Wireless Personal Communications (2022) 125:1857–1881

https://doi.org/10.1007/s11277-022-09637-7

Mobility Models for Internet of Vehicles: A Survey

M. Kezia1 · K. V. Anusuya1

Accepted: 7 February 2022 / Published online: 28 February 2022

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022

Abstract
The Internet of Vehicles framework largely relies on Vehicular Ad-hoc Network (VANET)
to achieve the vision of connected, smart cars with mobility as a core component. Since
deploying and testing a VANET in real-time is expensive, simulations are an essential
tool as part of research on vehicular communication. Moreover, including an appropriate
mobility model for a successful simulation of VANET is quite challenging because of its
less realistic nature. This paper presents a survey on vehicular mobility models, emphasiz-
ing the realistic nature of vehicular movements and their corresponding challenges.

Keywords Connected car · Internet of vehicles · Mobility models · Vehicular Ad-hoc

network

1 Introduction

The Internet of Vehicles (IoV) is a branch of Internet of Things (IoT) that connects smart
cars on the go to improve transportation safety and efficiency in real-time. IoV supports
inter-vehicle communication, warning signals, online vehicle status checking, intelligent
navigation, and rescue via cloud platform and wireless connectivity [1]. The IoV architec-
ture comprises five types of vehicular communications, namely, Vehicle-to-Vehicle (V2V),
Vehicle-to-Roadside unit(V2R), Vehicle-to-Infrastructure (V2I) of mobile networks, Vehi-
cle-to-Personal devices or Pedestrians (V2P), and Vehicle-to-Sensors (V2S), as shown in
Fig. 1. Specifically, wireless access technologies such as IEEE 802.11p, Wi-Fi, 4G/LTE,
CarPlay, Near Field Communication(NFC), and Media Oriented System Transfer (MOST)
are used to establish unique connections in IoV, as depicted in Table 1 [2].
V2V and V2R links play an essential role in the IoV framework through IEEE 802.11 p,
also known as Vehicular Ad hoc Network (VANET). VANET allows self-organized, high-
mobility vehicles to communicate without demanding any specialized infrastructure.
Owing to high mobility, VANET is confronted with dynamic network topology, short-
lived connectivity, Spatio-temporal diversity in traffic, speed restrictions, roadblocks, and

* M. Kezia
1907nl05@psgtech.ac.in
K. V. Anusuya
kva.ece@psgtech.ac.in
1
ECE Department, PSG College of Technology, Coimbatore, Tamil Nadu, India

13
Vol.:(0123456789)
1858 M. Kezia, K. V. Anusuya

Fig. 1  Communication architecture in IoV

Table 1  Communication types in Communication type Technology

IoV architecture
V2V and V2R IEEE 802.11p
V2I Wi-Fi and 4G/LTE
V2P CarPlay/NFC
V2S MOST/Wi-Fi

atmospheric influences [3]. Hence, vehicular networks have evolved with improved proto-
cols and standards to address these challenges during network establishment.
Vehicular network performance is usually evaluated using actual testbed experiments
or software-based simulations. Testbeds analyze protocols or standards in real vehicles.
However, a real field test in VANET is not economically or technically feasible due
to varying topologies. Thus, simulation-based analysis by incorporating real vehicular
attributes is more feasible than actual tests for flexibility and cost. As a result, to gener-
ate a realistic simulation environment, the VANET requires the concept of a mobility
model [4].
The mobility model typically constitutes the movement patterns of nodes and spec-
ifies the changes that occur over time in acceleration, speed, position, and direction.
Moreover, it analyses how mobility affects the performance of a protocol. The valid-
ity of simulation results depends on the imitation of real-world scenarios by a mobility
model. Therefore, employing a realistic mobility model plays a vital role in VANET and
self-driving vehicles, trajectory predictions, and other applications of IoV. The subse-
quent sections discuss the basic building blocks of realistic mobility models and their
distinct types with the scope in VANET simulations.

13
Mobility Models for Internet of Vehicles: A Survey 1859

Fig. 2  Basic blocks in a realistic mobility model

2 Mobility Models

Mobility models describe the vehicular movements and their local properties (loca-
tion, speed and acceleration) in mobile network simulations. The fundamental units of
a mobility model are motion constraints and traffic generators [5]. Motion constraints
reveal the vehicle’s degree of freedom based on a topological map. Alternatively, traffic
generator creates and deals with interactions between cars based on the environment. As
depicted in Fig. 2, a realistic mobility model should consider:

• Topological maps comprising different lanes, intersections, and streets with associated
density and velocities.
• Driver behavior patterns in handling static and dynamic obstacles. Also, mutual inter-
actions model for preferred path selection, overtaking, and traffic jams.
• Deceleration and acceleration models between attraction or repulsion points as vehicles
do not abruptly break and move.
• Different times, such as rush hours and regular hours to observe heterogeneous traffic
density.

Since the existing mobility models focus on mimicking different real-world scenarios
mentioned earlier, they differ as per [6]. Based on the type of mobility scenario, these
models can be classified as random, flow, traffic, trace, survey, behavioral, and learning-
based models, as shown in Fig. 3. Here, the mobility models that are majorly used for
evaluating the performance of the VANET routing protocols are discussed.

13
1860 M. Kezia, K. V. Anusuya

Fig. 3  Classification of mobility models

2.1 Random Models

Random mobility models are renowned for their stochastic properties in simulations of
the entire mobile networks. Here, nodes move randomly, and hence mobility parameters
such as speed and distance originate from the random process. Further, these models
imitate mobility patterns for specific applications while maintaining their stochastic
properties intact [7]. The limited interaction between vehicles is a significant drawback
in these models compared to their simplicity and ease of implementation. Since these
models aren’t realistic for vehicular network simulations, we selected only the basic ran-
dom models.

2.1.1 Random Way Point (RWP) Model

Initially, it samples a randomly selected destination (waypoint) d with constant velocity

v, followed by a pause time [8]. The next destination and velocity values are independent
of the previous values as it is memoryless. The nodes cannot model velocity slowdown
without previous speed in the memory; thus, sudden stops and sharp turns are inevitable.
Likewise, the assumption of constant velocity between waypoints is unrealistic in vehicular
mobility.

2.1.2 Random Walk (RWalk) Model

As an alternative to RWP, without considering the pause time, each node randomly gen-
erates azimuth θ and the journey time t instead of sampling a destination [9]. After the
determined time interval, each node receives new values of velocity and direction. But,

13
Mobility Models for Internet of Vehicles: A Survey 1861

sudden turns, absence of a particular destination, and pauses are rarely observed in vehicu-
lar movements.

2.1.3 Random Direction Mobility (RDM) Model

Nodes in this model move randomly with the direction and speed until they reach the
boundary line [10]. A node selects future direction and speed after pausing at the edge.
The nodes are distributed uniformly during the simulation since no nodes accumulate at
the center. However, RDM suffers from an average speed decay problem that reduces the
node’s velocity over time.

2.1.4 Reference Point Group Mobility (RPGM) Model

This model mimics the movement of a group of nodes in an area. Each group includes a
leader, and the members are evenly distributed within a limited range. According to [11],
the leader moves (vleader (t) and θleader(t)) along predefined paths at a limited speed as per
the RDM model. The group member’s movement (vmember(t) and θmember(t)) is slightly dif-
ferent from the leader’s, as illustrated in (1, 2).
vmember (t) = vleader (t) + 𝜇.𝜈max .dev, 0 ≤ 𝜈max .dev ≤ 1 (1)

θmember (t) = θleader (t) + 𝜇.θmax .dev, 0 ≤ θmax .dev ≤ 1 (2)

where vmax. dev and θmax. dev represent the maximum deviation ratio of the respective
maximum velocity and heading of the group leader and µ is the stochastic parameter. It
is a memory-based model and controls the speed and direction of the entire group. How-
ever, the problem of sudden alterations in the movement persists in this model. Table 2 and
Fig. 4 give the comparative summary and basic movement patterns of the above random
models respectively.
To enhance the realism of random models, the movement of cars is restricted on street
graphs such as Freeway and Manhattan models. While the former restricts the vehicle
movement on several bidirectional multilane freeways, the latter limits the movements to
urban grids, as in Fig. 5 [4].
Both these models facilitate an essential interaction between cars. However, such inter-
action is too limited to impact traffic flow. In both models, vehicular mobility is modeled
according to the following rules.
vi (t + Δt) = v(t) + 𝜏. a. Δt (3)

if Δxi (t) > Δxsafe then vi (t) ≤ vi+1 (t) ∀i (4)

Equation (3) illustrates that the present speed temporally depends on the previous speed.
From (4), it is clear that the velocity of vehicle i is less than that of the preceding one i + 1.

2.2 Flow Models

Flow models provide extensive modeling details by considering the physical interaction
between different vehicles and the road environment [12]. These models consider vehicular
mobility as flow and are most commonly used in traffic safety applications. The literature

13
 1862

13
Table 2  Comparative summary of random models
Random models Velocity Direction/ Des- Pause Time Entity Group Car-to-car Memory Based Map-based Obstacles
tination interaction

RWP ✓ ✓ ✓ ✓ ☒ ☒ ☒ ☒ ☒
RWalk ✓ ✓ ✓ ✓ ☒ ☒ ☒ ☒ ☒
RDM ✓ ✓ ✓ ✓ ☒ ☒ ☒ ☒ ☒
RPGM ✓ ✓ ✓ ☒ ✓ ☒ ✓ ☒ ☒
M. Kezia, K. V. Anusuya
Mobility Models for Internet of Vehicles: A Survey 1863

(a) RWP (b) RWalk (c) RDM (d) RPGM

Fig. 4  Movement Patterns in random mobility models

presents three categories of flow models based on microscopic, macroscopic, and meso-
scopic approaches.

2.2.1 Microscopic Model

Microscopic models describe the mobile parameters of a particular car with respect to
other vehicles. It generally specifies acceleration or Deceleration to maintain a safe dis-
tance or time headway from another vehicle. At the same time, it offers a high degree of
precision and advanced computational complexity.

2.2.1.1 Car Following Model (CFM) The car-following model is among the most popular
driver models [5]. It focuses on safe distances between vehicles and safe headways based on
car dynamics or a combination of both to prevent accidents from happening. In CFM, the
safe distance between the following vehicle and the vehicle ahead is defined by Pipe’s rule.
The safe distance between two cars is at least the length of the vehicle for every 16.1 km/h
(10.41 miles/h) speed of the following car. According to this rule, Collision Avoidance (CA)
is based on the safety distance formula as in (5).

Δxsafe vi = L + T.vi + Ψ.v2i (5)

( )

L refers to vehicle length, T, the safe time headway (also the reaction time), and ψ. vi2,
the braking distance. ψ is an adjusting parameter function on the Deceleration of both vehi-
cles. Figure 6 from [13] illustrates the general schema of car-following models.
The Gipps model [14] is a popular CFM, based on CA or the Safety Distance approach. It
prevents accidents with the preceding and following vehicles even at sudden halts. Also, the
stimulus-response approach is introduced by a group of researchers from General Motors [15].
Response = sensitivity ∗ stimulus (6)
The GM model defines a vehicle’s acceleration at time t as a function of the speed and dis-
tance between two cars at time t − T.
dvi (t) Δvi (t − T)
= c.vm
i (t) (7)
dt Δxil (t − T)

Recently emerging Connected Autonomous Vehicles (CAV) push the study of CFM to
a new level. The interaction between Human Driver Vehicle (HDV) and a higher autopilot
differs from the traditional car following operation. Because CAVs and HDVs have different

13
1864 M. Kezia, K. V. Anusuya

Fig. 5  Street Graphs: a Freeway model and b Manhattan model

Fig. 6  General schema of car-following models, adapted from [13]

Fig. 7  Speed change in CAV follower and HDV follower, adapted from [17]

distributions of mobility properties, especially CAVs have high speed and enormous dis-
tance headway over HDVS. In [16], a car-following experiment conducted in mixed traffic
uses speed difference and distance headway between the front and rear vehicles, front vehicle
acceleration as per (8), (9), (10)

an (t) = 𝛿 ⋅ an−1 (t) + 𝜂 ⋅ sn (t) − s∗n (t) + 𝜃 ⋅ vn−1 (t) − vn (t) (8)
[ ] [ ]

(9)
[ ]
sn (t) = yn−1 (t) − yn (t)

v (t) ⋅ Δv (t)
s∗n (t) = s0 + 1 + U dhobs ⋅ Tvn (t) + n √ n
� � ��
(10)
2 ab

13
Mobility Models for Internet of Vehicles: A Survey 1865

The cooperation between leader vehicle and following CAV or HDV vehicle is demon-
strated in Fig. 7 [17] and Fig. 8 [18].

2.2.1.2 Intelligent Driver Models (IDM) The IDM models are also based on stimulus-response
and are predominantly used in the VanetMobiSim traffic simulator [19]. This model computes
the instantaneous acceleration and Deceleration of vehicles. A difference between the current
gap ∆xi and the desired gap δ serves as a stimulus to the next car, which is either trying to catch
up with a pulling-away leader or slowing down with a closing-in leader [20].

afree = a ⋅ 1 − vi ∕vdes (11)

[ ]
i

Fig. 8  CAV and HDV followers’ acceleration with respect to the distance headway, adapted from [18]
)2
aint = −a ⋅ 𝛿 Δxi (t) (12)
( /

It is clear from Eqs. (9), (10) that the instantaneous acceleration obtained based on
deterministic stimuli is inadequate and thus, fails to model irrational behaviors at inter-
sections. Hence, in the IDM extended model for Intersection Management (IDM_IM), the
vehicle maintains current speed for the green light and forcefully decelerates to stop for a
red light at the intersections [21]. IDM_IM model is extended for lane changing situations
as IDM_LC model [22]. A vehicle can change lanes and overtake another vehicle when the
vehicle is not moving at the driver’s desired speed and crosses the minimum safe distance
between two consecutive vehicles as per (13), (14).

al − a ± abias > p acurr + anew − alnew − alcurr + athr (13)

( )

alnew > −asafe (14)

Recently proposed Autonomous Vehicle based IDM (AV_IDM) in [23] computes

human-like driver behavior for AV using LiDAR measurements. It considers surrounding
vehicles in four directions (front, rear, right and left) to calculate average velocities in lon-
gitudinal and lateral directions in the mixed traffic environment.

2.2.1.3 Krauss Model The Krauss model, which computes future speed at time step t + ∆t
by a vehicle, can be categorized under the pure stimulus-response approach [24]. Further,
this model tries to mimic human sporadic and irrational reactions with stochastic parameter

13
1866 M. Kezia, K. V. Anusuya

µ. Krauss model defines the following equations by considering maximum speed, maximum
acceleration, and Deceleration as inputs.
safe
vi (t + Δt) = vi+1 (t) ⋅ TΔvi (t)∕(2 ⋅ b + T) (15)

[ ]
safe
vdes
i (t + Δt) = min v max , vi (t) + a ⋅ Δt ⋅ vi
(t + ΔT) (16)

vi (t + Δt) = max 0, vdes (17)

[ ]
i (t + Δt) − 𝜇

xi (t + Δt) = xi (t) + vi (t) ⋅ Δt (18)

The traffic simulator SUMO [25] uses the Krauss model.

2.2.1.4 Wiedemann Models The Wiedemann model under the psycho-physical category
considers perceptual psychology for a stimulus–response. Different drivers can send out
multiple individual responses to the same stimulus possibly. Hence, the Wiedemann
model distinguishes four driving states in a driver: Accessible driving mode, Approach-
ing mode, Following mode, and Braking mode and controls its response to a similar
stimulus [26]. This model is used in the commercial traffic simulator VISSIM [27].

2.2.1.5 Cellular Automata (CA) Models The CA models are discrete in time and space
with reduced computational complexities in contradiction to CFM models. These models
can imitate drivers’ reactions to the environment. Here, the traffic system has a lattice of
equal size cells. The size is chosen based on the movement of a single vehicle from one
cell to the next in one step time (∆t), and the movement can be controlled using rules.
Nagel and Schreck Enberg (N-SCHR) model is the most popular CA model [28] with the
following rules:

(19)
( )
Acceleration ∶ vi (t + 1) = min vmax , vi (t)

(20)
( )
Breaking ∶ vi (t + 1) = min vi (t), Δxi (t)

(21)
( ( ))
Randomization ∶ vi (t + 1) = rand 0 … vmax − 1 with probability p

Moving ∶ xi (t + 1) = xi (t) + vi (t + 1) (22)

here many vehicles limit the maximum speed and cause unrealistic brakes. Nevertheless,
the model can be used efficiently for vehicular traffic in TRANSIMS [29] traffic simulator
from the US Los Alamos Laboratory.

2.2.2 Macroscopic Model

In macroscopic models, quantities like vehicle flow m(x, t), velocity v(x, t) and den-
sity 𝜌(x, t) are modeled for a small road segment x at time t, from hydrodynamics phe-
nomena. While density represents the expected number of vehicles in x at time t, the

13
Mobility Models for Internet of Vehicles: A Survey 1867

expected number of cars passing x during the time interval [t, t + ∆t] represents the flow.
Velocity is the expected speed of vehicles in x. The Eq. (23) describes the relationship
among flow, speed, and density, and the Eq. (24) represents vehicular density in x that
varies with the incoming and outgoing flows in x.
v(x, t) = m(x, t)∕𝜌(x, t) (23)

𝜕𝜌 (x, t)∕𝜕t + 𝜕m (x, t)∕𝜕x = 0 (24)

Under this category, the most popular Light hill–Whitham–Richard (LWR) fluid model
[30] expresses the velocity as a density function.
v(x, t) = v(𝜌(x, t)) (25)
Due to its rearranged computational intricacies, the LWR model can be applied for
large-scale traffic simulations and kinematic waves design. However, the accuracy levels
are found lesser than those obtained in microscopic models. Also, it lacks in speed adap-
tion effects in the urban environment while transiting from a sparse area to a dense area
and vice versa. LWR model extensions [31] relax this assumption by adding inertia to
the speed. Further, the numerical implementation of macroscopic models also exists with
approximations of continuum models in spatial and temporal aspects. The advantages of
fluid motion models are the gross characterization of vehicular mobility and potentially
reduced computational load.

2.2.3 Mesoscopic Models

The mesoscopic approach represents gross traffic characteristics at the macroscopic level
and individual interaction between vehicles at the microscopic level. Recently, automatic
cruise control and automatic driving have been studied using this model. The mesoscopic
approach can model the headway distribution and density of a cluster of vehicles at a spe-
cific time and space. The vehicle behavior is also simultaneously controlled.
The Prigogine and Herman introduced the Gas-kinetic traffic flow model [32] offers
better performance than LWR or Payne models under sparse traffic, where the interaction
between vehicles is rare or absent. In this model, Phase Space Density (PSD) (x, v, t) rep-
resents the distribution function of cars, which is the expected number of vehicles in an
interval dx, dv, dt around x, v, t. The macroscopic term (x, t) from PSD is defined as
∞

∫0 (26)
𝜌(x, t) = 𝜌(x, v, t)dv

Similarly, the macroscopic speed from PSD is calculated as

∞

∫0 (27)
v(x, t) = v(x, v, t)dv

Queue model [33] is another mesoscopic approach, which considers a road segment
between two junctions as a FIFO queue. Each queue has a length lqueue, capacity C, free
flow speed v0, number of vehicles already in the queue nvehicle and the number of lanes nlane.
The travel time along the road is expressed as,

13
1868 M. Kezia, K. V. Anusuya

Fig. 9  Trip planning, adapted

from [34] Home

Shopping
Restaurant
Mall

Park

(28)
( )
ttravel = f lqueue , v0 , nlane , ncars

Another expression for ttravel (in homogeneous traffic), lqueue/v0 neglects the impact of
the vehicular density on speed and offers a good approximation of motion patterns. A vehi-
cle i enters a queue whenever it accesses the corresponding road. When the time expires,
vehicle i is extracted and entered into the next road queue. Restrictive measures are added
at this stage, because each road has finite incoming and outgoing capacities. A queue can-
not accept a car if it exceeds its incoming capacity. Further, a vehicle cannot leave a queue
if the maximum outgoing capacity is exceeded or unless it finds another line that would
accept the car.

2.3 Traffic Models

As flow models focus on the study of vehicle movements, traffic models work on a car’s
immediate neighborhood. According to flow models, an intersection slows down or decel-
erates the vehicle at turnings. Therefore, the flow models are highly recommended for
small-scale traffic flow studies, like lane merging impacts on flow. In contrast, traffic mod-
els study city-wide evacuation policy and the global journey from Origin to Destination
(OD). Also, these models design intersection policy (traffic light, stop sign, etc.) and turn-
ing policy (stochastic turn, pre-computed turns) in large-scale mobility [6].
Traffic models are categorized into trip and path models. Detailed trip models describe
the sequence of vehicles’ visiting points, while a path model describes car movement from
origin to destination. Additionally, departure time affects the nodes’ chosen path and tran-
sition probability. Modeling trip and path is far from random, as humans select OD points
not randomly. Moreover, OD points are chosen based on habits for optimal paths (fastest
path, least traffic path, etc.).

2.3.1 Trip Planning

In trip planning, vehicles from origin to destination are modeled using three differ-
ent approaches: random trip, stochastic turn, and the OD matrix. There is no correlation
between source and destination in random trips (refer to Sect. 2.1). A stochastic turn selects
a new direction at each intersection or other waypoint rather than a destination. The OD
matrix approach uses Points of Interests (PoI) and runs on the transition probabilities of
each PoI to correlate various trips, as shown in Fig. 9 [34].

13
Mobility Models for Internet of Vehicles: A Survey 1869

Table 3  OD matrix with Origin Dest.

transition probabilities
Home Restaurant Park Shopping mall

Home – 20% 30% 50%

Restaurant 40% – 20% 40%
Park 30% 30% – 40%
Shopping mall 50% 25% 25% –

Fig. 10  Path planning, adapted

from [34]

(a) Agent-centric path planning (b) Flow-centric path planning

Fig. 11  Agent centric path planning versus flow centric path planning, adapted from [35]

Transition probabilities represent the tendency of a vehicle to move from one PoI to
another. A specific OD matrix might be given to each car with certain transition prob-
abilities, or all vehicles might share a standard OD matrix as depicted in Table 3. Usually,
surveys are the primary source of information to identify OD points and estimate transition
probabilities. For instance, based on a survey, the OD matrix in Table 3 gives a transi-
tion probability of 50%, 25%, and 25% from shopping mall to home, restaurant, and park,
respectively. Trip models are typically assigned to a flow of vehicles to reduce complexities
in modeling large-scale mobility patterns.

2.3.2 Path Planning

Path planning assigns the sequence of directions to each vehicle once the origin and desti-
nation points are decided. A preferred optimization function of the driver, such as shortest
path, fastest path, less crowded path, etc., is used to reach its destination. For example, the
car can choose the shortest path from the three indicated paths in Fig. 10 [34].
Path selection for an extensive set of vehicles over a large area is quite challenging and
requires algorithms from graph theory with dynamic weights as the key parameters. For
instance, Dijkstra algorithms calculate link weights by distance, speed, density, habits,
and so forth. Scalability is an issue during simulations of many vehicles over high-concept

13
1870 M. Kezia, K. V. Anusuya

urban maps. To address this challenge, path planning may be described from the micro-
scopic or macroscopic angle in the simulation environment [5], as depicted in Fig. 11 [35].
Microscopic or agent-centric path planning controls individual vehicles and creates
one distinct path per vehicle. Regardless of computational complexity, a traffic accident
can be mathematically modeled here to reflect the impact on immediate and other vehi-
cles. This approach is employed in MATSim [36] and VanetMobiSim [19].
Flow-centric path planning, known as macroscopic path planning, increases scalabil-
ity by building a subset of paths for the vehicle flow. This plan reduces the compu-
tational complexity in pathfinding, as the number of ways is significantly lesser than
the number of vehicles. However, it cannot control a single vehicle for specific traffic
situations. The pre-computation of paths for the re-routing of cars under dynamic traffic
evolution is another limitation in this model. However, traffic simulators such as SUMO
[25], VISSIM [27], Aimsun [37], and CORSIM [38] use this model owing to its high
scalability. The comparative summary of traffic models is depicted in Table 4.

2.4 Trace and Survey Based Models

2.4.1 Trace Models

Many synthetic models fail to be feasible due to a complex mobility trace calibration. Thus,
trace and survey models were created to conserve time by capturing movement patterns
from real vehicles. Mobile nodes visit various locations through measurement campaigns
to collect these traces [39]. However, the availability of vehicle traces is limited and time-
constrained (possible usage is only after 6 months or a year). Additionally, patterns that
aren’t seen directly are hard to extrapolate. For example, an extrapolated bus trace cannot
be applied to personal vehicles. Yet, few traces discussed in this section can validate and
analyze new protocols and applications that have profound similarities in realistic vehicu-
lar motion. The differences between synthetic models and mobility traces are depicted in
Table 5 [39].

2.4.1.1 San Francisco Taxi Trace [40] A 30-day cab spotting project contains traces of
500 taxis in the San Francisco Bay area. GPS receivers trace geographic coordinates,
time stamps, and vehicle identifications on each cab. Even if a graph shows a taxi’s move-
ments, mobility traces can also be retained and available to the public.

2.4.1.2 UMass DieselNet [41] Vehicular DTN trace was developed by the University of
Massachusetts for transport research using a bus system in Amherst, USA. The 40 buses
in DieselNet had a desktop computer, including a GPS receiver and a 40 GB HDD.
802.111b radio Access Point (AP) allows traffic to access media and connections within
the AR range.

2.4.1.3 Shanghai Traces [42] The Shanghai Taxi GPS System collected the traces of
4316 taxis within 24 h. It contains Vehicle’s ID, longitude and latitude coordinates,
timestamp, driving speed, direction, and operational status. Data centers receive every
trace via the onboard GPS-enabled device.

13
 Table 4  Summary of Traffic models
Traffic models Correlation between Controls indi- Controls vehi- Complexity Car-to-car Simulators
Mobility Models for Internet of Vehicles: A Survey

source and destination vidual vehicle cles flow interaction

Trip Random trip ☒ ✓ ☒ ☒ ☒ NS-2 and NS-3

Stochastic turn ☒ ✓ ☒ ☒ ☒ NS-2 and NS-3
OD matrix ✓ ✓ ✓ ✓ ☒ MATSim [36] and VanetMobiSim
Path Agent-centric ✓ ✓ ☒ ✓ ✓ MATSim [36] and VanetMobiSim [22]
Flow-centric ✓ ☒ ✓ ✓ ✓ SUMO [25], VISSIM [27], Aimsun
[38], and CORSIM
1871

13
1872 M. Kezia, K. V. Anusuya

Table 5  Differences between Characteristics Synthetic mobility Mobility traces

synthetic mobility model and model
realistic mobility traces
Scalable Yes No
Reality Low High
Time overhead Low High
Deployment cost Low High
Storage memory Low High
Computation overhead High Low
Complexity High Low

Table 6  Comparative summary of trace models

Name of trace Mobility type Number of nodes Duration Device type Network

San Francisco Traces Vehicle (Taxi) 500 30 days GPS GPRS

UMass DieselNet Vehicle (Bus) 40 60 days Ha-Com Open 802.11b
Traces Brick with
AP
Shanghai Traces Vehicle (Taxi) 4316 24 h GPS GPRS
Morocco traces Vehicle (heterogene- 36 210 days GPS GPRS
ous)
Chicago Traces Vehicle (Bus) 1,736,431 18 days AVL GPRS

2.4.1.4 Morocco Traces [43] In these traces, GPS tracked the vehicle’s location in real-time
and sent the information to the central server according to event preferences. The traces from
36 heterogenous cars were collected in six months (approximately 214 days). It includes
recording, monitoring, reporting, alert management, and statistics generation features.

2.4.1.5 Chicago Traces [44] Chicago Transport Authority (CTA) bus tracker API provided
bus traces for 18 days. The Automated Vehicle Location (AVL) systems on the bus can track
the location of the bus and send position updates to the central server at CTA.

2.4.1.6 GATech Vehicular Traces [45] It contains GPS vehicular movement traces collected
during a network connectivity campaign on a highway segment.
The comparative summary of trace models is presented in Table 6.

2.4.2 Survey Models

On the other hand, surveys provide vital statistics about macroscopic mobility to calibrate
synthetic models and create new ones. The statistics of visitors’ commuting time, lunch-
time, travel distance, and preferred recreational situations help to generate pseudo-random
or deterministic behavior in cities. This category includes the University of Delaware
(UDEL), agenda-based, and Multi-agent Microscopic Traffic Model (MMTS). US Depart-
ment of Labor and Statistics, corporate research, and urban planning communities con-
ducted car mobility studies in the UDEL model [46].

13
Mobility Models for Internet of Vehicles: A Survey 1873

Fig. 12  Driver Behavior Modeling, adapted from [50]

The agenda-based mobility model [47] uses a personal plan for each node, including
social activities and geographic movements. This model collects activity distributions,
occupation distributions, and dwell-time distributions from the US National Household
Travel Survey. Ceitin et. al. proposed MMTS [48] to generate public and private vehicu-
lar traffic over accurate road maps of Switzerland within 24 h. This model used data
from census and other local or national mobility surveys or statistics. In a survey, only
coarse-grain mobility is provided; calibration of complex synthetic models and survey
models provides more detailed and realistic mobility representations.

2.5 Behavioral Model

In many synthetic models, driver reactions in specific situations are not anticipated. Fur-
thermore, drivers are not programmed to obey instructions on every occasion. Neglect-
ing human behaviors in designing vehicular mobility negatively impacts VANET appli-
cations. Traffic safety applications, for example, warning drivers to avoid accidents rely
on the driver’s compliance with traffic signals.
Generally, the behavioral model designs the driver’s response using either stimulus-
response or learning patterns based on the environment. This class incorporates inher-
ent advantages and disadvantages in modeling driver movements. The driver behavior
model can be developed in the sensing, reasoning, and application layers, respectively
[49]. Data is collected from vehicles, drivers, and the environment in the sensing phase.
As shown in Fig. 12 [50], the data is then fed into a reasoning engine to develop specific
service applications.
Initial attempts at modeling human behavior with stimulus-response utilized the
Wiedemann model described in the previous section. For example, Legendre et al. [51]
calculated movement vectors according to the sum of attraction and repulsion forces
between the target destination and obstacles in vehicular traffic. The complexity of such
computations can increase as the vehicle movements are prone to rapid changes at every
step.
Balmer used a multi-agent behavioral model [52] to explain how behaviors change with
learning curves. Each agent of this model reacts differently to the same stimuli depending

13
1874 M. Kezia, K. V. Anusuya

on their perception of the environment. Balmer et al. developed the multi-agent traffic sim-
ulator MATSim [36], which combines a dynamic agent behavior model and a mesoscopic
queue model for simulation. MatSim begins with a plan for the database and then uses a
mesoscopic queue model to execute each objective simultaneously. Plans are then scored
based on specific criteria. A strategy is then developed based on the scores. The process
(execution, scoring, strategy, plan change) is iterated until the strategy’s aim is fulfilled.
Espié et al. [53] model uses non-normative motion patterns to obtain more realistic
mobility patterns when the driver fails to respect traffic rules. Some more behavioral mod-
els with their applications are depicted in Table 7.

2.6 Learning Model

Rule-based techniques have lost their efficiency and precision when dealing with critical
traffic circumstances. This is due to the increased complexity of traffic mobility in modern
society. In order to improve the realism of VANET simulations, learning methods have
recently been incorporated into the community of mobility models. As a result, machine
learning techniques have become more popular in complex traffic settings to achieve vehi-
cle motion patterns.
Deep transport discussed in [62] based on deep Long-Short Memory Learning (LSTM)
method understands human mobility and transportation patterns from big and heterogene-
ous data. It automatically predicts a person’s future movements and transportation transi-
tion at the city level. For traffic speed prediction in [63], Look-up Convolution Recurrent
Neural Network (LC-RNN) is used to learn time-series patterns and spatial features of traf-
fic dynamics.
The learning-based model in [64] produces reliable virtual traffic flow in the SUMO
simulator by training a neural network with the observed real-world traffic data. A neural
network learns the motion pattern from accurate traffic data as shown in Fig. 13a [65],
instead of flow attributes at the macroscopic level. Further, the learned mobility model can
characterize a few line-changing actions of highway vehicles. However, the credibility of
the simulation decreases due to the accumulation of errors over a certain period.
A Hidden Markov Model (HMM) is also introduced in [64] to enhance the previous
model. HMM represents the temporal dependency of a future driving behavior relating
to the present behavior. A specification procedure by HMM has been added between the
data and neural network before sending the data to form a demonstration of examples
for learning as depicted in Fig. 13b [65]. The initialization of HMM requires a par-
tial knowledge of the pre-defined driving state with its probability representation on the
emitted observations. This defined probability distribution can be achieved by adapting
the traditional platoon car-following model. Thus, the hidden state with a definite driv-
ing pattern is predefined in the model. Finally, specific details are given to the neural
network to obtain the final mobility model for simulation. Since more precise data is
used in neural network training, the model’s increased performance increases the cred-
ibility of the simulation compared to the previous learning model.
In [65], A Mandatory Lane Changing (MLC) model based on Recurrent Neural Net-
works (RNN) was analysed for V2V links in a highway scenario. Microscopic mobility
metrics such as velocity, acceleration, desired minimum gap and safe gap between vehi-
cles are used to trian RNN hidden layers. RNN activates output in various lane changing
actions such as Free Lane Change (FLC), Forced Lane Change (FoLC), Cooperative

13
 Table 7  Recent behavioral models with their applications
Reference Modeling approaches Applications

Ahmed Amer et al.[54] The driver behavior is modeled after the yellow indicator (Dilemma Zone) and updated until the final ADAS (avoids accidents)
decision
Hussein dia et al. [55] (1) Compliance of drivers with the information model, Smart mobility, Sustainable transport
(2) Modeling drivers’ delay tolerance thresholds, and
Mobility Models for Internet of Vehicles: A Survey

(3) Modeling driver route utility

Bing et al. [56] Profile-Price-Profit (PPP) framework: Insurance pricing strategy
Predicts drivers’ behavior in the real-world, large-scale vehicle data and insurance data
Yassine et al. [57] Detects dangerous driving behaviors on the road and alerts other vehicles by disseminating warning mes- ADAS (avoids accidents)
sages on time
Manuel et al. [58] It includes finite reaction times, spatial and temporal anticipation, driving and estimation errors, and Anticipates future traffic situations
distractions caused by human drivers
Alberto Rossi et al. [59] Taxi drivers’ behavior and geographic information are modeled for the next destination Reduces traffic jam
Xiaoyuan Wang et al.[60] The Support Vector Machine (SVM) theory makes driver intention recognition model based on their emo- AI-driven interaction in vehicles
tions
Amirath et al. [61] Next route: predicts the future route of drivers in personal and global modules without loss of information Smart cities, ITS, Traffic congestion
forecasting, etc
1875

13
1876 M. Kezia, K. V. Anusuya

(a) Flow chart of Neural Network (NN) based method

(b) Flowchart of enhanced NN-based method with HMM

Fig. 13  Learning Models, adapted from [65]

lane change (CLC) based on safety gap between vehicles. The comparative overview of
mobility models is depicted in Table 8

3 Conclusion

The Internet of Vehicles paradigm promotes traffic safety and efficiency by sharing
information among vehicles through the VANET structure. Due to the high mobility of
cars, a proper choice of mobility model that reflects the real vehicular traffic is crucial
while testing VANET with critical performance. This paper presents the comparative
classification of various mobility models that are used in the simulation of vehicular
networks and their intrinsic limitations. Learning models that evolved recently can pre-
dict undefined vehicle behavior under dynamic traffic conditions. Further, incorporating
various types and scales of traffic data in urban and highway situations would improve
the authenticity of learning models.

Author Contributions KM have been involved in drafting the manuscript and revising it critically for impor-
tant literature review and AKV have edited the manuscript and given final approval of the version to be
published.

Funding No funding was received for conducting this study.

Data Availability Data sharing not applicable to this article as no datasets were generated or analysed during
the current study.

13
 Table 8  Comparative overview of mobility models
Mobility models Ease of Interaction between a Realistic Applicable Scenario Simulators Advantages DIsadvantages
Implemen- traffic and veicle move- movement
tation ments Patterns

Random models ✓ ☒ ☒ Urban ns-2, ns-3 *Flexible to use for differ- *Not realistic (Low preci-
ent scenarios sion)
Highway *Sudden alterations in the
movement
Flow models ✓ ✓ ✓ Urban VANETMobi- Useful for small scale traf- Can not model intersection
Highway Sim, SUMO, fic studies scenario
VISSIM,
Mobility Models for Internet of Vehicles: A Survey

TRANSIMS
Traffic models ✓ ✓ ✓ Urban SUMO, Useful for large scale Precomputation of paths
Highway VISSIM, traffic studies with under DTA is challenging
AIMSUN, intersections
CARSIM,
MATSIM
Trace models ☒ ☒ ✓ Urban SUMO Realistic nature (high Long time required for data
precision) collection
Highway
Survey models ☒ ☒ ✓ Urban – Models macroscopic flow Provides only coarse-grain
Highway with real time statistics mobility
Behavior models ☒ ✓ ✓ Urban MATSim Models human driving Complexity in designing
Highway pattern human response to par-
ticular stimuli
Learning models ☒ ✓ ✓ Urban – Predicts undefined vehicle Data collection is time
Highway behavior consuming
1877

13
1878 M. Kezia, K. V. Anusuya

Declarations
Conflict of interest The authors declare that they have no potential conflict of interest in relation to the study
in this paper.

References
1. Contreras-Castillo, J., Zeadally, S., & Guerrero-Ibanez, J. A. (2018). Internet of vehicles: Architec-
ture, protocols, and security. IEEE Internet of Things Journal, 5(5), 3701–3709. https://​doi.​org/​10.​
1109/​jiot.​2017.​26909​02
2. Kaiwartya, O., Abdullah, A. H., Yue, C., Ayman, A., Mukesh, P., Chin-Teng, L., & Liu, X. (2016).
Internet of vehicles: Motivation, layered architecture, network model, challenges, and future
aspects. IEEE Access, 4, 5356–5373. https://​doi.​org/​10.​1109/​access.​2016.​26032​19
3. Gasmi, R., & Aliouat, M. (2019). Vehicular Ad Hoc NETworks versus internet of vehicles - A com-
parative view. International Conference on Networking and Advanced Systems (ICNAS), 2019, 1–6.
4. Bai, F., & Helmy, A. (2004). A survey of mobility models. Wireless ad hoc networks (pp. 16–55).
University of Southern California.
5. Harri, J., Filali, F., & Bonnet, C. (2009). Mobility models for vehicular ad hoc networks: a survey
and taxonomy. IEEE Communications Surveys & Tutorials, 11(4), 19–41. https://​doi.​org/​10.​1109/​
SURV.​2009.​090403
6. Jérôme, H. (2010). Vehicular mobility modelling for VANET. In Vehicular applications and inter-
networking technologies (pp. 107–156).
7. Bai, F., Sadagopan, N., & Helmy, A. (2003). The IMPORTANT framework for analyzing the
impact of mobility on performance of routing protocols for Adhoc networks. Ad Hoc Networks,
1(4), 383–403. https://​doi.​org/​10.​1016/​s1570-​8705(03)​00040-4
8. Josh, B., Maltz, D. A., Johnson D. B. (1998). A performance comparison of multi-hop wireless ad
hoc network routing protocols. In ACM/IEEE international conference on Mobile computing and
networking (pp. 1–13).
9. Gaikwad, D. S. & Zaveri, M. (2011). A Novel mobility model for realistic behaviour in Vehicular
Ad Hoc Networks In 11th IEEE international conference on computer and information technology
(pp. 597–602).
10. Safaei, B., Mohammadsaleh, A., Khoosani, K. T., Zarbaf, S., Monazzah, A. M. H., Samie, F.,
Bauer, L., Henkel, J., & Ejlali, A. (2020). Impacts of mobility models on RPL-based mobile IoT
infrastructures: An evaluative comparison and survey. IEEE Access, 8, 167779–167829. https://​doi.​
org/​10.​1109/​ACCESS.​2020.​30227​93
11. Bai, F., Sadagopan, N., & Helmy, A. (2003). Important: a framework to systematically analyze the
impact of mobility on performance of routing protocols for ad hoc networks. In Proceedings of the
22nd annual joint conference on the IEEE computer and communications societies (INFOCOM
’03) (Vol. 2, pp. 825–835).
12. Hoogendoorn, S. P., & Bovy, P. H. L. (2001). State-of-the-art of vehicular traffic flow modelling.
Proceedings of the institution of mechanical engineers. Part I: Journal of Systems and Control
Engineering, 215(4), 283–303. https://​doi.​org/​10.​1177/​09596​51801​21500​402
13. Harri, J., Filali, F., & Bonnet, C. (2009). Fig. 10: General schema for car following models. Mobil-
ity models for vehicular ad hoc networks: A survey and taxonomy. IEEE Communications Surveys
& Tutorials, 11(4), 19–41.
14. Gipps, P. G. (1981). A behavioural car-following model for computer simulation. Transportation
Research Part B: Methodological, 15(2), 105–111. https://​doi.​org/​10.​1016/​0191-​2615(81)​90037-0
15. Chandler, R. E., Herman, R., & Montroll, E. W. (1958). Traffic dynamics: Studies in car following.
Operations Research, 6(2), 165–184. https://​doi.​org/​10.​1287/​opre.6.​2.​165
16. Ding, S., Chen, X., Fu, Z., & Peng, F. (2021). An extended car-following model in connected and
autonomous vehicle environment: Perspective from the cooperation between drivers. Journal of
Advanced Transportation. https://​doi.​org/​10.​1155/​2021/​27391​29
17. Ding, S., Chen, X., Fu, Z., & Peng, F. (2021). Figure 12: Speed variation during the following pro-
cess. (a) Speed variation of CAV follower. (b) Speed variation of HDV follower. An extended car-
following model in connected and autonomous vehicle environment: Perspective from the coopera-
tion between Drivers. Journal of Advanced Transportation. https://​doi.​org/​10.​1155/​2021/​27391​29
18. Ding, S., Chen, X., Fu, Z., & Peng, F. (2021). Figure 13: Variation of acceleration and dis-
tance headway during the following process. (a) CAV follower. (b) HDV follower. An extended

13
Mobility Models for Internet of Vehicles: A Survey 1879

Car-following model in connected and autonomous vehicle environment: Perspective from the
cooperation between drivers. Journal of Advanced Transportation. https://​doi.​org/​10.​1155/​2021/​
27391​29
19. VanetMobiSim Project official website. http://​vanet.​eurec​om.​fr
20. Treiber, M., Hennecke, A., & Helbing, D. (2000). Congested traffic states in empirical observations
and microscopic simulations. Physical Review E, 62(2), 1805–1824. https://​doi.​org/​10.​1103/​physr​
eve.​62.​1805
21. Liu, H., Sun, D., & Zhao, M. (2016). A model prediction control based framework for optimization of
signaled intersection: A cyber-physical perspective. Optik, 127(20), 10068–10075. https://​doi.​org/​10.​
1016/j.​ijleo.​2016.​07.​094
22. Tian, J., & Meng, F. (2020). Comparison survey of mobility models in vehicular ad-hoc network
(VANET). In IEEE 3rd international conference on automation, electronics and electrical engineering
(AUTEEE) (pp. 337–342). doi:https://​doi.​org/​10.​1109/​AUTEE​E50969.​2020.​93155​83
23. Sharath, M. N., & Velaga, N. R. (2020). Enhanced intelligent driver model for two-dimensional motion
planning in mixed traffic. Transportation Research Part C: Emerging Technologies, 120, 102780.
https://​doi.​org/​10.​1016/j.​trc.​2020.​102780
24. Krauss, S., Wagner, P., & Gawron, C. (1997). Metastable states in a microscopic model of traffic flow.
Physical Review E, 55(5), 5597–5602. https://​doi.​org/​10.​1103/​physr​eve.​55.​5597
25. SUMO Official Website. https://​www.​eclip​se.​org/​sumo/
26. Wiedemann, R.. (1974). Simulation des Straenverkehrsflusses, PhD thesis, Schriftenreihe des Instituts
fur Verkehrswesen der Universitat, Karlruhe 8, Germany, 1974.
27. VISSIM official website. https://​www.​ptvgr​oup.​com/​en/​solut​ions/​produ​cts/​ptv-​vissim/
28. Nagel, K., & Schreckenberg, M. (1992). A cellular automaton model for freeway traffic. Journal de
Physique I, 2(12), 2221–2229. https://​doi.​org/​10.​1051/​jp1:​19922​77
29. Smith, L., Beckman, R., & Baggerly, K. (1995). https://​code.​google.​com/p/​trans​ims/
30. Lighthill, M. J., & Whitham, G. B. (1995). On kinematic waves: II. A theory of traffic flow on long
crowded roads. Proceedings of the Royal Society of London A, 229, 317–345.
31. Jin, W. L. (2013). A multi-commodity Lighthill–Whitham–Richards model of lane-changing traf-
fic flow. Transportation Research Part B: Methodological, 57, 361–377. https://​doi.​org/​10.​1016/j.​trb.​
2013.​06.​002
32. Munjal, P., & J.pahl. (1961). An analysis of the Boltzmann-type statistical models for multi-lane traffic
flow. Transportation Research, 3(1), 90112–90121.
33. Cetin, N., Nagel, B. A., & K (2003). A large-scale agent based traffic microsimulation based on queue
model. In Swiss transport research conference (STRC). Monte Vertia.
34. Jérôme, H. (2010). Figure 5.1: The multilayer modeling concept of flow, path, and trip modeling as
addressed in this chapter Vehicular mobility modelling for VANET. Vehicular Applications and Inter-
Networking Technologies, 107–156.
35. Jérôme, H. (2010). Figure 5.15: Agent-centric versus flow-centric path planning. Vehicular mobility
modeling for VANET. Vehicular Applications and Inter-Networking Technologies, 107–156.
36. MATSim official website. (2009). http://​www.​matsim.​org
37. AIMSUN official website. (2009). http://​www.​aimsun.​com
38. CORSIM: Microscopic traffic simulation model. https://​ops.​fhwa.​dot.​gov/​traff​i cana​lysis​tools/​corsim.​
htm
39. Batabyal, S., & Bhaumik, P. (2015). Mobility models, traces and impact of mobility on opportunistic
routing algorithms: A survey. IEEE Communications Surveys & Tutorials, 17(3), 1679–1707. https://​
doi.​org/​10.​1109/​COMST.​2015.​24198​19
40. The Cab spotting Project. (2006). https://​stamen.​com/​work/​cabsp​otting/
41. CRAWDAD Data Set Dieselnet/UMass. https://​crawd​ad.​org/​umass/​diesel/​20080​914/
42. Feng, H., & Youji, X. (2016). An empirical study on evolution of the connectivity for VANETs based
on taxi GPS traces. International Journal of Distributed Sensor Networks. https://​doi.​org/​10.​1155/​
2016/​25804​65
43. Ibadah, N., Minaoui, K., Rziza, M., Oumsis, M., & Benavente-Peces, C. (2018). Smart collection of
real-time vehicular mobility traces. Future Internet, 10, 78.
44. Michael, D., Tobias, P., Wolf-Bastian, P., Lars, W. (2010). A new mobility trace for realistic large-
scale simulation of bus-based DTNs. In Proceedings of the 5th ACM workshop on challenged networks
(CHANTS ’10) (pp. 71–74). Association for Computing Machinery. doi:https://​doi.​org/​10.​1145/​18599​
34.​18599​50
45. Fujimoto, R., Guensler, R., Hunter, M., et al. (2006). CRAWDAD data set GAtech/vehicular (v. 2006–
03–15). http://​crawd​ad.​cs.​dartm​outh.​edu/​gatech/​vehic​ular

13
1880 M. Kezia, K. V. Anusuya

46. UDel Models for Simulation of Urban Mobile Wireless Networks. (2007). http://​udelm​odels.​eecis.​
udel.​edu/
47. Zheng, Q., Hong, X., & Liu, J. (2006). An agenda-based mobility model. In 39th IEEE annual simula-
tion symposium (pp. 188–195). Huntsville, AL, USA. doi:https://​doi.​org/​10.​1109/​ANSS.​2006.​11
48. Realistic Vehicular Traces from the Multi-agent Microscopic Traffic Simulator (MMTS). (2006).
https://​www.​lst.​inf.​ethz.​ch/​resea​rch/​ad-​hoc/​reali​stic-​vehic​ular-​traces.​html
49. Najah, A. A., & Hatem, A. Z. (2016). Driver behavior modeling: Developments and future directions.
International Journal of Vehicular Technology. https://​doi.​org/​10.​1155/​2016/​69527​91
50. Ali, N. A., & Abou-zeid, H. (2016). Figure 1: Driver behavior modeling (DBM): Sensing, applica-
tions, and future systems. Driver behavior modeling: Developments and future directions. Interna-
tional Journal of Vehicular Technology. https://​doi.​org/​10.​1155/​2016/​69527​91
51. Legendre, F., Borrel, V., De Amorim, M. D., et al. (2006). Reconsidering microscopic mobility mod-
elling for self-organizing networks. IEEE Network Magazine, 20(6), 4–12. https://​doi.​org/​10.​1109/​
MNET.​2006.​273114
52. Balmer M. (2007). Travel demand modeling for multi-agent traffic simulations: Algorithms and Sys-
tems. Ph.D. thesis. ETH Zurich, Switzerland.
53. Doniec, A., Espie, S., Mandiau, R., & Piechowiak, S. (2006). Non-normative behavior in multi-agent
system: Some experiments. In IEEE/WIC/ACM international conference on intelligent agent technol-
ogy (pp. 30–36). doi:https://​doi.​org/​10.​1109/​IAT.​2006.​96
54. Amer, A., Rakha, H., El-Shawarby, I. (2011). Agent-based behavioral modeling framework of driver
behavior at the onset of yellow indication at signalized intersections. In 14th international IEEE con-
ference on intelligent transportation systems (ITSC) (pp. 1809–1814). doi:https://​doi.​org/​10.​1109/​
ITSC.​2011.​60828​87
55. Dia H., & Panwai, S. (2014). Intelligent mobility for smart cities: Driver behaviour models for assess-
ment of sustainable transport. In: IEEE fourth international conference on big data and cloud comput-
ing (pp. 625–632). doi:https://​doi.​org/​10.​1109/​BDClo​ud.​2014.​50
56. He, B., Zhang, D., Liu, S., Liu, H., Han, D., & Ni, L. M. (2018). Profiling driver behavior for personal-
ized insurance pricing and maximal profit. In IEEE international conference on big data (Big Data)
(pp. 1387–1396). doi:https://​doi.​org/​10.​1109/​BigDa​ta.​2018.​86224​91
57. Hernafi, Y., Ben Ahmed, M., & Bouhorma, M. (2016). An approaches’ based on intelligent transpor-
tation systems to dissect driver behavior and smart mobility in smart city. In 4th IEEE international
colloquium on information science and technology (CiSt) (pp. 886–895). doi:https://​doi.​org/​10.​1109/​
CIST.​2016.​78050​13
58. Lindorfer, M., Mecklenbräuker, C. F., & Ostermayer, G. (2018). Modeling the imperfect driver: Incor-
porating human factors in a microscopic traffic model. IEEE Transactions on Intelligent Transportation
Systems, 19(9), 2856–2870. https://​doi.​org/​10.​1109/​TITS.​2017.​27656​94
59. Rossi, A., Barlacchi, G., Bianchini, M., & Lepri, B. (2020). Modeling taxi drivers’ behaviour for the
next destination prediction. IEEE Transactions on Intelligent Transportation Systems, 21(7), 2980–
2989. https://​doi.​org/​10.​1109/​TITS.​2019.​29220​02
60. Wang, X., Guo, Y., Bai, C., Yuan, Q., Liu, S., & Han, J. (2020). Driver’s intention identification with
the involvement of emotional factors in two-lane roads. IEEE Transactions on Intelligent Transporta-
tion Systems. https://​doi.​org/​10.​1109/​TITS.​2020.​29958​37
61. Amirat, H., Lagraa, N., Fournier-Viger, P., & Ouinten, Y. (2020). NextRoute: A lossless model for
accurate mobility prediction. Journal of Ambient Intelligence and Humanized Computing, 11(7),
2661–2681. https://​doi.​org/​10.​1007/​s12652-​019-​01327-w
62. Xuan, S., Hiroshi, K., & Ryosuke, S. (2016). Deeptransport: prediction and simulation of human
mobility and transportation mode at a citywide level. In Proceedings of the twenty-fifth international
joint conference on artificial intelligence (IJCAI’16) (pp. 2618–2624). AAAI Press.
63. Lv, Z., J. Xu, Kai Zheng, Hongzhi Yin, Pengpeng Zhao and X. Zhou. (2018). LC-RNN: A deep learn-
ing model for traffic speed prediction, IJCAI, 3470–3476.
64. Jian, Z. (2018). deep learning-based vehicular mobility models for intelligent transportation systems.
Ph.D Thesis, Automatic Control Engineering. Ecole Centrale de Lille.
65. Jian, Z. (2018). Figure 3.9: Differences of the flow chart between enhanced method and the original
one. Deep learning-based vehicular mobility models for intelligent transportation systems. Ph.D The-
sis, Automatic Control Engineering. Ecole Centrale de Lille.
66. Naskath, J., Paramasivan, B., & Aldabbas, H. (2021). A study on modeling vehicles mobility with
MLC for enhancing vehicle-to-vehicle connectivity in VANET. Journal of Ambient Intelligence and
Humanized Computing, 12(8), 8255–8264. https://​doi.​org/​10.​1007/​s12652-​020-​02559-x

13
Mobility Models for Internet of Vehicles: A Survey 1881

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

M. Kezia is pursuing a Ph.D. under National Doctoral Fellowship

Scheme initiated by AICTE in the Department of Electronics and
Communication Engineering at PSG College of Technology, Coim-
batore, India. Her main research areas are Vehicular Ad hoc network
and Machine Learning. She obtained her M. Tech. and B.Tech. in
Electronics and Communication Engineering from Jawaharlal Nehru
Technological University (JNTU), Anantapur, India and Yogi Vemana
University (YVU), Kadapa, India respectively.

K. V. Anusuya received her Ph. D in Information and Communication

Engineering from Anna University, Chennai and M.E degree from
Bharathiar University, Coimbatore in 2010 and 2004 respectively. She
is an Associate Professor in the Department of Electronics & Commu-
nication Engineering, PSG College of Technology, Coimbatore. She
has 27 years of experience from the Textile Electronics Industry and
Academics. Her research interests include, Wireless Embedded sys-
tems, Vehicular Networks, Wireless Sensor Networks, Cyber security
system models, Connected and Autonomous cars. She received RPS
grant from AICTE for the design and development of network hard-
ware for streaming video traffic and instrumental in setting up the
Industry sponsored Centres of Excellence at PSG College of Technol-
ogy. She is a life member of SSI and ISTE.

13