sensors

Article
Path Planning and Formation Control for UAV-Enabled Mobile
Edge Computing Network
Kheireddine Choutri 1 , Mohand Lagha 1 , Souham Meshoul 2, * , Samiha Fadloun 3

1 Aeronautical Sciences Laboratory, Aeronautical and Spatial Studies Institute, Blida 1 University,
Blida 0900, Algeria
2 Department of Information Technology, College of Computer and Information Sciences,
Princess Nourah bint Abdulrahman University, P.O. Box 84428, Riyadh 11671, Saudi Arabia
3 Ecole Nationale Supérieure d’Informatique (ESI), Alger 16309, Algeria
* Correspondence: [email protected]

Abstract: Recent developments in unmanned aerial vehicles (UAVs) have led to the introduction
of a wide variety of innovative applications, especially in the Mobile Edge Computing (MEC) field.
UAV swarms are suggested as a promising solution to cope with the issues that may arise when
connecting Internet of Things (IoT) applications to a fog platform. We are interested in a crucial
aspect of designing a swarm of UAVs in this work, which is the coordination of swarm agents in
complicated and unknown environments. Centralized leader–follower formations are one of the most
prevalent architectural designs in the literature. In the event of a failed leader, however, the entire
mission is canceled. This paper proposes a framework to enable the use of UAVs under different
MEC architectures, overcomes the drawbacks of centralized architectures, and improves their overall
performance. The most significant contribution of this research is the combination of distributed
formation control, online leader election, and collaborative obstacle avoidance. For the initial phase,
the optimal path between departure and arrival points is generated, avoiding obstacles and agent
Citation: Choutri, K.; Lagha, M.; collisions. Next, a quaternion-based sliding mode controller is designed for formation control
Meshoul, S.; Fadloun, S. Path and trajectory tracking. Moreover, in the event of a failed leader, the leader election phase allows
Planning and Formation Control for agents to select the most qualified leader for the formation. Multiple possible scenarios simulating
UAV-Enabled Mobile Edge real-time applications are used to evaluate the framework. The obtained results demonstrate the
Computing Network. Sensors 2022, capability of UAVs to adapt to different MEC architectures under different constraints. Lastly, a
22, 7243. https://doi.org/10.3390/ comparison is made with existing structures to demonstrate the effectiveness, safety, and durability
s22197243 of the designed framework.
Academic Editors: Weizhe Zhang
and Ibrahim A. Elgendy Keywords: Mobile Edge Computing; UAV; formation control; leader election; path planning;
obstacle avoidance; artificial intelligence; multi-agent systems
Received: 25 August 2022
Accepted: 20 September 2022
Published: 24 September 2022

Publisher’s Note: MDPI stays neutral 1. Introduction

with regard to jurisdictional claims in
The majority of Internet of Things (IoT) applications require a number of computa-
published maps and institutional affil-
tional tasks that cannot be performed locally on IoT devices due to the vast amount of data
iations.
that are generated and need to be processed in real time. Because of their greater processing
and storage capacity, external devices in centralized servers in remote clouds or at the edge
are used to perform these IoT tasks, which require offloading operations. Connecting an
Copyright: © 2022 by the authors.
Internet of Things network to a Mobile Edge Computing (MEC) platform, on the other
Licensee MDPI, Basel, Switzerland. hand, could be a challenging endeavor. The use of unmanned aerial vehicles (UAVs) is one
This article is an open access article possible solution to this issue. A survey concerned UAV-Enabled Mobile Edge Computing
distributed under the terms and for IoT Devices can be found on [1]. Depending on their role in the network, UAVs can
conditions of the Creative Commons be used as fog nodes [2] or as relays to improve the connectivity of a wireless network [3].
Attribution (CC BY) license (https:// In [4], a swarm of UAVs is employed in the fog layer to detect objects in real time videos.
creativecommons.org/licenses/by/ UAVs with onboard computing resources can provide offloading services to nearby
4.0/). mobile users (MUs). Within the constraints of the available computing power, numerous

Sensors 2022, 22, 7243. https://doi.org/10.3390/s22197243 https://www.mdpi.com/journal/sensors

Sensors 2022, 22, 7243 2 of 18

strategies were proposed to reduce the processing time among MUs. While [5] offers a
compute offloading strategy based on Deep Deterministic Policy Gradient (DDPG), Ref. [6]
uses Reinforcement Learning (RL) to enhance the Quality-of-Service for each terminal user.
Depending on the size and weight of the UAV, work offloading in UAV-enhanced edges
can be challenging. Due to the influence that UAV-reliant networks may have on QoS and
performance, it is vital that they be designed with energy efficiency in mind. The authors
of [7] offer an approach that simultaneously optimizes the offloading of tasks and the
UAV’s flying path. Two objective functions addressing energy-efficient offloading and safe
path planning are proposed in [8] as part of a constrained multi-objective optimization
problem for UAVs. The authors of [9] are primarily concerned with path planning under
the paradigm of edge computing. In addition, the authors of [10] present an algorithm for
creating a feasible task migration path, taking into account the migration distance between
UAVs, the load condition of UAVs, and environmental characteristics. The authors of [11]
describe a cloud-based motion planning method for coordinating the safe movement of a
large number of mobile robots.
The mobility of UAVs and their ever-increasing computational power paved the path
for their usage in a wide variety of novel applications. Recently, a significant amount of
work and attention has been focused on their application in the support of mobile edge
computing systems. Several different MEC systems that are aided by UAVs have been
proposed [12]. On the other hand, it has been reported that a single UAV-based solution
might not be adequate to complete the mission due to the limited life of its battery and
the limited computing power it possesses. A swarm of drones has many advantages
over a single drone, including increased reliability and decreased operation time. As a
consequence, there has been a surge in the use of swarms of drones rather than a single
drone. Dealing with a swarm of drones gives rise to a number of challenges that need to be
addressed among which, are multi-agent systems, path planning, and formation control.
Multi-agent systems have received much attention from the scientific community in
recent years. Multiple collaborating UAVs can complete more challenging jobs and achieve
more complicated objectives than a single UAV. Various techniques and designs have been
proposed in the literature, including behavior-based [13], virtual structure [14], potential
field [15], and leader–follower [16–18]. In the centralized leader–follower (L-F) scenario,
a “leader” agent has the reference motion that the other agents, “followers”, track it. To
facilitate collaboration, the leader conveys its state to the followers via an appropriate
communication channel or link. Consequently, a leader’s failure will result in mission
failure. The leader election is one of the recently researched challenges in leader–follower
settings. In most cases, the leader is assumed to be a specific agent picked at the start of
the job; this event is known as a static leader election. Ref. [19] offers a new leader election
approach based on an adaptive/reliable network structure. Ref. [20] offers a distributed
leader election method that does not require direct inter-agent communication. For online
leader election, the authors of [21] suggest a fully-decentralized adaptive technique capable
of selecting the best leader among the existing leader’s neighbors on a regular basis. This
author describes a distributed online leader election model. The swarm agents are adaptive
to unexpected leader failure and changes in network topology.
In formation control, the motion of UAVs is tightly restricted to preserve the formation
topology. Consensus algorithms have been explored extensively in the literature for this
purpose [22,23]. Numerous control mechanisms were proposed for UAV formation control,
with the underlying theory described in [24]. Ref. [25] suggests a second-order consensus
algorithm to follow a specified external reference in their work, whereas [26] frames
the formation control problem as a position control problem to be solved. In contrast,
spacecraft formation vehicles use a mechanism for robust attitude control, as described
in [23]. The prior control strategies were able to sustain the formation, requiring a position
estimate for both the leader and the followers. In addition, communication limitations
pertaining to the proposed typologies were not considered. A consensus-based attitude
Sensors 2022, 22, 7243 3 of 18

formation controller is utilized for this study. The formation topology is then preserved
with minimal data interchange, and the controller is resilient to external disturbances.
Avoiding other UAVs and natural obstacles is essential for mission success and safe
operation. There have been numerous proposed solutions to this problem. The avoidance
of obstacles by unmanned aerial vehicles is examined in [27]. In [28], the methods of conflict
identification and resolution for a multi-UAV collision-avoidance system are investigated.
Therefore, a modified technique is given for tentacle construction and collision avoidance
for multiple UAVs operating in unstructured environments. For use in the actual world
without the assistance of a pilot, Ref. [29] developed a system that can automatically navi-
gate and avoid obstacles. Moreover, Ref. [30] describes a decentralized control system based
on behavior, which employs a sliding mode controller and artificial potential functions.
Although other works have addressed the issue of avoiding obstacles when flying in a UAV
formation, nobody has yet optimized the resulting trajectory.
This research offers a new paradigm for multi UAVs, one that makes distributed
formation control operable across a wide range of MEC architectures, unlike previous
work in [31–34]. Therefore, algorithms for optimal trajectory tracking, obstacle avoidance,
and leader selection were incorporated into the resulting framework. The suggested frame-
work removes the problems associated with conventional leader–follower development
and produces better results. The following are the foundational requirements of the method
we propose:
• R1: Multi-agent systems (MAS): Multi-agent systems are computational systems
composed of a large number of interacting computer elements known as agents.
To take advantage of the decentralized structure of the multi-agent system, the agent
must be provided with some autonomy. When we state that an agent is autonomous,
we infer that it can collect data by interacting with other agents and its environment,
and then make decisions based on this information.
• R2: Formation control: According to the proposed method, swarms have a certain
topology, such as a rectangle, a diamond, etc. This allows the formation topology to be
maintained with little data sharing and makes the controller more resilient to external
disturbances.
• R3: Leader election: In centralized formations, there is often one leader of the swarm;
nevertheless, if a single error happens, the entire mission is aborted. In this situation,
the other agents must select a new leader.
• R4: Obstacle avoidance: In centralized formations, there is often one leader of the
swarm; nevertheless, if a single error happens, the entire mission is aborted. In this
situation, the other agents must select a new leader.
The remainder is organized as follows: In Section 2, the system is introduced with
a brief overview of graph theory and MEC architectures. The problem of path planning
and obstacle avoidance is formulated in Section 3. Section 4 is dedicated to formation
control. The formalization of the dynamic model using quaternions is described first.
Then, a suitable SMC controller is developed to preserve the topology of the formation.
Finally, the formation transformation for leader election is described. Section 5 examines the
simulation results, proposing numerous scenarios and comparing performance. Section 6
presents conclusions and future potential work.

2. Preliminaries
2.1. Mec Architectures
Depending on the role of UAVs in the network, three distinct MEC topologies can be
distinguished. In each scenario, the MEC server and MUs are the primary components,
which are modeled as data analysis centers located closer to the users on the network’s
edge. Using wireless networking and accessible communication technologies, links are
built between the different users and servers. Internet is then used to link the MEC server
to cloud-based data analysis centers [1].
Sensors 2022, 22, 7243 4 of 18

• Assisted MEC: This architecture is often used to provide services following a natural
disaster or bombardment-related infrastructure damage. As depicted in Figure 1,
the UAV assists mobile users by acting as an aerial MEC server-enabled base station.
Each user transfers its computationally intensive activities to one or many UAVs
for processing. Therefore, UAVs with long-lasting batteries and powerful CPUs are
necessary for this architecture. In addition, this architecture is typically employed to
satisfy QoS requirements by optimizing the overall energy consumed by the MUs.

Figure 1. Assisted MEC architecture.

• Cellular-Connected MEC: Figure 2 illustrates these kind of architectures. During a

mission, UAVs are viewed as aerial users with computationally intensive tasks, such as
path planning and data analysis. Due to limited onboard processing capability, UAVs
offload computation to an MEC server on a Ground Base Station (GBSs). In comparison
to the previous architecture, the UAVs deployed in this manner have limited batteries
and possessors, but they must conduct intense computation tasks.

Figure 2. Cellular-Connected MEC architecture.

Sensors 2022, 22, 7243 5 of 18

• Relayed MEC: As shown in Figure 3, the UAVs in this final architecture serve as relays
to help the MUs in offloading their intensive computation tasks to the MEC server of
the GBSs. Therefore, none of the UAVs include an MEC server. This architecture is
intended to enable long-distance communication links between the MUs and the MEC
server in the event that other regular links are interrupted.

Figure 3. Relayed MEC architecture.

2.2. System Presentation

The proposed framework consists of a swarm of UAVs assisting different MEC archi-
tectures in a distributed formation. We assume that only the leader is aware of the mission
task and all possible formation types in the L-F approach. Followers are expected to collect
data from other UAVs and the surrounding environment. The leader takes decisions on spe-
cific movements, planned trajectory, and allocation of tasks. The framework is constructed
according to the flow chart shown in Figure 4.

START

Path
planning

Formation
control

Leader Obstacles
failure dectection Else

Leader Split Computation

election topology offloading

Switch Obstacles END

topology avoidance

Figure 4. Proposed framework for UAVs swarm monitoring under distributed L-F formations.
Sensors 2022, 22, 7243 6 of 18

The mission is first planned by the user, who specifies the departure and arrival points,
the number of UAVs, the formation topology, and the employed MEC architecture. The path
planning stage then generates the best path between the departure and arrival points.
The control space is optimized, while some constraints in the output and state spaces
are applied. This includes physical constraints, actuator limitations, and environment
obstacles. Next, the formation control phase aims to preserve the formation’s topology
and accommodate any topological variations caused by constraints. Both the leader and
the followers have an autopilot system for attitude stabilization and trajectory tracking.
As a result, the leader receives the required position to control command for the autopilot
in order to arrive at the desired location. The main objective of the UAVs is to offload
their computation tasks to/from the MEC server depending on the selected architecture.
However, if a failure happens during the mission, the framework includes a safety stage to
cope with it. If the leader is unable to complete the task, the followers elect a new leader.
The technique seeks to select the most qualified UAV for leadership and switch topologies
based on the number of remaining UAVs. Furthermore, if an external obstacle is detected,
the leader instructs the followers to split the topology, avoid the obstacle, and return to the
original topology.
In this study, we represent the network topology of the UAV swarm using graph
theory notations. Nodes in an undirected graph represent network communication nodes,
while edges represent communication links between nodes. In the case of UAV swarms,
nodes represent individual UAVs and edges represent inter-UAV links, such as wireless
communications or sensors. We assume that all communication links are bidirectional.

2.3. Graph Theory

Abstractly, an undirected graph G = (ν, ε) is defined by a node set v with cardinality
n, the number of nodes in the graph, and an edge set E comprised of pairs of nodes, where
nodes νi and νj are adjacent if {νi , νj } ∈ ε ⊆ [v]2 . One special family of undirected graphs
are tree graphs, denoted by the set Γ, where all two-node pairs are connected by exactly
one simple path, that is, a connected graph without cycles. A spanning tree of a graph is a
tree sub graph that connects all vertices in the graph.
The neighborhood set N (vi ) of a node νi is composed of the set of nodes adjacent to νi .
The scalar d = (νi , νj ) is the minimum path length, induced by the graph G, between nodes
νi and νj . The adjacency matrix of G is a positive matrix denoted by G A = [ωija ] ∈ Rnn ,
where ωija represents the entry of the i-th row and j-th column of matrix G A with ωija = 1 if
(i, j) ∈ E, and ωija = 0, otherwise. The degree δi of node νi is the number of adjacent nodes.
The degree matrix G D is a diagonal matrix with δi at matrix element (i, i ). We also define
the diagonal matrix G L = diag{ω1l ,. . . , ωnl } representing the status of the agent—if ωil = 1,
then agent i is a leader. Otherwise, if ωil = 0, then agent i is a follower.
The Laplacian matrix graph is defined as L = G D − G A ∈ Rnn . It plays an important role
in the dynamics of the network. An important feature of this matrix is that it is a (symmetric)
positive semi-definite matrix. The spectrum is ordered as 0 = λ1 ( L) ≤ λ2 ( L) ≤ . . . ≤ λn ( L).
The interaction positive matrix G for the L-F formation is defined as follows:

G = GD − G A + GL = L + GL (1)
Based on graph theory, the L-F consensus’ desired trajectory for each agent can be
given by:
1
x̄id = ∑ j∈ Ni (( x j + dij )
| Ni |
i follower
1 (2)
x̄id = | Ni +1| ∑ j∈ Ni
( (( x j + dij ) + r + di0 ) i leader
where dij = di0 − d j0 is the inter-distance. Once an observation is available, Equation (2) is
reformulated based on the normalized interaction matrix G e as follows:
Sensors 2022, 22, 7243 7 of 18

xi − x̄id xi − xid
   
.. e ⊗ I2 ) ..
 = (G (3)
  
 .  . 
xn − x̄nd xn − xnd

For this study, a normalized interaction matrix is considered for L-F formations as
indicated in Equation (4):
Ge = ( G D + G L ) −1 · G (4)

Extra edges can be added to the proposed topology as shown in Figure 5. The added
two edges and the same vertices constitute the graph Ge = (ν, (e)e ). Then, Ge and G
e yield:

e = G + Ge
G (5)

Figure 5. Leader–follower formation topology.

3. Path Planning Stage

The main objective of the path planning stage is to generate the optimal path between
the departure and arrival points. A collision avoidance system should often be incorpo-
rated into the control architecture to protect UAVs from causing damage to themselves and
surrounding objects (other vehicles, persons, and infrastructures). Thus, obstacle avoidance
techniques are used to tackle problems with trajectory optimization challenges. The major-
ity of issues in the literature are addressed using dynamic programming, with trajectory
optimization occurring throughout the challenge. In this study, however, the obstacle
avoidance problem is seen as a Constrained Optimization (CM) problem, with the objective
of optimizing the problem’s key parameters: the UAVs desired inter-distance (dd ) and the
radius of the obstacle avoidance zone (Rmax ).
Next, an energy-aware communication and computation resource is incorporated for
the aim of minimizing the total energy consumption. First, the energy consumed by MUs
and UAVs while processing tasks is included in the task processing energy consumption.
According to capacitance theory [10], the processing energy consumption of an MU or
UAV is mostly governed by the CPU performance of the electronic device. Secondly,
the communication energy consumption is made up of the energy consumed by the MU’s
offload task, as well as the energy consumed by switching topologies between UAVs.
Finally, the problem of minimizing the energy consumption by engines and electronic
devices is also considered. To summarize, the total energy consumption of all UAVs in the
system during a time step t can be defined by Equation (6):
t
EU AV = ∑ t
ECom,n t
+ ETran,n t
+ EFly,n (6)
n∈ N
Sensors 2022, 22, 7243 8 of 18

where:
t
ECom is the energy consumption for processing offloaded tasks of UAV n;
t
ETran is the transmission energy consumption of MU m offload tasks to UAV n;
t is the flight energy consumption of UAV n.
EFly
The optimization procedure is guided by an aggregation of two objective functions,
namely the minimal distance traveled (e p ) and the minimum energy (E). Consequently,
the objective of this work is to minimize traveled distance and energy consumption, while
avoiding obstacles and collisions. This can be formulated as follows:
RT
min Φ = 0 (e Tp · W · e p + E T · Q · E)dt
s.t.
E ≤ Emax (7)
R ≥ Rmax
d = dd
where W and Q are the weighting matrix. Traditional methods can be used to solve the
problem. Karush–Kuhn–Tucker conditions are utilized because of inequality limitations.

4. Formation Control Stage

4.1. UAV Model
Quadrotors are a subset of Vertical Take-off and Landing (VTOL) UAVs. Their ability
to hover, move forward, and perform vertical takeoffs and landings make them nonlinear
and under-actuated systems. This work supposes an attitude representation under quater-
nions approach to avoid singularities presented in Euler angle presentation. Let the unit
quaternion q(t) ∈ H, q̄(t) ∈ R3 , q0 (t) ∈ R be defined with:

q(t) = q0 (t) + q̄(t) = q0 (t) + [q1 (t), q2 (t), q3 (t)] T (8)

with q0 (t) a given quaternion and q̄(t) the complex and q0 (t) the scalar parts of q(t).
The unit quaternion should satisfy:

q0 (t)2 + kq̄(t)k = 1 (9)

Finally, the quadrotors dynamic model is:

F
p̈ = q ⊗ ⊗ q∗ + ḡ (10)
m
1
q̇ = q⊗w (11)
2
ω̇b = J −1 (τ − ωb ∗ Jωb ) (12)
The quaternion derivative is derived based on the angular velocity ω (t). The designed
control inputs are given by Equation (13) with k T as the thrust constant, k D the the drag
constant, and l the distance from the motor axis to the quadrotors center of mass:

k T (ω12 + ω22 + ω32 + ω42 )

   
U1
U2   lk T (ω 2 − ω 2 − ω 2 + ω 2 ) 
 = 1 2 3 4  (13)
U3   lk T (ω 2 + ω 2 − ω 2 − ω 2 ) 
1 2 3 4
U4 k D (−ω12 + ω22 − ω32 + ω42 )
Based on Lyapunov stability analysis, the position and velocity control laws can be
derived as indicated in [26] as follows:
vd = c1P e1 + ṗd (14)
Sensors 2022, 22, 7243 9 of 18

p̈ = e1 ( I − c21P ) + e2 (c1P + c2P ) + p̈d (15)

where e1 = pd − p is the position error, e2 = vd − v is the velocity error, and c1P , c2P are
positive constant coefficients.

4.2. Formation Control

After following the leader for the same or different altitude, the formation controller’s
objective is to maintain the X-Y topology. This is accomplished by maintaining a specific
distance d and angle α between the leader and each follower. Thus:

d x = −( X L − X F ) cos ψL − (YL − YF ) sin ψL

dy = ( X L − X F ) sin ψL − (YL − YF ) cos ψL (16)

where the X and Y coordinates of the desired distance are d x and dy , respectively. For the
purposes of this study, an SMC controller is used to maintain formation topology in the
face of external disturbances and environmental uncertainties. The x and y control errors
should satisfy the following constraints:

lim k ex k=k ddx − d x k= 0

t→∞
lim k ey k=k ddy − dy k= 0 (17)
t→∞

where ddx and ddy are the desired distance in the X and Y coordinates, respectively.
For this purpose, first-order SMC is introduced. The time varying surface S(t) is
selected by s(e; t) = 0, with:

d
s(e, t) = ( + λ ) n −1 e (18)
dy

Then, the derivative function is given by:

ṡ = ë + λė
1 d 2
s ≤ −η | s | (19)
2 dy

Finally, the formation topology is controlled for each agent as follows:

Ẍ Fi = Ẍ L + λ x ( Ẋ L − Ẋ Fi )
ŸFi = ŸL + λy (ẎL − ẎFi ) (20)

Once the position control problem is solved, a transformation to attitude control can
be obtained. This means a minimum sharing data between agents:

θ Fi = θ L + λθ (θ̇ L − θ̇ Fi )
φFi = φL + λφ (φ̇L − φ̇Fi ) (21)

with λθ > 0 and λφ > 0 being selective gains for the formation control.

4.3. Formation Transformation

This section describes the scenario when one of the swarm agents unexpectedly fails.
In such situations, the agent performs an emergency landing to avert crashes, and the other
agents must switch their topologies. Additionally, when the failed agent is a leader, a leader
election procedure is required. This is carried out as outlined in Algorithm 1.
Sensors 2022, 22, 7243 10 of 18

Algorithm 1 Leader Election

N: Number of agents, A: list of N agents where each one has an ID from 1 to N, T = { T ( N ),
T ( N − 1), . . . , T (1)}: list of N topology according to the agent’s number, P( T ): list of N paths
according to each topology.
1: idL = 1
2: L = AidL
3: Y = f ( Batt, Health, Mission)
4: while true do
5: Tex = T ( N )
6: Pex = P( Tex )
7: L = [ Tex , T, Pex , P, A, idL]
8: ALLReady = 0
9: for i ← 1 N 1 do
10: if i 6= idL then
11: send(i, ready = f alse) by L
12: wait(i, ready(i )) = true by L
13: ALLReady = ALLReady + 1
14: end if
15: end for
16: if AllReady = N − 1 then
17: for i ← 1 N 1 do
18: if i 6= idL then
19: send( Pex (i ), i ) by L
20: end if
21: end for
22: if Failure = idL then
23: IdY = 0
24: for j ← 1 N 1 do
25: if i 6= idL then
26: Statei = election
27: idY = max (Y ( Ai ), IdY )
28: end if
29: end for
30: L = AidY
31: idL = idY
32: end if
33: A.remouveElement( A Failure )
34: N = N−1
35: end if
36: end while

First of all, a set of N UAVs are identified from 1 to N. The desired topology (T)
depends on the UAVs number. Each topology has its related paths (P( T )). The head of
the UAV list (A1 ) is considered a leader L. The leader is supposed to have the knowledge
about: the UAVs IDs, the desired topology (Tex ), path (Pex ), and all the possible paths and
topologies (P and T). The performance function (Y) depends on the battery remaining time,
the UAV’s health (i.e., no failure presented), and the percentage of mission executing.
The leader sends a ready message to all UAVs before the mission to determine whether
or not they are prepared (line 9). Once all UAVs are operational (line 15), the leader follows
the reference path and maintains formation topology (line 18).
A test loop is required during the mission to test the state of each drone. If one of
the UAVs fails (line 21), an emergency landing occurs to avoid a crash. Depending on the
failing agent, one of two scenarios can occur. If the failing agent is a leader, the election
procedure begins (line 25), and the agent with the best performance (Y) is chosen as the
leader (line 26–31). If the failing agent is a follower, it will be deleted from the list of UAVs
(line 32). Depending on the number of remaining UAVs, the intended topology will be
switched in both scenarios (return to line 3).
Sensors 2022, 22, 7243 11 of 18

5. Simulation Results
The simulation results for quadrotor creation using Matlab/Simulink are shown in
this section. Many scenarios have been run based on formation control and the many
limitations that can occur during a mission. All of the following scenarios make use of
four quadrotor UAVs. The communication lines between all of the UAVs are expected
to be secure. The goal is for all of the UAVs to be able to switch formations at any time.
A summary of the simulated cases is provided below:
1. Scenario 1: In this scenario, UAVs are deployed as fixed-position stationary nodes
that serve as communication relays. Four UAVs under rectangular topology take
off from various points and fly to a predetermined altitude. The goal is to keep the
optimal fixed position for maximum network connectivity.
2. Scenario 2: The mission to be executed during this scenario is that every UAV acts
as an aerial MEC server. Nonetheless, one of the swarm agents has an unanticipated
engine failure. The remaining UAVs must address this situation. Leader selection and
topology switching are implemented.
3. Scenario 3: The third scenario simulates the UAVs as mobile nodes in a Cellular-
Connected MEC. Each UAV is designed to follow a desired path with different hov-
ering positions to serve for IoT devices. The Swarm must travel to its intended
destination while avoiding external obstacles and agents collisions.

5.1. Scenario 1: Relayed MEC

This scenario aims to test the ability of the designed framework to deploy the UAVs
in a Relayed MEC. For this case, no leader is designed, and all the UAVs have knowledge
about the desired path to follow. The UAVs from 1 to 4 start from the initial positions
F1 (0) = [10; −10; 0] T , F2 (0) = [−10; −10; 0] T , F3 (0) = [−10; 10; 0] T and F4 (0) = [10; −10; 0] T ,
respectively, and meet each other at an altitude z = 10m, while keeping a rectangular sep-
aration of 10 m. An extra edge controller is added to assure the convergence over the
meeting points.
Figure 6 illustrates the success of UAVs in relaying MUs to the MEC server, as all
quadrotors were able to reach the intended offloading spots and cover the entire proposed
region. The formation errors between UAVs 1–2 and 3–4 converge to zero in less than
six seconds. Due to the synchronization of the altitude control for all swarm agents,
the error in altitude is kept at zero throughout the course of the mission. Such position
hold accuracy, while preserving separation distances, would assist MUs in offloading their
computationally intensive tasks and preventing connection overlap.

Figure 6. Relayed MEC under distributed formation.

Sensors 2022, 22, 7243 12 of 18

5.2. Scenario 2: Assisted MEC

In this scenario, the swarm is supposed to start from the following positions:
F1 (0) = [50; 0; 0] T , F2 (0) = [40; −10; 0] T , F3 (0) = [60; −10; 0] T , and F4 (0) = [50; −20; 0] T .
Every UAV is denoted as an MEC server to mobile users among its trajectory. For the
mission, the leader is to achieve the desired point Pf ( x0 , y0 , z0 ) = [50; 100; 10] T . The swarm
is supposed to hold the diamond topology with a separation distance of 10 m between the
agents. At t = 25 s, an engine failure occurs to the leader. According to Algorithm 1, the
leader lands to avoid the crash, a new leader is then elected, and the formation is switched
from diamond to triangular. Depending on the elected leader, three possible cases can be
carried out:

5.2.1. Case 1
Let us suppose that Follower 1 is elected as a new leader. As depicted in Figure 7,
the swarm was able to switch its topology from diamond to triangular immediately following
the leader’s emergency landing. Figure 8 demonstrates that the position transfer between
the new leader and Follower 2 took only 5 s, but the separation distance between the new
leader and Follower 3 (10 m) was strictly observed during the switching process. After
t = 30 s, the new triangular topology was totally established.

Figure 7. Leader election Case 1.

Figure 8. Leader election Case 1—tracking errors.

Sensors 2022, 22, 7243 13 of 18

5.2.2. Case 2
In this case, Follower 2 is supposed to be elected as a new leader. Figures 9 and 10
illustrate the simulation case. The obtained results are similar to those obtained in the first
case. The triangular topology was achieved in t = 30 s, and the position’s switching between
the new leader and Follower 3 was made in 5 s. The separation distance between the new
leader and Follower 1 was also maintained during the switching operation (Figure 10).

Figure 9. Leader election Case 2.

Figure 10. Leader election Case 2—tracking errors.

5.2.3. Case 3
For this last case, Follower 3 is supposed to be elected as the new leader. As shown in
Figure 11, Follower 3 is positioned behind the leader, thus, there is no need to switch posi-
tions. Figure 12 demonstrates that the separation distance was maintained with Follower 3
and Follower 1 with high accuracy.
Sensors 2022, 22, 7243 14 of 18

Figure 11. Leader election Case 3.

Figure 12. Leader election Case 3—tracking errors.

5.3. Scenario 3: Cellular-Connected MEC

The UAVs in this scenario are considered as mobile nodes in a cellular-connected MEC
server. The four UAVs are supposed to start from the following positions: F1 (0) = [50; 65] T ,
F2 (0) = [60; 75] T , F3 (0) = [50; 85] T , and F4 (0) = [40; 75] T . The leader’s objective is to
get to the target spot: Pf ( x0 , y0 ) = [50; 15] T and avoid the obstacle (R = 10 m) situated at
O1 ( x0 , y0 ) = [50; 50] T . The agents are separated by 10 m. The diamond configuration is
expected to be held by the swarm. Otherwise, it is possible to split the topology, avoid the
obstacles, and go back to the initial formation.
In order to conserve energy, the swarm was able to circumvent the circular barrier
by forming a distributed configuration (Figure 13). The swarm was split into two teams,
each with two leaders: one created by the original leader and follower 1, and the other by
follower 2 (new leader) and follower 3 (Figure 14). The formation error of teams 1 and 2 is
shown below. It can be seen that the separation distance (10 m) between the two UAVs was
respected with excellent precision in both the x- and y-directions. This demonstrates the
formation controller’s excellent performance.
Sensors 2022, 22, 7243 15 of 18

Figure 13. Obstacles avoidance.

Table 1 summarizes the results of the previous scenarios. As previously demonstrated,

the UAV swarm proved capable of enabling diverse MEC topologies and accommodating
all possible constraints. While tracking the created path with the least possible error/delay,
the convergence time was very reasonable. The architecture is also ideal for long-range
missions due to the energy consumption strategy.

Table 1. Performance results.

MEC Architecture Relayed Assisted Cellular-Connected

Convergence time (s) 10 50 120
Traveled distance (m) 10 70 80
Total energy consumption (%) 5 15 20

Figure 14. Obstacles avoidance tracking error.

Sensors 2022, 22, 7243 16 of 18

5.4. Comparative Study

To show the usefulness of the proposed framework, a comparison with state-of-the art
methods was conducted. The comparison of the architectures’ key features is presented in
Table 2, from which we can deduce that our suggested framework constitutes a comprehen-
sive effort in comparison to other current works. First, the distributed architecture solves
the shortcomings of the centralized architecture since leader and follower interaction is
taken into account. In addition, the algorithms utilized for path planning and formation
control are user-friendly, suitable for a large number of robots, and can be implemented
within an embedded system. Our framework is the only one that has strengthened safety
precautions by including a leader election process.

Table 2. Article architecture comparison.

Author Wu et al. (2020) Wen et al. (2019) Tran et al. (2021) Proposed Framework
Formation Centralized Decentralized Distributed Distributed
Type UAV UGV UAV/UGV UAV
Vehicle
Number 8 4 3 4
Path planning PSO APF NI CO
Strategy Position consensus Position consensus Velocity consensus Attitude consensus
Formation control
Controller MPC Robuste H ∞ NI SMC
Safety precautions N/C Switching Switching Switching/Leader election

In addition, the performance of formation control results may be found in Table 3.

We can argue that the employment of attitude-based control based on SMC controller has
significantly enhanced the L-F configuration’s precision, speed, and stability.

Table 3. Formation control performance comparison.

Author Wu et al. (2020) Wen et al. (2019) Tran et al. (2021) Proposed Framework
Rise time (s) 5 3 5 4
Over shoot % 0 0 5 0
Setting time (s) 10 5 10 5
Switching time (s) N/C 4 5 1
x N/A 0.3 0.1 0.05
Tracking error (m)
y N/A 0.3 0.1 0.05

6. Conclusions
The issue of multi-UAV-enabled MEC architectures in a distributed formation was
investigated in this study. In the case of centralized UAV formation, the leader is the sole
agent who is fully aware of the mission’s details. As a result, every probable failure has an
impact on mission execution. To overcome this issue, a new framework of quadrotors with a
distributed L-F configuration was presented. The framework is divided into several stages,
each of which takes into account the relationship between the leader and the followers.
Throughout the formation control stage, a consensus-based attitude control was used to
maintain the formation topology with minimal data sharing. An SMC controller was used
to monitor the optimal generated path with the minimum possible error/delay. During the
trajectory generation phase, a collaborative obstacle avoidance technique was developed
for safety and to accommodate the multiple environmental constraints. The swarm’s agents
were able to switch topology, avoid obstacles, and return to the desired formation without
colliding. Furthermore, in response to an agent’s unexpected death, the agents were able
Sensors 2022, 22, 7243 17 of 18

to elect new leadership. The leader election phase was designed to indicate topological
changes caused by external obstacles. In comparison to the vast majority of comparable
suggested works, all produced results were deemed satisfactory. Although this study
focused on the use of UAVs under various MEC architectures, the technical computing
approach can be generalized and applied to many other IoT specifications and access con-
trol strategies, such as networking and data gathering approaches, educational platforms,
healthcare systems, transportation services, and many other real-world applications. There-
fore, incorporating the proposed architecture into an MEC-based real-world application
would be an intriguing endeavor.
Author Contributions: Conceptualization, K.C.; data curation, K.C. and S.F.; methodology, S.M.;
software, K.C.; supervision, M.L.; validation, K.C., M.L., S.M., and S.F.; writing—original draft, K.C.;
writing—review and editing, S.M. All authors have read and agreed to the published version of
the manuscript.
Funding: This work is supported by Princess Nourah Bint Abdulrahman University Researchers
Supporting Project number (PNURSP2022R196), Princess Nourah Bint Abdulrahman University,
Riyadh, Saudi Arabia.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No data available.
Acknowledgments: The authors would like to acknowledge the Princess Nourah Bint Abdulrah-
man University Researchers Supporting Project number (PNURSP2022R196), Princess Nourah Bint
Abdulrahman University, Riyadh, Saudi Arabia.
Conflicts of Interest: text The authors declare no conflict of interest.

References
1. Abrar, M.; Ajmal, U.; Almohaimeed, Z.M.; Gui, X.; Akram, R.; Masroor, R. Energy efficient UAV-enabled mobile edge computing
for IoT devices: A review. IEEE Access 2021, 9, 127779–127798. [CrossRef]
2. Faraci, G.; Grasso, C.; Schembra, G. Fog in the clouds: UAVs to provide edge computing to IoT devices. ACM Trans. Internet
Technol. TOIT 2020, 20, 1–26. [CrossRef]
3. Kim, S.; Oh, H.; Suk, J.; Tsourdos, A. Coordinated trajectory planning for efficient communication relay using multiple UAVs.
Control. Eng. Pract. 2014, 29, 42–49. [CrossRef]
4. Rahbari, D.; Mahtab Alam, M.; Le Moullec, Y.; Jenihhin, M. Edge-to-Fog Collaborative Computing in a Swarm of Drones. In
International Conference on Model and Data Engineering; Springer: Berlin/Heidelberg, Germany, 2021; pp. 78–87.
5. Wang, Y.; Fang, W.; Ding, Y.; Xiong, N. Computation offloading optimization for UAV-assisted mobile edge computing: A deep
deterministic policy gradient approach. Wirel. Netw. 2021, 27, 2991–3006. [CrossRef]
6. Chang, H.; Chen, Y.; Zhang, B.; Doermann, D. Multi-UAV mobile edge computing and path planning platform based on
reinforcement learning. IEEE Trans. Emerg. Top. Comput. Intell. 2021, 6, 489–498. [CrossRef]
7. Guo, H.; Liu, J. UAV-enhanced intelligent offloading for Internet of Things at the edge. IEEE Trans. Ind. Inform. 2019, 16, 2737–2746.
[CrossRef]
8. Peng, C.; Huang, X.; Wu, Y.; Kang, J. Constrained Multi-Objective Optimization for UAV-Enabled Mobile Edge Computing:
Offloading Optimization and Path Planning. IEEE Wirel. Commun. Lett. 2022, 11, 861–865. [CrossRef]
9. Tropea, M.; De Rango, F.; Nevigato, N.; Bitonti, L.; Pupo, F. Scare: A novel switching and collision avoidance process for
connected vehicles using virtualization and edge computing paradigm. Sensors 2021, 21, 3638. [CrossRef]
10. Gong, C.; Wei, L.; Gong, D.; Li, T.; Feng, F. Energy-Efficient Task Migration and Path Planning in UAV-Enabled Mobile Edge
Computing System. Complexity 2022, 2022. [CrossRef]
11. Song, K.T.; Sun, Y.X. Coordinating multiple mobile robots for obstacle avoidance using cloud computing. Asian J. Control. 2021,
23, 1225–1236. [CrossRef]
12. Miao, Y.; Hwang, K.; Wu, D.; Hao, Y.; Chen, M. Drone Swarm Path Planning for Mobile Edge Computing in Industrial Internet of
Things. IEEE Trans. Ind. Inform. 2022, 1–11. doi: 10.1109/TII.2022.3196392. [CrossRef]
13. Jiang, C.; Fang, Y.; Zhao, P.; Panneerselvam, J. Intelligent uav identity authentication and safety supervision based on behavior
modeling and prediction. IEEE Trans. Ind. Inform. 2020, 16, 6652–6662. [CrossRef]
14. Jiang, Y.; Wang, S.; Lei, L. A Multi-UAV Formation Maintaining Method Based on Formation Reference Point. J. Phys. Conf. Ser.
2020, 1621, 012014. [CrossRef]
Sensors 2022, 22, 7243 18 of 18

15. He, Y.; Liu, J.; Tong, E.; Niu, W.; Huang, X.; Zhou, Y.; Li, C.; Chang, L. Adapt Swarm Path Planning for UAV Based on Artificial
Potential Field with Birds Intelligence Extensions. In International Symposium on Security and Privacy in Social Networks and Big
Data; Springer: Berlin/Heidelberg, Germany, 2020; pp. 340–354.
16. Liu, H.; Lyu, Y.; Zhao, W. Robust visual servoing formation tracking control for quadrotor UAV team. Aerosp. Sci. Technol. 2020,
106, 106061. [CrossRef]
17. Parrany, A.M.; Alasty, A. Decentralized aggregation and leader-following control of a swarm of quadcopters with nonlinear
under-actuated dynamics. Aerosp. Sci. Technol. 2020, 107, 106317. [CrossRef]
18. Ai, X.; Yu, J. Flatness-based finite-time leader–follower formation control of multiple quadrotors with external disturbances.
Aerosp. Sci. Technol. 2019, 92, 20–33. [CrossRef]
19. Brust, M.R.; Strimbu, B.M. A networked swarm model for UAV deployment in the assessment of forest environments. In
Proceedings of the 2015 IEEE Tenth International Conference on Intelligent Sensors, Sensor Networks and Information Processing
(ISSNIP), Singapore, 7–9 April 2015; pp. 1–6.
20. Shames, I.; Teixeira, A.; Sandberg, H.; Johansson, K.H. Distributed leader selection without direct inter-agent communication.
In Proceedings of the 2nd IFAC Workshop on Distributed Estimation and Control in Networked Systems, NecSys’ 10, Annecy,
France, 13–14 September 2010; pp. 221–226.
21. Franchi, A.; Giordano, P.R. Online leader selection for improved collective tracking and formation maintenance. IEEE Trans.
Control. Netw. Syst. 2018, 5, 3–13. [CrossRef]
22. CHOUTRI, K.; LAGHA, M.; DALA, L.; LIPATOV, M. Quadrotors UAVs Swarming Control Under Leader-Followers Formation.
In Proceedings of the 2018 22nd International Conference on System Theory, Control and Computing (ICSTCC), Sinaia, Romania,
10–12 October 2018; pp. 794–799.
23. Xu, M.; Fang, Y.; Li, J.; Zhao, X. Chattering Free Distributed Consensus Control for Attitude Tracking of Spacecraft Formation
System with Unmeasurable Angular Velocity. Int. J. Control. Autom. Syst. 2020, 18, 2277–2288. [CrossRef]
24. Hou, Z. Modeling and formation controller design for multi-quadrotor systems with leader-follower configuration. Ph.D. Thesis,
Université de Technologie de Compiègne, Compiègne, France, 2016.
25. Stojkovi, I. Formation Control of Robotized Aerial Vehicles Based on Consensus-Based Algorithms. FME Trans. 2017, 45, 559–564.
[CrossRef]
26. Choutri, K.; Lagha, M.; Dala, L. Multi-layered optimal navigation system for quadrotor UAV. Aircr. Eng. Aerosp. Technol. 2019, 92,
145–155. [CrossRef]
27. Pham, H.; Smolka, S.A.; Stoller, S.D.; Phan, D.; Yang, J. A survey on unmanned aerial vehicle collision avoidance systems. arXiv
Preprint 2015, arXiv:1508.07723.
28. Lao, M.; Tang, J. Cooperative Multi-UAV Collision Avoidance Based on Distributed Dynamic Optimization and Causal Analysis.
Appl. Sci. 2017, 7, 83. [CrossRef]
29. Misir, O.; Gökrem, L. Flocking-Based Self-Organized Aggregation Behavior Method for Swarm Robotics. Iran. J. Sci. Technol.
Trans. Electr. Eng. 2021, 45, 1427–1444. [CrossRef]
30. Dang, A.D.; La, H.M.; Nguyen, T.; Horn, J. Distributed Formation Control for Autonomous Robots in Dynamic Environments.
CoRR 2017, arXiv:1705.02017.
31. Singh, Y.; Bibuli, M.; Zereik, E.; Sharma, S.; Khan, A.; Sutton, R. A novel double layered hybrid multi-robot framework for
guidance and navigation of unmanned surface vehicles in a practical maritime environment. J. Mar. Sci. Eng. 2020, 8, 624.
[CrossRef]
32. Di Lillo, P.; Pierri, F.; Antonelli, G.; Caccavale, F.; Ollero, A. A framework for set-based kinematic control of multi-robot systems.
Control. Eng. Pract. 2021, 106, 104669. [CrossRef]
33. Liu, T.; Han, D.; Lin, Y.; Liu, K. Distributed multi-UAV trajectory optimization over directed networks. J. Frankl. Inst. 2021,
358, 5470–5487. [CrossRef]
34. de Freitas, E.P.; Basso, M.; da Silva, A.A.S.; Vizzotto, M.R.; Corrêa, M.S.C. A Distributed Task Allocation Protocol for Cooperative
Multi-UAV Search and Rescue Systems. In Proceedings of the 2021 International Conference on Unmanned Aircraft Systems
(ICUAS), Athens, Greece, 15–18 June 2021; pp. 909–917.