IAES International Journal of Robotics and Automation (IJRA)

Vol. 13, No. 2, June 2024, pp. 205~219

ISSN: 2722-2586, DOI: 10.11591/ijra.v13i2.pp205-219  205

Distributed and autonomous multi-robot for task allocation and

collaboration using a greedy algorithm and robot operating
system platform

Abderrahmane Tamali1, Nourredine Amardjia1, Mohammed Tamali2

1
LIS Laboratory, Department of Electronics, Ferhat Abbas University Setif 1, Setif, Algeria
2
ENERGARID Laboratory, Department of Electrical Engineering, Tahri Mohammed Bechar University, Bechar, Algeria

Article Info ABSTRACT

Article history: Research investigations in the realm of micro-robotics often center around
strategies addressing the multi-robot task allocation (MRTA) problem. Our
Received Jan 24, 2024 contribution delves into the collaborative dynamics of micro-robots deployed
Revised Feb 28, 2024 in targeted hostile environments. Employing advanced algorithms, these robots
Accepted Apr 23, 2024 play a crucial role in enhancing and streamlining operations within sensitive
areas. We adopt a tailored GREEDY approach, strategically adjusting weight
parameters in a multi-objective function that serves as a cost metric. The
Keywords: objective function, designed for optimization purposes, aggregates the cost
functions of all agents involved. Our evaluation meticulously examines the
3D digitization MRTA efficiency for each micro-robot, considering dependencies on factors
Collaboration such as radio connectivity, available energy, and the absolute and relative
Light detection and ranging availability of agents. The central focus is on validating the positive trend
Multi-robot associated with an increasing number of agents constituting the cluster. Our
Multi-robot task allocation methodology introduces a trio of micro-robots, unveiling a flexible strategy
Robot operating system aimed at detecting individuals at risk in demanding environments. Each micro-
Simultaneous localization and robot within the cluster is equipped with logic that ensures compatibility and
mapping cooperation, enabling them to effectively execute assigned missions. The
implementation of MRTA-based collaboration algorithms serves as an
adaptive strategy, optimizing agents' mobility based on specific criteria related
to the characteristics of the target site.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Abderrahmane Tamali
LIS Laboratory, Department of Electronics, Setif 1 University-Ferhat Abbas
Setif, Algeria
Email: [email protected]

1. INTRODUCTION
Robotic technology has the potential to replace human workers in high-risk scenarios, reducing the
potential for human and financial loss. Recent research has led to progress in developing autonomous
techniques and devices capable of operating without direct human intervention, including working online and
collaborating with humans in hazardous environments. A major focus of our research is the study of the
optimal number of robots per task of a cluster (multi-robot teamwork) under specifications and particular
potentials. The primary objective is to refine control methods for multi-robot in hostile areas. The importance
of this research is underlined by its significant professional benefits and potential applications. Micro-robots
are expected to play a key role in improving task distribution and execution, as well as facilitating the
transmission of relevant information and collected data [1]–[4]. The challenging transmission process makes
use of affordable means and logic strategies, such as radio communication, within the cluster.

Journal homepage: http://ijra.iaescore.com

206  ISSN: 2722-2586

To facilitate simulation, we have chosen to utilize the open-source robotics operating system (ROS)
solution along with its associated tools and modules [5], [6]. ROS will streamline the process of translating
both conceptual and physical models into reality. An illustrative instance within this field has been developed
by [7].
Multi-robot task allocation (MRTA) problems are intricate case studies. In operational research
(OR) and for NP-class problems characterized by numerous constraints and a single objective function,
special approaches are used, such as differential methods and AI-based algorithms. Our decision-making
process adheres to specific criteria to adopt a proactive and constructive approach: i) form and structure
(dimensions, degrees of freedom), ii) data transmission (means, persistence), iii) economy and energy
autonomy (energy security), iv) equipment carried out as payload (sensors), and v) embedded logic.
The collected data is vital for the cluster. Each agent undergoes an evolution process through a
collaborative treatment approach. Our goal is to ensure efficient completion of the task in the shortest time,
while maintaining the importance of the collected data and utilizing dedicated resources.
The main objective addressed by Rekleitis et al. [8] is the optimization of the generated map by
minimizing errors at the estimation level, contingent on the robot's capabilities and the attached payload.
Task coordination for optimal multi-robot evolution through a task-based multi-robot task allocation (MRTA)
optimal assignment problem (OAP) is emphasized by Gerkey and Matarić [9]. Additionally, Lee et al. [10]
focus on developing a metric for estimating fault levels within a swarm of robots, while Zhang et al. [11]
introduce a model based on data correlation, specifically the correlated random walk model, to efficiently
approximate task searching time for distributions of multi-robot systems in large arenas.
Recent literature, including [12] and [13], explores bio-inspired techniques for collaboration and
sharing state information between pursuing agents and fast evaders. Collaboration necessitates meeting
specific criteria, such as sharing each agent's state information and utilizing onboard resources to complete
tasks. The context of a hostile site adds complexity, demanding strategies with mathematical intricacy [14],
[15], influenced by ecosystem characteristics (multi-robot setups, tasks, region of interest, and resources). In
the case of a hostile site with intricate morphology, the challenge intensifies, relying on capabilities like
finding a direct path. The objective is to derive optimal solutions [16] for the routes to be followed,
minimizing time to reach the rescue target [17].
In the context of industrial sites, the focus shifts dramatically, with an emphasis on completing
surveillance in the most relevant manner for incident localization [18]. Other research concentrates on the
capabilities that micro-robots must possess to acquire cognitive abilities, enabling them to navigate and
evolve on the study site using reinforced learning methods [19]. Regarding morphology, concrete examples
like MIT's Cheetah 3 [20] or ANYBOTICS' ANYmal from ETH Zurich [21] meet the requirements of
degrees of freedom in challenging sites with heightened aggressiveness.

2. METHOD
Various tools, including proprietary OnShape assembly and open-source Phobos add-on in blender,
are utilized for preparatory tasks before simulating under ROS. XACRO, simulation description format
(SDF), or unified robotics description format (URDF) scripts are essential for developing 3D robot models
with joints for Gazebo/RVIZ simulation. The JETBOT prototype by NVIDIA serves as the modeled micro-
robot with updated URDF (SDF) to fit study needs. Elements at this stage are crucial for setting up a
compatible simulation scene in line with our approach:
− The actual space is defined by a specific location (the dimensions of an apartment with spatial
limitations) in 3D [5], [10], [18].
− The proposed cluster of multi-robot is a formation of three similar agents [22].
− Possibility of the heterogeneous case [23]–[25].
− Target, an injured human (3D model of a human, target behavior complexity).
The standard packaging used in this research are as follows. i) Economy and energy autonomy for a
given payload (implementation of ROS battery plug-in as shown in Figure 1 within the URDF joints with its
compatible compiled library). ii) Ensure stability in communication links using the ROS radio connectivity
plug-in with the cluster. iii) Enhance collaborative capabilities for improved performance by implementing
necessary plug-ins, rules, and algorithms, as illustrated in Algorithm 1. iv) Algorithm enhancements-based
AI (AI Logic code integrated). v) Minimum agent's size and volume (for swarm extension purposes).
The scene requirements in this research are: i) The cluster incorporates three micro-robots (agents)
denoted as 𝜇𝑅𝑖 |𝑖𝜖[1,𝑅] , each equipped with motorized wheels for locomotion (differential, driver). ii) Every
agent unit is furnished with transmission capabilities for data transmission and maintaining communication
links. iii) The autonomy of every agent is supported by an onboard battery, providing sufficient power for its
activities within the cluster. iv) Every agent possesses an integrated AI algorithmic foundation, empowering

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  207

it to make decisions for both collaborative and individual purposes. v) Every agent unit within the cluster is
outfitted with essential equipment, including a Camera, LiDAR (light detection and ranging), a motor driver,
actuators, and sensors as shown in Figure 2. vi) The micro-robot group operates in two modes: Supervised
mode and collaboration mode.

Figure 1. Nominal current discharge characteristic (0.65A)

Algorithm 1. MRTA, Greedy oriented search algorithm

Begin
Initializations:
Let Sk ← ∅
Let K0 ← k’ {∀ k’ ∈ [1;R] , R: number of robots}
Let Tk ← ∅
(𝑃 𝑘 )∗ 𝑔𝑒𝑡 𝑎𝑙𝑙 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑡𝑎𝑠𝑘𝑠 𝑖𝑓 𝑡ℎ𝑒𝑦 𝑒𝑥𝑖𝑠𝑡
Let 𝑃𝑖𝑘 = { 𝑖
∅ 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
for (k ∈ [1; R]) do
Sk= area(k)
if (k ≠ K0) then
Let 𝑆𝐾0 ← (𝑆𝐾0 - 𝑆𝐾0 ∩ Sk)
end if
end for
T ← task(K0)|𝑆 𝐾0
i ← 1
while (𝑖 ≤ |𝑇|) do
𝐾 𝐾
𝑡𝑖 0 ← Eval(𝑓𝑖 0 |𝑆𝐾0 ) {ith objective function}
𝐾0 𝐾0 𝐾
if (𝑇𝑖 (𝑡𝑖 ) ⊂ 𝑃𝑖 0 ) then
𝐾
𝑡𝑖 0 ← 𝑁𝑜𝑛𝑒
end if
i ← i+1
end while
𝐾
𝑇𝐾0 ← 𝑇𝑗 0 |𝑆𝐾0 ,𝐵𝑒𝑠𝑡(𝑡 𝐾0 ) {for j ∈ [1;m], m is set of specific tasks}
𝑗
if (𝑇𝐾0 ≠ ∅) then
𝐾 𝐾
𝑃𝑖 0 ← 𝑃𝑖 0 + 𝑇𝐾0
else
T ← ∅
for (k ∈ [1; R]) do
if (k ≠ K0)do
T ← T + task(k)|𝑆𝑘 ,𝑘 ≠ 𝐾0
i ← 1
while (𝑖 ≤ |𝑇|) do
𝐾
𝑡𝑖𝑘 ← Eval(𝑓𝑖 0 |𝑆𝑘,𝑘 ≠ 𝐾0 )
if (𝑇𝑖 (𝑡𝑖 ) ⊂ 𝑃𝑖𝑘 ) then
𝑘 𝑘

𝑡𝑖𝑘 ← 𝑁𝑜𝑛𝑒
end if
i ← i+1
end while
𝑇 𝑘 ← 𝑇𝑗𝑘 |𝑆𝑘 ,𝐵𝑒𝑠𝑡(𝑡𝑘 ),𝑘 ≠ 𝐾0 {for j ∈ [1;m] and m
𝑗
end if is set of specific tasks}
end for
𝑇𝐾0 ← Best(Tk)|k∈{1,2,…,R}
𝑃𝑖𝑘 ← 𝑃𝑖𝑘 + 𝑇𝐾0 |k∈{1,2,…,R}
end if
End.

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
208  ISSN: 2722-2586

Figure 2. AL Mustaksheef3D, wheeled robot developed

We aim to effectively locate and aid victims in challenging situations through Collaborative MRTA.
Our approach involves the implementation of 3D digitization utilizing LiDAR or a depth camera, ensuring
optimal outcomes for Search and simultaneous localization and mapping (SLAM) purposes. The ROS
environment serves to manage the context, and, in addition to interoperability, ROS can engage with various
platforms dedicated to simulating MRTA problems [26]–[30]. The ROS.MSG embedded module facilitates
message exchange among micro-robots, with inertial measurement unit (IMU) data conversion to ODOM
(odometry) enabling the validation and voting of agent moves in areas that are still unoccupied as shown in
Figure 3.

Figure 3. ROS basics and concepts

2.1. Distributed MRTA use case and problem statement

Achieving coordinated teamwork among robots with real-time task distribution necessitates a
decentralized framework [31] that covers resilient robot awareness, low-level motion control, and high-level
task scheduling [32], [33]. Effective location management in multi-robot networks is critical for
collaboration, where decision-making and communication play pivotal roles in mission distribution,
presenting substantial challenges in data exchange among robots and the operating station [34]. In
approaches that are self-organized and decentralized, individual robot nodes independently make decisions
with limited regard for other agents [35]. Decisions in these approaches are frequently influenced by natural
or real-world phenomena, as highlighted in [3], including bees swarm, market strategy, swarm intelligence,
and ant colony. These inspirations give rise to complex collective behavior arising from local interactions
among numerous agents with straightforward behaviors. In such methodologies, sensors play a pivotal role
by actively collecting local knowledge for sharing within the cluster, as emphasized by Stolzle et al. [7] and
Ball et al. [36]. The collaborative use of sensors facilitates the accumulation of necessary knowledge
pertaining to an overarching goal in these procedures as shown in Figure 4.
Robots need to be able to understand the tasks they have to perform by collecting data through
sensors and using specific code to make decisions. Broadband, range, power, and data rates are critical to
system performance. Point-to-point communication is the most basic form, and the choice of transmission
medium is a function of the type of information being exchanged. Robot evolution on hostile sites can, in
some manner, be assumed to be a progression of the vehicle forming a path on a surface contoured by a set of
N points in a space defined in a plane delimited by a closed polygon, where 𝑃 = 𝑃1 , 𝑃2 , … , 𝑃𝑁 define a poly-
point or a set of N points as shown in Figure 5.

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  209

Figure 4. Cluster architecture

Figure 5. N-gon of a scanned area

Normally each point is located on the plan (Г) by its Cartesian coordinates 𝑥𝑖 and 𝑦𝑖 . A one-line
equation can be written in the form of (1).

𝑦 = 𝑎. 𝑥 + 𝑏 (1)

and are two parameters related to the jth line (slope and y-intercept) in the Polyline defined by P
components. By using each couple of points coordinates, the related and parameters are obtained
according to (using Cramer's rule).
𝑦1 − 𝑦2
𝑎= (2)
𝑥1 − 𝑥2
𝑥1 𝑦2 − 𝑥2 𝑦1
𝑏= (3)
𝑥1 − 𝑥2

So, (1) can be like follows for a line equivalent (L 1) between two points 𝑃1 (𝑥1 , 𝑦1 ) and 𝑃2 (𝑥2 , 𝑦2 ). y is given
by (4).

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
210  ISSN: 2722-2586

𝑥1 ≤𝑥<𝑥2 ,∀𝑥1 −𝑥2 <0

𝑦1 − 𝑦2 𝑥1 𝑦2 − 𝑥2 𝑦1 𝑥2≤𝑥<𝑥1 ,∀𝑥1 −𝑥2 >0
⋅𝑥+ |
𝑦 = {𝑥1 − 𝑥2 𝑥1 − 𝑥2 𝑦1≤𝑦<𝑦2,∀𝑦1−𝑦2 <0 (4)
𝑦2 ≤𝑦<𝑦1 ,∀𝑦1 −𝑦2 >0
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Now, we need to construct a Polygon equation using a combination of multiple-line equations. We have
points, which implies that the number of line equations is 𝑁(𝐿1 , 𝐿2 , … . , 𝐿𝑁 ). The n-gon's (polygon) formula
is given by (5).
𝑁−1

𝑌 = 𝑌1 + 𝑌2 +⋅⋅⋅⋅⋅⋅ +𝑌𝑁 = ∑ 𝑌𝑖 + 𝑌𝑁 (5)

𝑖=1

Witch is 𝑌1 , 𝑌2 , … . , 𝑌𝑁 or simply 𝑌𝐼 , 𝑌𝑁 where 𝑖 ∈ ℕ natural strictly positive number and, the set 𝑌𝐼 with 𝑖 ∈ ℕ
represent each line’s equation:
𝑥𝑖 ≤𝑥<𝑥𝑖+1 ,∀𝑥𝑖 −𝑥𝑖+1 <0
𝑦𝑖 − 𝑦𝑖+1 𝑥𝑖 𝑦𝑖+1 − 𝑥𝑖+1 𝑦𝑖 𝑥𝑖+1≤𝑥<𝑥𝑖 ,∀𝑥𝑖−𝑥𝑖+1>0
⋅𝑥+ |
𝑌𝑖 = 𝑦 = {𝑥𝑖 − 𝑥𝑖+1 𝑥𝑖 − 𝑥𝑖+1 𝑦𝑖 ≤𝑦<𝑦𝑖+1 ,∀𝑦𝑖 −𝑦𝑖+1 <0 (6)
𝑦𝑖+1 ≤𝑦<𝑦𝑖 ,∀𝑦𝑖 −𝑦𝑖+1 >0
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Equation valid for 1 ≥ 𝑖 ≥ 𝑁 − 1, and the last line (𝐿𝑁 ):

𝑥1 ≤𝑥<𝑥𝑁 ,∀𝑥1 −𝑥𝑁 <0
𝑦1 − 𝑦𝑁 𝑥1 𝑦𝑁 − 𝑥𝑁 𝑦1 𝑥𝑁 ≤𝑥<𝑥1 ,∀𝑥1−𝑥𝑁 >0
⋅𝑥+ |
𝑌𝑁 = 𝑦 = {𝑥1 − 𝑥𝑁 𝑥1 − 𝑥𝑁 𝑦1≤𝑦<𝑦𝑁,∀𝑦1−𝑦𝑁<0 (7)
𝑦𝑁 ≤𝑦<𝑦1 ,∀𝑦1 −𝑦𝑁 >0
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

For 𝑥, 𝑦, 𝛼 ∈ ℝ and 𝑓(𝑥) = 𝑦 = 𝛼, The area inside the irregular polygon can be defined as the result of:

Area Y = ∫ α|f(x)=α=Y={x′ ;x′′ ;...;xn ;xn+1} dα (8)

ℝ
where xn ,xn+1 ∈γ(x)

Consider expressions like 𝑛 = 2𝑚 + 1|𝑚∈𝑁 , where xn<xn+1, and γ(x) representing the variation domain of
the variable x. Consequently, the explored area by the kth robot can be denoted as Area(k), where k=1,2, …,
R, and R is the total number of cluster agents. We introduce the relation gk(α,x), which satisfies the condition
f(x)=α=Y={x';x'';x''';.;.;xn;xn+1}. This relation, denoted as gk, aids in determining whether the robot k is
situated within the designated Area(k) or not. T0 assess whether a robot is outside or inside an area, we can
analyze four zone-shaped situations illustrating the most probable cases, as depicted in Figures 6(a), 6(b),
6(c), and 6(d).
In the first scenario as shown in Figure 6(a), robot1 is situated in zone Z, while the other robots
(robot2 and robot3) are not. Formulating this situation involves expressions such as
g1(α1,x)=g2(α2,x)=Z{x',x''} and g3(α,x)= ø . Therefore, by comparing 𝑥𝑅𝑘 with x' and x'', we can draw the
following conclusions:
− If gk(α,x)=ø ⇒ the robotk is outside zone Z.
− If x'≤𝑥𝑅𝑘 ≤x'' and gk(α,x)≠ø ⇒ the robotk is within zone Z.
− If 𝑥𝑅𝑘 <x' or x''<𝑥𝑅𝑘 and gk(α,x)≠ø ⇒ the robotk is outside zone Z.
In the second scenario as shown in Figure 6(b), we have g1(α1,x)=g2(α2,x)=g3(α3,x)=Z{x',x'',x''',x''''}.
robot2 is within the area where 𝑥 ′ ≤ 𝑥𝑅2 ≤ 𝑥 ′′, while the others are not. Specifically, 𝑥 ′′ < 𝑥𝑅2 < 𝑥 ′′′ and
𝑥𝑅2 < 𝑥 ′ , leading to the following determinations:
− If 𝑥 𝑛 ≤ 𝑥𝑅𝑘 ≤ 𝑥 𝑛+1 | and gk(α,x)≠ø ⇒ the robotk is within the area.
𝑛=2𝑚+1,∀𝑚∈𝑁
− If 𝑥 𝑛 < 𝑥𝑅𝑘 < 𝑥 𝑛+1 | and gk(α,x)≠ø ⇒the robotk is outside the area.
𝑛=2𝑚+2,∀𝑚∈𝑁
The third situation as shown in Figure 6(c) encompasses two singular cases, where
g1(α1,x)={x',x'',x'''} and g2(α2,x)={x'}. The robot's presence in the area can only be determined in these cases
where |gk|=3 for robot1 and |gk|=1 for robot2 :
− If 𝑥 ′ ≤ 𝑥𝑅𝑘 ≤ 𝑥 ′′′ and gk(α,x)≠ø where |gk|=3 ⇒ the robotk is in the area.

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  211

− If 𝑥 ′ = 𝑥𝑅𝑘 and gk(α,x)≠ø where |gk|=1 ⇒ the robotk is in the area.

In the last case as shown in Figure 6(d), it is impossible to ascertain whether the robot is in the zone
or not using the relation gk, where gk(α,x)≠ø, and |gk|=n|n=2m+5,∀m ∈ℕ.

(a) (b)

(c) (d)

Figure 6. Potential scenarios for the robot's posture (a) simple case with 3 robots, (b) complex case with
3 robots, (c) singular case, and (d) insoluble case

2.2. Simulation cases and success factors

Before initiating our simulations, we define our context and our environment. To simulate
collaboration and explore to locate a victim, a workstation is employed. This workstation features an HP
ProBook x360 435 G7 with an AMD Ryzen 7 PRO 4750U processor, AMD Radeon RX Vega 7 graphics, 32
GB DDR4 Kingston RAM, and a 1TB Samsung SSD Drive. The simulation utilizes ROS [37] on Linux
Ubuntu 20.04, enhancing node communication and facilitating message passing between robots and
computers. Additionally, JetBot, powered by NVIDIA Jetson Nano and accessible at NVIDIA, supports
sensors and implements RNN for object detection and collision avoidance. Furthermore, JetBot is capable of
connecting through various radio links.

2.2.1. Optimization based algorithm

The next flowchart illustrates the overall process as shown in Figure 7. The cluster collects data
related to the ROI's context, investigates frontiers around each micro-robot, generates map fragments, and
shares them within the cluster. If tasks are completed, processes can be disposed of, otherwise, the
preparation step is repeated. In a practical scenario, each micro-robot needs to meet the optimal condition
defined by the cost function fcostR in (9).
− Radio connectivity to the sink or to all cluster Fcon/AP which guarantees data exchange and/or maintain a
link (AP for access point).
− Economic and energy autonomy Fauton quantifies the battery lifespan for a particular task.
− Absolute availability Disp/all, Denotes the 'OK state' of the unit within the cluster, signifying its
capability to proficiently carry out the task.

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
212  ISSN: 2722-2586

− The relative availability, denoted as Disp/Res, signifies the accessibility of the payload, which is a
specific resource crucial for a particular task.
𝜇𝑅
𝑓𝑐𝑜𝑠𝑡 = 𝑘1 ∗ 𝐹𝑐𝑜𝑛⁄𝐴𝑃 + 𝑘2 ∗ 𝐹𝑎𝑢𝑡𝑜𝑛 + 𝑘3 ∗ 𝐷𝑖𝑠𝑝 𝐴𝑙𝑙 + 𝑘4 ∗ 𝐷𝑖𝑠𝑝 ⁄𝑅𝑒𝑠
(9)
where k1+k2+k3+k4=1

Figure 7. Flowchart of our approach

The weights ki for i∈[1,4] symbolize the contributions of each specific part to the agent's functional
cost, as defined earlier. Our goal is to discover the optimal outcome (10) for the function fcost for every agent
in the cluster, representing the best solution. Subsequently,

𝑜𝑝𝑡 𝜇𝑅𝑗
𝐹𝑐𝑜𝑠𝑡 = 𝑀𝑎𝑥|𝑗∈[1,𝑅[ {𝑓𝑐𝑜𝑠𝑡 , 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠} (10)

The cluster's optimal estimate is indicated by its capability to declare an 'OK' status for availability,
surpassing the minimum required condition for executing assigned tasks. We introduce Fcon/AP as the reduced
ref
effective availability of at least one link with a predefined access point. If Fcon represents the reference
threshold for a Wi-Fi connection to an access point and Fcon(t) signifies the current connectivity level of an
agent to the access point, then Fcon/AP is defined as the ratio of the instant connectivity Fcon(t) to the reference
ref
threshold Fcon This is expressed as (11):

ref
Fcon/AP = 100 ∗ (Fcon(t)/Fcon ) (11)

Therefore, Fauton signifies the battery autonomy, indicating the remaining energy in Ah needed by the battery
to provide adequate power for the agent to successfully accomplish the assigned task as a singular unit within
the cluster. Here, we represent τ as the estimated time required for the assigned task, I as the actual
discharged current of the battery in Ampere (A), and C as the battery capacity, acting as a current source for a
specified duration in Ampere-hour (Ah). Hence, the expressions are as (12).

𝐹𝑎𝑢𝑡𝑜𝑛 = 𝐶/𝜏 ∗ (1/𝐼 − 10/𝑃𝑐ℎ𝑎𝑟𝑔𝑒 ) (12)

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  213

In estimating Disp/all and Disp/Res, the availability of a cluster member is assessed based on its effective
suitability for any given task. Relative availability for a resource indicates the agent's capacity to have the
specified resource (payload) ready at the designated time when a task is assigned. On the other hand, absolute
availability encompasses the sum of all relative availabilities (of Nres resources), signifying that all of the
agent's resources are ready for use. The relative availability is set to 1 when there is positive feedback upon
querying the resource and 0 otherwise. Therefore:

1 𝑖𝑓 𝑡ℎ𝑒 𝑡𝑎𝑟𝑔𝑒𝑡𝑒𝑑 𝑟𝑒𝑠𝑜𝑢𝑟𝑐𝑒 𝑖𝑠 𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑙𝑒

𝐷𝑅𝑒𝑙/𝑅𝑒𝑠 = { (13)
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Absolute availability is given by (14).

𝑁𝑟𝑒𝑠

𝐷𝑖𝑠𝑝 𝐴𝑙𝑙
= ⋂(𝐷𝑅𝑒𝑙/𝑅𝑒𝑠 𝑖 ) (14)
𝑖=1

The agent's ability to fully intervene in a task depends on the logic AND connection of all relevant
availability functions. The weighting coefficients ki are selected using various methods to optimize the
objective function's final result. We adopt a bio-inspired approach to determine these coefficients.

2.3. Collaboration by greedy algorithm for distributed MRTA

We introduce a greedy algorithm in Algorithm 1 [3] designed for the MRTA problem with
broadcast messaging. This algorithm is configured to allocate agents based on the optimal assignment to a
task that maximizes the performance-to-cost ratio (Vik/dik). Here, Vik represents the performance of the kth
agent for the accomplishment of the ith task, and dik denotes the Euclidean distance between the agent and the
target/assigned location (8) and (9).

dki = √(xi − xk )2 + (yi − yk )2 + (zi − zk )2 (15)

𝑇𝑖𝑘 = 𝑚𝑎𝑥 (𝑉𝑖𝑘 ⁄𝑑𝑖𝑘 ) = 𝑚𝑎𝑥 (𝑉𝑖𝑘 𝜂𝑖𝑘 ) 𝑤𝑖𝑡ℎ 𝜂𝑖𝑘 = 1⁄𝑑𝑖𝑘 (16)
𝑖∈𝑇 𝑖∈𝑇

Here, task i is the chosen task for the kth agent from the entire set of potential tasks, and T represents the set
of viable tasks within the kth agent's scope out of the total M available tasks. The algorithm, identified as
Algorithm 1, operates as follows: Initially, we define surface variables explored by each robotk, where 𝑆𝑘 |𝑘∈𝑅
denotes the set of boundary points of the presently scanned surface. Additionally, fik|k represents an objective
function (cost function) for each robotk, with taski being the sub-mission stored in Tik Furthermore, Tik is the
collection of tasks that have not been assigned yet and are included in T, representing the total available tasks
(Ti ∈ T). Concurrently, Tk for k ∈ R is the selected task for the robot, and Pik denotes the locations that have
already been assembled. Here, k indicates the robot that needs to update its bid at the current stage.
At the start, no tasks are assigned, so Tk=ø for all robots k∈ ℝ. In each step, one task is allocated to
a single robot independently as shown in Figure 8, following decentralization. Thus, we need |T| steps,
the number of tasks for the robot, to complete its state. At each iteration i, after removing conflicting parts
with areas explored by other robots, all robots 𝑘 ∈ 𝑅 submit an offer (tki, Tki). Each robot k selects task Tk
from the list of non-located tasks Tik to maximize its objective function fik. Upon gathering all the bids, we
discover a superior optimal gain for the collective objective: the multiplicative success of group F (17).
Through bidding, we efficiently choose the optimal task pair-robot combination for the greatest overall
benefit [38], [39].

𝐹 = 𝐵𝑒𝑠𝑡{𝑇 𝑘} ∏[𝑓𝑘 (𝑇 𝑘 )] (17)

𝑘∈𝑅
𝑘∈𝑅

When a robot is surrounded by other robots, it can be misled about its capabilities, hindering its
development and exploration. This situation can impede the robot's ability to effectively navigate and
fulfill its tasks. To address this issue and maintain high performance even when surrounded ( Tk=ø), our
objective is to enhance the robot's cost function by utilizing the neighboring robots' available spaces until
it can operate independently. By incorporating information from surrounding robots, the robot can make
more informed decisions and adapt its behavior, accordingly, ultimately improving its overall performance
and autonomy.

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
214  ISSN: 2722-2586

Figure 8. JetBot during domain exploration

3. RESULTS AND DISCUSSION

The simulation scenario is illustrated in Figure 9. It is characterized by various constraints that
evaluate the efficiency and reliability of our approach in navigating a cluster of micro-robots, both with and
without collaboration. The simulation scene consists of rooms (isolated spaces) containing furniture
(obstacles). Micro-robots work together to rescue multiple victims as shown in Figure 10, to minimize time
and enhance reliability. The simulation illustrates three severity levels involving 1-3 micro-robots as shown
in Figure 11.

Figure 9. Model area of the simulation Figure 10. Scene with target location

Figure 11. Space segmentation

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  215

We make the following assumptions for the simulation:

− The scene's domain and dimensions remain constant throughout the investigation.
− The target position within the scene does not significantly affect the simulation's duration.
− The agents (micro-robots) are assumed to be homogeneous, possessing identical characteristics.
− Connectivity, Payload, and Autonomy conditions are satisfactory for all units within the cluster.
The simulation's outcomes are presented in the subsequent Figures 11 to 15 and Tables 1 to 8. The
plan involves finding the time to reach each victim, pinpointing the longest time to reach two victims, and
averaging the maximum values. We focus on the maximum and minimum average durations. The regions in
Figure 15 show designated areas in RGB colors within the All Area. These areas come from segmenting the
total ROI using the greedy algorithm. The process is repeated three times. The first iteration involves a single
robot over ten attempts, and the results are in Table 1. Where time distribution by number of attempts is
given (with trend function) in the next chart as shown in Figure 16.

Figure 12. The used robot model Figure 13. GAZEBO Scene Simulator

Figure 14. Robots while performing a mission Figure 15. Agents ROI updates

Table 1. Simulation time to find the two victims

(in seconds)
Attempts Target 1 Target 2 To mission's end
1st 55 102 102
2nd 70 30 70
3rd 20 150 150
4th 93 35 93
5th 64 86 86
6th 46 90 90
7th 77 112 112
8th 95 150 150
9th 9 76 76
10th 110 123 123 Figure 16. Duration taken by a single robot to reach
Mean 63.9 95.4 105.2 the target on the initial attempt

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
216  ISSN: 2722-2586

The process involves ten simulations, each done three times. Results are assessed across five
different scenarios (cases I to V). In each test, we note the time to meet all goals until finding two targets
(victims), with and without considering cooperation.
− Scenario I: single robot (refer to Table 2).
− Scenario II: two robots without collaboration (refer to Table 3).
− Scenario III: two robots with collaboration (refer to Table 4).
− Scenario IV: three robots without collaboration (refer to Table 5).
− Scenario V: three robots with collaboration (refer to Table 6).

Table 2. Simulation time of one robot to find the two Table 3. Two robots without collaboration algorithm
victims (in seconds) average time (sec)
target1 target2 duration target1 target2 duration
Exp.1 80.2 37.5 90.7 Exp.1 72 38.2 85.3
Exp.2 67.5 99.5 120.6 Exp.2 72 61.6 95.2
Exp.3 63.9 95.4 105.2 Exp.3 62.4 40.3 66.9

Table 4. Two robots with collaboration algorithm Table 5. Three robots without collaboration
average time (sec) algorithm average time (sec)
target1 target2 duration target1 target2 duration
Exp.1 27.2 21.7 34.2 Exp.1 72.6 32.2 76.5
Exp.2 33.8 30.4 47.5 Exp.2 57.8 39.3 69.1
Exp.3 60.1 38.9 41.7 Exp.3 71.2 29 73.9

Table 6. Three robots with collaboration algorithm average time (sec)

target1 target2 duration
Exp.1 22.7 7.9 22.7
Exp.2 18.3 21.5 24
Exp.3 18.4 14.1 20.7

Aggregate outcomes are compiled as the min and max values for potential consolidation as shown in
Tables 7 and 8. Table 7 presents the average time durations across various experimental conditions, offering
insights into the comparative efficiency of different strategies. In contrast, Table 8 delineates the minimum
and maximum simulation times required to locate two victims, providing a comprehensive view of
performance variability under different scenarios.

Table 7. Time duration average by experiment

Exp.1 Exp.2 Exp.3
One Robot 90.7 120.6 105.2
2NoCollab 85.3 95.2 66.9
2WCollab 34.2 47.5 41.7
3NoCollab 76.5 69.1 73.9
3WCollab 22.7 24 20.7

Table 8. Simulation time to find the two victims (in seconds)

One 2NoCol 2WCol 3NoCol 3WCol
Min 90.7 66 34.2 69.1 20.7
Max 120.6 95.2 62.1 76.5 24

Collaboration increasingly impacts the duration of time as shown in Figures 17 and 18. These
figures vividly illustrate the significant influence of collaboration on time duration, underscoring its growing
importance in various contexts. Additionally, as the number of agents increases, the duration decreases,
particularly when a cooperative strategy is enabled, highlighting the efficiency gains achieved through
collaborative efforts, especially in scenarios involving larger groups.

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  217

Figure 17. Progression of time duration through Figure 18. Max and min time duration by experiment
experimentation until the victim is found

The ROI was evaluated in two ways: by pinpointing the targets and by 3D scanning the area for
potential reconnaissance. This was done using an enhanced version of a horizontal LiDAR (RPLiDAR
A1M8) as part of the new features on our AL Moustaksheef3D platform, a robotic payload for 3D SLAM
currently in development. The enhanced LiDAR as shown in Figures 19 and 20 was tested on a wheeled
machine and a Drone. The collected data as shown in Figure 21 is provided in point cloud (PCL) of “.pcd”
format. This model integrates two LiDARs, with the first used for geolocation and the second for 3D vision.
Scene reproduction yields relative data for estimating facts in the study area. Point-cloud data is used for
scene reconstruction. Software like CloudCompare, MeshLab, Blender, and Gimp enable background
processing on point cloud data. The micro-cobot agents integrate APIs into their embedded logic, employing
robust algorithms for identifying specific targets in hostile environments. Using an MRTA approach with the
enhanced Greedy algorithm, our strategy optimizes target identification through cluster agent cooperation,
significantly reducing task time. This collaborative approach requires substantial processor capabilities, with
basic calculations centralized for collaboration and decentralized for independent agent decision-making
post-task assignment.

Figure 19. Figure 20. The 3D model Figure 21. ROI SLAM Segmentation, on
JetBot/AlMustaksheef3D of the new LiDAR RVIZ and Gazebo

4. CONCLUSION
The project aims to design collaborative robots with physical capability and intelligence to work
effectively in challenging environments. These cobots will help reduce the workload of human response or
rescue teams by identifying targets. They will also help in areas where the density of robots and technicians
poses a risk to human safety.
Our research focuses on micro-robots collaborating with humans, leading to the concept of
micro-cobots. This aims to reduce payload and minimize bottlenecks in tasks. Increasing micro -robot
agents enhances investigation speed and target search. The results prove our initial consideration and
confirm the predefined hypotheses. The results obtained from the simulation explain that the cooperation
significantly improves the cluster's progress in the search mission. Collaboration prevents erratic
micro-robot behavior, reduces duration, and prevents system collapse. Fewer agents lead to more
dispersion and chaos. A moderate number of investigators (1<NR≤3) is preferable. Swarm techniques with
collaboration are used to manage the situation.

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)
218  ISSN: 2722-2586

REFERENCES
[1] F. Montori, L. Bedogni, and L. Bononi, “A Collaborative Internet of Things Architecture for Smart Cities and Environmental
Monitoring,” IEEE Internet of Things Journal, vol. 5, no. 2, pp. 592–605, Apr. 2018, doi: 10.1109/JIOT.2017.2720855.
[2] Y. Dong, L. Zhang, and Y. Liu, “Experimental study on control strategy of cobot micro operation force hoisting system,”
Gaojishu Tongxin/Chinese High Technology Letters, vol. 17, no. 5, pp. 493–497, 2007.
[3] A. El shenawy, K. Mohamed, and H. M. Harb, “Exploration Strategies of Coordinated Multi-Robot System: A Comparative
Study,” IAES International Journal of Robotics and Automation (IJRA), vol. 7, no. 1, p. 48, Mar. 2018, doi:
10.11591/ijra.v7i1.pp48-58.
[4] I. El Mallahi, J. Riffi, H. Tairi, A. Ez-Zahout, and M. A. Mahraz, “A Distributed Big Data Analytics Model for Traffic Accidents
Classification and Recognition based on SparkMlLib Cores,” Journal of Automation, Mobile Robotics and Intelligent Systems,
vol. 16, no. 4, pp. 62–71, Oct. 2022, doi: 10.14313/jamris/4-2022/34.
[5] X. Wang, X. Ma, X. Li, X. Ma, and C. Li, “Target-biased informed trees: sampling-based method for optimal motion planning in
complex environments,” Journal of Computational Design and Engineering, vol. 9, no. 2, pp. 755–771, Apr. 2022, doi:
10.1093/jcde/qwac025.
[6] S. Michieletto, S. Ghidoni, E. Pagello, M. Moro, and E. Menegatti, “Why teach robotics using ROS,” Journal of Automation,
Mobile Robotics and Intelligent Systems, pp. 60–68, Jan. 2014, doi: 10.14313/jamris_1-2014/8.
[7] M. Stolzle, T. Miki, L. Gerdes, M. Azkarate, and M. Hutter, “Reconstructing Occluded Elevation Information in Terrain Maps
with Self-Supervised Learning,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 1697–1704, Apr. 2022, doi:
10.1109/LRA.2022.3141662.
[8] I. M. Rekleitis, G. Dudek, and E. E. Milios, “Multi-robot collaboration for robust exploration,” in Proceedings-IEEE
International Conference on Robotics and Automation, 2000, vol. 4, pp. 3164–3169, doi: 10.1109/ROBOT.2000.845150.
[9] B. P. Gerkey and M. J. Matarić, “Multi-robot task allocation: Analyzing the complexity and optimality of key architectures,” in
Proceedings - IEEE International Conference on Robotics and Automation, 2003, vol. 3, pp. 3862–3868, doi:
10.1109/robot.2003.1242189.
[10] S. Lee, E. Milner, and S. Hauert, “A Data-Driven Method for Metric Extraction to Detect Faults in Robot Swarms,” IEEE
Robotics and Automation Letters, vol. 7, no. 4, pp. 10746–10753, Oct. 2022, doi: 10.1109/LRA.2022.3189789.
[11] Y. Zhang, D. Boley, J. Harwell, and M. Gini, “A Correlated Random Walk Model to Rapidly Approximate Hitting Time
Distributions in Multi-robot Systems,” in Lecture Notes in Networks and Systems, vol. 577 LNNS, Springer Nature Switzerland,
2023, pp. 724–736.
[12] X. Fu, Y. Zhang, J. Zhu, and Q. Wang, “Bioinspired cooperative control method of a pursuer group vs. a faster evader in a limited
area,” Applied Intelligence, vol. 53, no. 6, pp. 6736–6752, Jul. 2023, doi: 10.1007/s10489-022-03892-8.
[13] G. Merkulov, M. Weiss, and T. Shima, “Virtual Target Approach for Multi-Evader Intercept,” in 2022 European Control
Conference, ECC 2022, Jul. 2022, pp. 1491–1496, doi: 10.23919/ECC55457.2022.9838427.
[14] A. Giusti, G. C. Maffettone, D. Fiore, M. Coraggio, and M. di Bernardo, “Distributed control for geometric pattern formation of
large-scale multirobot systems,” Frontiers in Robotics and AI, vol. 10, Sep. 2023, doi: 10.3389/frobt.2023.1219931.
[15] S. Wang, Z. Li, H. Gao, K. Shan, J. Li, and H. Yu, “Agile Running Control for Bipedal Robot Based on 3D-SLIP Model
Regulation in Task-Space,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and
Lecture Notes in Bioinformatics), vol. 13455 LNAI, Springer International Publishing, 2022, pp. 505–516.
[16] P. Fong, C. Fang, and J. He, “Optimal Attack Against Coverage Path Planning in Multi-robot System,” in Lecture Notes in
Electrical Engineering, vol. 934 LNEE, Springer Nature Singapore, 2023, pp. 1760–1772.
[17] L. E. Parker, “Multiple Mobile Robot Teams, Path Planning and Motion Coordination in,” in Encyclopedia of Complexity and
Systems Science, Springer New York, 2009, pp. 5783–5800.
[18] S. H. Kim, S. G. Yoon, S. H. Chae, and S. Park, “Economic and environmental optimization of a multi-site utility network for an
industrial complex,” Journal of Environmental Management, vol. 91, no. 3, pp. 690–705, Jan. 2010, doi:
10.1016/j.jenvman.2009.09.033.
[19] A. Yahya, A. Li, M. Kalakrishnan, Y. Chebotar, and S. Levine, “Collective robot reinforcement learning with distributed
asynchronous guided policy search,” in IEEE International Conference on Intelligent Robots and Systems, Sep. 2017, vol. 2017-
September, pp. 79–86, doi: 10.1109/IROS.2017.8202141.
[20] J. Di Carlo, P. M. Wensing, B. Katz, G. Bledt, and S. Kim, “Dynamic Locomotion in the MIT Cheetah 3 Through Convex Model-
Predictive Control,” in IEEE International Conference on Intelligent Robots and Systems, Oct. 2018, pp. 7440–7447, doi:
10.1109/IROS.2018.8594448.
[21] M. Bjelonic et al., “Offline motion libraries and online MPC for advanced mobility skills,” International Journal of Robotics
Research, vol. 41, no. 9–10, pp. 903–924, Jun. 2022, doi: 10.1177/02783649221102473.
[22] K. H. Lee, O. Nachum, T. Zhang, S. Guadarrama, J. Tan, and W. Yu, “PI-ARS: Accelerating Evolution-Learned Visual-
Locomotion with Predictive Information Representations,” in IEEE International Conference on Intelligent Robots and Systems,
Oct. 2022, vol. 2022-October, pp. 1447–1454, doi: 10.1109/IROS47612.2022.9981952.
[23] A. Das, O. Naroditsky, Z. Zhu, S. Samarasekera, and R. Kumar, “Robust visual path following for heterogeneous mobile
platforms,” in Proceedings - IEEE International Conference on Robotics and Automation, May 2010, pp. 2431–2437, doi:
10.1109/ROBOT.2010.5509699.
[24] M. Debord, W. Honig, and N. Ayanian, “Trajectory Planning for Heterogeneous Robot Teams,” in IEEE International
Conference on Intelligent Robots and Systems, Oct. 2018, pp. 7924–7931, doi: 10.1109/IROS.2018.8593876.
[25] Z. G. Saribatur, V. Patoglu, and E. Erdem, “Finding optimal feasible global plans for multiple teams of heterogeneous robots
using hybrid reasoning: an application to cognitive factories,” Autonomous Robots, vol. 43, no. 1, pp. 213–238, Mar. 2019, doi:
10.1007/s10514-018-9721-x.
[26] J. Collins, S. Chand, A. Vanderkop, and D. Howard, “A review of physics simulators for robotic applications,” IEEE Access, vol.
9, pp. 51416–51431, 2021, doi: 10.1109/ACCESS.2021.3068769.
[27] Y. Yu, Z. Miao, X. Wang, and L. Shen, “Cooperative circumnavigation control of multiple unicycle-type robots with non-
identical input constraints,” IET Control Theory and Applications, vol. 16, no. 9, pp. 889–901, Mar. 2022, doi:
10.1049/cth2.12275.
[28] C. Calderón-Arce, J. C. Brenes-Torres, and R. Solis-Ortega, “Swarm Robotics: Simulators, Platforms and Applications Review,”
Computation, vol. 10, no. 6, p. 80, May 2022, doi: 10.3390/computation10060080.
[29] M. Garzón et al., “Using ROS in multi-robot systems: Experiences and lessons learned from real-world field tests,” in Studies in
Computational Intelligence, vol. 707, Springer International Publishing, 2017, pp. 449–483.

IAES Int J Rob & Autom, Vol. 13, No. 2, June 2024: 205-219
IAES Int J Rob & Autom ISSN: 2722-2586  219

[30] C. Pinciroli et al., “ARGoS: A modular, parallel, multi-engine simulator for multi-robot systems,” Swarm Intelligence, vol. 6, no.
4, pp. 271–295, Nov. 2012, doi: 10.1007/s11721-012-0072-5.
[31] M. U. Arif and S. Haider, “A Flexible Framework for Diverse Multi-Robot Task Allocation Scenarios Including Multi-Tasking,”
ACM Transactions on Autonomous and Adaptive Systems, vol. 16, no. 1, pp. 1–23, Mar. 2022, doi: 10.1145/3502200.
[32] P. Fraisse, D. Andreu, R. Zapata, J. P. Richard, and T. Divoux, “Remote decentralized control strategy for cooperative mobile
robots,” in 2004 8th International Conference on Control, Automation, Robotics and Vision (ICARCV), 2004, vol. 2, pp. 1011–
1016, doi: 10.1109/icarcv.2004.1468982.
[33] M. Dihya, M. Moufid, B. Chemseddine, and B. Moussaab, “Switched time delay control based on neural network for fault
detection and compensation in robot,” IAES International Journal of Robotics and Automation (IJRA), vol. 10, no. 2, p. 91, Jun.
2021, doi: 10.11591/ijra.v10i2.pp91-103.
[34] W. Łabuński and A. Burghardt, “Software for the Control and Monitoring of Work of a Collaborative Robot,” Journal of
Automation, Mobile Robotics and Intelligent Systems, vol. 2021, no. 3, pp. 29–36, May 2021, doi: 10.14313/JAMRIS/3-2021/16.
[35] Y.-C. Wu, J.-W. Lee, and H.-C. Wang, “Robots for search site monitoring, suspect guarding, and evidence identification,” IAES
International Journal of Robotics and Automation (IJRA), vol. 9, no. 2, p. 84, Jun. 2020, doi: 10.11591/ijra.v9i2.pp84-93.
[36] M. G. Ball, B. Qela, and S. Wesolkowski, “A review of the use of computational intelligence in the design of military surveillance
networks,” in Studies in Computational Intelligence, vol. 621, Springer International Publishing, 2016, pp. 663–693.
[37] L. Joseph and J. Cacace, “Mastering ROS for robotics programming : design, build, and simulate complex robots using Robot
Operating System,” 2015.
[38] M. Otte, M. J. Kuhlman, and D. Sofge, “Auctions for multi-robot task allocation in communication limited environments,”
Autonomous Robots, vol. 44, no. 3–4, pp. 547–584, Jan. 2020, doi: 10.1007/s10514-019-09828-5.
[39] D. Tihanyi, Y. Lu, O. Karaca, and M. Kamgarpour, “Multi-robot task allocation for safe planning against stochastic hazard
dynamics,” Jun. 2023, doi: 10.23919/ECC57647.2023.10178126.

BIOGRAPHIES OF AUTHORS

Abderrahmane Tamali received a master’s degree in Telecommunications

Systems from Tahri Mohammed Bechar University. Currently pursuing a Ph.D. in Networks
and Telecommunications at Ferhat Abbas Setif1 University, he excels in both substantive and
group work environments. He possesses a diverse skill set, including programming, electronic
application development, manipulation of Linux environments, development of robotic
machines, utilization of ROS framework, and IoT applications. In addition to his academic
achievements, Abderrahmene holds a senior IT technician diploma with a specialization in
computer network maintenance. He is proficient in computer networks and is adept at
troubleshooting and maintaining network systems. He can be contacted via email at
[email protected] or [email protected].

Nourredine Amardjia received an engineering degree in electronics in 1982

from ENPA (Ecole Nationale Polytechnique d'Alger), Algeria, his Master of Science degree in
electrical engineering in 1985, from Fairleigh Dickinson University, New Jersey, USA, and
his State Doctorate (Ph.D.) in communications in 2007 from University of Setif, Algeria. He
joined the Electronics Department, University of Setif as an assistant professor in 1986. He
was granted the level of associate professor in 2007. He is a member of the LIS laboratory, the
University of Setif. His research interests include discrete transforms, image processing, filter
design techniques, systolic architectures, and fast algorithms for signal processing
applications. He can be contacted at [email protected].

Mohammed Tamali graduated from USTO-MB as a state engineer in Electrical

engineering, he received his M.Sc. in 1996 in Energetical physics from Bechar University and
the Ph.D. degree from UST-MB of Oran, Algeria in 2007. In 2013, he became a professor in
Electrical Engineering. He is head of the SimulIA research team at the ENERGARID Lab. His
fields of interest included power electrical systems, scientific computing tools, sustainable
development, environmental studies applied to distributed electrical network optimization, and
system theory application on power systems. He worked as a research professor at the University
of Bechar from 1986 until today. He can be contacted at [email protected].

Distributed and autonomous multi-robot for task allocation and collaboration … (Tamali Abderrahmane)