2021 IEEE 22nd International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM)

Multi-Platform Hardware In The Loop (HIL)

Simulation for Decentralized Swarm
Communication Using ROS and GAZEBO
Saran Khaliq∗ , Shahzeb Ahsan∗ , and M.Danish Nisar†
∗ Swarm Robotics Lab, NCRA, University of Engineering & Technology (UET), Taxila.
† Sir
Syed CASE Institute of Technology (SSCIT), Islamabad, Pakistan.
Email: {saran.khaliq,shahzeb.ahsan}@uettaxila.edu.pk, † {[email protected]}
∗

Abstract—Swarm of robots is a group of multiple autonomous swarm independent of a single point of failure. Hence, if one
agents, collaborating with each other to achieve collective mis- agent fails, it does not affect any other agent in the swarm,
sions such as surveillance and tracking. In Unmanned Aerial
making the system more robust.
Vehicles (UAVs) swarms, robust and decentralized communica-
tion between the agents is required. To this end, Software-In- Considerable work has been done on UAV swarms with
The-Loop (SIL) is performed before the actual flight to avoid both centralized and decentralized communication. Elkilany
the risk of failure. However, single-platform-based simulations et al. proposed a neural network (NN) model of decentralized
can neither fully encounter nor resolve the communication formation control algorithm [7]. The experiment is done using
challenges. In this work, we target to incorporate the real-time
three turtlebot3 swarm running ROS, whereas the algorithm
communication challenges such as network failure, faced by
multi-agent systems via Hardware-In-The-Loop (HIL) simulation was implemented in MATLAB. The proposed NN algorithm
using Robot-Operating-System (ROS), Gazebo and off the shelf performs fine in searching and tracking various trajectories.
communication modems. A Ground-Control-Station (GCS) is However, decentralization of swarm communication through
developed to monitor and supervise autonomous UAVs missions. a single ROS-master remains unresolved.
The multi-platform HIL completely emulates the multi-agent
Several algorithms can be found on leader-follower forma-
system and the communication between the agents, thereby
reducing the risk of swarm failure. tion. Souza et. al. proposed two strategies based on leader-
Index Terms—Decentralized Communication, Swarm Commu- follower formation [8]. This work is implemented on ROS
nication, Multi-agent, Software-In-The-Loop (SIL), Hardware- framework and accessed on Gazebo simulator, thus highlight-
In-The-Loop (HIL), ROS, Gazebo. ing the behavior of drones in the terms of response time and
accommodation time of the formation. Although they have
I. I NTRODUCTION done experiments on gazebo using ROS, the implementation of
decentralized swarm communication is not fully investigated.
Lately, the swarm of UAVs, consisting of multiple UAVs Petracek experimented with vision-based self-localization
collaborating to perform different missions, are being used in in aerial swarms [9]. They have proposed a simple method
military, industrial, agriculture, emergency rescue, and numer- to navigate, control and stabilize aerial robots while having
ous other applications. Formation control and path planning information about only the local agent, thus making it de-
are the most important tasks for the swarm to achieve its centralized. As they have used ROS for implementing the
goal. These critical missions motivate researchers to develop swarm, but their communication is still dependent on a single
robust architecture to maintain specific formations and to avoid ROS-master. Indriyanto et al. developed a basic simulation-
obstacles and collisions. Hence, we need robust and faultless based swarming structure for search, monitoring, and mapping
communication among the swarm nodes. applications [10]. They have developed a centralized but
Mainly, two methods are used for the swarms to com- modular UAVs solution which updates the mission if any of
municate with each other, which are either centralized and the UAV’s link gets broken. Therefore, the rest of the UAVs
decentralized. The centralized communication method needs complete the mission with the trade-off of additional time.
each robot to communicate through one central agent, which However, this setup is yet to be tested on hardware.
could be a ground control station (GCS), as pursued in [1]–[6]. Lamping et al. worked on a multi-agent UAV system based
A consequence of centralized control is that if the GCS link on ROS [11]. They have experimented with control and su-
breaks, the whole swarm fails. This single point of failure is pervision algorithms on multiple UAVs and created a software
highly undesirable for swarm missions. On the other hand, in package as flyMASTER. They have done both SIL and HIL
a decentralized approach, the robots communicate with each and created a GCS to control and monitor the swarm. Their
other using mesh connectivity where each robot is connected system can work on any flight controller that can be integrated
to one or multiple robots within its range. This makes the with Micro Aerial Vehicle Link (MAVLink) communication

978-1-6654-2263-5/21/$31.00 ©2021 IEEE 310

DOI 10.1109/WoWMoM51794.2021.00057
 Registration ROS Registration
MASTER

MAVROS
Node 1 Messages Node 2
Publisher Subscriber

Publish
Rosbridge_server Subscribe
Topic
Flight Onboard Roslibjs
Controller Computer Node.js
Modem
Fig. 2: ROS architecture

sharing of information without relying on a single link. ROS

Fig. 1: System Description supports different programming languages i.e. C++, Python,
JAVA, Lisp, and other famous tools such as Lab-VIEW. It
provides domain-specific tools and libraries to smooth the
protocol. Thus, experimenting on multiple UAVs having PX4 process of development. ROS architecture can be seen in
and Pixhawk flight controllers. However, their SIL consists Figure 2.
of a single platform simulating multiple drones ultimately
depending upon a single ROS-master. B. Gazebo
In our work, we propose a decentralized approach on multi- Gazebo is a simulator that provides a three-dimensional
platform HIL simulation of UAV swarm. We handpicked the (3D) environment implanted with both indoor as well as
linking libraries/ tools and developed a ROS network of UAVs outdoor environments, containing 3D models of robots and
that does not depend upon a single ROS-master, making the accompanying sensors to set up. It comes with real-world
communication link more robust. We have also tested our models with their GPS information [13]. Many simulators can
proposed HIL simulation and monitored respective swarm be used for the modeling and simulating UAV hardware-in-
behaviors on GCS. The main features of our work are as the-loop system. The jMavSim [14] is a simple open-source
follows: used for simulating a lightweight multi-rotor UAV system.
A. Emulating Individual UAV However, there are some limitations regarding simulating more
complex models. Other simulators that can be used for HIS
• Each UAV is simulated in a physically isolated system such as FlightGear [15], Vrep [16], GEFS [17], Xplane [18],
equipped with a ROS-Gazebo simulation. and MAV3DSim [19] offers a good simulation platform but
• Each system is attached to a wireless modem. a significant amount of effort is required to test high-level
B. Multi-Platform communication control algorithms.
To avoid these limitations, we chose the most popular open-
• Communication between physically isolated systems con-
source simulation software in the robotics community, Gazebo
nected via modems.
[20]. The main advantage of this simulator is the easiness to set
• Immune to team-member failure i.e. out-of-network, mal-
up a 3D virtual environment, modeling robots, and their sen-
functioning.
sors, and defining the communication protocol. Furthermore,
C. Ground Control Station (GCS) its open dynamics engine (ODE) is useful to perform real-time
• Monitoring the sensory data of each UAV. experiments accurately [21]. Thus, providing real-time sensor
• Visualizing the video stream of UAVs. readings in a dynamic simulation environment for the UAVs.
• Geo-locating all UAVs on a single 2D map. C. MAVROS
A UAV is equipped with a Flight Control System (FCS) to
II. SYSTEM ARCHITECTURE
control different components of a UAV. FCS uses a specific
Our multi-platform HIL relies on existing software and tools protocol named MAVLink. It is a library explicitly designed
such as ROS, Gazebo, and MAVROS. We employ Microhard for limited bandwidth applications i.e. UAVs. Micro Aerial
modems for wireless communication between UAVs. The Vehicle Robotic Operating System (MAVROS) is a dedicated
complete architecture of a testing platform can be seen in ROS package that works as a bridge between MAVLink and
Figure 1. Individual components are briefly explained below. ROS in order to provide high-level control over FCS using
ROS [22].
A. ROS
When it comes to programming robots, researchers face lots D. Ground Control Station (GCS)
of complexities to develop software. ROS, however, is a set of Rosbridge is a web-socket server, which lays an extra layer
libraries that provides the researchers with the opportunity to to the ROS. This bridge allows the communication of non-ROS
fast-track product development by reducing software complex- applications with ROS applications. Roslibjs is a javascript-
ity [12]. ROS provides a distributive architecture, which allows based ROS library that is used in applications particularly

311
 UAV 1
Modem1

Platform 1

Geo location
Sensor data
Battery status
UAV 2 Ground Control
GCS Station
Modem2 Modem
Roslibjs Open source
Platform 2 Javascript Map

Fig. 4: Ground Control Station

UAV N
Modem N
Mesh
Platform N Network
and feed these control inputs to the 6DOF model of the quad-
Fig. 3: Communication of multi-UAVs on different systems rotor UAV. The dynamic model contains the translational and
rotational motion for the quad-rotor UAV which is defined as
roll φ, pitch θ, yaw ψ, and x, y, z respectively. The model is
using IP sockets, consenting them to get interfaced with ROS implemented and explained in [26], which is described by the
applications [23] [24]. following equations
We have developed a GCS, a webpage using Roslibjs
which monitors the sensor’s data, GPS location, and respective 
φ̈ = M1xx I2 − Jr θωr + Myy − Mzz ψ̇ θ̇




swarm formations and flocking on a 2D world map provided 
θ̈ = M1yy I3 − Jr φωr + Mzz − Mxx φ̇ψ̇





by leaflet OpenStreetMap contributors [25]. The customized 

ψ̈ = M1zz I4 + Mxx − Myy θ̇φ̇




GCS web page can be seen in Figure 4. (4)


ẍ = cos θ cos φ sin θ + sin θ sin φ Im1
III. UAV MATHEMATICAL MODEL 


ÿ = cos θ sin φ sin θ − sin θ cos φ Im1

z̈ = cos φ cos θ Im1 − g

We used quad-rotor UAV having a 6-DOF model in the
gazebo software, in which we have to control the angular where Mxx , M yy, Mzz are moments of inertia around the
velocity of each motor. Two diagonal motors of the UAV frame axis; ωr is the sum of all angular velocities; Jr is the
rotate in the clockwise direction and the other two rotate in inertia of the motor; m is the total mass and g is gravitational
the anticlockwise direction. Motor force and the moment are constant. Global positioning system (GPS) and the Inertial
produced around its axis by providing thrust and torque such measurement unit (IMU) sensor modules are used in each of
as the quad-rotor models in Gazebo. The GPS data contains the
Fi = Cf − ωi2 (1) longitude (δ), latitude (λ), and altitude (z), whereas IMU data
contains the information of the angular velocity (va ), linear
Mi = Cm − ωi2 (2) acceleration (a), and orientation (ξ). Gazebo uses physics
library to provide us the current position (x, y, z), IMU data
where Cf is the motor thrust constant, Cm is the motor torque (va , a, ξ) and the initial location (δ0 , λ0 , z0 ). Based on the
constant and ωi is the angular velocity of ith motor. Therefore, above data the current location is calculated as [27],
the control inputs i.e. I1 , I2 , I3 , I4 can be defined as

δ1 = arcsin cos d sin λ0 + γx sin d cos δ0



ω12
     
I1 Cf Cf Cf Cf 
  
y sin d
 I2   −l1 Cf −l1 Cf l1 C f l1 C f   ω22  λ1 = λ0 + arctan γ cos d cos δ0 −x sin d sin δ0 (5)
 I3  =  l2 Cf
    
ω32

−l2 Cf −l2 Cf l2 C f 
  

 z1 = z0 + z
I4 −Cm Cm −Cm Cm ω42
(3) p
where γ = x2 + y 2 , d = R γ
and R is the radius of Earth.
where l1 and l2 are the distances along the x and y-axis
to UAV’s center of gravity. This motor control function is
described in the motor plugin in the gazebo. These motors IV. P ROPOSED M ETHODOLOGY
are derived using two methods, first is to take the signal from
A. Emulating individual UAVs
the controller and use the inverse function of Equation 3 to
get the angular velocity of all motors, which is further used A Gazebo simulator is required to simulate a UAV in
for the rotation of the propellers. The second is to use force a virtual world. In Gazebo, MAVLink is responsible for
and moments for each motor (I1 , I2 , I3 , and I4 ) from Eq. 3 establishing a connection between the FCS and its onboard

312
 4/20/2021 4/20/2021 SWARM ROBOTS - GCS SWARM ROBOTS - GCS

GROUND CONTROL STATION

GROUND CONTROL STATIO
Menu Menu UAV 1
+ +
Dashboard (index.html) Dashboard (index.html) Start

Leader − Leader −
Follower Follower
Geolocation Geolocation
(Geolocation.html) (Geolocation.html)

UAV 1 UAV 1
Connection : Closed Connection : Closed

Battery : Battery : t2 t2
UAV 2
Lat: Lat:
Start
Lon: Lon:
t1 t1
UAV 2 UAV 2
Connection : Closed Connection : Closed
4/20/2021 SWARM ROBOTS - GCS
Battery : Battery :

Lat: Lat:
GROUND CONTROL STATION
Microhard modems
Lon: Lon:

UAV 3 UAV 3 UAV 3

Fig. 5: Communication of two UAVs simulating
Menu on Gazebo at
Connection : Closed
+
Connection : Closed (a) Initial point (b) Mid flight Start
two separate platforms
Battery :
Dashboard (index.html)
Battery :
Leader −
Lat: Lat:
Follower
Lon: Geolocation Lon:

computer. To give high-level control, MAVROS creates a (Geolocation.html)

bridge between FCS and ROS. UAV 1

Connection : Closed
The ROS messages of UAV Battery simulated
:
in Gazebo are
t2
published on a certain topic through
Lat: MAVROS, which are Leaflet
(https://leafletjs.com) | © MapTiler (https://www.maptiler.com/copyright/)
Leaflet©(https://leafletjs.com)
OpenStreetMap contributors (https://www.openstreetmap.org/copyright)
| © MapTiler (https://www.maptiler.com/copyright/) © OpenStreetMap contributors (https:

forwarded to a web socket usingLon:ros bridge. The modems t1

attached to each platform will create aUAV 2

mesh network so
Connection : Closed
that the UAVs emulated on multiple platforms can share
Battery :
information. Lat:

Lon:
B. Establishing communication between UAVs and with GCS
UAV 3
via modems Connection : Closed (c) Destination point
UAVs are simulated on Gazebo-simulators
Battery : on three different
Lat:
platforms physically isolated communicating over a mesh net- Fig. 6: GCS showing the formation and trajectory of UAVs,
Lon:
work of modems. Each UAV is assigned a unique IP address
file:///home/dc/testing/version A/index.html file:///home/dc/testing/version A/inde
mission points A, B, C and D.
x.html

so that the UAVs can communicate with each other using these
addresses in a network as shown in Figure 3.
V. E XPERIMENTATION AND R ESULTS
A GCS using Roslibjs can fetch information through an Leaflet (https://leafletjs.com) | © MapTiler (https://www.maptiler.com/copyright/) © OpenStreetMap contributors (https://www.openstreetmap.org/copyright)

IP socket. GCS being the part of the network can subscribe to Each platform in the network is installed with Ubuntu 18.04,
the published topic provided the IP address of the particular ROS melodic, Gazebo9 simulator, and an open-source ROS
system with Gazebo-simulator. The communication of GCS package provided by Intelligent Quads (IQsim) community
with other UAVs simulated on physically isolated systems can [29]. IQsim provides high-level control of UAVs to develop
be seen in Figure 3. intelligent drone applications. The gazebo is loaded with a
The GCS will receive the Global Position System (GPS) model of a runway at Canberra Model Aircraft Club Flying
coordinates from UAVs in Gazebo which can be visualized on Field, located in Symonstion, Australia. Each platform is
a 2D Map, thanks to Roslibjs. This functionality is important further connected to wireless modems through Ethernet for
to see the collaboration as the UAVs are in the same world their communication. The wireless modems are sending their
but on different platforms. Similarly, data from sensors are
file:///home/dc/testing/version A/index.html
data between the platforms and the GCS with a transmission
monitored which are published by ROS on topics and are power of 7 dBm. UAVs simulated in a Gazebo on two
subscribed on GCS. physically isolated platforms can be seen communicating with
C. The architecture of Decentralized Communication each other in Figure 5. Similarly, more platforms can be added
to increase the number of UAVs in the network and their
Microhard modem pMDDL2450 provides a built-in feature
collective information can be seen on GCS.
of mesh connectivity, which helps in creating a decentralized
To investigate the communication between UAVs over a
communication network for UAVs swarm operations [28].
mesh network, a leader-follower formation is created. The
Each platform is connected to every other platform via a mesh
four platforms, one leader-UAV, and three follower-UAVs are
network through this modem. If any of the nodes leaves or
emulated on the same runway in Gazebo. The mission points
joins the network, the rest of the nodes will communicate in-
are only dictated to the leader to fly from point A to point
dependently in a network. If any node is not directly connected
D as shown in Figure 6a. Initially UAVs maintain a specific
to the other node, the modem uses its proprietary routing
distance, resulting in a diamond formation. The three followers
algorithms to send the data using relay nodes. The complete
go along with the leader while maintaining a specific distance
architecture of decentralized communication is shown in Fig-
based on IMU readings of their leader until the leader reaches
ure 3.
its destination point. These platforms are communicating GPS

313
 Subscribing
topic of UAV1
from network

Node.js

/mavros_global_position_local

/drone1_gps_local

Fig. 7: ROS-node graph of Leader Fig. 8: ROS-node graph of Follower

-30 -30

t1

t2

t1

t2
UAV2 UAV1
UAV3 UAV3
-40 -40
coordinates and IMU readings with each other and GCS UAV4
GCS
UAV4
GCS
over a mesh network through modems. The GPS coordinates -50 -50

RSSI (dBm)

RSSI (dBm)
assigned by the gazebo to all the UAVs are monitored on -60 -60

GCS throughout the mission. The formation of UAVs at initial -70 -70

point A, between B and C, and at a final point D can be seen -80 -80

in Figure 6. The formation remains the same throughout the -90

0 50 100 150
-90
0 50 100 150
mission, as the UAVs are maintaining equal distance on the Time (seconds) Time (seconds)

basis of shared information in the network. (a) RSSI of UAVs with respect to (b) RSSI of UAVs with respect to
Multiple ROS nodes are running in a particular UAV. UAV1 UAV1 UAV2
is acting as a leader and knows the mission. UAV1 needs to -30 -30
t1

t2

t1

t2
share its ROS nodes with other UAVs in the mesh network -40
UAV1
UAV2
-40
UAV1
UAV2
UAV4 UAV3
through a web socket server. We are using ROS node to send GCS GCS
-50 -50
information of leader to followers (UAV2, UAV3 & UAV4)
RSSI (dBm)

RSSI (dBm)
-60 -60
after subscribing the topic from rosbridge and then publish
-70 -70
the same message on respective IP sockets of UAVs in the
-80 -80
network. The ROS node graph can be seen in Figure 7.
Similarly, a ROS-node graph of follower UAVs can be seen in -90
0 50 100 150
-90
0 50 100 150
Time (seconds) Time (seconds)
Figure 8. The UAVs acting as the follower will continuously
receive information from the UAV acting as a leader and will (c) RSSI of UAVs with respect to (d) RSSI of UAVs with respect to
UAV3 UAV4
keep on following the UAV1.
Received Signal Strength Indicator (RSSI) is recorded from Fig. 9: RSSI of each UAV with respect to other UAVs. Time
the modems connected with the platforms. We see that the t1 and t2 shows the interval in which UAV1 is disconnected
RSSI of the UAV4 platform is degraded with respect to the with UAV4
UAV1 platform over time due to channel impairments by
including a metallic plate between them. During the time
interval of 73 to 101 seconds, the RSSI goes below -90 dBm platform, which is acting as a relay and providing a path to
and the link between UAV1 and UAV4 gets broken.This is UAV4 to remain in the network with other UAV’s. Thus, all
shown in Figure 9 where t1 and t2 donate the time instances the platforms stay in the network throughout the simulation
where link gets broken and restored respectively. Figure 9a period. Therefore, to be part of the network, each platform
shows that while the link of UAV1 with UAV2,3 remain intact needs at least one direct link with any other platform in the
throughout, link with UAV-4 gets broken during t1 and t2 . network.
Similar observations can be made from Figure 9b, 9c. In The UAV platform is considered to be out of the network
Figure 9d, it can be seen that the link of UAV4 breaks down when it is not in direct connection with all of the other UAV
with UAV1 and UAV2 while it remains connected with UAV3. platforms and the GCS. As the mission information is only
Moreover, we see that even in the interval when the follower known to the leader UAV, its disconnection will result in all
UAV4 platform is disconnected from the leader UAV1, UAV4 follower UAVs stopping at their positions. The follower UAVs
still follows the path of UAV1 by maintaining a specified will remain in the network even in the absence of the leader
distance as shown in Figure 6b. This is because of the UAV3 UAV. To continue the mission, the remaining follower UAVs

314
can select a new leader which is not in the scope of this paper [13] “Gazebo,” 2018, date accessed 03-10-2020. [Online]. Available:
http://gazebosim.org/
and maybe discussed in future work. [14] R. Wandarosanza, B. R. Trilaksono, and E. Hidayat, “Hardware-in-
the-loop simulation of uav hexacopter for chemical hazard monitoring
mission,” in 2016 6th International Conference on System Engineering
VI. C ONCLUSIONS and Technology (ICSET), 2016, pp. 189–193.
[15] Y. A. Prabowo, B. R. Trilaksono, and F. R. Triputra, “Hardware in-
In this work, we have proposed and demonstrated a novel the-loop simulation for visual servoing of fixed wing uav,” in 2015
International Conference on Electrical Engineering and Informatics
multi-platform hardware-in-the-Loop (HIL) emulation mech- (ICEEI), 2015, pp. 247–252.
anism for realistic modeling of swarm communication. The [16] E. Rohmer, S. P. N. Singh, and M. Freese, “V-rep: A versatile and
proposed mechanism consists of ROS and Gazebo running on scalable robot simulation framework,” in 2013 IEEE/RSJ International
Conference on Intelligent Robots and Systems, 2013, pp. 1321–1326.
physically isolated platforms connected via wireless modems [17] Geo-Fs, “Geo-fs free online flight simulator,” date accessed 01-03-2021.
allowing us to explore real-time communication and its la- [Online]. Available: https://www.geo-fs.com/
tency. Moreover, the HIL is implemented using the principles [18] H. KORKMAZ, O. B. ERTİN, C. KASNAKOĞLU, and ünver
KAYNAK, “Design of a flight stabilizer system for a small
of decentralized communication, thereby avoiding a single fixed wing unmanned aerial vehicle using system identification,”
point of failure, which is crucial to the successful operation of IFAC Proceedings Volumes, vol. 46, no. 25, pp. 145–149,
critical missions that the swarm is intended to pursue. Results 2013, 1st IFAC Workshop on Advances in Control and
Automation Theory for Transportation Applications. [Online]. Available:
and discussion of a simple leader-follower swarm formation https://www.sciencedirect.com/science/article/pii/S147466701535223X
have been provided as an example. [19] I. Lugo-Cárdenas, S. Salazar, and R. Lozano, “The mav3dsim hardware
in the loop simulation platform for research and validation of uav
controllers,” in 2016 International Conference on Unmanned Aircraft
ACKNOWLEDGMENT Systems (ICUAS), 2016, pp. 1335–1341.
[20] Gazebo, “Player/stage/gazebo,” 2017, date accessed 01-03-2021.
[Online]. Available: https://sourceforge.net/projects/playerstage/
This research work is carried out at Swarm Robotics Lab [21] C. E. Agüero, N. Koenig, I. Chen, H. Boyer, S. Peters, J. Hsu, B. Gerkey,
under National Centre for Robotics and Automation (NCRA). S. Paepcke, J. L. Rivero, J. Manzo, E. Krotkov, and G. Pratt, “Inside
The NCRA is funded by Higher Education Commission the virtual robotics challenge: Simulating real-time robotic disaster
response,” IEEE Transactions on Automation Science and Engineering,
(HEC), Pakistan. The authors acknowledge the NCRA and vol. 12, no. 2, pp. 494–506, 2015.
HEC for the technical and financial support. [22] “mavros - ros wiki,” 2018, date accessed 21-10-2020. [Online].
Available: http://wiki.ros.org/mavros
R EFERENCES [23] “roslibjs - ros wiki,” 2018, date accessed 05-11-2020. [Online].
Available: http://wiki.ros.org/roslibjs
[1] L. Chaimowicz and V. Kumar, Aerial Shepherds: Coordination among [24] K. Alisher, K. Alexander, and B. Alexandr, “Control of the mobile robots
UAVs and Swarms of Robots, 01 2007, pp. 243–252. with ros in robotics courses,” vol. 100, p. 1475–1484, 2015.
[2] M. Egerstedt and X. Hu, “Formation constrained multi-agent control,” [25] O. contributors, “leaflet,” 2020, date accessed 05-11-2020. [Online].
Robotics and Automation, IEEE Transactions on, vol. 17, pp. 947 – 951, Available: https://leafletjs.com/
01 2002. [26] J.-J. Xiong and G.-B. Zhang, “Global fast dynamic
[3] W. Kowalczyk, “Target assignment strategy for scattered robots building terminal sliding mode control for a quadrotor uav,” ISA
formation,” in Proceedings of the Third International Workshop on Robot Transactions, vol. 66, pp. 233–240, 2017. [Online]. Available:
Motion and Control, 2002. RoMoCo ’02., 2002, pp. 181–185. https://www.sciencedirect.com/science/article/pii/S0019057816303627
[4] R. O. Saber, W. B. Dunbar, and R. M. Murray, “Cooperative control of [27] J. P. Snyder, “Map projections: A working manual,”
multi-vehicle systems using cost graphs and optimization,” in Proceed- Washington, D.C., Tech. Rep., 1987, report. [Online]. Available:
ings of the 2003 American Control Conference, 2003., vol. 3, 2003, pp. http://pubs.er.usgs.gov/publication/pp1395
2217–2222 vol.3. [28] Microhard, “Microhard pmddl2450,” date accessed 06-11-2020.
[5] S. Zelinski, T. Koo, and S. Sastry, “Optimization-based formation [Online]. Available: http://microhardcorp.com/pMDDL2450.php
reconfiguration planning for autonomous vehicles,” 10 2003, pp. 3758 [29] E. Johnson, “Intelligent quads community,” 2020, date
– 3763 vol.3. accessed 05-11-2020. [Online]. Available: https://github.com/Intelligent-
[6] Z. Cao, M. Tan, S. Wang, Y. Fan, and B. Zhang, “The optimization Quads/iq tutorials
research of formation control for multiple mobile robots,” 02 2002, pp.
1270 – 1274 vol.2.
[7] B. Elkilany, A. Ali, A. Fath El Bab, and H. Ishii, “A proposed
decentralized formation control algorithm for robot swarm based on an
optimized potential field method,” Neural Computing and Applications,
05 2020.
[8] A. M. de Souza Neto and R. A. F. Romero, “A decentralized approach
to drone formation based on leader-follower technique,” in 2019 Latin
American Robotics Symposium (LARS), 2019 Brazilian Symposium on
Robotics (SBR) and 2019 Workshop on Robotics in Education (WRE),
2019, pp. 358–362.
[9] P. Petráček and M. Saska, “Decentralized aerial swarms using vision-
based mutual localization,” 11 2018.
[10] T. Indriyanto, A. Rizki, M. Hariyadin, M. Akbar, and A. Syafi, “Cen-
tralized swarming uav using ros for collaborative missions,” vol. 2226,
04 2020, p. 030012.
[11] A. Lamping, J. Ouwerkerk, N. Stockton, K. Cohen, M. Kumar, and
D. Casbeer, “Flymaster: Multi-uav control and supervision with ros,”
06 2018.
[12] W. Garage, “Robot operating system,” 2007, date accessed 05-11-2020.
[Online]. Available: https://www.ros.org/about-ros/

315