A SwarmBased Flocking Control Algorithm for Exploration and

Coverage of Unknown Environments

FREDY MARTINEZ, ANGELICA RENDON, FERNANDO MARTINEZ

Facultad Tecnológica

Universidad Distrital Francisco José de Caldas

Carrera 7 No. 40b53, Bogotá

COLOMBIA

Abstract:  The exploration of unknown environments can be beneficial for a variety of applications, such as in

spection of industrial equipment, environmental monitoring, or search and rescue missions. In order to tackle this

problem, swarm robotics has emerged as a promising approach due to its ability to leverage the collective behavior

of a group of robots to explore an area efficiently. This paper proposes a swarmbased control algorithm for ex

ploration and coverage of unknown environments. The algorithm utilizes shortrange distributed communication

and sensing among agents, with no central unit, to coordinate the swarm’s navigation and search tasks. This sens

ing is prioritized in the outermost agents of the swarm to reduce processing and energy costs, and these positions

can be rotated with other agents in the swarm. The formation rules that keep the system cohesive are simple and

independent of the individual robot characteristics, enabling the use of heterogeneous agents. The performance

of the proposed strategy is demonstrated through experiments in coverage and search tasks, and compared with

other swarm strategies. The results show the effectiveness of the proposed algorithm for exploration and coverage

of unknown environments. The research presented in this paper has the potential to contribute to the development

of more efficient and effective swarmbased exploration and coverage strategies.

KeyWords:  Coverage, exploration, flocking control, robotics, swarmbased, unknown environments

Received: November 28, 2022. Revised: May 9, 2023. Accepted: June 13, 2023. Published: July 19, 2023.

1 Introduction

The exploration of unknown environments is an

essential aspect of many applications, including en

vironmental monitoring, crop analysis and monitor

ing, healthcare, industrial equipment inspection, and

search and rescue missions, military applications, and

space exploration, [1], [2]. Robotics has become an

increasingly popular approach to explore unknown

environments, with swarm robotics emerging as a

promising technique due to its ability to utilize the

collective behavior of a group of robots to explore an

area efficiently, [3]. Swarmbased control algorithms

have been developed to enable the coordination of a

swarm’s navigation and search tasks in an unfamiliar

environment.

Swarm Robotics is a technique that replaces a

single robot with a group of agents of simpler and

cheaper design with limited processing and sensing

capabilities, but with the ability to selforganize into a

multiagent system inspired by behaviors observed in

nature, [4]. One such behavior is the Flocking Behav

ior, derived from migratory birds and modeled from

basic behaviors that enable functional replication of

these systems, [5], [6]. Numerous experiments have

demonstrated the flocking dynamics’ ability to solve

problems in complex environments, such as aquatic

and aerial tasks, [7], [8].

The basic rules of behavior for achieving flocking

in a multiagent system were postulated by Reynolds

in 1986, [5]. These rules establish criteria for each

agent’s motion strategy in the system to avoid col

lision, perform velocity matching according to its

neighbors’ motion, and define a flocking axis. The

performance of each behavior is measured by met

rics on the group’s area and polarization, [9], [10].

While the flocking strategy prioritizes the distance be

tween agents, there is another strategy, line flocking,

in which agents move in a Vshape, similar to geese

migrating. However, the basic rules of cluster flock

ing can also give rise to line flocking, [11].

In light of the aforementioned challenges, the pri

mary aim of this research is to propose, implement,

and evaluate a novel swarmbased control algorithm

for the exploration and coverage of unknown envi

ronments. More specifically, we seek to address the

issues of efficient task coordination, robustness and

resilience, resource utilization, and adaptability of

the swarm system. We achieve this through the use

of distributed communication and sensing strategies,

coupled with rotation of roles within the swarm to

optimize processing and energy use. Furthermore,

the proposed algorithm is designed to function effec

tively with heterogeneous agents, thereby enhancing

its practical applicability in realworld scenarios. It is

our intention to demonstrate, through rigorous experi

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

195

Volume 18, 2023

mental trials, that our proposed approach not only im

proves upon the limitations of current swarm strate

gies, but also contributes substantively to the ongoing

evolution of swarm robotics.

Despite the growing interest in autonomous robots

for exploring unknown environments, current ap

proaches have limitations that can hinder their ef

fectiveness. For example, singlerobot approaches

may have poor scalability, limited coverage area, and

struggle to adapt to complex environments, while

swarmbased approaches face challenges such as ro

bustness and resilience, uncertainty, and efficient co

ordination of navigation and search tasks, [12]. To ad

dress these limitations, this paper proposes a swarm

based control algorithm that utilizes shortrange dis

tributed communication and sensing among agents to

coordinate the swarm’s tasks in an unfamiliar envi

ronment, [13]. By prioritizing sensing in the outer

most agents of the swarm, the proposed algorithm re

duces processing and energy costs, [14]. Moreover,

the simple formation rules that keep the system co

hesive are independent of individual robot character

istics, allowing for the use of heterogeneous agents.

This paper contributes to the development of more

efficient and effective swarmbased exploration and

coverage strategies.

Exploring unknown environments using swarm

based robotics poses several technical challenges that

need to be addressed for effective performance, [15],

[16]. Firstly, swarmbased systems must ensure the

robustness and resilience of the multiagent system

to maintain the collective behavior despite individual

agent failure or communication loss, [17]. Secondly,

swarmbased systems require efficient coordination

of navigation and search tasks, which can be compli

cated by the uncertainty and dynamic nature of the

environment, [18]. Thirdly, efficient use of available

resources, such as processing power and energy, is

critical for swarmbased systems, especially for long

term operations, [3]. Finally, the design and imple

mentation of swarmbased systems must be flexible

enough to adapt to different environments, tasks, and

agent characteristics, as well as to integrate with ex

isting systems.

The paper is organized as follows. Section 2 de

scribes the context and relevant previous research.

Section 3 presents preliminary concepts, including the

interaction between the robots, the flocking rules, and

and the communication strategy between agents. Sec

tion 4 presents the strategy of our algorithm, as well

as the implementation considerations. Section 5 sum

marizes the results achieved in the laboratory tests.

Section 6 presents further extensions and concludes

the paper.

2 Background

Autonomous robots can be utilized for various ap

plications, including environmental monitoring, crop

analysis and monitoring, industrial equipment inspec

tion, and search and rescue missions. These ap

plications typically require robots to operate in un

known environments, and thus require effective ob

stacle avoidance and exploration strategies. Rein

forcement learning (RL) based control has been stud

ied as a potential solution for obstacle avoidance

in autonomous underwater vehicles (AUVs), [19].

However, standard onestep Qlearning based con

trol has a tendency to explore all possible actions at

a given state, which may lead to an increased num

ber of collisions. Modified Qlearning based control

approaches have been proposed to deal with this prob

lem in unknown environments, [20].

Target search control of AUVs in underwater en

vironments has also been studied using deep RL and a

grid map of the search environment, [21]. In order to

enhance the area coverage of unmanned aerial vehicle

(UAV) swarms, a novel mobility model that combines

an Ant Colony algorithm with chaotic dynamics has

been presented, [22]. This model has been extended

by the addition of a collision avoidance mechanism

and testing its efficiency in terms of area coverage by

UAV swarm.

Cooperative exploration strategies have been pro

posed for multiple mobile robots, which reduce

the overall task completion time and energy costs

compared to conventional methods, [3]. Hex

decompositionbased coverage planning algorithms

have also been proposed for unknown, obstacle

cluttered environments, [23]. In addition, a novel

swarmbased control algorithm for exploration and

coverage of unknown environments, while maintain

ing a formation that permits shortrange communica

tion, has been proposed, [24].

Furthermore, agents cannot visit arbitrary loca

tions in many applications due to a priori unknown

safety constraints, which has led to the development

of efficient density learning algorithms for solving the

coverage problem while preserving the agents’ safety,

[25]. The development of effective obstacle avoid

ance and exploration strategies is crucial for the suc

cess of autonomous robots in a variety of applications.

3 Problem Formulation

In recent years, multiagent systems have become

increasingly popular due to their ability to perform

tasks more efficiently than a single agent system. A

multiagent system consists of nagents, which can be

robots in their physical implementation, operating in a

closed environment W ⊂ R2that is bounded by ∂W.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

196

Volume 18, 2023

The environment includes finite obstacles that form a

set Oof areas that are inaccessible to the robots, and

a free navigation space Edefined by W − O. The

behavior of the system is a result of the interaction of

each agent, which operates autonomously without a

central control unit. The behavior is determined by a

set of basic rules that guide the agent to move towards

the destination point while avoiding collisions with

obstacles and other agents and maintaining formation.

These rules are specified by the agent’s velocity ˙

xi(t)

and orientation θi (t), which are tuned based on the

system’s behavioral policy and sensed information in

the environment. However, not all agents perform en

vironmental sensing, only those in the outer layers of

the system, while others maintain communication and

sensing with other agents.

In more mathematical detail, each agent is repre

sented as aifor i∈1,2, ..., n where nis the total num

ber of agents in the system. Each agent is associated

with a state vector siconsisting of its position xi(t)

and orientation θi (t). The velocity vector ˙

xi(t)is

derived from the state vector and is used to determine

the agent’s movement in the next time step based on

the system’s behavioral policy and the agent’s local

sensing information.

Task allocation within the system is another crit

ical aspect that we aim to address in this research.

Since the agents in the system can be heterogeneous,

the task allocation should account for the varying ca

pabilities and available resources of each agent. We

propose a dynamic task allocation approach in which

the roles of agents, such as sensing the environment

and communicating with other agents, are rotated

based on their current state and the overall task re

quirements. The effectiveness of this task allocation

strategy is quantified through measures like system

efficiency, task completion time, and the utilization

of agent resources.

The performance of our multiagent system is

evaluated based on several measures. These include

the overall coverage of the environment, the time

taken to complete the task, and the robustness of the

system in terms of its ability to adapt to agent fail

ures or loss of communication. We also examine the

resource usage of the system in terms of the com

putational and energy demands placed on the agents.

Through our proposed control algorithm, we aim to

improve these performance measures, demonstrating

the efficacy of our approach in the context of explo

ration and coverage of unknown environments.

One of the advantages of a multiagent system is

that it allows for the formation of a swarm of robots

that can be heterogeneous. This means that each agent

can have different capabilities and sensors, allowing

for a more efficient distribution of tasks. Addition

ally, the distributed architecture of the system enables

each agent to exchange information with neighboring

agents directly without relying on a central processing

unit. This architecture is robust as the loss of a subset

of agents does not significantly impact the overall per

formance of the system in completing the task. How

ever, implementing such a system on lowcost hard

ware with limited sensing and communication capa

bilities presents a significant challenge. Nonetheless,

the potential benefits of multiagent systems make

them an attractive research area in robotics.

4 Methods

The research problem concerns the coverage of an

observable but unknown environment, with restric

tions on processing and communication capabilities

of swarm robots. To conduct our experiments, we

built a system consisting of simple differential robots

assembled with 3 mm acrylic plates, equipped with

servo motors that provide a maximum linear displace

ment speed of 37.4 cm/s, and an Espressif Systems

ESP32 microcontroller (Fig. 1). We use an RPLI

DAR A1M8R6 as a distance sensor. The environ

ment’s limits and obstacles Oare preestablished, but

their position within those limits is unknown. To ad

dress this problem, we propose a distributed (decen

tralized) solution using Bluetooth Low Energy (BLE)

v4.2, which has a maximum transmission speed of 1

Mbps and an approximate range of 50 m. The use

of BLE technology allows for shortrange commu

nication and reduced power consumption, making it

a suitable option for lowcost and lowpower robots.

The ESP32 microcontroller has builtin BLE support,

making it an ideal platform for implementing our so

lution. The use of decentralized communication and

control reduces the need for a centralized control unit,

providing a robust and faulttolerant system that can

continue operating even if some robots fail. This ap

proach enables each robot to make decisions based

on local readings and interactions with neighboring

robots, facilitating the coverage of the unknown envi

ronment with limited resources.

To ensure high precision and efficiency in the cov

erage task, our scheme utilizes a twodimensional lo

cal coverage matrix that is stored in the memory of

each robot’s ESP32 microcontroller. This matrix is

composed of cells, whose size defines the resolution

with which the environment is divided into regions,

and which is related to the actual size of the robots

(robot’s shadow in the environment). Each row of the

matrix corresponds to a section of the region, and each

column corresponds to a robot. The size of the matrix

is defined in advance in the robot’s code and consid

ers both the number of robots in the environment and

the expected distance between agents of the system.

If the robot is in an undiscovered or covered cell, it

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

197

Volume 18, 2023

Figure 1: Differential robot used in laboratory tests

is marked with a 1; if the robot detects an obstacle in

the cell, the cell is marked with a 2; otherwise, it is

marked with a 0. This local coverageobstacle matrix

of the agent is called the CO matrix (Fig. 2).

To reduce network load and minimize the im

pact of issues such as loss of lineofsight between

pairs of robots, when the distance between two agents

is smaller than the communication radius, they ex

change their local CO matrices. This localized shar

ing of CO matrices is facilitated by swarming move

ments since the robots are continually pulled together

due to the cohesion force that controls their move

ment. The advantage of using a local CO matrix is

that it enables each robot to have a detailed map of the

environment that can be used to plan its movements

and avoid obstacles, without the need for a central

ized processing unit. Moreover, the use of shortrange

communication facilitated by the swarm movement

reduces the impact of communication delays and fail

ures, increasing the overall efficiency of the coverage

task.

The robots attempt to navigate to the nearest unex

plored region according to the gradient of the search

sensor. The search sensor is the sensor installed on the

robot capable of detecting the particular signal emit

ted by the navigation target point (such as a gas sig

nal in the case of gas leak detection). This signal can

propagate in different ways in the environment ac

cording to the nature of the signal itself. For practical

purposes, in this study, it is assumed that the signal

power decreases linearly as it propagates, so its in

tensity at the target point is maximum, and minimum

in the concentric circle around the target point with

radius equal to the propagation distance. This behav

Figure 2: An example of a local coverageobstacle

CO matrix of the agent

ior forms the search gradient in the environment. The

value of the gradient detected by the robot, together

with the unexplored regions, sets the navigation route

for the robots at the edges of the swarm. This route

is established only by the robots at the front of the

system, identified by the direction of the system’s ad

vance. The robots in the other peripheries behave like

internal robots until the system changes its direction

of advance. The rest of the system navigates main

taining the organized structure of the system. The

distances between robots are programmed in advance

in the microcontroller and controlled through the Li

DAR and CO matrices of each agent. Over time, and

depending on the battery capacity in the robots, the

robots at the periphery of the swarm exchange posi

tions with the robots inside the system.

To minimize collisions between agents, a repul

sion parameter is incorporated based on the distances

detected by the LiDAR, which takes into account the

distance between agents, their speed, and the CO ma

trix of the robot. This parameter is crucial in environ

ments where there are obstacles or when robots are

navigating in close proximity to one another. It helps

to maintain the desired separation distance between

agents, which is necessary to prevent collisions while

ensuring efficient exploration and coverage. The re

pulsion parameter is dynamically adjusted based on

the current sensor readings, and it is integrated into

the agents’ decisionmaking process to ensure safe

and effective navigation. Additionally, the repulsion

parameter can be adjusted to meet specific environ

mental conditions, such as in highly cluttered envi

ronments, where a higher repulsion parameter may be

needed to avoid collisions.

The proposed swarmbased control algorithm was

designed using a variety of elements, including sen

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

198

Volume 18, 2023

sors, communication modules, and actuators. The

selection criteria for each of these elements were

based on their availability, robustness, accuracy, en

ergy consumption, and cost. The advantages of each

of these elements were considered when selecting the

best components for the prototype. The performance

of the proposed algorithm was evaluated using exper

iments in coverage and search tasks. The most im

portant characteristics and parameters used to charac

terize the system performance included energy con

sumption, accuracy, and efficiency.

The complete algorithm is summarized below:

1. Initialize the robots with the following:

• Differential drive system using servo mo

tors

• RPLIDAR A1M8R6 distance sensor

• ESP32 microcontroller with builtin BLE

support

• 2D local coverage matrix (CO matrix) with

cells of defined size

2. Define the environment limits and obstacles O,

and mark them on the CO matrix.

3. Implement decentralized communication and

control using BLE technology.

4. Initialize the CO matrix with values of 0 for

undiscovered or covered cells.

5. Begin exploration of the environment with

swarm of robots by following the steps:

• Calculate the gradient of the search sensor

and locate the nearest unexplored region.

• Calculate the navigation route for the robots

at the edges of the swarm based on the gra

dient and unexplored regions.

• The robots at the front of the swarm navigate

according to the established route.

• The robots in the other peripheries of the

swarm navigate maintaining the organized

structure of the system.

• The distances between robots are controlled

through the LiDAR and CO matrices of each

agent.

• Over time and depending on battery capac

ity, robots at the periphery of the swarm ex

change positions with the robots inside the

system.

6. Implement a repulsion parameter to minimize

collisions between agents.

• The repulsion parameter takes into account

the distance between agents, their speed,

and the CO matrix of the robot.

• The repulsion parameter is dynamically ad

justed based on the current sensor readings

to ensure safe and effective navigation.

• In highly cluttered environments, the repul

sion parameter is adjusted to avoid colli

sions.

7. When the environment is fully covered, end the

exploration process.

5 Findings and results

Four performance metrics are used to evaluate the

algorithm: two related to exploration and coverage

performance, and two related to selforganizing for

mation control. The first metric, coverage percentage

(CP), measures the percentage of the known region

covered by any robot of the system. CP is defined as

the ratio between the number of covered cells and the

total number of cells to be covered in the task. The

second metric, turnaround time (TT), summarizes the

coverage performance in terms of the time taken for

the robots to indirectly cover the known environment,

including mapping all unknown obstacles. The third

metric, grouping metric (G), estimates how closely a

set of robots are clustered together. Gis defined as the

summation of the average distances between agents in

all the robots in the system at a given moment, relative

to the population size. Finally, the fourth metric, or

der metric (C), computes how similarly the robots are

aligned. Cis defined as the summation of the average

relative velocities between agents in all the robots in

the system at a given moment, relative to the popula

tion size.

We conducted multiple experiments in a 2×3 m

rectangular environment, with different obstacle con

figurations Oand population sizes, while ensuring

sufficient free space for robot movement. Search

tasks involved different types of destination points,

including zero signal emission, one target point with

signal, and two or three target points with different

magnitudes of emitted signals. For simplicity, we

used two types of signals, sound and visible light from

LEDs. In the experiments, we recorded both the sys

tem’s ability to find the target points (task success)

and the mean and standard deviation of the four met

rics (Table 1).

Although the robots used in the experiments have

the same design (Fig. 1), they were constructed by the

research group, resulting in operational differences

between the prototypes. All robots moved at differ

ent maximum speeds despite using the same type of

wheels and servo motors, and even the behavior of the

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

199

Volume 18, 2023

Table 1: Comparative performance of the algorithm

against different test conditions. The obstacles re

mained constant in these results

two wheels on the same robot was different. These

functional variations served to evaluate the perfor

mance of the system with a heterogeneous structure,

deliberately forcing the agents’ velocities and behav

ior to be markedly different. In our experimental

setup, the robots are controlled by an onboard ESP32

microcontroller. We used 360degree LiDAR sensors

for distance detection.

The robots’ movement in the environment was

captured by a digital camera, and the frames were pro

cessed externally with Python. The location informa

tion was transmitted to the robots every 50 ms. These

tests were performed indoors; GPS mounted on each

robot is envisaged to replace this form of localization

in outdoor environments. The CO matrices for each

robot were shared with neighboring robots at an up

date rate of 25 ms. The proposed algorithm was im

plemented on each robot, ensuring that if one of the

robots fails due to hardware problems, the remaining

robots can still complete the task. In our experimental

simulations, the same settings were implemented us

ing Python. For measurement purposes, we assume

that once all cells in the local map of any robot are

covered directly or indirectly, a message completed

task is sent to the remaining robots, and the experi

ment ends.

The benefits of our study are multifaceted, as

we tackle critical issues in swarm robotics, particu

larly the exploration and coverage of unknown en

vironments. Firstly, our proposed algorithm exhib

ited strong performance metrics in all the conducted

tests. The consistent 100% coverage of the environ

ment demonstrates the algorithm’s robustness and re

liability, a significant advantage for realworld appli

cations where complete coverage is essential.

Secondly, the algorithm’s successful operation

amidst the heterogeneity of the swarm  due to op

erational differences in robots  validates its capabil

ity to handle disparities in individual agents. This is

a significant stride towards more practical and cost

effective swarm systems where agent uniformity can’t

be guaranteed.

Thirdly, our proposed dynamic task allocation

strategy and selforganizing formation control exhibit

great promise in enhancing the efficiency of swarm

systems. It effectively minimizes resource usage and

optimizes task completion time while maintaining the

robustness of the system in case of individual robot

failure.

Finally, the utilization of a distributed control ap

proach enables scalability and robustness, allowing

the swarm to handle a larger and more complex en

vironment without a significant performance drop or

the need for an overarching centralized control unit.

This research, therefore, paves the way for advance

ments in realworld applications like environmental

monitoring, search and rescue missions, and space ex

ploration.

Despite variations in total task times for similar

conditions, all developed tests achieved a 100% cov

erage of the environment (CP = 100% in all cases).

The differences in times are not significant at the 95%

confidence level.

6 Discussion

The research presented in this paper proposes a

distributed, decentralized solution for the coverage of

an unknown environment using a swarm of simple

differential robots with limited processing and com

munication capabilities. The proposed approach uti

lizes Bluetooth Low Energy (BLE) communication

technology to achieve shortrange communication be

tween robots, reducing power consumption and en

abling decentralized communication and control. The

use of a local CO matrix enables each robot to have

a detailed map of the environment that can be used to

plan its movements and avoid obstacles, without the

need for a centralized processing unit. The proposed

algorithm incorporates a repulsion parameter to min

imize collisions between agents and enable safe and

effective navigation.

The experimental results show that the proposed

algorithm achieved a 100% coverage of the environ

ment in all cases. The proposed algorithm was eval

uated using four performance metrics: coverage per

centage (CP), turnaround time (TT), grouping met

ric (G), and order metric (C). The evaluation demon

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

200

Volume 18, 2023

strated that the proposed algorithm achieved high pre

cision and efficiency in the coverage task.

The proposed algorithm has several advantages

over centralized approaches. First, decentralized

communication and control reduce the need for a cen

tralized control unit, providing a robust and fault

tolerant system that can continue operating even if

some robots fail. Second, the use of shortrange com

munication facilitated by the swarm movement re

duces the impact of communication delays and fail

ures, increasing the overall efficiency of the cover

age task. Third, the use of a local CO matrix enables

each robot to have a detailed map of the environment,

which can be used to plan its movements and avoid

obstacles, without the need for a centralized process

ing unit. Fourth, the incorporation of a repulsion pa

rameter minimizes collisions between agents and en

ables safe and effective navigation.

The algorithm provides a distributed, decentral

ized solution for the coverage of an unknown envi

ronment with limited resources. The use of short

range communication, local CO matrices, and a re

pulsion parameter enables efficient and precise cover

age while reducing the impact of communication de

lays and failures and minimizing collisions between

agents. The experimental results demonstrate the ef

fectiveness and efficiency of the proposed algorithm,

making it a promising approach for various applica

tions in different fields. Future work includes testing

the proposed algorithm in outdoor environments and

integrating it with GPS localization for improved ac

curacy and performance.

7 Conclusion

The main findings and contributions of this pa

per include the proposal of a novel swarmbased

control algorithm that enables the coordination of a

swarm’s navigation and search tasks in an unfamiliar

environment. The proposed algorithm utilizes short

range distributed communication and sensing among

agents, with no central unit, and prioritizes sensing in

the outermost agents of the swarm to reduce process

ing and energy costs. The algorithm also maintains

simple formation rules that are independent of indi

vidual robot characteristics, allowing for the use of

heterogeneous agents. This research has the poten

tial to contribute to the development of more efficient

and effective swarmbased exploration and coverage

strategies. Possible future work includes the imple

mentation of the proposed algorithm in larger envi

ronments and with more complex tasks.

Despite the promising results, our research is not

without its limitations. Our experiments were con

ducted in a relatively controlled indoor environment.

While this was sufficient for validating the proposed

algorithm, realworld scenarios can be considerably

more challenging and unpredictable. External fac

tors like weather conditions, signal interferences, and

unstructured terrain can significantly impact the per

formance of the swarm, suggesting a need for further

robustness in our algorithm to tackle such uncertain

ties. Another limitation is that our algorithm, in its

current state, may not perform as effectively when

scaled up for larger environments or more complex

tasks. It’s important to note that as the number of

robots and the complexity of tasks increase, the need

for efficient and robust communication among robots

also increases. The impact of communication delays,

packet loss, and other network issues will need to be

addressed in future iterations of the algorithm.

In light of these limitations, a few improvements

are suggested. Firstly, introducing adaptive behavior

in the algorithm, allowing the swarm to adjust their

formation and exploration strategy based on the im

mediate environment, could enhance the performance

of the swarm in complex and dynamic environments.

Secondly, it could be beneficial to incorporate ma

chine learning techniques to enable the swarm to learn

from past experiences and continuously improve their

performance over time.

As for future work, we aim to test the proposed

algorithm in larger and more complex environments,

including outdoor settings with realworld challenges.

We also plan to extend our research to explore how the

swarm can adaptively respond to dynamic changes

in the environment and optimize their strategy in

realtime. Furthermore, integration of our algorithm

with more advanced communication technology to

improve network resilience and robustness is another

avenue we are keen to explore.

Acknowledgment:

This work was supported by the Universidad Distri

tal Francisco José de Caldas, in part through CIDC,

and partly by the Facultad Tecnológica. The views

expressed in this paper are not necessarily endorsed

by Universidad Distrital. The authors thank the re

search group ARMOS for the evaluation carried out

on prototypes of ideas and strategies.

References:

[1] A. Q. Li, “Exploration and mapping with

groups of robots: Recent trends,” Current

Robotics Reports, vol. 1, no. 4, pp. 227–237,

2020. DOI: 10.1007/s43154-020-00030-5.

[2] V. Serebrenny and M. Shereuzhev, “Depen

dence of dynamics of multirobot system

on control architecture,” in Robotics: Indus

try 4.0 Issues &amp New Intelligent Control

Paradigms, Springer International Publishing,

2020, pp. 125–132. DOI: 10.1007 / 978- 3-

030-37841-7_10.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

201

Volume 18, 2023

[3] J. Hu, H. Niu, J. Carrasco, B. Lennox,

and F. Arvin, “Voronoibased multirobot au

tonomous exploration in unknown environ

ments via deep reinforcement learning,” IEEE

Transactions on Vehicular Technology, vol. 69,

no. 12, pp. 14 413–14 423, 2020, ISSN: 1939

9359. DOI: 10.1109/tvt.2020.3034800.

[4] Y. Kantaros and G. J. Pappas, “Scalable active

information acquisition for multirobot sys

tems,” in 2021 IEEE International Conference

on Robotics and Automation (ICRA), IEEE,

2021. DOI: 10 . 1109 / icra48506 . 2021 .

9561244.

[5] C. W. Reynolds, “Flocks, herds and schools:

A distributed behavioral model,” ACM SIG

GRAPH Computer Graphics, vol. 21, no. 4,

pp. 25–34, 1987, ISSN: 00978930. DOI: 10.

1145 / 37402 . 37406. [Online]. Available:

https : / / dl . acm . org / doi / 10 . 1145 /

37402.37406.

[6] F. Martínez, “Review of flocking organization

strategies for robot swarms,” Tekhnк, vol. 18,

no. 1, pp. 13–20, 2021, ISSN: 16928407.

[Online]. Available: https : / / revistas .

udistrital.edu.co/index.php/tekhne/

article/view/19257.

[7] R. BailonRuiz, A. BitMonnot, and S. Lacroix,

“Realtime wildfire monitoring with a fleet of

UAVs,” Robotics and Autonomous Systems,

vol. 152, no. 1, p. 104 071, 2022, ISSN: 0921

8890. DOI: 10 . 1016 / j . robot . 2022 .

104071. [Online]. Available: https://www.

sciencedirect . com / science / article /

abs/pii/S0921889022000355.

[8] T. Elmokadem and A. V. Savkin, “A method

for autonomous collisionfree navigation of a

quadrotor UAV in unknown tunnellike envi

ronments,” Robotica, vol. 40, no. 4, pp. 835–

861, 2021, ISSN: 02635747. DOI: 10.1017/

s0263574721000849.

[9] A. A. Paranjape, S.J. Chung, K. Kim, and

D. H. Shim, “Robotic herding of a flock

of birds using an unmanned aerial vehicle,”

IEEE Transactions on Robotics, vol. 34, no. 4,

pp. 901–915, 2018, ISSN: 15523098. DOI:

10 . 1109 / tro . 2018 . 2853610. [Online].

Available: https : / / ieeexplore . ieee .

org/document/8424544.

[10] F. Martínez, “Turtlebot3 robot operation for

navigation applications using ROS,” Tekhnк,

vol. 18, no. 2, pp. 19–24, 2021, ISSN:

16928407. [Online]. Available: https : / /

revistas . udistrital . edu . co / index .

php/tekhne/article/view/19261.

[11] J. Yang, R. Grosu, S. A. Smolka, and A.

Tiwari, “Love thy neighbor: Vformation as

a problem of model predictive control (in

vited paper),” en, in 27th International Con

ference on Concurrency Theory (CONCUR

2016), Schloss Dagstuhl  LeibnizZentrum

fuer Informatik GmbH, Wadern/Saarbruecken,

Germany, 2016, 4:1–4:5. DOI: 10 . 4230 /

LIPICS . CONCUR . 2016 . 4. [Online]. Avail

able: https://drops.dagstuhl.de/opus/

volltexte / 2016 / 6189 / pdf / LIPIcs -

CONCUR-2016-4.pdf.

[12] B. Aminof, A. Murano, S. Rubin, and F.

Zuleger, “Verification of agent navigation in

partiallyknown environments,” Artificial In

telligence, vol. 308, no. 103724, p. 103 724,

2022, ISSN: 00043702. DOI: 10 . 1016 /

j . artint . 2022 . 103724. [Online]. Avail

able: https : / / www . sciencedirect .

com / science / article / abs / pii /

S0004370222000649.

[13] M. S. Talamali, A. Saha, J. A. R. Marshall, and

A. Reina, “When less is more: Robot swarms

adapt better to changes with constrained com

munication,” Science Robotics, vol. 6, no. 56,

eabf1416, 2021, ISSN: 24709476. DOI: 10 .

1126/scirobotics.abf1416.

[14] E. Latif, Y. Gui, A. Munir, and R. Parasuraman,

“Energyaware multirobot task allocation in

persistent tasks,” arXiv, pp. 1–6, Dec. 31, 2021.

arXiv: 2112.15282v1 [cs.RO].

[15] L. Bobadilla, F. Martinez, E. Gobst, K. Goss

man, and S. M. LaValle, “Controlling wild mo

bile robots using virtual gates and discrete tran

sitions,” in 2012 American Control Confer

ence (ACC), IEEE, 2012, pp. 1–8. DOI: 10 .

1109 / acc . 2012 . 6315569. [Online]. Avail

able: https : / / ieeexplore . ieee . org /

document/6315569.

[16] J.G. Dong, “Avoiding collisions and pattern

formation in flocks,” SIAM Journal on Ap

plied Mathematics, vol. 81, no. 5, pp. 2111–

2129, 2021, ISSN: 00361399. DOI: 10.1137/

21m1390141.

[17] S. He, R. Xu, Z. Zhao, and T. Zou, “Vision

based neural formation tracking control of

multiple autonomous vehicles with visibility

and performance constraints,” Neurocomput

ing, vol. 492, no. 1, pp. 651–663, 2022, ISSN:

09252312. DOI: 10.1016/j.neucom.2021.

12.056. [Online]. Available: https://www.

sciencedirect . com / science / article /

abs/pii/S0925231221018944.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

202

Volume 18, 2023

[18] D. N. M. Hoang, D. M. Tran, T.S. Tran, and

H.A. Pham, “An adaptive weighting mech

anism for reynolds rulesbased flocking con

trol scheme,” PeerJ Computer Science, vol. 7,

no. 2021, e388, 2021, ISSN: 23765992. DOI:

10.7717/peerj-cs.388.

[19] P. Bhopale, F. Kazi, and N. Singh, “Reinforce

ment learning based obstacle avoidance for au

tonomous underwater vehicle,” Journal of Ma

rine Science and Application, vol. 18, no. 2,

pp. 228–238, 2019, ISSN: 19935048. DOI:

10.1007/s11804-019-00089-3.

[20] X. Cao, C. Sun, and M. Yan, “Target search

control of AUV in underwater environment

with deep reinforcement learning,” IEEE Ac

cess, vol. 7, no. 1, pp. 96 549–96 559, 2019,

ISSN: 21693536. DOI: 10 . 1109 / access .

2019.2929120.

[21] E. Vidal, N. Palomeras, K. Istenič, J. Hernán

dez, and M. Carreras, “Twodimensional

frontierbased viewpoint generation for explor

ing and mapping underwater environments,”

Sensors, vol. 19, no. 6, p. 1460, 2019, ISSN:

14248220. DOI: 10.3390/s19061460.

[22] J. Dentler, M. Rosalie, G. Danoy, P. Bouvry,

S. Kannan, M. A. OlivaresMendez, and H.

Voos, “Collision avoidance effects on the mo

bility of a UAV swarm using chaotic ant colony

with model predictive control,” Journal of In

telligent and Robotic Systems, vol. 93, no. 1

2, pp. 227–243, 2018, ISSN: 09210296. DOI:

10.1007/s10846-018-0822-8.

[23] X. Kan, H. Teng, and K. Karydis, “Online

exploration and coverage planning in un

known obstaclecluttered environments,”

arXiv, Jun. 30, 2020. arXiv: 2006 . 16460v1

[cs.RO].

[24] V. P. Tran, M. A. Garratt, K. Kasmarik, and

S. G. Anavatti, “Frontierled swarming: Ro

bust multirobot coverage of unknown environ

ments,” arXiv, pp. 1–13, Nov. 29, 2021. arXiv:

2111.14295v2 [cs.RO].

[25] M. Prajapat, M. Turchetta, M. N. Zeilinger, and

A. Krause, “Nearoptimal multiagent learning

for safe coverage control,” arXiv, pp. 1–44,

Oct. 12, 2022. arXiv: 2210.06380v1 [cs.LG].

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed in the present

research, at all stages from the formulation of

the problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflicts of Interest

The authors have no conflicts of interest to

declare that are relevant to the content of this

article.

Creative Commons Attribution License 4.0

(Attribution 4.0 International , CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.20

Fredy Martinez, Angelica Rendon, Fernando Martinez

E-ISSN: 2224-2856

203

Volume 18, 2023