-AODV)
metrics such as average end-to-end delay, packet delivery ratio for enhancing multicast routing in VANETs, based on
and the consumed bandwidth. Moreover, it would also be Multicast Ad-hoc On-demand Distance Vector (MAODV) [102]
interesting to investigate how TACR can be applied to urban was developed in [103]. MAV-AODV uses an ant colony
and rural scenarios. system to improve multicast routing. The main goal of MAV-
AODV is to find and to enhance multicast routes from the
2) Multi-objective Ant Colony Optimization source to multiple destinations. By making use of vehicular
Ant-based algorithms have also been deployed to mobility information, this proposal builds stable multicast
disseminate messages in VANETs trying to optimize multiple group trees (such as a tree that gathers all multicast group
criteria. members proposed initially by MAODV), and uses Ant
Colony Optimization (ACO) to increase the multicast tree’s
In [101], an Ant-based algorithm applied ant intelligence lifetime.
behaviors to find the near-optimal converged topology set up
by trusted mobile nodes and links in V2V communications. First, beacon messages between vehicles in the same range
This topology helps to achieve an efficient routing to achieve are communicated to allow a vehicle to predict the mobility of
QoS. A distributed approach was proposed in that each node other vehicles (position vector and velocity vector of the
collects information from the local neighboring nodes by vehicle), as well as the lifetime of each link [103]. Second, to
sending and receiving ant packets from each active node to discover a route from the source to the multicast group, a route
find the appropriate nodes to construct a routing path. Each request message called Ant-RREQ-J is broadcasted to find any
vehicle in the network could interact with other nodes by member node of target’s multicast group (similar to when an
sending and receiving ant packets as control packets to ant tries to find its food). Each Ant-RREQ-J sent stores the
discover the expected topology where each node is assumed to lower lifetime value of a link of the route estimated by beacon
be traversed at most once. messages. This lifetime field is updated each time a new node
is visited. In addition, the hop count to the destination node is
During the initialization step each link is initialized by a also stored in the Ant-RREQ-J message.
pheromone value to express the link quality based on the
available bandwidth information, delay information, capacity After finding a member of the multicast group, an Ant-
and cost information, and connectivity information, as RREP message is generated and is sent to the source node
multiple evaluated criteria. From one randomly selected node, through the reverse route in a unicast manner. Before
a set of ‘M’ ants are dispatched in the network, where each ant generating an Ant-RREP, the node examines the traversed
chooses its adjacent node in a probabilistic manner based on route quality by calculating the route pheromone using Ant-
the quality of adjacent links indicated by pheromone values. RREQ-J lifetime and hop count fields.
Upon receiving the ant from one node, the pheromone This link quality information represents the quality of the
values along the path are increased. When the link’s lifetime route found and is used by the source node to predict and
expires or no more packets go through, the pheromone on this make the best routing decision. Transporting this quality
link is reduced. Consequently, a more frequented link has information is similar to ants when they deposit pheromone
more chances of being selected by other ants and vice versa. If along the route back.
the destination node is reached by all M ants, one round of
iterations is considered as completed, then the ants will stop MAV-MAODV showed better performance than MAODV
searching and sleep there (i.e. there is no going back to the in terms of maximum end-to-end delay and packet delivery
source node). Next, another round of iterations is rerun when ratio. Nevertheless, MAV-MAODV incurred important
the pheromone information from the previous iteration is used, routing overheads when the transmission range was fixed to
and so on until a predefined number of searching iterations is 350m. Future work may focus on evaluation of MAV-
reached. Therefore, the route with maximum pheromone is MAODV using realistic mobility models and non-urban
scenarios as well.

14

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

B. Particle Swarm Optimization for VANET Routing

There are some research activities inspired by Particle
Swarm Optimization (PSO) which try to enhance network
connectivity and VANET routing process. We note that the
proposed PSO-based VANET routing protocols which are of
multi-objective type are explained below. For example, the
authors of [34] proposed a parallel Particle Swarm
Optimization algorithm (pPSO) to search the optimal
configuration of the VANET unicast routing protocol AODV.
pPSO is a variation of PSO in which different criteria in the
fitness function are calculated in parallel manner. Except
parallel GA-based algorithms cited in subsection V.B, we note
that pPSO is the only parallel bio-inspired algorithm Figure 7. pPSO optimization of AODV parameterization
developed for VANET routing. The goal of this work was to
congestion with the prediction of vehicle speed and network
improve multiple objectives of AODV’s QoS parameters such
congestion. However, particle swarm optimization is used to
as packet delivery ratio, normalized routing load, and end-to-
adjust the parameters of the membership functions in the
end delays.
proposed fuzzy logic systems. Specifically, the fuzzy speed
pPSO starts with the representation of solutions (particles) prediction mechanism estimates the vehicle’s speed for the
with different AODV parameters: HELLO INTERVAL, ACTIVE next time period based on the distance between two
ROUTE TIMEOUT, MY ROUTE TIMEOUT, NODE TRAVERSAL TIME, consecutive vehicles, the driver’s age, and the current speed of
MAX RREQ TIMEOUT, NET DIAMETER, ALLOWED HELLO LOSS,
the vehicle. Therefore, it is quite easy to determine whether
REQ RETRIES, TTL START, TTL INCREMENT, and TTL THRESHOLD
(more details can be found in [96]). Next, an initial population the vehicle is within the communication range of its neighbors
is generated using a uniform initialization to ensure the by computing the distances of the vehicle and its neighbors in
diversity of the initial particles (solutions) over different areas next time period.
of the search space. In case the vehicle’s position is found to be out of the
During each iteration of pPSO, all particles move in the communication range of its neighbors in the next time period,
search space by adjusting their velocities and positions (update the vehicle can initiate a backup route construction process (by
pi position & speed). The movement of each particle is piggybacking a link break warning message to its neighbors)
influenced by its current fitness value and its best known local to prevent link failure caused by the mobility of vehicles.
position. In other words, the particle moves to the best
The fuzzy congestion detection module is used by nodes to
position (new solution) using its velocity (local change) if its
fitness is improved. pPSO proposed a swarm evaluation of determine whether congestion might occur and this module is
multiple parallel instances (i.e. particles) using multiple ns-2 similar to the fuzzy speed prediction module. In particular, the
simulators installed on several computing units. As result, the inputs are the packet queue length, the expected count of the
required time to obtain the QoS metrics will be dramatically packets travelling through the vehicles and its neighbors, and
reduced. In this way, the best particle (called the leader) can the expected number of the vehicles during the next time
be reached iteratively. This process is repeated until a stopping period. The output of the congestion detection module is the
criterion is satisfied. Figure 7 outlines the process of pPSO congestion indicator. If congested, the node sends piggyback
optimization. congestion information to its neighbors to indicate the
occurrence of congestion. Once the message is received by its
pPSO-based AODV, PSO-based AODV and the original
downstream neighbor, the downstream node re-initiates the
AODV configurations were compared. The results confirm
route discovery process to construct a new route to the
that pPSO-based AODV computed configuration outperforms
the others (PSO-based AODV and original AODV) for destination. In this study, the authors proposed that each
performance metrics such as packet delivery ratio and end-to- vehicle is able to execute its individual PSO algorithm in
end delay. However, pPSO configuration incurred high ordered to adjust the two parameters, mean and variance of the
routing overheads. The complexity of pPSO configuration can inputs in the membership functions for the fuzzy speed
be further investigated especially for large-scale VANETs prediction module and fuzzy congestion detection module.
which can increase the end-to-end delay due to the additional The experimental results obtained illustrate the effective-
computation time. ness and feasibility of the proposed schemes compared to
In PSO-based VANET routing, we can find [104] in which some traditional routing protocols such as AODV and
the authors proposed a novel mechanism for VANET routing MAODV in terms of packet delivery ratio and end-to-end
that offers alternative unicast routing paths whenever a broken delay especially for high speed vehicles. The fuzzy-based
link or a congestion event occurs along the routing path. To approach can be further evaluated and improved in order to
achieve this, two machine learning techniques are proposed: a reduce the routing overhead for urban scenarios especially
fuzzy logic system which predicts broken links and network when the network density is high.

15

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

C. Bees Colony Optimization for VANET Routing end delay and packet delivery ratio over both AODV and
DSDV along with an acceptable routing overhead. However,
In addition to ACO and PSO, bees colony optimization has
QoSBeeVANET can flood the network by control packets
been applied to improve VANET routing. Two sub-categories
especially when link failures occur.
of this kind of algorithms can be found in the literature: mono
and multi-objective bees colony optimization. To overcome the drawbacks of QoSBeeVANET, the
authors of [36] proposed Hybrid Bee swarm Routing (HyBR)
1) Mono-objective Bees Colony Optimization
protocol for VANET. HyBR is unicast and multipath routing
In this sub-category, the authors of [10] proposed a protocol which guarantees that the stringent requirements
topology-based unicast routing protocol called Quality of (such as end-to-end delay, packet delivery ratio and
Service Bee Swarm routing protocol for VANET normalized overhead load) of VANET safety applications are
(QoSBeeVANET). This protocol was inspired by the bee met. As a hybrid routing protocol, a topology-based routing
swarm when bees search the food source. QoSBeeVANET is a procedure is performed when the network is dense (e.g., city-
reactive, multipath routing protocol designed for VANETs based VANET) and a geography-based routing procedure is
which guarantees the Quality of Service in terms of applied if the network density is low, as shown in figure 8. It
performance metrics such as latency and bandwidth. Two
is worth pointing out that HyBR can apply both topology-
main procedures were defined to disseminate data packets
based and geography-based schemes in VANET based on the
from a source node to a destination node.
network density (high and low densities are determined by a
Route discovery is the first step in QoSBeeVANET in threshold [107]). The topological procedure is inspired by the
which the source launches a request route packet called bees’ communication behavior when bees search for food.
forward scout in its neighborhood to find the destination. The More specifically, the source node generates, clones, and
authors proposed a novel broadcasting mode called stochastic sends route request packets (called forward scouts) to a limited
broadcasting which aims to decrease the number of routing number of its neighbors. Each neighbor repeats the same
packets exchanged to reduce network congestion. Stochastic process until the destination is reached or until a node is
broadcasting is the diffusion of routing packets (forward encountered that knows the route to the destination.
scouts are created and cloned by the source node) to only a
limited number of neighbors. If the path is discovered, a route reply (called backward
scout) is generated and is disseminated to the source node
Similarly, after receiving the forward scout, each neighbor
through the reverse route. We note that multiple routes can be
regenerates, clones, and sends this scout to its neighbors in a
stochastic broadcasting fashion until the scout reaches the discovered to forward messages. Concerning the geographical
destination or encounters an intermediate node that knows the procedure, a genetic algorithm is called to find the optimal
route to the destination. After the forward scout reaches the route among several ones found using the geographic
destination, a backward scout is created and is routed toward coordinates of network nodes. The authors proposed to
the source node via the reverse route. This discovery step is execute the GA by the source node starting with an initial
performed by taking into considerations QoS requirements population which has been randomly generated. This
such as the average end-to-end delay, the average throughput, population consists of a set of paths from the source to the
and the packet delivery ratio. The route found is saved in the destination with different intermediate nodes. Each route is
routing table of the source node that contains a set of entries evaluated and considered optimal if its entire geographical
which correspond to all known paths. For each entry, the distance is the shortest one.
destination, the next hop node to the destination and the hop
count to reach the destination are stored. The second step is HyBR GA uses a two-point crossover to generate two
about route maintenance which uses refresh packets that are offsprings after selecting two parents using RWS. Each
periodically sent between nodes in a neighborhood to ensure offspring can be mutated using a two-point mutation operator.
the stability of connections. If a link is broken, the detector This process is rerun until an optimal route is obtained. The
node informs all concerned nodes saved in its routing table authors have simulated HyBR using a realistic mobility model
about this broken link using a special packet called an error with different densities, and various vehicles’ speeds. The
scout. In addition, any node can detect QoS requirements results obtained demonstrated reduced end-to-end delay, high
violations such as reduced bandwidth or delay variations. This
node then sends an error scout to all concerned nodes (nodes
located along the path of the QoS degradation).
The nodes affected remove all invalid routes from their
routing tables and initiate a new QoS route discovery, if
needed. QoSBeeVANET has been simulated using ns-2 using
a realistic mobility model of the downtown area of the city of
Malaga in Spain [105] and its performance was compared
against the reactive routing protocol AODV and a proactive
routing protocol Destination-Sequenced Distance Vector
(DSDV) [106]. QoSBeeVANET improved the average end-to- Figure 8. HyBR Transmission based on the Network Density

16

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

TABLE II
SWARM INTELLIGENCE FOR VANET ROUTING
(NOTATION. +: SUITABLE, -: LESS SUITABLE, M: MEDIUM, L: LOW, N/A: NOT AVAILABLE)
Evaluated objectives Performance characteristics
Unicast/ multicast
Sub- Routing Mobility Packet
Algorithm Transmission Scalability Robustness Complexity Routing
category Type Criteria decision model Delay Delivery Bandwidth
mode overhead
ratio
[33]
Distributed
MAR- Lifetime Unicast Urban M N/A L + M + N/A
decision
DYMO
Mono No
-obj propagation
[100] Distributed
of malicious Unicast Highway M N/A L N/A M + N/A
TACR decision
Ant Colony vehicle
Optimization messages
Bandwidth,
Distributed
[101] delay, cost, Unicast Highway + N/A L + + + N/A
decision
Multi connectivity
[103] -obj
Lifetime, Distributed
MAV- Multicast Urban - N/A L + + - N/A
hop count decision
AODV
Packet
delivery
[34] Distributed
Particle ratio, Unicast Urban - N/A L + + + N/A
pPSO decision
Swarm Multi Overhead,
Intelligence -obj delay
Adjusting
[104] system Unicast Source node Pathway - N/A L + + M N/A
parameters

[10]
Distributed
QoSBee Delay Unicast Urban - N/A L + + + N/A
decision
VANET
Mono
-obj Shortest
Bees Colony [36] path between Distributed Urban &
source and Unicast M + L + + M N/A
Optimization HyBR decision highway
destination
Cost, delay,
[108] Multi jitter,
bandwidth Multicast Source node Urban M N/A L + N/A N/A +
BLA -obj

packet delivery ratio over AODV and GPSR routing protocols population consisting of only N bees. Next, W workers (the W
applied to VANETs. The routing overhead incurred by HyBR worst solutions) are chosen to execute the food foraging. Each
is close to those obtained by AODV and GPSR protocols. worker tries to find the best food (best solution) in its region
2) Multi-objective Bees Colony Optimization with the help of other recruited bees (by comparing neighbor
solutions). Only the best bee in each region is selected to form
Contrary to VANET unicast routing studied using the next bee population (consisting of the new queen, new D
QoSBeeVANET and HyBR, [108] proposed multi-objective drones, and new W workers). This process is rerun until a
and multicast routing algorithm called Bee Life Algorithm stagnation state is reached when the fitness no longer
(BLA). The main goal of this proposal is to find the optimal improves during some iterations.
tree called the multicast tree from the source to multiple
destinations according to four objectives, such as minimum BLA was applied to the multicast routing problem for
cost, reduced delays, decreased jitter and maximum bandwidth VANETs and its performance was compared with GA [54],
usage of links. Moreover, the expected tree should satisfy the Bees Algorithm [109], and the Marriage Bee Optimization
algorithm [110] as optimization algorithms. The results
three QoS constraints (namely, delay, jitter, and a minimum
obtained show the efficiency and the performance of BLA in
bandwidth) requested by VANET applications. To address this
terms of the combination of four objectives, namely delay,
multicast routing problem, and to satisfy the multiple QoS jitter, bandwidth and cost of found multicast routes and
constraints, BLA imitates the two major behaviors of bees: reduced computational time complexity. In the future, other
reproduction and food foraging. important metrics such as the packet loss and routing overhead
First, BLA constructs an initial population of bees which can also be investigated. In addition, BLA could be applied to
consists of N bees (solutions) chosen at random in the search address the multicast routing problem in a scalable VANET.
space. This population contains one queen, D drones and W To summarize, the complexity of VANET routing based
workers sorted following its fitness. One iteration of the bees’ on swarm intelligence is also improved to be linear and
population consists of two bee behaviors: reproduction and realistic. However, except for the study in [36], past studies
food foraging. did not consider robustness of network connectivity. Also,
most prior research works did not consider bandwidth as a
The reproduction is performed by the queen and the D
quality of service metric. Moreover, as cited for GA-based
drones using crossover and mutation operators similar to the studies, rural mobility models are a strong aspect which
natural mating of bees (called mating-flight). Next, the queen should be evaluated in future research to obtain more realistic
starts breeding N broods (i.e., N new solutions) which have results. Table 2 summarizes the properties of proposed
been sorted with the previous population bees to form a new VANET routing based on swarm intelligence.

17

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

VII. OTHER BIO-INSPIRED APPROACHES FOR VANET

ROUTING
In addition to evolutionary algorithms and swarm
intelligence categories, other bio-inspired approaches (as
shown in figure 9) have also been proposed and can be
grouped into two categories: natural immune systems and
algorithms inspired by human social behavior are influenced
Figure 9. Other Bio-inspired Approaches for VANET Routing
by bio-inspired phenomena such as genetics [111].
an area with promising outcomes.
A. Natural Immune Systems
Second, to find an information patch in a higher density
A novel harvesting strategy was proposed in [63] as a region in terms of meta-data, the agent moves through a 3-
topology-based unicast routing protocol known as Datataxis state transition diagram called Lévy walk based on the Lévy
which is inspired by the foraging behavior of Escherichia coli walk behavior [113]. Lévy walk is often applied for space with
bacteria [38], an active component in the natural immune high density of meta-data especially when there is no prior
system. Considered as a mono-objective proposal, Datataxis is knowledge of target locations or the mean spacing between
a topology-based unicast routing protocol that tries to harvest successive targets greatly exceeds the searcher’s perceptual
meta-data such as timestamp and location about particular range. Lévy walk is defined as a random walk (comprising of
events such as gas leak, explosion, attack trajectory, etc. random sequences of movement segments of length L) in
Datataxis was proposed to effectively cover a vehicular urban which the step-lengths have a probability distribution. This
sensing environment and to diffuse the collected meta-data distribution is heavy-tailed because large-length values are
using a protocol called MobEyes [112]. Datataxis uses a more prevalent than would be present within other random
multi-agent system in which agents are allowed to move to distributions such as Poisson or Gaussian. It has been reported
various network places (nodes) called information patches. that Lévy walk characterizes the movement patterns of a
These agents try to find a new information concentration with diverse range of animals such as albatrosses, deer,
high density. The main idea is based on the meta-data density bumblebees, ants, spider mites, jackals, and bacterial motion
estimation per road segment. The harvested information is [114].
transmitted to particular (irregular) nodes (such as police cars)
to enable sensing applications such as traffic reporting, Third, to avoid harvesting meta-data twice, the stigmergy-
environmental monitoring, and distributed surveillance. The based prevention mechanism is applied. This mechanism
authors of this work designed a multi-agent coordination prevents conflicts when an agent encounters some other agents
system to harvest meta-data and to route data packets using in the same area. In this case, the agent allows the new agents
the following biological phenomena. to move to other regions using a conflict resolution algorithm
so that its harvest task’s work is not duplicated. More
First, the harvesting strategy allows agents to move toward specifically, the authors have borrowed the idea of stigmergy
an area where the density in nutrients is high. Similarly, the for indirect communication between agents to mark the areas
authors used as indicator of information density the number of that they have already covered for information. Consequently,
meta-data carriers (regular nodes) actually transporting meta- the agents can move out of the area that has already been
data, in a road segment. The aim is to find a patch that searched for by other agents. This idea was inspired by the
contains a large number of “useful” meta-data carriers with ants’ behavior when they search their food. Ants used
information not yet harvested by either the same or another stigmergy to find and to remember found food source by
cooperating harvesting agent. Similarly, the harvesting laying a certain quantity pheromone [115]. This information
strategy combines a greedy heuristic with random search. helps to learn about nest mates from the route toward the food
More specifically, data harvesting agents try to move toward source.
an area where the density of new information is high (greedy
heuristic), and before they detect enough concentration of new In order to diffuse newly generated meta-data, a node
information, they can roam around (random search) to detect periodically advertises its neighbors. When a neighbor node
TABLE III
HUMAN BEHAVIOR AND NATURAL IMMUNE SYSTEM FOR VANET ROUTING
(NOTATION. +: SUITABLE, -: LESS SUITABLE, L: LOW, N/A: NOT AVAILABLE)
Evaluated objectives Unicast/ Performance characteristics
Sub- multicast Routing Mobility
Scalability Robustness Complexity Packet
category Algorithm Transmission decider model Routing
Type Criteria Delay Delivery Bandwidth
mode overhead
ratio
Harvesting meta-
Natural [63]
Mono data of events Distributed
Immune Data Unicast Urban + + L N/A N/A - N/A
-obj occurred at decision
System taxis
vehicular

making reliable
Human and secure
[37] Mono Distributed
Social routing Unicast Urban + N/A L + + - N/A
FAST -obj decision
Behavior decision towards
destination

18

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

network determines the type of the other nodes from the lists
of friends, or friends-of-friends, or no friends.
The friendship level corresponds to the transmission
requirements. This distribution of node friendship is derived
using a node’s prior knowledge of real-time vehicular traffic.
Therefore, the source node broadcasts a packet only to its
friends in its neighborhood. After that, the friend that is closer
to the destination and available at the street intersection
forwards the message to the next available node. However, if
there is no next node available at the street intersection to
forward the packet then the current node in the street will hold
the message if and only if it can reach at intersection before
Figure 10. Datataxis Harvested Meta-data Diffusion
the Time-To-Live (TTL) expires, otherwise the message is
forwarded to the next available node in the same street.
receives a packet it stores the packet in its local meta-data
databases. For example, figure 10 depicts the operation of The intersection node makes a decision to forward the
meta-data diffusion in the case of two sensors C1 and C2, message to the best suitable route towards destination using a
installed on two vehicles. C1 and C2 sensors periodically fuzzy inference system [107]. This decision is based on prior
advertise a new meta-data packet SC1,1 and SC2,1 respectively. global knowledge of real-time vehicular traffic concerning the
At time T-t4, vehicle of C2 encounters vehicle of C1 (at an traffic-density: friends, friends-of-friends, and non-friends
encounter point), and they exchange these advertisements. information on each road segment. For each node, this system
Nodes keep track of the freshness of meta-data packets by consists of inputs (friends, friends-of-friends, and non-
using a sliding window with the maximum window size of a friends), and membership functions are used to determine if
fixed expiration time. In addition, since a single advertisement the network is Sparse, Medium, Dense in terms of friendships.
packet may contain multiple meta-data records, it is possible As for the output, the greater the friendship density value, the
to define a packet sensing location as the average position of better the optimal node becomes (corresponding to a more
all meta-data in the packet. When a packet expires or the efficient route).
packet originator moves away more than a threshold distance
FAST has been simulated and compared with GPSR,
from the packet sensing location, the packet is automatically
GPCR, RBVT-R, and intersection-based Greedy Traffic-
disposed.
Aware Routing (GyTAR) protocol [116] and Road-Based
For urban scenarios and using the Manhattan mobility using Vehicular Traffic routing -Reactive protocol (RBVT-R)
model, the three algorithms (bacteria harvesting strategy, Lévy [117], and has shown an improvement in the average delay,
walk strategy and stigmergy-based strategy) have been and packet delivery ratio for VANETs with 100, 150 and 200
validated through a set of simulations to demonstrate the nodes. However, constructing friendship lists can increase
efficiency of the bio-inspired behavior of harvesting agents routing overheads. FAST is also more appropriate to urban
and, its robustness for a wide range of various parameters scenarios because it is based on forwarding packets using
including density distribution, aggregated number of harvested intersection nodes which are less used for rural VANETs.
Meta-data and harvesting rate per road segment over time. The
As summarized in table 3, natural immune systems and
simulation results obtained also showed the efficient
human social behavior approaches contributed to reducing
distribution in the movement of multiple agents. However, the
routing complexity to linear. Nevertheless, previous studies
authors conducted their experiments using a theoretical
dealt only with mono-objective problems. In addition, routing
mobility pattern (i.e., the Manhattan model) only for an urban
overhead was not taken into consideration and rural mobility
area in which vehicles moved with limited speeds. Moreover,
models were also not considered but would be interesting to
this proposed approach suffers from an exchange of control
investigate in future bio-inspired VANET routing research.
packets which can cause network congestions.
B. Human Social Behavior influenced by Biological VIII. REINFORCEMENT LEARNING FOR BIO-INSPIRED VANET
Phenomena ROUTING: A UNIFIED FORMAL MODEL
The social behavior of humans on the road influenced by Previous bio-inspired protocols focus mostly on an
some biological phenomena such as genetics or other factors objective function based on distinct bio-inspired modules.
such as emotions, hypnosis, coercion, has been used as an Moreover, most of these previous works suffer from the major
inspiration in the development of a geography-based routing drawback that the programmer should input user parameters.
protocol for VANETs called Fuzzy-Assisted Social-based For example, bio-inspired routing mechanism was proposed
rouTing (FAST) protocol [37]. FAST is based on a fuzzy for discovering paths using GA in AMR [83]. However,
inference system that leverages a friendship mechanism to alternative unicast paths searching mechanism were proposed
make optimal and secure routing decisions from the source to in [104] using fuzzy logic system whenever a broken link or a
the destination using prior global knowledge of real-time congestion event occurs along a path. On the other hand, PSO
vehicular traffic. A secure routing decision was considered as was applied in [32] to adjust OLSR user parameters.
the single objective of FAST. To achieve this, each node in the

19

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

Therefore, we discuss in this section a proposal for a

unified formal model (based on the Reinforcement Learning
(RL) paradigm) for bio-inspired multi-modular approaches
aimed at finding multiple solutions represented by a global
objective function. Our goal is to make this bio-inspired
VANET routing a unified model, which allows the discovery
of routes, maintaining paths, tuning parameters, optimizing
routing etc. considered as a part of some global function to be
built. This model performs an iterative process using RL
Figure 11. Reinforcement Learning Model
which is sufficient for this kind of process, especially when we
have no a priori knowledge about the global function to be simple parametric functions. Each of these elementary
built. In addition, RL is more appropriate for bio-inspired functions is built by a separate module that ensures an action
methods compared to traditional exact methods, where RL is of a local routing decision executed by each agent at node
focused on on-line performance, without any sub-optimal level. To achieve this, continuous reinforcement learning in
actions explicitly introduced. VANET involves changing the parameters of the network
model using an adaptive update rule as follows:
The proposed model is designed from Vapnik theory [118],
and adapted based on the framework proposed in [11]. RL w(n)=w(n-1) - (n) F(x(n), w(n-1)) (2)
modeling is very suitable to bio-inspired VANET routing
compared to conventional routing because of the expected where w(n) is called the global objective function, denotes
effectiveness of RL processing to adapt to the dynamics of the parameters of the VANET at time n, x(n) is the current
VANET problems. In fact, RL does not request any prior network situation and  is the modification step. F is either
explicit learning in its initial step which is very similar to bio- the gradient of the global objective function or a heuristic rule.
inspired VANET routing process, so the network parameters x can be defined by a probability density function p(x), and w
(i.e., initial nodes locations, speeds, VANET density, links, the parameters of the learning system. The concept behind our
bandwidth, latency etc.) are given in a stochastic manner and proposal is based on the fact that each iteration uses an
cannot be assumed in advance. Consequently, a real time instance of a real flow instead of a finite training set. The goal
routing decision will be taken at the node level without any of this learning system consists of finding the minimum of a
prior hypothesis. function C(w) called the expected risk function [120] and the
Moreover, in contrast to traditional distributed algorithms, average update is therefore a gradient algorithm which directly
reinforcement learning is considered a reactive approach optimizes C(w). As proposed in [121], for a given state of the
which ensures the execution of the desired task in a highly network, we can define a local objective function J(x,w) which
dynamic network environment [119]. Therefore, each measures how well our network behaves on state x based on
variation in the vehicular network involves quick perception, the fact that our knowledge of the random process comes from
decision, and performing the appropriate reaction. These a series of observations {xi} of the variable x. Usually, the
reasons motivate our proposal. expectation of the local objective function over the space X of
the concept to learn:
A. Steps and Formalism of the Proposed Model
The reinforcement learning model is known as a finite-state
machine, represented by an agent, which interact with its 
C(w)= J ( x, w) p ( x ) dx = Ex{J(x,w) }
x
(3)
stochastic environment. The main goal of the RL agent is to
perform an optimal action using the environment through a Most often we do not know the explicit form of p(x) and
learning process. The learning process is performed using a therefore C(w) is unknown. We are thus only able to calculate
reinforcement function which has as inputs the current state J(x,w) from the observations {xi}.
(St) of the environment and the reward of the last transition
(Rt), giving rise to a reinforcement signal as an output, as A necessary condition of optimality for the parameters of the
shown in figure 11. This reinforcement signal determines the system is:
next action (at) to be sent to the environment, according to a
policy of RL. Finally, the environment enters into a new state C(w) = Ex{wJ(x,w)}=0 (4)
(St+1), with a new reward (Rt+1) which measures the goodness
of this new state. where is the gradient operator. Since we do not know
The vehicular ad-hoc network can be viewed as an RL but only the result of , which corresponds to
environment, where each node is considered as RL agent the current network state, we cannot use classical optimization
trying to guarantee an improved bio-inspired routing by techniques to obtain the optimum. One solution is to use an
optimizing one or multiple criteria through the reinforcement adaptive algorithm as follows:
function which is unified by the proposed model described
below. Consequently, a global function is built by multiple w (n)=w(n-1) - (n) w J(x(n),w(n-1)) (5)

20

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

where  is the gradient step. Here, Eq. 2 and Eq. 5 are

similar because learning aims to minimize an objective
function. The convergence proof of the online gradient
algorithm from this formula can be found in [121] where it is
proven that this formulation is particularly adequate for
describing adaptive algorithms that simultaneously process an
observation and learn to perform better. In next sub-section, an
application of the proposed unified model in the case of bio-
inspired VANET routing based on a global bi-objective
function of two evolving parameters (end-to-end delay and the
end-to-end residual bandwidth (BW)) is presented.
B. Study of a VANET Case
For the VANET case, our goal is to optimize a global bi- Figure 12. Multipath VANET Routing
objective function of VANET routing based on two criteria:
the end-to-end delay (DY) and the end-to-end residual optimal solutions. The selected solution provides the best path
bandwidth (BW). In other words, the best end-to-end delay in terms of end-to-end delay and end-to-end residual
(DY) and the optimal end-to-end residual bandwidth (BW) are bandwidth, where it is impossible to find one better path
the objectives of the proposed unified model. Therefore, the without making at least one criterion worse off. Hence, the
optimal path among all possible paths from a source node to a user favors one criterion to select best_h among non-
destination node can be found to forward data packets. dominated solutions (i.e. the Pareto optimal solutions).

First, we propose to apply HyBR routing protocol inspired Once the choice of the next hop is made, node s can
by Bees colony systems presented in our prior work [36]. therefore update its reinforcement functions (a global
HyBR provides for each source node s, all possible multiple objective function of end-to-end delay and end-to-end residual
paths towards each destination d through every next hop h bandwidth) as follows:
which is within the transmission range of node s. The multiple ′ , , 1 ,
paths are maintained in the routing table of s. (7)
′ , , 2 ,
To find the best parameters of the optimal path, our RL where αDY and ηDY are the packet transmission times between h
model proposes that the routing table saves a set of end-to-end and s and the delay learning rate respectively. On the other
delay values DYh (s, d), as well as, a collection of end-to-end hand, αBW and ηBW are the residual bandwidth between h and s
residual bandwidth values BWh (s, d) for each destination d and the bandwidth learning rate respectively.
and for every immediate next hop h. Moreover, node h can
estimate the bandwidth and the delay throughout the So, the new estimations , and , can be
remainder path (from node h to the destination) by exchanging written as:
control packets with its neighbor nodes, then its extra- ′ , , 1 1
(8)
neighbor nodes and so on until the destination is reached as ′ , , 1 2
shown in figure 12.
When we compare the formal model given by Eq. 5, we
Upon receipt of the packet, node h responds to node s with
observe the similarity between parameters ηDY and DYh
an estimate of its best bi-criteria (DY and BW) to reach the
representing the gradient step γ and the function J
destination d by sending the reinforcement learning signal
respectively. It is same for ηBW and BWh which represent the
, .
gradient step γ and the function J respectively.
More specifically, the reinforcement signal (s,d)
consists of a set of two types of information namely, minimum IX. OPPORTUNITIES AND FUTURE TRENDS
delay and maximum residual bandwidth from node h to Bio-inspired approaches are powerful and efficient. They
destination node d through all possible neighbors of node h, have strong potential to address some of the challenges
calculated using the following reinforcement functions (local associated with VANET routing. In this section, we present
objective functions): some of the opportunities and future trends that bio-inspired
approaches can efficiently address for VANET routing
1 ∑ ∈ , because of their dynamic characteristics.
s,d (6)
2 ∈ ,
A. Vehicular Network Scalability
After receiving all reinforcement signals (all Bio-inspired methods can efficiently handle a large number
, , ∈ ), node s selects the best of VANET nodes in contrast to traditional routing methods
next hop (best_h) to use to transmit data packets. Selection of because they are based on local and limited cooperation and
best_h is based on the bi-criteria function found using Pareto communication among nodes when searching for routes. More
21

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

specifically, bio-inspired methods are able to increase the withstand against external and internal perturbations [126],
network’s lifetime as well as the QoS performance metrics using a specific topological structure of biological systems
such as bandwidth and packet delivery ratio under high loads called modularity. This modularity structure helps to localize
when new nodes are added. These bio-inspired methods are damages and prohibits malicious effects from spreading
also capable of minimizing transmission delays and errors. throughout the entire system [127]. By exploiting the
robustness and modularity associated with bio-inspired
The idea is to manage the entire vehicular network by
techniques, we can guarantee some level of VANET
dividing it to a set of sub-networks in order to determine an
performance, especially for heterogeneous networks to cope
efficient global behavior in terms of QoS through multiple
with their heterogeneity in terms of structure (used
interactions of local behaviors of sub-networks which have
heterogeneous topologies), or functionality (applied
local and limited networking performance. However, the
heterogeneous protocols).
design of scalable routing protocols which are reliable with
increasing size of the network remains a significant challenge. To summarize, the strongest aspect reinforced by bio-
To address this challenge, researchers need to investigate inspired approaches for VANET routing is the complexity of
routing protocols that attempt to maintain a small data finding routes from source nodes to destinations used to
structure such as the routing table, used to find the outgoing disseminate and forward data packets in the network.
link information when the number of nodes in the VANET is Consequently, routing overhead and the average end-to-end
very large [122] [123]. delay are reduced. However, there are certain design criteria
and additional evaluation metrics which should be investigated
B. Self-organized Control of VANET’s Dynamic Nature
in future bio-inspired approaches for VANET routing. These
Bio-inspired approaches can be investigated to control the metrics include the robustness of network connectivity,
dynamic properties of VANET using self-organizing scalability in terms of network density, and the need to
solutions, which are adaptive to frequent VANET changes consider more realistic vehicular mobility patterns like rural
caused by nodes’ mobility and network failures and areas. Moreover, we suggest in further research works, to
perturbations. VANET changes include vehicles’ positions, study VANET routing as a multi-objective and multi-
directions and velocities, number of nodes in the network, as constraint problem. In other words, more quality of service
well as specific network conditions such as congestion, packet metrics should be taken into account in the objective function
drop, security etc. In contrast to traditional methods, biology- like transmission cost, bandwidth, routing overhead, jitter,
based systems and methods usually show the ability to packet delivery ratio, and others, where some of these
effectively adapt to constantly changing environments. Based objectives can be restricted by the user and explained in the
on a large number of dynamic, autonomous, and distributed computational studies by multiple constrains (e.g. minimum
components, most biological systems can generate effective used bandwidth, maximum accepted jitter etc.).
entire behaviors through local interactions as a result of self-
organization [124]. As result, bio-inspired methods can better We also suggest that the VANET community of researchers
manage undesirable networking events without the apply other bio-inspired methods which are not applied to
intervention of the human operator. solve routing problems for VANETs and could give promising
results. In this context, relevant bio-inspired methods such as
C. Reducing Resource Requirements and Complexity of Evolution Strategy [128], Evolutionary Programming [129],
Message Exchanges Genetic Programming [130], Estimation of Distribution
Bio-inspired methods can be applied to reduce the Algorithm [131], Differential Evolution [132], and Cultural
complexity of message exchange when finding the best routes Algorithms [133] may prove very useful.
in VANET. Indeed, bio-inspired techniques applied to ad hoc
networks can considerably minimize the network resources X. CONCLUSION
(such as memory and network bandwidth) consumed during
We have presented in this paper a survey on bio-inspired
message delivery which ultimately lead to reducing the
approaches applied to VANET routing which support ITS
complexity of message exchanges. Consequently, message
applications. In fact, it has been shown that the bio-inspired
latency can be decreased and the message delivery rate [125]
methods are more efficient for large-scale vehicular networks,
can be increased. Here, bio-inspired methods are based on the
which disseminate data packets with low complexity. Also,
local behavior and reaction of each independent individual.
bio-inspired VANET routing is Adaptive and robust which
Dissemination of messages for VANET can also take
maintains an improved routing performance despite network
advantage of biology inspiration to achieve low-complexity
disruptions. After presenting the VANET routing background
routing compared to traditional routing methods.
and existing standards, we described the basic concepts and
D. Robustness operations of recently proposed bio-inspired algorithms and
We argue that bio-inspired approaches have tremendous techniques reported in the literature which include genetic
potential in providing robust, resilient, and adaptive vehicular algorithms, ant colony optimization, particle swarm
networks. Robustness is the most commonly observed feature optimization, bee colony optimization, natural human system
in all living organisms which have evolved themselves to and fuzzy logic off human brain.

22

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

In this survey, we have identified three bio-inspired [9] J.-S. Park, U. Lee, S. Y. Oh, M. Gerla, and D. S. Lun, “Emergency
related video streaming in VANET using network coding,” in Proc.
categories for routing in VANETs and they include ACM VANET’06, Los Angeles, CA, USA, 2006, pp. 102-103.
evolutionary algorithms, Swarm Intelligence and other bio- [10] S. Bitam, and A. Mellouk, “QoS Swarm bee routing protocol for
inspired approaches for VANET routing. For each category, vehicular ad hoc networks,” in Proc. IEEE ICC’11, Kyoto, Japan, 2011,
pp. 1-5.
research studies have been classified, evaluated and compared [11] A. Mellouk, S. Hoceïni, and S. Zeadally, “Design and performance
with the most important routing and computational criteria, analysis of an inductive QoS routing algorithm,” Comput. Commun, vol.
namely complexity, scalability, robustness, mobility model 32, no. 12, pp. 1371-1376, 2009.
used, and the QoS routing performances (i.e. delay, packet [12] Y. W. Lin, Y. S. Chen, and S. L. Lee, “Routing protocols in vehicular ad
hoc networks: a survey and future perspectives,” J. Inform. Sci. Eng.,
delivery ratio, routing overhead and bandwidth). To determine vol. 26, no. 3, 2010.
the convergence of bio-inspired approaches applied to [13] O. Abedi, M. Fathy, and J. Taghiloo, “Enhancing AODV routing
VANET routing, we have proposed a unified formal model protocol using mobility parameters in VANET,” in Proc. IEEE/ACS
AICCSA, Doha, Qatar, 2008, pp. 229-235.
based on reinforcement learning system. This model aims at [14] J. Santa, M. Tsukada, T. Ernst, O. Mehani, and A. F. Gómez-Skarmeta,
finding multiple solutions represented by a global objective “Assessment of VANET multi-hop routing over an experimental
function representing performance criteria of routes discovery, platform,” Int. J. Internet Protocol Technol., vol. 4, no. 3, pp. 158-172,
2009.
maintaining, and optimization etc. It is worth mentioning that [15] B. Karp, and H. T. Kung, “GPSR: greedy perimeter stateless routing for
the proposed model is more suitable for bio-inspired methods wireless networks,” in Proc. IEEE/ACM MobiCom, Boston, MA, USA,
compared to traditional exact methods, where reinforcement 2000, pp. 243-254.
[16] C. Lochert, H. Hartenstein, J. Tian, H. Füßler, D. Hermann, and M.
learning is focused on on-line performance, without any sub- Mauve, “A routing strategy for vehicular ad hoc networks in city
optimal actions explicitly introduced. environments,” in Proc. IEEE Intell. Vehicles Symp. Columbus, OH,
USA, 2003, pp. 156–161.
Finally, lessons learned, some opportunities, and future [17] S. S Wang, and Y. S Lin, “PassCAR: A passive clustering aided routing
trends of bio-inspired methods for routing in VANET are protocol for vehicular ad hoc networks,” Comput. Commun., vol. 36, no.
discussed. We found that bio-inspired approaches can improve 2, pp. 170-179, 2013.
[18] F. Dressler, and O. B. Akan, “A survey on bio-inspired networking,”
the computational criteria and enhance the QoS performances Comput. Netw., vol. 54, no. 6, pp. 881-900, 2010.
of VANET routing. However, we also discovered that recent [19] G. Liu, and C. Ji, “Scalability of network-failure resilience: analysis
efforts take into consideration only a few of the using multi-layer probabilistic graphical models,” IEEE/ACM Trans.
Netw., vol. 17, no. 1, pp. 319-331, 2009.
aforementioned criteria and lacked the others. Thus, we [20] R. Ramanathan, R. Allan, P. Basu, J. Feinberg, G. Jakllari, V. Kawadia,
recommend that researchers develop bio-inspired VANET and J. Freebersyser, “Scalability of mobile ad hoc networks: Theory vs
routing approaches based on the proposed unified formal practice,” in Proc. IEEE MILCOM, 2010, pp. 493-498.
[21] A. Ovcharenko, D. Ibanez, F. Delalondre, O. Sahni, K. E. Jansen, C. D.
model (i.e. reinforcement learning system) which addresses Carothers, and M. S. Shephard, “Neighborhood communication
different routing aspects. paradigm to increase scalability in large-scale dynamic scientific
applications,” Parallel Computing, vol. 38, no. 3, pp. 140-156, 2012.
[22] F. Neumann, and C. Witt, “Bio-inspired computation in combinatorial
ACKNOWLEDGEMENTS optimization: algorithms and their computational complexity,” in Proc.
We thank the anonymous reviewers for their constructive ACM Int. Conf. Genetic Evol. Computation, 2012, pp. 1035-1058.
[23] J. J. Hopfield, “Neural networks and physical systems with emergent
suggestions and comments which helped us to improve the collective computational abilities,” in Proc. Nat. Academy of Sci. of the
content, quality, and presentation of this paper. USA, vol. 79 no. 8 pp. 2554–2558, 1982.
[24] G. Bin, L. Zhe, Z.-J. Wei, “A Dynamic-cluster energy-aware routing
algorithm based on neural structure in the wireless sensor networks,” in
REFERENCES Proc. IEEE Int. Conf. Comput. Inform. Technol., 2005, pp. 401-405.
[1] S. Zeadally, R. Hunt, Y.-S. Chen, A. Irwin, and A. Hassan, “Vehicular [25] C. List, C. Elsholtz, and T. D. Seeley, “Independence and
ad hoc networks (VANETs): status, results, and challenges,” interdependence in collective decision making: an agent-based model of
Telecommun. Syst. vol. 50, no 4, pp. 217-241, 2012. nest-site choice by honeybee swarms,” Philosoph. Trans. Roy. Soc. B:
[2] A. L. Beylot, and H. Labiod, “CONVOY: a new cluster-based routing biological sci., vol. 364, no 1518, pp. 755-762, 2009.
protocol for vehicular networks,” in Vehicular Networks: Models and [26] J. Kennedy, and R. C. Eberhart, “Particle swarm optimization,” in Proc.
Algorithms, 1st ed., London, UK: John Wiley & Sons, 2013, ch. 3,pp. IEEE Int. Conf. Neural Netw., 1995, pp. 1942–1948.
91-139. [27] F. Heppner, and U. Grenander, “A stochastic nonlinear model for
[3] T. Okabe, T. Shizuno, and T. Kitamura, “Wireless LAN access network coordinated bird flocks,” Washington DC, USA: Amer. Assoc.
system for moving vehicles,” in Proc. 10th IEEE Symp. Comput. Advancement Sci., 1990.
Commun, 2005, pp. 211-216. [28] A. Haghighat, K. Faez, M, Dehghan, A, Mowlaei, and Y, Ghahremani,
[4] J. Santa, A. F. Gómez-Skarmeta, and M. Sánchez-Artigas, “Architecture “GA-based heuristic algorithms for QoS based multicast routing,”
and evaluation of a unified V2V and V2I communication system based Knowledge-Based System, vol. 16, pp. 305–312, 2003.
on cellular networks,” Comput. Commun., vol. 31, no. 12, pp.2850-2861, [29] F. Ducatelle, “Adaptive routing in ad hoc wireless multi-hop networks,”
2008. Ph.D. dissertation, Inform. Fac., Della Svizzera Italiana Univ., Lugano,
[5] F. Li, and Y. Wang, “Routing in vehicular ad hoc networks: A survey,” Switzerland, 2007.
IEEE Trans. Veh. Technol., vol. 2, no. 2, pp. 12-22, 2007. [30] A. Larhlimi, S. Blachon, J. Selbig, and Z. Nikoloski, “Robustness of
[6] Research and Innovative Technology Administration (RITA), Intell. metabolic networks: A review of existing definitions,” Biosystems, vol.
Transportation Syst. Available: http://www.itsoverview. its.dot.gov/. 106, no. 1, pp. 1-8, 2011.
[7] E. Schoch, F, Kargl, M. Weber, and T. Leinmuller, “Communication [31] H. Saleet, R. Langar, K. Naik, R. Boutaba, A. Nayak, and N. Goel,
patterns in VANETs,” IEEE Commun. Mag. vol. 46, no. 11, pp. 119- “Intersection-based geographical routing protocol for VANETs: a
125, 2008. proposal and analysis,” IEEE Trans. Veh. Technol., vol. 60, no. 9, pp.
[8] H. Hartenstein, K. P. Laberteaux, “A tutorial survey on vehicular ad hoc 4560–4575, 2011.
networks,” IEEE Commun. Mag., vol. 46, no 6, p. 164-171, 2008. [32] T. Clausen, and P. Jacquet, “The optimized link-state routing (OLSR)
protocol,” IETF RFC 3626, 2003.

23

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

[33] S. L. O. B. Correia, J. C. Junior, and O. Cherkaoui, “Mobility-aware ant [59] M. Dorigo, “Optimization, learning and natural algorithms,” Ph.D.
colony optimization routing for vehicular ad hoc networks,” in Proc. dissertation, Dept. Elect., Polytechnic Univ. Milano, Italy, 1992.
IEEE WCNC, 2011, pp. 1125-1130. [60] S. Bitam, M. Batouche and E-G. Talbi “A survey on bee colony
[34] J. Toutouh, and E. Alba, “Parallel swarm intelligence for VANETs algorithms,” in Proc. IEEE IPDPS, NIDISC workshop, Atlanta, GA,
optimization,” in Proc. IEEE 3PGCIC, 2012, pp. 285-290. USA, pp. 1-8.
[35] C. Perkins, E. M. Belding-Royer, S. Das, “Ad hoc on-demand distance [61] L. A. Zadeh, “Fuzzy sets,” Information and control, vol. 8, no. 3, pp.
vector (AODV) routing protocol,” draft-perkins-manet-aodvbis-00.txt, 338-353, 1965.
2003. [62] E. H. Mamdani, “Application of fuzzy logic to approximate reasoning
[36] S. Bitam, A. Mellouk, and S. Zeadally, “HyBR: A Hybrid bio-inspired using linguistic synthesis,” IEEE Trans. Comput., vol. 100, no. 12, pp.
bee swarm routing protocol for safety applications in vehicular ad hoc 1182-1191, 1977.
networks (VANETs),” J. Syst. Architecture, vol. 59, no. 10-B, pp. 953- [63] U. Lee, E. Magistretti, M. Gerla, P. Bellavista, P. Lio, and K.-W. Lee
967, 2013. “Bio-inspired multi-agent data harvesting in a proactive urban
[37] R. H. Khokhar, R. M. Noor, K. Z. Ghafoor, C. H. Ke, and M. A. Ngadi, monitoring environment,” Adhoc Netw., vol. 7, no. 4, pp. 725-741, 2008.
“Fuzzy-assisted social-based routing for urban vehicular environments,” [64] J. Yick, B. Mukherjee, and D. Ghosal, “Wireless sensor network
J. Wireless Commun. Netw. EURASIP, vol. 2011, no. 1, pp. 1-15, 2011. survey,” Comput. Netw., vol. 52, no. 12, pp. 2292-2330, 2008.
[38] H. Berg, and D. Brown, “Chemotaxis in escherichia coli analysed by [65] M. Saleem, G. A. Di Caro, and M. Farooq, “Swarm intelligence based
three-dimensional tracking,” Nature, vol. 239, pp. 500–504, 1972. routing protocol for wireless sensor networks: Survey and future
[39] A. K. Saha, D. B. Johnson, “Modeling mobility for vehicular ad-hoc directions,” Inform. Sci., vol. 181, no. 20, pp. 4597-4624, 2011.
networks”, in Proc. ACM Int. workshop on Veh. ad hoc netw., 2004, pp. [66] R. V. Kulkarni, A. Forster, and G. K. Venayagamoorthy,
91-92. “Computational intelligence in wireless sensor networks: A survey,”
[40] H. Saleet, O. Basir, R. Langar, and R. Boutaba, “Region-based location- IEEE Commun. Surveys Tuts., vol. 13, no. 1, pp. 68-96, 2011.
service-management protocol for VANETs,” IEEE Trans. Veh. [67] H. Yang, F. Ye, and B. Sikdar, “A swarm-intelligence-based protocol
Technol., vol. 59, no. 2, pp. 917-931, 2010. for data acquisition in networks with mobile sinks,” IEEE Trans. Mobile
[41] R. Nagel, and S. Eichler, “Efficient and realistic mobility and channel Comput., vol. 7, no. 8, pp. 931-945, 2008.
modeling for vanet scenarios using omnet++ and inet-framework,” in [68] B. Zhang and H. Mouftah, “Energy-aware on-demand routing protocols
Proc. SIMUTools, Marseille, France, 2008, pp. 98. for wireless ad hoc networks,” Wireless Netw., vol. 12, no. 4, pp. 481-
[42] M. Hossain, and M. Atiquzzaman, “Stochastic properties and application 494, 2006.
of city section mobility model,” in Proc. IEEE Globecom, Honolulu, HI, [69] D. B. Johnson, D. A. Maltz, and J. Broch, “DSR: the dynamic source
USA, 2009, pp. 1-6. routing protocol for multi-hop wireless ad hoc networks,” in Ad hoc
[43] F. Bai, N. Sadagopan, and A. Helmy, “The IMPORTANT framework Netw., C. Perkins, Ed. Boston, MA, USA: Addison-Wesley, 2001, ch. 5,
for analyzing the impact of mobility on performance of routing protocols pp. 139-172.
for adhoc networks,” Ad Hoc Netw., vol. 1, no. 4, pp. 383-403, 2003. [70] J. Sun, W. Fang, X. Wu, Z. Xie, and W. Xu W, “QoS multicast routing
[44] X. Hong, M. Gerla, G. Pei, and C. Chiang, “A group mobility model for using a quantum-behaved particle swarm optimization algorithm,” Eng.
ad hoc wireless networks,” in Proc. 2nd ACM Int. workshop on Appl. Artificial Intell., vol. 24, pp. 123–131, 2011.
Modeling, analysis and simulation of wireless and mobile syst, 1999, pp. [71] S. A. Bitaghsir, and F. Hendessi, “An intelligent routing protocol for
53-60. delay tolerant networks using genetic algorithm,” Smart Spaces and
[45] B. Zhou, K. Xu, and M. Gerla, “Group and swarm mobility models for Next Generation Wired/Wireless Networking, pp. 335-347, 2011.
ad hoc network scenarios using virtual tracks,” in Proc. IEEE MILCOM, [72] C. O’rourke, and S. B. Johnson, “Mobile ad hoc networking revamps
vol. 1, 2004, pp. 289-294. military communications,” J. of Military Electron. and Comput., pp. 34-
[46] N. Meghanathan, “A simulation study on the impact of mobility models 42, 2011.
on the network connectivity, hop count and lifetime of routes for ad hoc [73] G. A. Di Caro, F. Ducatelle, and L. M. Gambardella, “AntHocNet: an
networks,” Informatica, vol. 34, no. 2, 2010, pp. 207-221. adaptive nature-inspired algorithm for routing in mobile ad hoc
[47] S. Bitam, and A. Mellouk, “Markov-history based modeling for realistic networks,” Eur. Trans. Telecommun., vol. 16, no. 5, pp. 443–455, 2005.
mobility of vehicles in VANETs,” in Proc. IEEE VTC-Spring, Dresden, [74] S. Scellato, I. Leontiadis, C. Mascolo, P. Basu, and M. Zafer,
Germany, 2013, pp. 1-5. “Evaluating temporal robustness of mobile networks,” IEEE Trans.
[48] Federal Communications Commission (FCC) of USA, “Amendment of Mobile Comput., vol. 12, no. 1, pp. 105-117, 2013.
the Commission’s Rules Regarding Dedicated Short-Range [75] D. S. Callaway, M. E. Newman, S. H. Strogatz, and D. J. Watts,
Communication Services in the 5.850-5.925 GHz Band,” Tech. Rep. “Network robustness and fragility: Percolation on random graphs,”
FCC-03-324, FCC, Dec. 2003. Physical Review Letters, vol. 85, no. 25, pp. 5468-5471, 2000.
[49] Standards of IEEE, Available: http://standards.ieee.org/IPR/ disclaimers. [76] R. Albert, H. Jeong, and A.-L. Barabasi, “Error and Attack Tolerance of
html/. Complex Networks,” Nature, vol. 406, no. 6794, pp. 378-382, 2000.
[50] B. J. Kenney, “Dedicated short-range communications (DSRC) [77] L. Ma, and P. A. Iglesias, “Quantifying robustness of biochemical
standards in the United States,” in Proc IEEE 99.7, 2011, p. 1162-1182. network models,” BMC bioinformatics, vol. 3, no. 1, pp. 38-50, 2002.
[51] D. Johnson, A. Menezes, and S. Vanstone, “The elliptic curve digital [78] Z. Zi, , K. H. Cho, M. H. Sung, X. , Xia, J. Zheng, and Z. Sun, “In silico
signature algorithm (ECDSA)”, Int. J. Inform. Security, vol. 1, no. 1, p. identification of the key components and steps in IFN-γ induced JAK-
36-63, 2001. STAT signaling pathway,” FEBS letters, vol. 579, no. 5, pp. 1101-1108,
[52] E. G. Talbi, Metaheuristics: from design to implementation. Hoboken, 2005.
NJ: Wiley, 2009. [79] T. Yamada, and P. Bork, “Evolution of biomolecular networks - lessons
[53] C. J. Huang, Y. T. Chuang, Y. J. Chen, Y. C. Luo, and D. X. Yang, from metabolic and protein interactions,” Nature Reviews Molecular
“Constructing alternative routes for VANET using machine learning Cell Biology, vol. 10, no. 11, pp. 791-803, 2009.
techniques,” in Proc. 7th Int. Conf. AIT, Taichung, Taiwan, 2007. [80] A. O. Tarakanov, “Immunocomputing for intelligent intrusion
[54] J. H. Holland, “Genetic algorithms,” Scientific American, vol. 267, no. 1, detection,” IEEE Comput. Intell. Mag., vol. 3, no. 2, pp. 22-30, 2008.
1992, pp. 66-72. [81] N. Atreas, C. Karanikas, and A. Tarakanov, “Signal processing by an
[55] T. Back, Evolutionary algorithms in theory and practice, evolution immune type tree transform,” Artificial Immune Systems, vol. 2787, pp.
strategies, evolutionary programming, genetic algorithms. Oxford 111–119, 2003.
University Press, USA, 1996. [82] R. Horn, and Ch. Johnson, Matrix Analysis. Cambridge University
[56] A. H. Wright, “Genetic algorithms for real parameter optimization,” Press, 1986.
Foundations of Genetic Algorithms, vol. 1, pp. 205-218, 1991. [83] H. Saleet, R. Langar, O. Basir, and R. Boutaba, “Adaptive message
[57] C. Berek, and C. Milstein, “Mutation drift and repertoire shift in the routing with QoS support in vehicular ad hoc networks,” in Proc. IEEE
maturation of the immune response,” Immunological reviews, vol. 96, Globecom, Honolulu, Hawaii, USA, 2009, pp. 1-6.
no. 1, pp. 23-41, 1987. [84] P. C. Cheng, J. T. Weng, L. C. Tung, K. C. Lee, M. Gerla, and J. Haerri,
[58] Z. Michalewicz, and C. Z. Janikow, “Handling constraints in genetic “GeoDTN+ Nav: A hybrid geographic and DTN routing with navigation
algorithms,” in Proc. 4th Int. Conf. Genetic Algorithms, vol. 151, 1991, assistance in urban vehicular networks,” in Proc. 5th Annu. Int. Conf.
pp. 157. Mobile Ubiquitous, Dublin, Ireland, 2008.

24

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/COMST.2014.2371828, IEEE Communications Surveys & Tutorials

[85] G. Danoy, B. Dorronsoro, P. Bouvry, B. Reljic, and F. Zimme, “Multi- [111] J.W. Gilger, and S. L. Hershberger, “Human behavioral genetics:
objective optimization for information sharing in VANETs,” in Proc. synthesis of quantitative and molecular approaches,” Human Biology,
Advances Inform. Technol., pp. 58–70, 2009. vol. 70, no. 20, pp. 156, April 1998.
[86] L. Hogie, M. Seredynski, F. Guinand, and P. Bouvry, “A bandwidth- [112] P. Bellavista, E. Magistretti, U. Lee, and M. Gerla, “Standard integration
efficient broadcasting protocol for mobile multi-hop ad hoc networks,” of sensing and opportunistic diffusion for urban monitoring in vehicular
in Proc. ICN/ICONSMCL, Los Alamitos, CA, USA, pp. 71-79, 2006. sensor networks: the MobEyes architecture,” in Proc. IEEE Int. Symp.
[87] L. Hogie, F. Guinand, and P., Bouvry, “The madhoc metropolitan Adhoc Industrial Electron., Vigo, Spain, 2007, pp. 2582-2588.
network simulator,” Available: http://www-lih.univ-lehavre.fr/~hogie/ [113] P. Lévy, Theorie de l’addition des variables aléatoires. Gauthier-Villars,
madhoc/. Paris, 1954.
[88] J. Durillo, “The jMetal framework,” Available: http:// neo.lcc.uma.es/ [114] D. Sims, D. Righton, and J. W. Pitchford, “Minimizing errors in
software/metal/. identifying Lévy flight behavior of organisms,” J. Animal Ecology, vol.
[89] M. Slavik, and I. Mahgoub, “Applying machine learning to the design of 76, no. 2, pp. 222–229, 2007.
multi-hop broadcast protocols for VANET,” in Proc. IEEE IWCMC, [115] D. E. Jackson, and F. L. Ratnieks, “Primer: communications in ants,”
Istanbul, Turkey, 2011, pp. 1742-1747. Current Biology, vol. 16, no. 15, pp. 570–574, 2006.
[90] M. Slavik, L., Mahgoub, and F. N. Sibai, “The distance-to-mean [116] M. Jerbi, S.M. Senouci, T. Rasheed, and Y. Ghamri-Doudane, “Towards
broadcast method for vehicular wireless communication systems,” in efficient geographic routing in urban vehicular networks,” IEEE Trans.
Proc. IEEE Int. Conf. Innovations Inform. Technol., 2011, pp. 371-374. Veh. Technol., vol. 58, no. 9, pp. 5048-5059, 2009.
[91] J. Toutouh, S. Nesmachnow, and E. Alba, “Fast energy-aware OLSR [117] J. Nzouonta, N. Rajgure, G. Wang, and C. Borcea, “VANET routing on
routing in VANETs by means of a parallel evolutionary algorithm,” city roads using real-time vehicular traffic information,” IEEE Trans.
Cluster Computing, pp. 435-450, 2013. Veh. Technol., vol. 58, no. 7, pp. 3609–3626, 2009.
[92] H. Cheng, and S. Yang, “Genetic algorithms with immigrants schemes [118] V. Vapnik, The nature of statistical learning theory. Springer, 2000.
for dynamic multicast problems in mobile ad hoc networks,” Eng. [119] A. Mellouk, S. Hoceini, and S. Zeadally, “A state-dependent time
Applications AI., vol. 23, no. 5, pp. 806-819, 2010. evolving multi-constraint routing algorithm,” ACM Trans. Autonomous
[93] M. Dorigo, V. Maniezzo, and A. Colorni, “Ant system: optimization by and Adaptive Syst., vol. 8, no. 1, 2013.
a colony of cooperating agents,” IEEE Trans. Syst., Man, Cybern. B, [120] Y. Tsypkin, Foundations of the theory of learning systems. Academic
Cybern. vol. 26, no. 1, pp. 29–41, 1996. Press, New York, 1973.
[94] I. Chakeres and C. Perkins, “Dynamic manet on-demand (DYMO) [121] L. Bottou, “Stochastic learning”, Advanced Lectures on Machine
routing,” IETF draft-ietf-manet-dymo-21, 2010. Learning, 2004, pp. 146-168.
[95] C. Sommer, and F. Dressler, “The DYMO routing protocol in VANET [122] G. Ding, and B. Bhargava, “Peer-to-peer file-sharing over mobile ad hoc
scenarios,” in Proc. IEEE VTC-2007 Fall, Baltimore, MD, USA, 2007, networks,” in Proc. 2nd IEEE Ann. Conf. Pervasive Comput. and
pp. 16–20. Communications Workshops, 2004, pp. 104-108.
[96] C. Perkins, E.M. Belding-Royer, and S. Das, “Ad hoc on-demand [123] J. Yu, T. Kim, W. Na, H. Bae, Y. Lee, J., Lee, and J., Hur, “Fully-
distance vector (AODV) routing,” IETF RFC 3561, 2003. distributed multicast routing protocol for IEEE 802.15. 8 peer-aware
[97] J. Härri, C. Bonnet, and F. Filali, “Kinetic mobility management applied communications,” in Proc. IEEE ICOIN, 2014, pp. 64-69.
to vehicular ad hoc network protocols,” Comput. Commun., vol. 31, no. [124] G. A. Di Caro, F. Ducatelle, and L. M. Gambardella, “BISON: biology-
12, pp. 2907–2924, 2008. inspired techniques for self-organization in dynamic Networks,”
[98] Network simulator, ns-2, Available: http:// www.isi.edu/ nsnam/ns/. Available: http://www. cs. unibo. it/bison/publications/KIbison.pdf,
[99] J. Martins, S. Correia, and J. Celestino, “Ant-DYMO: A bio-inspired 2005.
algorithm for MANETS,” in Proc. IEEE ICT, 2010, pp. 748–754. [125] A. Vahdat, and D. Becker, “Epidemic routing for partially connected ad
[100] R. R. Sahoo, R. Panda, D. K. Behera, and M. K. Naskar, “A trust based hoc networks,” Duke University, Tech. Rep. CS-200006, 2000.
clustering with Ant Colony Routing in VANET,” in Proc. IEEE [126] H. Kitano, “Biological robustness,” Nature Reviews Genetics, vol. 5, no.
ICCCNT, Coimbatore, India, 2012, pp. 1-8. 11, pp. 826-837, 2004.
[101] F. Chiang, and J. I. Agbinya, “Bio-inspired topology convergence [127] S. Eum, S. I. Arakawa, M. Murata, “Toward bio-inspired network
algorithms in resource-constrained VANETs,” in Proc. Australian robustness-Step 1. Modularity,” in Proc. 2nd IEEE Bio-Inspired Models
Telecommun. Netw. Applications Conf., 2007, pp. 1-6. of Netw. Inform. Comput. Syst., Bionetics, 2007, pp. 84-87.
[102] E. Royer, and C. E. Perkins “Multicast operation of the ad-hoc on- [128] H. G. Beyer, and H. P. Schwefel “Evolution strategies–A comprehensive
demand distance vector routing protocol”, in Proc. ACM/IEEE Annu. introduction,” Natural Comput., vol. 1, no. 1, pp. 3-52, 2002.
Conf. Mobile Comput. and Netw., 1999, pp. 207-218. [129] A. E. Eiben, and J. E. Smith, Introduction to evolutionary computing.
[103] A. B. Souza, J. Celestino, F. A. Xavier, F. D. Oliveira, A. Patel, and M. Springer, 2003.
Latifi, “Stable multicast trees based on Ant Colony optimization for [130] J. R. Koza, Genetic programming: on the programming of computers by
vehicular Ad Hoc networks,” in Proc. IEEE ICOIN, 2013, pp. 101-106. means of natural selection. vol. 1, MIT press. 1992.
[104] C. J. Huang, Y. T. Chuang, D. X. Yang, I. F. Chen, Y. J. Chen and K. [131] P. Larrañaga, and A. J. Lozano, “Estimation of distribution algorithms:
W. Hu, “A mobility-aware link enhancement mechanism for vehicular A new tool for evolutionary computation,” vol. 2, Springer, 2002.
ad hoc networks,” EURASIP J. on Wireless Commun. and Netw., [132] P. Rocca, G. Oliveri, and A. Massa, “Differential evolution as applied to
Springer, vol. 2008, no. 24, 2008. electromagnetics,” IEEE Antennas Propag. Mag. Vol. 53, no. 1, pp. 38-
[105] University of Malaga, NEO research group, VANETs projects, “Malaga 49, 2011.
urban scenario, Available: http://neo.lcc.uma.es/ staff/ jamal /portal/ [133] R. Reynolds, and M. Ali, “Embedding a social fabric component into
?q=content/ malaga-scenario/. cultural algorithms toolkit for an enhanced knowledge-driven
[106] C. E. Perkins, and P. Bhagwat, “Highly dynamic destination-sequenced engineering optimization,” Int. J. Intelligent Comput. and Cybern., vol.
distance-vector routing (DSDV) for mobile computers,” Comput. 1, no. 4, pp. 563-597, 2008.
Commun. Review 24, no. 4, pp. 234-244, 1994.
[107] K. Z. Ghafoor, K. A. Bakar, M. van Eenennaam, R. H. Khokhar, and A.
J. Gonzalez, “A fuzzy logic approach to beaconing for vehicular ad hoc
networks,” Telecommun. Syst., pp. 1-11, 2011, pp. 139-149.
[108] S. Bitam, and A. Mellouk, “Bee life-based multi constraints multicast
routing optimization for vehicular Ad hoc networks,” J. Netw. Comput.
Applications, vol. 36, no. 3, pp. 981-991, 2013.
[109] D. T. Pham, E. Kog, A. Ghanbarzadeh, S. Otri, S. Rahim, and M. Zaidi,
“The bees algorithm – A novel tool for complex optimisation problems,”
in Proc. 2nd Int. Virtual Conf. Intelligent PROduction Machines and
Syst., Oxford, 2006, pp. 454-459.
[110] H. A. Abbass, “A single queen single worker honey-bees approach to 3-
SAT,” in Proc. ACM GECCO, San Francisco, USA, 2001, pp. 807-814.

25

1553-877X (c) 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.