mathematics

Article
Edge-Weighted Consensus-Based Formation Control with
Collision Avoidance for Mobile Robots Based on Multi-Strategy
Mutation Differential Evolution
Jesus Hernandez-Barragan 1 , Tonatiuh Hernandez 2 , Jorge D. Rios 1 , Marco Perez-Cisneros 3
and Alma Y. Alanis 1, *

1 Departamento de Innovación Basada en la Información y el Conocimiento, Centro Universitario de Ciencias

Exactas e Ingenierías, Universidad de Guadalajara, Guadalajara 44430, Mexico;
[email protected] (J.H.-B.); [email protected] (J.D.R.)
2 Departamento de Ciencias Computacionales, Centro Universitario de Ciencias Exactas e Ingenierías,
Universidad de Guadalajara, Guadalajara 44430, Mexico; [email protected]
3 Departamento de Electro-Fotónica, Centro Universitario de Ciencias Exactas e Ingenierías,
Universidad de Guadalajara, Guadalajara 44430, Mexico; [email protected]
* Correspondence: [email protected]

Abstract: An edge-weighted consensus-based formation control strategy is presented for mobile

robots. In the edge-weighted strategy, a desired formation pattern is achieved by adjusting gain
weights related to the distance between robots. Moreover, the edge-weighted formation control
exploits the properties of weighted graphs to allow the formation to rotate and adapt its shape to
avoid collision among robots. However, formation patterns are commonly defined by biases with
respect to the centroid of the consensus rather than gain weights. This work proposes to optimize
the gain weights in edge-weighted graphs, given a formation pattern in terms of biases. A multi-
strategy mutation differential evolution algorithm is introduced to solve the optimization problem.
Simulation and real-world experiments are performed considering multi-robot systems composed of
differential drive robots. Additionally, the experimental setup includes Turtlebot3® Waffle Pi robots
Citation: Hernandez-Barragan, J.;
Hernandez, T.; Rios, J.D.;
and an OptiTrack® motion capture system for control purposes. The experimental results verify the
Perez-Cisneros, M.; Alanis, A.Y. effectiveness of the proposed approach.
Edge-Weighted Consensus-Based
Formation Control with Collision Keywords: edge-weighted control; consensus-based control; differential evolution; evolutionary
Avoidance for Mobile Robots Based algorithms; mobile robots
on Multi-Strategy Mutation
Differential Evolution. Mathematics MSC: 93-10
2023, 11, 3633. https://doi.org/
10.3390/math11173633

Academic Editor: Zhijia Zhao

1. Introduction
Received: 28 July 2023 Cooperative work is the main advantage of multi-agent systems. A network of
Revised: 16 August 2023
mobile robots is used for many applications, such as search-and-rescue (SaR) operations,
Accepted: 18 August 2023
cooperative transportation of objects, collaborative exploration, and mapping, along with
Published: 23 August 2023
others [1,2]. There is more than one strategy to solve the formation problem. Most of the
time is solved as a consensus problem in which the states of all the agents converge to
a common value [3]. In the case of mobile robots, this value defines the centroid of the
Copyright: © 2023 by the authors.
desired formation.
Licensee MDPI, Basel, Switzerland. Several works deal with formation problems based on the Laplacian properties [4].
This article is an open access article The most common algorithms exploit the following control law x̄˙ = −Lx̄, where x̄ ∈ Rn
distributed under the terms and corresponds to the position of the ith robot translated by a bias with respect to the centroid
conditions of the Creative Commons of the formation, and L ∈ Rn×n stands for the Laplacian matrix. One of the main advan-
Attribution (CC BY) license (https:// tages of this approach is that the formation is achieved even in the presence of delays in
creativecommons.org/licenses/by/ the communication if the communication topology is static, undirected, and connected [5].
4.0/). Different approaches using a bias-based strategy have been proposed. For example, in [6],

Mathematics 2023, 11, 3633. https://doi.org/10.3390/math11173633 https://www.mdpi.com/journal/mathematics

Mathematics 2023, 11, 3633 2 of 19

a hybrid technique to drive a group of mobile robots to a desired formation is presented,

and Particle Swarm Optimization is applied for the exploration of unknown environments.
In [7], a full consensus-formation is reached by a switching control law. Other exam-
ples under this strategy use single-integrator dynamics [8], pinning control of complex
networks [9], and persistency of excitation to deal with non-holonomic constraints [10].
Nevertheless, solving the formation problem through a bias-based strategy does not
void collisions while moving in unknown environments [11], and incorporates artificial
potential fields for collision avoidance, producing an unstable behavior in the presence
of communication delays [12]. The weighted graph strategy overcomes these drawbacks.
In [11,13], an edge-weighted graph strategy is implemented to include collision avoidance
among the robots.
This paper extends the results presented in [11], where an edge-weighted formation
control is proposed. In this strategy, the distance between the ith robot and the jth robot is
given by the minimum of the edge-tension function; thus, the choice of a constant gain in
this function defines the desired distance for each couple of robots. However, a desired
formation is usually given in cartesian coordinates rather than in terms of a gain related to
the distance between robots. Optimization algorithms are applied under the edge-weighted
graph scheme to find the best constant gain value corresponding to the desired formation
given in cartesian coordinates.
Evolutionary Algorithms (EAs) have been widely used for intelligent optimization in
many research areas such as electrical and electronics, path planning, trajectory design and
tracking, automation control systems, interdisciplinary applications, and formation con-
trol [14,15]. In [15], the authors discussed the application of EAs to engineering problems.
Particularly, the Differential Evolution (DE) algorithm stands out in the automation control
and engineering civil areas.
Differential Evolution (DE) is commonly used to solve complex optimization prob-
lems [16]. Its flexibility and versatility have promoted several customized variants to solve
real-world problems. Recently, many works have modified the conventional DE algorithm
to enhance its effectiveness and efficiency [17]. In [17], the authors presented a recent
review based on the state-of-the-art DE variants. The strengths and weaknesses of several
DE modifications were carefully reviewed, followed by their potential application in solv-
ing real-world engineering problems. Among the reviewed variants, the multi-strategy
different dimensional mutation differential evolution version stands out for its enhanced
converge speed and exploitation search capacities, overcoming stagnation in local optima
regions [18]. All these benefits motivated the authors to use the multi-strategy mutation
scheme to deal with the drawbacks of conventional DE.
The contributions of this paper are summarized below: Given a desired formation in
cartesian coordinates, it is proposed to optimize the constant gains of the edge-weighted
graph, such that the edge-weight formation control drives the robots to the desired forma-
tion, keeping its capacities of collision avoidance among robots. Based on [18], a multi-
strategy mutation differential evolution (MMDE) algorithm is proposed to solve this prob-
lem. MMDE enhances convergence speed rates and avoids premature convergence of the
conventional DE algorithm. The proposed approach is called the MMDE-based formation
pattern for edge-weighted formation control.
This article is organized as follows: The next section presents the kinematics model
formulation for differential drive non-holonomic robots. Then, the edge-weighted forma-
tion control is described in Section 3. In Section 4, the description of the MMDE algorithm
is provided. Moreover, the proposed MMDE-based formation pattern strategy is presented
with details in Section 5. Simulation experiments and real-world implementation are per-
formed in Section 6. An analysis of the reported results and the future research directions
are given in Section 7. Finally, conclusions are presented in Section 8.
Mathematics 2023, 11, 3633 3 of 19

2. Kinematics Model Formulation

Consider a set of N differential drive non-holonomic robots, like the one depicted in
Figure 1, each of them modeled as the unicycle kinematics (1) under the assumption that
wheels are rolling without slippage, the steering axis is orthogonal to the xy-plane, and the
geometry center Q coincides with the center of mass. The kinematics model is given by

ẋi = cos(θi )vi

ẏi = sin(θi )vi (1)
θ̇i = ωi ,

where xi , yi ∈ R stands as the cartesian coordinates, θi ∈ R is the orientation with respect

to the z-axis, and vi and ωi ∈ R are the linear and angular velocities, respectively. To deal
with the non-holonomic constraints, the input/output linearization feedback in [19] is
employed. Defining point B, from Figure 1, located along the sagittal axis of the robot at a
distance bi from the contact point of the wheel with the ground, the cartesian coordinates
are given by

xib = xi + bi cos(θi )
(2)
yib = yi + bi sin(θi ),

with bi 6= 0. After evaluating the time derivative of (2) yields

 b     
ẋi cos(θi ) −bi sin(θi ) vi v
= = T ( θi ) i , (3)
ẏib sin(θi ) bi cos(θi ) ωi ωi

where the matrix T(θi ) is invertible as long as bi 6= 0.

𝑦ሶ 𝑖

𝑦𝑖𝑏 𝐵
𝜃𝑖
𝑦𝑖
𝑏𝑖

𝑥𝑖 𝑥𝑖𝑏 𝑥ሶ 𝑖
Figure 1. Differential drive robot model.
y y
To drive the robot, the input control (uix , ui ) is designed such as ẋib = uix and ẏib = ui .
Then, using the following input transformation
 x 
uix
   
vi u cos(θi ) sin(θi )
= T−1 (θi ) iy = y , (4)
ωi ui − sin(θi )/bi cos(θi )/bi ui
Mathematics 2023, 11, 3633 4 of 19

the input/output linearization feedback is achieved

ẋib = uix
y
ẏib = ui (5)
y
θ̇i = (ui cos(θi ) − uix sin(θi ))/bi .

The communication topology is modeled as an undirected, static, connected, and

weighted graph defined as a pair G = (V , E ), where E ⊆ V × V is the edge set of cardinality
M and V stands for the vertex set of cardinality N. Given a graph the Laplacian matrix,
L ∈ R N × N can be defined as
L = I · I >, (6)
where I ∈ R N × M is the incidence matrix whose components are given by
(
1 if vertex vi is incident with edge (i, j) ∈ E
iij = (7)
0 otherwise,

an edge (i, j) represents the bidirectional communication between agents i and j. By con-
struction, L1 N = 0, such that 1 N = [1, · · · , 1]> , L is symmetric, it has a unique zero
eigenvalue, and the rest of its spectrum is strictly positive [20].

3. Edge-Weighted Formation Control

The consensus problem for N agents to drive the robots to a final common state can be
solved with the Laplacian-based feedback method [11]. The feedback control is in the form

żi = − ∑ wij (z)(zi − z j ), (8)

j∈Ni

where zi := [ xib , yib ]> , Ni = {∀v j ∈ V | (i, j) ∈ E } is the Neighbors subset of the ith robot,
and wij (z) is a positive edge-weight function in terms of z := [z1> , . . . , z> N ]. Defining a
weight matrix as
W (z) = diag({wij (z) | (i, j) ∈ E }) ∈ R M× M , (9)
and the weighted Laplacian matrix as follows

LW (z) = I · W (z) · I > ; (10)

then the control law (8) can be recast as

ż = −(LW ⊗ I2 )z, (11)

where I2 ∈ R2×2 is an identity matrix, and ⊗ is the Kronecker product.

Regarding collision avoidance, the weight function is designed with a safety distance
parameter δ > 0 among the robots. Therefore, the edge-weight function is defined as
! !
1 2 k zij k − δ
wij (z) = αij − csch +1 , (12)
k ij k ij

where αij > 0 defines the inter-robot influence, zij := zi − z j stands for the position error,
and k ij > 0 is a constant related to the distance between robots.
The edge-weight function (12) was designed based on an edge-tension function that
guarantees collision avoidance between agents i and j, if the δ parameter is strictly greater
than the distance kzij k [11].
The distance between robots i and j is given by
q 
kzij k = δ + k ij acsch k ij . (13)
Mathematics 2023, 11, 3633 5 of 19

The selection of the k ij values determines the final formation. It means that the edge-
weight control requires the appropriate selection of the k ij values to drive the robots to a
desired formation. Thus, given a formation pattern in cartesian coordinates, the goal is to
adjust the k ij gain such that
khi − h j k − kzij k = 0, (14)
where hi is the desired bias of agent i, and h j is the desired bias of agent j, both biases
defined with respect to the centroid of the desired formation.
This work proposes to solve the problem statement in (14), as an optimization problem.
An objective function to deal with the optimization is described in the next subsection.
For further details of the Laplacian-based feedback method and collision avoidance,
the reader is referred to read [11].

Objective Function Formulation

A candidate solution xn composed of a set of knij gains is defined as

xn = {knij | (i, j) ∈ E }, (15)

where xn ∈ R M , since the cardinality of E defines the dimension of the optimization problem.
Based on the edge-weight graph strategy, each element knij ∈ xn is related to a distance
dij between agents i and j. The distance dijn (knij ) is computed as
n

q 
dijn (knij ) = δ + knij acsch knij . (16)

On the other hand, a desired distance dij∗ between agents i and j can be defined based
on the desired formation pattern in cartesian coordinates. Distance dij∗ is given by

dij∗ = khi − h j k, (17)

An objective function can be designed such that (dij∗ − dijn (knij )) → 0. Then, an error
= dij∗ − dijn (knij ) is defined, ∀(i, j) ∈ E .
eijn (knij )
Finally, the following objective function f (xn ) is proposed

1
f (xn ) = e ( x n ) T e ( x n ), (18)
M

where e(xn ) ∈ R M is a set of errors given by e(xn ) = {eijn (knij ) | (i, j) ∈ E }.

The optimization problem is expressed as

arg min f (xn ), subject to knij > 0, ∀ knij ∈ xn , (19)

xn

where each element knij ∈ xn is considered a feasible solution if knij > 0. Otherwise,
unfeasible solutions knij <= 0 cause system instability while performing the formation
control task.
In this work, the MMDE algorithm is used to solve this optimization problem stated
in (19).

4. Multi-Strategy Mutation Differential Evolution

DE is a population-based stochastic algorithm for global optimization
n [21].oThe DE al-
g g g
gorithm uses NP D-dimensional parameter vectors x1 , x2 , · · · , x NP , where
n o
g g g g
xn = xn1 , xn2 , · · · , xnD , and n = 1, 2, · · · , NP denotes the nth individual in the gth
generation. Every individual improves through three principal operations: mutation,
crossover, and selection, performed every generation.
Mathematics 2023, 11, 3633 6 of 19

In conventional DE, the mutation strategy DE/rand/1 is utilized for global optimiza-
tion, but its convergence ability is often insufficient with low convergence speed. On the
other hand, the mutation strategy DE/best/1 is utilized for local optimization, but it suffers
from premature convergence. Based on [18], this work presents the MMDE scheme to
enhance the convergence speed while avoiding premature convergence of the conventional
DE algorithm. The procedure of MMDE to solve minimizationn problemsois shown below.
g g g g
In the mutation operation, a mutant vector vn = vn1 , vn2 , · · · , vnD is generated for
each individual n according to
  
g g g


 xbest + F xr2 − x3 if 1 ≤ g ≤ G/4
  
 xrg + (1 − F ) xrg − xrg

if G/4 < g ≤ G/2

g 1
2 3
vn = g g g
 (20)


 x best + ( 1 − F ) x r2 − x 3 if G/2 < g ≤ 3G/2
  
 xrg + F xrg − xrg

if 3G/2 < g ≤ G,

1 2 3

g g g
where xr1 , xr2 , and xr3 are randomly selected individuals such as r1 , r2 , r3 ∈ {1, NP } and
n 6= r1 6= r2 6= r3 . The parameter F ∈ [0, 1] is called the amplification factor, and it
g g
controls the amplification of the differential variation (xr2 − xr3 ). Moreover, best is the best
individual in the population at generation g, and G is the total number of generations.
As shown in (20), the mutation scheme is divided into four generation units. The first
and the third generation promote exploitation search and speed convergence. The second
and fourth generation unitsn prevent population
o stagnation in local optima regions [18].
g g g g
A trial vector un = un1 , un2 , · · · , unD is generated based on the following binomial
crossover operation
 g
g vnm if rm ≤ CR or m = mrand
unm = g (21)
xnm otherwise,

where m = 1, 2, 3, · · · , D and CR ∈ [0, 1] is the crossover constant. rm and mrand are uniform
random numbers defined as rm ∈ [0, 1] and mrand ∈ {1, D }. The number mrand ensures
g g
that the trial vector un obtains at least one element from the mutation vector vn to avoid
evolutionary stagnation [22].
g g
In the selection operation, the trial vector un is compared to the actual vector xn based
g g
on the evaluation of an objective function f . If f (un ) yields a better solution than f (xn ),
g g g
then un replaces xn . Otherwise, xn is retained to the next iteration g + 1. The selection
scheme is defined as  g g g
g +1 un if f (un ) ≤ f (xn )
xn = g (22)
xn otherwise.
A detailed description of conventional DE and its modifications can be found in [17,21,22].

5. Description of the Mmde-Based Formation Pattern Algorithm

In consensus-based formation control strategies using bias hi , the formation patterns
are defined by translating (in cartesian coordinates) the position of the ith robot with respect
to the centroid of the formation. The advantage of bias-based strategies is the easy proposal
of formation patterns. However, these strategies do not allow the formation to rotate and
adapt its shape to avoid collisions [11].
In edge-weighted formation control, the formation patterns are defined by edge-
weighted values wij (z). Given the appropriate k ij values, ∀(i, j) ∈ E , the desired formation
is achieved without the definition of any bias, with the advantage of collision avoidance [11].
However, the choice of k ij values can be difficult to propose for some formation patterns.
Moreover, the more edge-weighted values wij the graph contains, the more difficult it will
be to propose the k ij values.
Mathematics 2023, 11, 3633 7 of 19

This work proposes to optimize the k ij values for the edge-weighted formation control,
∀(i, j) ∈ E , such that each robot i achieves the desired formation defined with cartesian
coordinates.
The description of the proposed approach is illustrated in Figure 2. The MMDE-
based formation pattern scheme consists of two stages: The first stage involves offline
optimization. In this phase, the MMDE algorithm determines the corresponding k ij values
based on a desired formation pattern hi with respect to the centroid of the formation.
In the second stage, the edge-weighted control strategy is employed to achieve the desired
formation using the calculated k ij values.

Desired formation pattern Edge-weighted

in Cartesian coordinates. formation control
𝐡1 𝑦 𝐡2 𝑘𝑖𝑗 𝑘12
𝐡𝑖
1 2 MMDE algorithm 1 2
𝑘13
𝑥 𝑘14 𝑘23
4 3 4 3
𝑘34
𝐡4 𝐡3
Offline optimization Online control
Figure 2. Proposed MMDE-based formation pattern scheme.

The following subsections provide a detailed description of the MMDE-based forma-

tion pattern algorithm presented in Figure 2.

Mmde-Based Formation Pattern Algorithm

This work proposes to optimize the objective function (18) using the MMDE algorithm.
The MMDE algorithm requires the initialization of each individual xn in the population.
Usually, individuals are randomly initialized considering lower and upper bounds.
To generate random individuals xn , it is proposed to use the following equation

knij = kmin max

ij + r ( k ij − kmin
ij ), ∀ knij ∈ xn , (23)

where kmin max are the lower and upper bounds, respectively. Moreover, r ∈ [0, 1] is a
ij and k ij
uniform random number. This boundary must be selected such as kmax
ij > kmin
ij > 0.
g
It is important to notice that some elements knij of the trial vector un may represent
unfeasible solutions due to the impact of the mutation operation. To deal with unfeasible
solutions, it is proposed to follow the next scheme
(
n
kmin max − k min ) if k n <= 0
ij + r ( k ij ij ij g
k ij = ∀ knij ∈ un , (24)
knij otherwise,

where each knij value is randomly computed to make it feasible. A summary of the MMDE
algorithm to adjust the k ij values, ∀(i, j) ∈ E for the edge-weighted formation control is
shown in Algorithm 1.
Mathematics 2023, 11, 3633 8 of 19

Algorithm 1 MMDE algorithm to adjust the k ij values, ∀(i, j) ∈ E for the edge-weighted
formation control
1: define each bias hi for each robot
2: set G, NP , F ∈ [0, 1] and CR ∈ [0, 1] parameters
3: set safety distance parameter δ and boundary [ k min max
ij , k ij ]
4: f (xn ) ← objective function defined in (18)
5: for each individual n ∈ {1, NP } do
6: define individual position xn = {knij | (i, j) ∈ E }
7: /* Perform initialization based on (23) */
8: knij = kmin max − k min ), ∀ k n ∈ x
ij + r ( k ij ij ij n
9: end
10: for each generation g ∈ {1, G } do
11: for each individual n ∈ {1, NP } do
g
12: /* Generate the mutant vector vn based on (20) */
13: randomly choose r1 , r2 , r3 ∈ {1, NP } such as n 6= r1 6= r2 6= r3
14: best ← find the best individual
15: if 1 ≤ g ≤ G/4 then  
g g g g
16: vn = xbest + F xr2 − x3
17: else if G/4 < g ≤ G/2  then 
g g g g
18: v n = x r1 + (1 − F ) x r2 − x r3
19: else if G/2 < g ≤ 3G/2 then 
g g g g
20: vn = xbest + (1 − F ) xr2 − x3
21: else if 3G/2 < g ≤  G then 
g g g g
22: v n = x r1 + F x r2 − x r3
23: end
g
24: /* Generate the trial vector un based on (21) */
25: randomly choose mrand ∈ {1, D }
26: for each dimensional parameter m ∈ {1, D } do
27: randomly choose rm ∈ [0, 1]
28: if rm ≤ CR or m = mrand then
g g
29: unm = vnm
30: else
g g
31: unm = xnm
32: end
33: end
g
34: /* Deal with unfeasible solutions knij <= 0, ∀ knij ∈ un based on (24) */
35: n
if k ij <= 0 then
36: knij = kmin max − k min )
ij + r ( k ij ij
37: end
38: /* Perform
  the  selection
 operation based on (22) */
g g
39: if f un < f xn then
g +1 g
40: xn = un
41: else
g +1 g
42: xn = xn
43: end
44: end
45: end
46: best ← find the best individual
47: report the optimal kbest
ij ∈ xbest values related to the best individual
Mathematics 2023, 11, 3633 9 of 19

6. Experimental Results
The applicability of the proposed approach is demonstrated through simulation and
real-world experiments. All parameter settings used for both experiments are provided below.
The settings of the MMDE algorithm are defined as follows: The amplification factor
F = 0.5 and the crossover constant CR = 0.5, as suggested in [17]. Moreover, a total of
G = 150 iterations and N = 30 population members were selected. With respect to the
edge-weight formation strategy settings, the safety distance was set to δ = 0.15, and the
inter-robot influence was fixed to α = 0.05. These parameters were experimentally selected.
For simplicity, the physical parameter bi is set to 0.08 meters for all robots. Additionally,
the linear and angular velocities (vi , ωi ) have been bounded to −0.2 ≤ vi ≤ 0.2 and
−0.9 ≤ ωi ≤ 0.9. This saturation is imposed to protect the equipment, especially for
real-world implementations.
All experiments are conducted as follows: First, a desired formation pattern is pro-
vided in cartesian coordinates hi with respect to the centroid of the formation. Then,
the MMDE algorithm optimizes offline the k ij values ∀(i, j) ∈ E for the edge-weighted
formation control. Finally, the edge-weighted formation control was performed online to
achieve the desired formation.
For real-world experimentation, up to five Turtlebot3 (Turtlebot3 is a registered trade-
mark of ROBOTIS Inc. USA) Waffle Pi mobile robots have been considered. The Turtlebot3®
Waffle Pi is a modular differential drive mobile robot, compact and customizable, pro-
grammable and ROS-based platform for education, research, and product development
applications [23] (see Figure 3).

Figure 3. Turtlebot3® Waffle Pi. Onboard cubes of silver reflective fabric are used for tracking purposes.

The ROS (ROS is a registered trademark of Open Robotics, Mountain View, CA, USA)
packages of Turtlebot3® provide access to send velocity control inputs and read its odometry.
However, odometry measures lack precision due to the sensor’s tell, terrain conditions,
and slipping. To deal with this problem, it is proposed to use an OptiTrack (OptiTrack is a
registered trademark of NaturalPoint Inc., Corvallis, OR, USA) tracking system to estimate
the poses of the robots with precision, instead of reading the robot’s odometry.
The OptiTrack® system is composed of six cameras Flex3 (OptiTrack Flex 3 is a regis-
tered trademark of NaturalPoint Inc., Corvallis, OR, USA) and a computer with the software
Motive for streaming. Tracking information is sent using ROS® packages. The space for
real-world implementations is presented in Figure 4.
Another computer is used to receive the poses of the robots from the OptiTrack® sys-
tem, and it sends the respective control signals to each Turtlebot3® . The proposed scheme
is programmed on this computer using MATLAB (MATLAB is a registered trademark of
Mathworks Inc., Torrance, CA, USA) and its ROS® toolbox.
Mathematics 2023, 11, 3633 10 of 19

Figure 4. OptiTrack® setup for real-world implementations. The six cameras Flex3® are installed on
the ceiling.

6.1. Simulation Experiments

The simulation experiments were carried out using a network of four robots inter-
connected as illustrated in Figure 5, where the dimensionality of the problem is D = 6.
Moreover, two tests were performed called Simulations 1 and 2 Simulation 2. The desired
formation patterns for the considered tests are shown in Figure 6.

𝑘12
1 2
𝑘13 𝑘24
𝑘14 𝑘23

4 3
𝑘34

Figure 5. Communication topology among robots for simulation experiments. Each k ij value must be
adjusted to define a desired formation.

𝑇 𝐡1 = 1 𝑇 𝑇 𝑇
𝐡2 = −1 1 1 𝐡1 = −0.5 1 𝐡2 = 0.5 1
2 1 1 2
𝑦 𝑦

𝑥 𝑥
4
𝐡4 = 0 𝑇
0

3 4 3
𝑇 𝑇 𝐡3 = 1.5 𝑇
𝐡3 = 0 −1 𝐡4 = −1.5 −1 −1

(a) (b)
Figure 6. Desired formation patterns in cartesian coordinates hi used for simulation. The xy frame
represents the centroid of the formation pattern. (a) Formation pattern for Simulation 1. (b) Formation
pattern for Simulation 2.

Table 1 reports the optimal k ij values of both simulation tests. In the case of Simula-
tion 1, the values k13 = k23 and k14 = k24 were expected to be equal due to the shape of its
desired formation pattern (see Figure 6). Similarly, the values k14 = k23 and k13 = k24 were
expected to be the same for Simulation 2.
Mathematics 2023, 11, 3633 11 of 19

Table 1. Optimization results of MMDE. The table reports the optimal k ij values. Moreover, f best is
the objective function evaluation at the final generation.

Test k12 k13 k14 k23 k24 k34 f best

Simulation 1 3.9042 4.8638 1.9926 4.8638 1.9926 1.0356 1.7184 × 10−25
Simulation 2 1.0356 7.7565 4.8638 4.8638 7.7565 8.7243 1.5517 × 10−24

Table 1 also shows the objective function evaluation at the final generation called
f best . The performance of MMDE is similar in both tests. After a total of 150 generations,
the algorithms report an objective function evaluation of 1.7184 × 10−25 and 1.5517 × 10−24 ,
for Simulations 1 and 2 2, respectively. The reported f best values are small enough to
consider the k ij values optimal for the edge-weighted formation control.
The convergence curves results for simulation experiments are illustrated in Figure 7.
These results exhibit the fast convergence speed of the MMDE algorithm. Moreover,
the algorithm also demonstrates its exploitation search capability.

Convergence curve result Convergence curve result

100 100

10-5 10-5
fitness value

fitness value
10-10 10-10

10-15 10-15

10-20 10-20

10-25 10-25
100 101 102 100 101 102
generations generations

(a) (b)
Figure 7. Convergence curve results of simulations. The fitness value represents the objective
function evaluation. (a) Convergence curve of Simulation 1. (b) Convergence curve of Simulation 2.

As it may be appreciated from Figure 8, the robots converge to the desired formation
pattern for Simulation 1. Figure 8a reports the position trajectories of each robot, in cartesian
coordinates. Additionally, the achieved formation pattern is drawn in Figure 8b. Notice
that the formation shape rotates to avoid collision among robots, reaching the desired
formation from Figure 6a.

Position trajectories in the Cartesian xy-plane Achieved formation pattern

3 3

1 1

0.5 0.5
4 4
y (m)

y (m)

1 bot 1 1
0 bot 2 0
bot 3
bot 4
-0.5 -0.5

-1 2 -1 2

-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5

x (m) x (m)

(a) (b)
Figure 8. Results of position trajectories and achieve formation pattern, in cartesian xy-plane of
Simulation 1. The initial robot’s positions are identified with asterisk markers. m: meters. (a) Position
trajectories of Simulation 1. (b) Achieved formation of Simulation 1.
Mathematics 2023, 11, 3633 12 of 19

The formation control results along time of Simulation 1 are reported in Figure 9.
The position trajectories show smooth displacements (see Figure 9a). The control signals
are also smooth, as can be seen in Figure 9b. Moreover, the robot reached its formation
after 30 s.
Position trajectories along time Input control signals along time
1.5 1

1
0.5
0.5

rad/s, m/s
m

0 0

-0.5
-0.5
-1

-1.5 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
time (s) time (s)

(a) (b)
Figure 9. Results of position trajectories and input control signals, along time of Simulation 1. m: meters,
s: seconds. (a) Position trajectories of Simulation 1. (b) Input control signals of Simulation 2.

Cartesian paths followed by the robots reached the desired formation (see Figure 10a).
Notice that Robot 2 and Robot 3 avoid a possible collision. Moreover, the achieved forma-
tion in Figure 10b is the one expected in Figure 6b. As can be observed, the final formation
is rotated since the formation adapts its shape to avoid collisions.

Position trajectories in the Cartesian xy-plane Achieved formation pattern

3
1.5
3
1
1
0.5 2
0.5 2

0
0
y (m)

y (m)

bot 1 1
1 bot 2 -0.5
-0.5 bot 3
bot 4
-1 -1

-1.5 -1.5
4 4

-2 -1 0 1 2 -1 0 1 2
x (m) x (m)

(a) (b)
Figure 10. Results of position trajectories and achieve formation pattern, in cartesian xy-plane of
Simulation 2. The initial robot’s positions are identified with asterisk markers. m: meters. (a) Position
trajectories of Simulation 2. (b) Achieved formation of Simulation 2.

Figure 11 presents the results of Simulation 2, which are the position trajectories and
the input control signals along time. Both the position displacements and the control signals
of each robot are smooth, as shown in Figure 11a and Figure 11b, respectively. In this case,
the robots reached the desired formation after 25 s.
Mathematics 2023, 11, 3633 13 of 19

Position trajectories along time Input control signals along time

2 1

1 0.5

rad/s, m/s
m 0 0

-1 -0.5

-2 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
time (s) time (s)

(a) (b)
Figure 11. Results of position trajectories and input control signals, along time of Simulation 2. m:
meters, s: seconds. (a) Position trajectories of Simulation 2. (b) Input control signals of Simulation 2.

6.2. Real-World Experiments

Two tests were performed in real-world experiments called Experiment 1 and Experi-
ment 2. Experiment 1 considers a network of three robots interconnected as illustrated in
Figure 12a, and Experiment 2 considers five interconnected robots, as shown in Figure 12b.
Notice that the dimensionality of these tests is quite different since D = 3 in Experiment 1,
while D = 10 in Experiment 2. Moreover, the desired formation patterns for Experiments
A and B, are included in Figure 13a and Figure 13b, respectively.

2 𝑘23
𝑘12
𝑘12 𝑘25
1 2 1 𝑘13 3
𝑘14
𝑘13 𝑘23
𝑘15 𝑘24 𝑘34
3
𝑘35
5 4
𝑘45
(a) (b)
Figure 12. Communication topology among robots for real-world experiments. Each k ij value
must be adjusted to define a desired formation. (a) Communication topology for Experiment 1.
(b) Communication topology for Experiment 2.

𝑇 𝑇 𝑇
𝐡3 = 0 1 𝐡5 = −0.75 0.75 𝐡1 = 0.75 0.75
3 5 1
𝑦
𝑦

𝑥 𝑥
2
𝑇
𝐡2 = 0 0

2 1 4 3
𝑇 𝐡1 = 1 −1 𝑇 𝑇 𝑇
𝐡2 = 0 −1 𝐡4 = −0.75 −0.75 𝐡3 = 0.75 −0.75

(a) (b)
Figure 13. Desired formation patterns in cartesian coordinates hi used for real-world experiments.
The xy frame represents the centroid of the formation pattern. (a) Formation pattern for Experiment 1.
(b) Formation pattern for Experiment 2.
Mathematics 2023, 11, 3633 14 of 19

Table 2 reports the optimal k ij values of Experiment 1. In this case, k12 6= k13 6= k23 as
expected (see Figure 13a). The result f best = 1.6435 × 10−34 is smaller than the previous
tests since the formation pattern is easier with lower dimensionality.

Table 2. Optimization results of MMDE in Experiment 1. The table reports the optimal k ij values.
Moreover, f best is the objective function evaluation at the final generation.

k12 k13 k23 f best

0.9414 4.6514 3.7149 1.6435 × 10−34

Table 3 shows the optimal k ij values of Experiment 2. The obtained results are
k13 = k15 = k34 = k45 , k12 = k23 = k24 = k25 , and k14 = k35 , as required by the de-
sired formation pattern in Figure 13b. Notice that f best = 1.1843 × 10−16 is the most
significant result among tests since the required formation pattern is more challenging to
solve. However, the optimal k ij values are still small enough for the formation control task.

Table 3. Optimization results of MMDE in Experiment 2. The table reports the optimal k ij values.
Moreover, f best is the objective function evaluation at the final generation.

k12 k13 k14 k15 k23 f best

1.0563 2.0904 4.1825 2.0904 1.0563
k24 k25 k34 k35 k45 1.1843 × 10−16
1.0563 1.0563 2.0904 4.1825 2.0904

The convergence curve results for real-world experiments are illustrated in Figure 14.
As can be seen, the convergence curve looks similar in both tests. These curves are also
similar to the simulation experiment results (see Figure 7). However, previous objec-
tive function evaluations suggest that more generations are required for more complex
communications topologies.

Convergence curve result Convergence curve result

100
100

10-10
10-5
fitness value

fitness value

10-20

10-10
-30
10

10-15
10-40
100 101 102 100 101 102
generations generations

(a) (b)
Figure 14. Convergence curve results of real-world experiments. The fitness value represents the
objective function evaluation. (a) Convergence curve of Experiment 1. (b) Convergence curve of
Experiment 2.

Figure 15 illustrates that robots reached the desired formation pattern for Experiment 1.
In this case, Figure 15a shows that position trajectories in the plane are smooth. Moreover,
there was no risk of collision. However, the edge-weight control still rotates the center of
the formation to best achieve the desired formation pattern (see Figure 15b).
Mathematics 2023, 11, 3633 15 of 19

Position trajectories in the Cartesian xy-plane Achieved formation pattern

1 1
1 1
2 2
0.5 0.5

bot 1
0 0
y (m)

y (m)
bot 2
bot 3

-0.5 -0.5

-1 -1

3 3

-1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1 1.5

x (m) x (m)

(a) (b)
Figure 15. Results of position trajectories and achieve formation pattern, in cartesian xy-plane of
Experiment 1. The initial robot’s positions are identified with asterisk markers. m: meters. (a) Position
trajectories of Experiment 1. (b) Achieved formation of Experiment 1.

Figure 16 presents the position trajectories and input control signals along time of
Experiment 1. As it may be appreciated from Figure 16a, trajectories reached the forma-
tion pattern around 20 s. Clearly, the control signals start saturated, but after 10 s their
performance becomes smooth (see Figure 16b).

Position trajectories along time Input control signals along time

1 1

0.5
0.5
rad/s, m/s

0
m

0
-0.5

-0.5
-1

-1.5 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
time (s) time (s)

(a) (b)
Figure 16. Results of position trajectories and input control signals, along time of Experiment 1. m:
meters, s: seconds. (a) Position trajectories of Experiment 1. (b) Input control signals of Experiment 1.

Regarding Experiment 2, the formation pattern results are provided in Figure 17. In this
case, the initial positions of robots 2 and 4 are proposed in a way that their trajectories cross
to show that the controller avoids the collision. However, the cartesian path followed by
the robots reported that the collision is avoided (see Figure 17a). Moreover, the formation
pattern is achieved successfully, as illustrated in Figure 17b.
Mathematics 2023, 11, 3633 16 of 19

Position trajectories in the Cartesian xy-plane Achieved formation pattern

1 1

0.5 5 0.5 5

bot 1
0 bot 2 0
y (m)

y (m)
2 bot 3 2
bot 4
bot 5
-0.5 -0.5

4 4

-1 3 -1 3

-0.5 0 0.5 1 1.5 -0.5 0 0.5 1 1.5

x (m) x (m)

(a) (b)
Figure 17. Results of position trajectories and achieve formation pattern, in cartesian xy-plane of
Experiment 2. The initial robot’s positions are identified with asterisk markers. m: meters. (a) Position
trajectories of Experiment 2. (b) Achieved formation of Experiment 2.

Figure 18 reports the formation control results along time of Experiment 2. The desired
formation is reached after 30 s. Figure 18a shows that the robots converge smoothly to
the desired formation pattern at around 30 s. However, input control signals reported in
Figure 18b are not smooth as previous tests. Although this test is more complex than the
previous ones, many drawbacks of the real-world implementations arose, such as non-
modeled dynamics, external perturbations, robot wear conditions, and imperfections in the
surface. Moreover, ROS® packages pass through a local network via Wi-Fi communications
involving loss of packages and delays. Despite all these drawbacks, the control strategy
succeeds in achieving the required formation.

Position trajectories along time Input control signals along time

1.5 1

1
0.5
rad/s, m/s

0.5
m

0
0

-0.5
-0.5

-1 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
time (s) time (s)

(a) (b)
Figure 18. Results of position trajectories and input control signals, along time of Experiment 2.
m: meters, s: seconds. (a) Position trajectories of Experiment 2. (b) Input control signals of Experiment 2.

Finally, the results of the real-world implementations using Turtlebot3® Waffle Pi

robots are included in Figure 19. The achieved formations patterns are highlighted with
green lines which approximate the reached formations.
Mathematics 2023, 11, 3633 17 of 19

(a) (b)
Figure 19. Achieved formation pattern in the real-world implementation using Turtlebot3® Waffle Pi
robots. Green lines approximate the reached formation pattern. Each number indicates the ith robot.
(a) Achieved formation of Experiment 1. (b) Achieved formation of Experiment 2.

7. Discussion
Desired formation patterns are often given in cartesian coordinates instead of provid-
ing gain weights since cartesian coordinates are easier to define with respect to the center of
the formation. The formation pattern choosing the gain weights in edge-weighted graphs
can be complicated, especially for complex robot networks. The proposed MMDE-based
formation pattern algorithm has successfully provided optimal gain weights given cartesian
coordinates, which was the main contribution of the paper.
It has been shown that the edge-weighted formation control exploits the properties
of weighted graphs to allow the formation to rotate and adapt its shape to avoid collision
among robots.
Moreover, all experiments were performed based on a complete graph topology.
However, not all the vertices are required to be connected, since the communication
topology only requires an undirected, static, connected, and weighted graph [11]. It is let
as a future work to study the impact of the MMDE-based formation pattern, varying the
vertices connections.
The performance of the edge-weight control scheme was admirable in real-world
implementations since ROS® packages pass through a local network via Wi-Fi communica-
tions involving loss of packages and delays. Additionally, the results of simulations and
real-world experiments indicate that the robots reached the desired formation patterns with
smooth trajectories in the xy-plane. However, the input control signals of the experiments
were not as smooth as in the simulation. This is mainly due to non-modeled dynamics,
external perturbations, loss of packages, and other drawbacks of the physical implementa-
tions which are not considered in simulated scenarios. Despite all these inconveniences,
the edge-weight control performs well.
In this work, the MMDE-based formation pattern algorithm is performed offline,
while the edge-weighted formation control drives the mobile robots online. It is left to
future work to propose an online optimization-based formation pattern. In this manner,
the desired formation pattern can be changed during online operation.
Moreover, the collision avoidance capacities of edge-weighted formation control can
be combined with pin control of complex networks, adding a dynamic topology without
restricting complete graph connection in complex networks [9].

8. Conclusions
This paper addressed an edge-weighted consensus-based formation control for dif-
ferential drive robots. An MMDE-based formation pattern algorithm was proposed to
provide the gain weights in edge-weighted graphs, given a desired formation in cartesian
coordinates. Then, the edge-weighted formation control drives the mobile robots to the
desired formation control, based on the adjusted gain weights. Simulation and real-world
experiments were performed to verify the effectiveness and applicability of the proposed
Mathematics 2023, 11, 3633 18 of 19

approach. Experiments considered a communication topology for multi-robot systems

composed of three, four, and five mobile robots.
For the MMDE-based formation pattern algorithm, MMDE demonstrated its exploita-
tion capacities with fast learning rates. MMDE has provided objective function evaluations
below to 1.1843 × 10−16 for a network of five robots, which was the most challenging ex-
periment. The smaller the objective function evaluations, the better precision the formation
achieved will have.
Moreover, the reported optimal gain weights were adequate in all experiments. With re-
spect to the edge-weighted formation control, the simulation and real-world implementa-
tion demonstrates the capacity of the control scheme to avoid collision among robots.
The applicability of the proposed method was tested with an experimental setup
composed of an OptiTrack® motion capture system and up to five Turtlebot3® Waffle Pi.
The main difference between simulation and real-world experiments was the computed
control signals. The control signals results of real experiments were not smooth as the
simulation results, due to non-modeled dynamics, external perturbation, imperfection in
the surface, and robot wear conditions. However, mobile robots converge to the desired
formation pattern performing smooth position trajectories in the xy-plane.
Finally, besides using different robots in the network topology, more future studies
can be proposed, including trajectory tracking in formation with collision avoidance in the
presence of dynamic objects.

Author Contributions: Conceptualization, J.H.-B. and T.H.; Data curation, J.D.R. and J.H.-B.; Funding
acquisition, A.Y.A. and M.P.-C.; Investigation and software, J.D.R. and T.H.; Methodology, J.H.-B.
and T.H.; Project administration and supervision, A.Y.A. and M.P.-C.; Validation, J.H.-B., T.H. and
J.D.R.; Visualization and Writing, J.H.-B., J.D.R., T.H. and A.Y.A. All authors read and agreed to the
published version of the manuscript.
Funding: This research was funded by CONAHCYT Mexico, through Project PCC-2022-319619.
Data Availability Statement: The data analyzed are available from the authors upon request.
Acknowledgments: The authors also thank Universidad de Guadalajara for their support in this re-
search.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

EAs Evolutionary Algorithms

DE Differential Evolution
MMDE Multi-strategy Mutation Differential Evolution
ROS Robot Operating System

References
1. Issa, B.A.; Rashid, A.T. A survey of multi-mobile robot formation control. Int. J. Comput. Appl. 2019, 181, 12–16.
2. Latora, V.; Nicosia, V.; Russo, G. Complex Networks: Principles, Methods and Applications; Cambridge University Press: Cambridge,
UK, 2017.
3. Jadbabaie, A.; Lin, J.; Morse, A.S. Coordination of groups of mobile autonomous agents using nearest neighbor rules. IEEE Trans.
Autom. Control. 2003, 48, 988–1001. [CrossRef]
4. Godsil, C.; Royle, G.F. Algebraic Graph Theory; Springer Science & Business Media: Berlin/Heidelberg, Germany, 2001; Volume 207.
5. Ren, W.; Beard, R.W.; Atkins, E.M. Information consensus in multivehicle cooperative control. IEEE Control. Syst. Mag. 2007,
27, 71–82.
6. Grandi, R.; Falconi, R.; Melchiorri, C. Coordination and control of autonomous mobile robot groups using a hybrid technique
based on particle swarm optimization and consensus. In Proceedings of the 2013 IEEE International Conference on Robotics and
Biomimetics (ROBIO), Shenzhen, China, 12–14 December 2013; IEEE: Piscataway, NJ, USA, 2013; pp. 1514–1519.
7. Jin, J.; Kim, Y.G.; Wee, S.G.; Gans, N. Consensus based attractive vector approach for formation control of non-holonomic mobile
robots. In Proceedings of the 2015 IEEE International Conference on Advanced Intelligent Mechatronics (AIM), Busan, Republic
of Korea, 7–11 July 2015; IEEE: Piscataway, NJ, USA, 2015; pp. 977–983.
Mathematics 2023, 11, 3633 19 of 19

8. Molinari, F.; Raisch, J. Efficient consensus-based formation control with discrete-time broadcast updates. In Proceedings of the
2019 IEEE 58th Conference on Decision and Control (CDC), Nice, France, 11–13 December 2019; IEEE: Piscataway, NJ, USA, 2019;
pp. 4172–4177.
9. Rios, J.D.; Ríos-Rivera, D.; Hernandez-Barragan, J.; Pérez-Cisneros, M.; Alanis, A.Y. Formation Control of Mobile Robots Based
on Pin Control of Complex Networks. Machines 2022, 10, 898. [CrossRef]
10. Nuno, E.; Loria, A.; Hernández, T.; Maghenem, M.; Panteley, E. Distributed consensus-formation of force-controlled non-
holonomic robots with time-varying delays. Automatica 2020, 120, 109114. [CrossRef]
11. Falconi, R.; Sabattini, L.; Secchi, C.; Fantuzzi, C.; Melchiorri, C. Edge-weighted consensus-based formation control strategy with
collision avoidance. Robotica 2015, 33, 332–347. [CrossRef]
12. Secchi, C.; Fantuzzi, C. Formation control over delayed communication networks. In Proceedings of the 2008 IEEE International
Conference on Robotics and Automation, Pasadena, CA, USA, 19–23 May 2008; IEEE: Piscataway, NJ, USA, 2008; pp. 563–568.
13. Falconi, R.; Sabattini, L.; Secchi, C.; Fantuzzi, C.; Melchiorri, C. A graph–based collision–free distributed formation control
strategy. IFAC Proc. Vol. 2011, 44, 6011–6016. [CrossRef]
14. Darvishpoor, S.; Darvishpour, A.; Escarcega, M.; Hassanalian, M. Nature-Inspired Algorithms from Oceans to Space: A
Comprehensive Review of Heuristic and Meta-Heuristic Optimization Algorithms and Their Potential Applications in Drones.
Drones 2023, 7, 427. [CrossRef]
15. Slowik, A.; Kwasnicka, H. Evolutionary algorithms and their applications to engineering problems. Neural Comput. Appl. 2020,
32, 12363–12379. [CrossRef]
16. Bilal; Pant, M.; Zaheer, H.; Garcia-Hernandez, L.; Abraham, A. Differential Evolution: A review of more than two decades of
research. Eng. Appl. Artif. Intell. 2020, 90, 103479. [CrossRef]
17. Ahmad, M.F.; Isa, N.A.M.; Lim, W.H.; Ang, K.M. Differential evolution: A recent review based on state-of-the-art works. Alex.
Eng. J. 2022, 61, 3831–3872. [CrossRef]
18. Xiao, P.; Zou, D.; Xia, Z.; Shen, X. Multi-strategy different dimensional mutation differential evolution algorithm. Proc. AIP Conf.
Proc. 2019, 2073, 020102 .
19. Siciliano, B.; Sciavicco, L.; Villani, L.; Oriolo, G. Force Control; Springer: Berlin/Heidelberg, Germany, 2009.
20. Ren, W.; Cao, Y. Distributed Coordination of Multi-Agent Networks: Emergent Problems, Models, and Issues; Springer:
Berlin/Heidelberg, Germany, 2011; Volume 1.
21. Storn, R.; Price, K. Differential evolution-a simple and efficient heuristic for global optimization over continuous spaces. J. Glob.
Optim. 1997, 11, 341. [CrossRef]
22. Simon, D. Evolutionary Optimization Algorithms; John Wiley & Sons: Hoboken, NJ, USA, 2013.
23. Amsters, R.; Slaets, P. Turtlebot 3 as a robotics education platform. In Robotics in Education: Current Research and Innovations 10;
Springer: Berlin/Heidelberg, Germany, 2020; pp. 170–181.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.