2024 International Conference on Unmanned Aircraft Systems (ICUAS)

June 4-7, 2024. Chania, Crete, Greece

A Modular UAV Hardware Platform for Aerial Indoor Navigation

Research and Development
Kyriakos M. Deliparaschos, Savvas G. Loizou, and Argyrios C. Zolotas

Abstract— This study introduces a specialised hardware plat- II. HARDWARE DESCRIPTION
form designed for indoor navigation, featuring a quadrotor
equipped with either a NVIDIA Jetson Nano or a Z-turn The finalised quadrotor project comprises diverse individ-
Zynq onboard computer. The onboard computer communicates ual components used to complete the project. This section
via ROS2 with the flight controller, the Inertial Measurement provides an overview of the key components chosen, accom-
Unit (IMU), Ultra-WideBand (UWB) localisation system, stereo
2024 International Conference on Unmanned Aircraft Systems (ICUAS) | 979-8-3503-5788-2/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICUAS60882.2024.10557007

panied by a concise analysis of the decision-making process.

camera, Light Detection and Ranging (LiDAR), and ultrasonic
sensors. The focus is on creating a low-cost modular Unmanned
Aerial Vehicle (UAV) system adaptable to various indoor
A. IMU
navigation applications. The modular design encompasses dif- An Inertial Measurement Unit (IMU) is a device that
ferent onboard computer platforms and sensor configurations, utilises a combination of accelerometer, gyroscope, and
allowing for easy adaptation to research experiment setups.
The objective is to facilitate the transition from simulated and magnetometer sensors to ascertain a body’s orientation, ve-
simplified laboratory experiments to deploying aerial robots locity, and gravitational forces. It is commonly employed
in challenging real-world conditions. The paper explores the alongside other sensors, such as those based on vision
hardware architecture and Robot Operating System 2 (ROS2)- or wireless technology, to improve the accuracy of pose
based communication system of the UAV and provides a weight estimation. In instances where other sensors are unavailable,
analysis and power estimation.
the IMU may function as the primary sensor. The IMU
I. INTRODUCTION delivers measurements of a quadrotor’s orientation through
the three Euler angles: roll, pitch, and yaw. The Pixhawk
Unmanned aerial vehicles (UAVs) have undoubtedly PX4 Flight Control Unit (FCU) incorporates three internal
opened new frontiers in research, particularly in aerial nav- sensor chips, primarily the Invensense MPU 6000 as the
igation [1], [2]. While there is extensive research efforts main 3-axis accelerometer/gyroscope, ST Micro L3GD20 3-
directed towards deploying UAV systems with various nav- axis 16-bit gyroscope, and ST Micro LSM303D 3-axis 14-bit
igation approaches (and in particular for the challenging accelerometer/magnetometer.
aspect of aerial indoor navigation), a substantial portion of
the proposed methodologies rely mostly on simulations [3]. B. FCU
Unfortunately, these approaches often fall short of meeting
The responsibility of controlling the drone’s movement
real-world deployment requirements. This paper introduces
and adjusting the power delivered to each motor is assigned
a low-cost modular UAV hardware platform developed to,
to the Flight Controller Unit (FCU). Flight controllers are
particularly, enable aerial indoor navigation research devel-
capable of measuring the drone’s level and speed, utilising
opment and system validation. By focusing on modularity,
this information to correct its orientation during flight. In
the platform provides a flexible and adaptable baseline for
this project, the chosen flight controller is the Readytosky
experiments, enabling researchers to customise and scale,
Pixhawk PX4 2.4.8 32-bit. This selection was made based
within reasonable payload requirements, their setups accord-
on the controller’s additional features beyond the standard
ing to the specific requirements of indoor environments. This
gyroscope and accelerometer sensors found in all flight
approach not only accelerates the rate of development but
controllers. The Pixhawk incorporates a barometer, magne-
also enables investigation of the precision and reliability
tometer, and IMU, enhancing the accuracy and consistency
of UAV navigation within indoor spaces. Through a com-
of flight performance. Furthermore, the Pixhawk is a 32-bit
prehensive exploration of its architecture, capabilities, and
controller with I2C and SPI interfaces and has the flexibility
potential applications, we aim to highlight how this platform
to run either the PX4 or ArduPilot flight stack.
can facilitate the advancement of UAV indoor navigation
research [4][5]. C. Onboard Computer
K. M. Deliparaschos is with the Electrical and Computer Engineering The quadrotor’s onboard computer, operated through
and Informatics Department, Cyprus University of Technology, Limassol, the Robot Operating System 2 (ROS2), manages essential
Cyprus k.deliparaschos@cut.ac.cy.
S. G. Loizou is with the Department of Mechanical Engineering and quadrotor functions, including image recognition from stereo
Materials Science and Engineering, Cyprus University of Technology, camera frames, processing ground Ultra Wide Band (UWB)
Limassol, Cyprus savvas.loizou@cut.ac.cy. sensor data, analysing point clouds from the LiDAR sensor,
A. C. Zolotas is with the Centre for Autonomous and Cyber-
Physical Systems, SATM, Cranfield University, Cranfield, UK integrating sensor measurements from the flight controller,
a.zolotas@cranfield.ac.uk. transmitting commands to the flight controller, and relaying

979-8-3503-5788-2/24/$31.00 ©2024 IEEE 398

Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
video to the ground station. When selecting a suitable com- features low power consumption (2.5W) and a lightweight
puter for our system, we sought a powerful yet lightweight design (138g).
and energy-efficient option. Our primary choice was the
NVIDIA Jetson Nano Development Kit, featuring a 4-core F. Ultrasonic Sensors
1.43GHz processor with 4GB of memory and a 128-core We selected a microcontroller, specifically the Atmel
Maxwell GPU [6]. Weighing 142 grams and using up to ATMEGA328P on an Arduino Pro Mini board, to oversee
15W at maximum power, its NVIDIA CUDA core GPU four laterally mounted HC-SR04 ultrasonic range sensors.
offers broader compatibility for software running object Directly invoking the ultrasonic sensor from the Jetson Nano
detection algorithms. This CPU-GPU combination ensures could result in suboptimal performance due to the Nano’s
ample computational power for running object detection real-time processing limitations, potentially leading to inac-
and other concurrently executed software. Additionally, the curate distance measurements. A more effective approach
Jetson Nano enables experimentation with NVIDIA-specific involves incorporating a compact microcontroller, such as
libraries for deep learning and computer vision, making it the Arduino Pro Mini, which, through I2C communication,
the preferred choice for our quadrotor application. can conduct distance measurements and relay results back to
Moreover, we integrated the MYIR Z-turn FPGA board the Jetson Nano, thereby enhancing the reliability of distance
[7] with an AMD (formerly Xilinx) Zynq-7020 SoC for real- calculations.
time point cloud processing. The Z-turn board, at least twice The chosen architecture, based on the Atmel AT-
as power-efficient compared to the Jetson Nano, enhances MEGA328P microcontroller on an Arduino Pro Mini board,
the processing capabilities for point cloud data, providing boasts an 8-bit 16 MHz design. This IC is well-suited for
an efficient solution for our quadrotor system. Utilising a driving ultrasonic range sensors and transmitting distance
dedicated Delaunay triangulation core [8], developed by the information to the processing module via UART. The choice
authors in prior work, on the FPGA portion (PL) of the Zynq is attributed to its compact form factor, low weight, mini-
SoC further enhances its ability to process point cloud data mal power consumption, extensive documentation, and user-
efficiently. Moreover, the Z-turn FPGA board features an on- friendly nature through the Arduino development platform.
board three-axis acceleration sensor and temperature sensor,
further augmenting its capabilities for various applications in G. WiFi Communication
our quadrotor system. To facilitate communication between the drone computer
and a ground control computer for sending video data, WiFi
D. Camera capability is essential. As the Jetson Nano lacks a built-in
WiFi module, we opted for the AC1300 Archer T3U Mini
Exploring various camera options for our system, one
Wireless MU-MIMO USB Adapter. Choosing a USB WiFi
possibility involved using a single, basic RGB camera to
module over a WiFi card was intentional, as the latter would
capture video data for an object detection algorithm, fa-
have necessitated the installation of a separate antenna for
cilitating the identification of object locations. However,
proper operation. The USB module, equipped with a small
we acknowledged that this setup might pose challenges
antenna, offered a more straightforward integration with the
in accurately determining the distance to most objects. To
overall system.
address this limitation, we opted for a ZED 2 stereo camera
The AC1300 Archer T3U USB module was specifically
[9], utilising the relative pixel differences between the two
selected for its compatibility with USB 3.0, available on
images to compute distances on the computer. The stereo
the Jetson Nano, allowing for data speeds of up to 4,800
camera connects via USB to the Jetson Nano. A drawback
Mbps—ten times faster than USB 2.0 ports. It delivers WiFi
of this approach is the substantial processing power required,
speeds of 400 Mbps on the 2.4GHz band and 867 Mbps
particularly when running a computationally intensive object
on the 5GHz band. The AC1300, combining 802.11ac WiFi
detection algorithm on the same system.
with USB 3.0, proves ideal for HD streaming and large file
Additionally, we considered a less computationally de-
downloads. Additionally, the AC1300 features MU-MIMO
manding approach by employing a LiDAR sensor instead
technology, enabling two simultaneous data streams, thereby
of two RGB cameras. Utilising the point cloud generated by
enhancing the throughput and efficiency of the entire network
the LiDAR sensor, we can employ the reconstructed surface
when paired with a compatible MU-MIMO router.
for object recognition while simultaneously determining the
distance to the objects. H. UWB Sensors
Ultra-Wideband (UWB) is a wireless communication tech-
E. LiDAR
nology that utilises short pulses with low energy over a
The light detection and ranging (LiDAR) sensor aids wide bandwidth, making it highly resistant to multipath in-
in detecting obstacles within the environment [10]. We’ve terference. Unlike conventional radio frequency identification
integrated a Hokuyo URG-04LX-UG01 2D LiDAR [11][12] (RFID) systems operating on single bands, UWB transmits
onto the UAV. This specific LiDAR offers a range from over a broad spectrum of radio frequencies, decreasing the
20mm to 5500m (with a 1mm resolution), covering a power spectral density. This characteristic allows UWB-
240°scan window with an angular resolution of 0.36°. It based systems to avoid interference with other RF signals.

399
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
 UWB’s unique features include the ability to precisely receiver’s function is to transmit signals to the controller
measure Time-of-Flight (ToF), facilitating accurate distance based on stick movements from the radio control transmitter,
estimation. This makes UWB a popular choice for local- enabling control over the quadcopter’s motion as required.
isation in indoor environments. Localisation using UWB
involves placing wireless transmitters, known as anchors, at III. SYSTEM DESCRIPTION
specific locations and using a UWB receiver, called a tag, on A. Hardware Architecture
the quadrotor to log the arrival times of UWB signals and The drone system architecture comprises several key
calculate the quadrotor’s position in space1 . components seamlessly integrated to ensure efficient and
The selected UWB system is based on the Pozyx system autonomous operation as shown in Fig.1. At its core is
[13], utilising the Decawave DWM1000 chip. the Jetson Nano, serving as the central processing unit re-
sponsible for high-level computations and decision-making.
I. Battery
Connected to the Jetson Nano is a flight controller, in this
The drone will be powered by a Lithium Polymer (LiPo) case, the Readytosky Pixhawk, which manages the drone’s
battery, connected to the power distribution board (PDB) to movement and motor control. Alternatively, the authors have
distribute power to the various components. The selected examined the use of the Z-turn FPGA board with a Zynq
battery is a 3S1P LiPo Battery (i.e. 3 cells connected in SoC as an alternative to the Jetson Nano, exploring different
series) with specifications of 11.1 volts, 4400 mAh, 30C possibilities for the system’s central processing unit.
discharge rate, and a weight of 340g. The choice of this Sensory input is facilitated by various sensors, including
battery was primarily determined by two key specifications: a stereo camera for image recognition, an Inertial Mea-
its total output voltage and weight. surement Unit (IMU) for orientation data, Ultra-WideBand
(UWB) sensors for localisation, and a LiDAR sensor for
J. DC-to-DC Converter
point cloud data. These sensors collectively provide essential
To supply power to the Jetson Nano through the drone’s information for navigation and obstacle avoidance.
power distribution board, we employed a buck converter2 . To power the system, a Lithium Polymer (LiPo) battery
The Jetson Nano is designed to receive a DC power supply of is utilised, connected to a power distribution board (PDB)
5V with a maximum current of 4 Amps, ensuring a sufficient that distributes power to all components. Additionally, a
power supply, especially during computationally demanding DC-to-DC converter ensures the Jetson Nano receives the
processes such as running an object detection algorithm, appropriate voltage from the drone’s power supply.
preventing a CPU shutdown. For communication, a WiFi module, specifically the
K. Motors AC1300 Archer T3U Mini Wireless MU-MIMO USB
Adapter, enables data exchange with a ground control station.
We selected four A2212 brushless DC motors (BLDC) The overall architecture is designed for modularity, allowing
rated at 1000kV (1000RPM/V)3 with 80% maximum ef- flexibility in adjusting configurations for different scenarios
ficiency. Each motor is powered by a 30A electric speed and experiments.
controller (ESC) and paired with 1045 (pitch 10 in, diameter
4.5 in) propellers (CW, CCW), producing 800g of thrust. B. Coordinate Systems
On a side note, brushless DC motors defy their name; To determine the location of the quadrotor in space and
they’re not DC motors but rather AC polyphase synchronous the relative locations of surrounding objects around it, we
motors. Through electronics, DC is converted to AC, generat- define two coordinate frames using the standard right-handed
ing a multiphase rotating magnetic field. In contrast, brushed robotics convention as shown in Fig.2. The Earth’s inertial
DC motors utilise brush contacts as electro-mechanical DC to frame {E} follows the East-North-Up (ENU) reference sys-
AC converters, while brushless DC motors employ electron- tem where +x axis points to the east, +y to the north and +z
ics to drive the coils with switched DC current, effectively points upwards based on the right-hand rule. The Body frame
producing AC. of the quadrotor {B} is coincident to the origin and thus to
the absolute position of the quadrotor, i.e. [x, y, z], while
L. RC Receiver
it follows the Forward-Left-Up (FLU) which gives forward
The Radiolink R8EF used in this UAV implementation horizontal, left horizontal and up vertical movement along
is an 8-channel receiver operating at a 2.4 GHz frequency, its +x, +y and +z axis, respectively.
connected to the quadcopter’s FCU. More precisely, the
SBUS/Channel 1 is linked to the Pixhawk RC input. This C. Quadrotor Aircraft Dynamical Model
1 The
The quadrotor is a naturally unstable non-linear complex
locations of the anchors can also be inferred if they are in a certain
formation and the initial position of the tag is known.
system composed of four rotors, as the name suggests. Each
2 A buck converter, also known as a step-down converter, is a DC-to-DC rotor is made up of a propeller and a motor that generates
converter that reduces voltage while simultaneously increasing current from an angular velocity !i , resulting in a thrust force fi , where
its input (supply) to its output (load). This converter falls within the category i refers to the number of the motor as depicted in Fig.2.
of switched-mode power supplies.
3 A 3-cell LiPo battery, with a nominal voltage of 11.1V, would theoreti- Two rotors rotate clockwise while the other two rotate coun-
cally cause a 1000kV motor to spin at 11,100 RPM without any load. terclockwise to prevent unwanted rotation of the quadrotor

400
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
 UAV
ECHO 4 X Ultrasonic
PWR
mC TRIG
Sensors (HC-
SR04)

I2C

USB 3.0
Stereo Camera
(ZED 2)
PWR DC-to-DC PWR
Converter
I2C
Battery Lidar-Lite V3
LiPo PWR Power Mission Computer
(3S1P) Distribution (NVIDIA Jetson Nano
Board or MYiR Z-turn) USB 3.0
WiFi Module
(Archer T3U)

USB 3.0 UWB Tag

(Pozyx)

UART

DSM/
SBUS
RC Receiver
PWR PWR DC-to-DC PWR FCU
Converter (Pixhawk PX4)
PWR
IMU Buzzer

PWM

CTRL
4X PWR Safety Switch
PWR
Brushless Electronic Speed Controller (ESC)
Motors PWM
(30Amp)
(1000KV)

Ground
Station
4 X UWB Anchors
PC running ROS2 RC Transmitter
(Pozyx)

Fig. 1. Quadrotor system architecture.

f4
body in the yaw ( ) direction (conservation of angular ω4
f3
momentum). The angular velocities of the rotors correspond z ω3
to specific rotational coordinates, ⌘ = ( , ✓, ) 2 R3 , and yaw (ψ)
move the quadrotor to different translational coordinates, roll (φ)
⇠ = (x, y, z) 2 R3 , in the Earth inertial frame E. The f1 x
ω1
orientation of the quadrotor is defined by the Euler angles , ΕΤ
{Β} Β
✓ and . (roll) is the angle around the x-axis, ✓ (pitch) is f2
ω2 z
the angle around the y-axis, and (yaw) is the angle around x
the z-axis. The translational coordinates x, y and z represent pitch (θ)
{Ε}
the centre of mass of the quadrotor relative to the Earth y
y

inertial frame. The translational and rotational equations of

motion for a quadrotor in Earth frame are described using
the Newton-Euler formalism [14], [15] and [16], respectively. Fig. 2. F450 quadrotor model with coordinate frame.
These equations take into account relatively small quadrotor
movement angles
fT I y I z ˙ ˙ ⌧x
ẍ = (c s✓c + s s ) , ¨ = ✓ + , to gravity. The terms ⌧✓ , ⌧ , and ⌧ are the pitch torque,
m Ix Ix roll torque, and yaw torque, respectively and they depend
fT I z I x ˙ ˙ + ⌧y ,
ÿ = (c s✓s c s ) , ✓¨ = on the angular velocities of the rotors ⌦i [17]. Furthermore,
m Iy Iy Ix , Iy , and Iz are the moments of inertia of the quadrotor’s
fT ¨= x I I y ˙˙ ⌧ z symmetric rigid body around its three axes.
z̈ = (c c✓) g, ✓+ ,
m Iz Iz A simplified linearised model can be obtained around
where m represents the mass of the quadrotor, fT represents the hovering equilibrium point by small-angle approxima-
the total thrust force, and g represents the acceleration due tion (i.e. assuming that the rotational angles of the system

401
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
are relatively small), which implies that ˙ , ✓, ˙ ˙ ' 0, and The uXRCE-DDS middleware is employed within the PX4
s ' , s✓ ' ✓, s ' , with c = 1, c✓ = 1, to enable the publication and subscription of uORB messages
and c = 1. The system state vector is defined as x = on the Jetson Nano, mimicking ROS2 topics. This facilitates
[x y z ✓ ẋ ẏ ż ˙ ✓˙ ˙ ]T . The hovering equilibrium a swift and dependable integration between PX4 and ROS2,
point, x̄ = [x̄ ȳ z̄ 0 0 0 0 0 0 0 0 0]T , is reached when simplifying the process for ROS2 applications to access UAV
the total thrust is a constant control input of fT = mg, information and issue commands. PX4 utilises an XRCE-
reflecting the force necessary to hover the quadrotor at an DDS implementation that takes advantage of eProsima Micro
arbitrary position (x̄, ȳ, z̄). The resulting linearised system XRCE-DDS [18]. The uXRCE-DDS middleware comprises
is described by a client operating on PX4 and an agent functioning on the
fT Jetson Nano. These components facilitate bidirectional data
ẍ = g✓, ÿ = g , z̈ = , exchange between them through a serial or UDP connec-
m
tion. Acting as a proxy for the client, the agent allows it
¨ = ⌧x , ✓¨ = ⌧y , ⌧
¨= z. (2)
Ix Iy Iz to both publish and subscribe to topics within the global
DDS data space. For PX4 uORB topics to be accessible
D. Camera-Based Localisation on the DDS network, it’s necessary to have the uXRCE-
Camera-based detection of landmarks is a well-established DDS client operating on PX4, connected with the micro
approach for robot localisation. The method involves util- XRCE-DDS agent running on the Jetson Nano. The PX4
ising a camera mounted on the quadrotor and a set of uxrce dds client manages the publication and subscription
landmarks whose positions are known with respect to the of designated uORB topics within the global DDS data
global reference frame. Analysing the objects detected by the space. On the Jetson Nano, the eProsima micro XRCE-DDS
camera facilitates the estimation of the quadrotor’s position. agent serves as a proxy for the client within the DDS/ROS2
Given that a monocular camera is used, depth information network.
about the landmarks can be obtained by analysing successive Figure 3 presents the communication architecture based on
frames. This process enables the estimation of the relative ROS2 between the UAV modules (onboard computer, FCU,
position between a landmark and the camera. Once the µC) and the ground station (computer and RC controller).
relative position of the camera is known, the relative position The IMU, UWB, LiDAR, stereo camera and ESC nodes send
of the quadrotor on which the camera is attached to, can be (i.e. publish) and receive (i.e. subscribe) data via a number
obtained and expressed in the global reference frame. of topics while the buzzer and safety switch nodes send and
receive data via services.
E. ROS2 Based Communication Architecture
The Robot Operating System (ROS), an open-source IV. RESULTS AND ANALYSIS
middleware, has found extensive application in robotics.
A. Weight Analysis
Despite its strengths, ROS isn’t well-suited for real-time
embedded systems due to its inability to meet real-time Given that the total takeoff weight (wT O ) of the electric
requirements and its limited compatibility with operating multirotor UAV is represented as [19]
systems. To overcome this challenge, ROS1 has undergone a
substantial overhaul, resulting in ROS2, which leverages the wT O = w0 + wb + wpl (3)
Data Distribution Service (DDS). DDS is better suited for
where w0 denotes the weight without the battery, consisting
real-time distributed embedded systems because of its diverse
of the frame, avionics, and propulsion system, wb indicates
transport configurations, such as deadline management and
the battery weight, and wp l represents the payload weight,
fault tolerance, as well as its scalability.
including the onboard computer, camera, LiDAR, ultrasonic
By using ROS2, a primary program can be subdivided into
sensors and UWB tag. The UAV’s total mass, excluding the
multiple subprocesses that operate concurrently and interact
battery, is a crucial factor in determining the battery size wb .
with each other. In ROS2, nodes represent processes, each
assigned with a distinct task aimed at enhancing the overall Table I lists the breakdown of weights for the different
system efficiency. Communication among these nodes occurs components utilised in the quadcopter design. The take off
through a series of topics and services. Topics act as bridges weight of the quadcopter amounts to 1̃.5 Kg with the Jetson
facilitating the transfer of information, known as messages, Nano or 1.4 Kg with the Z-turn FPGA board.
from one node to another. Nodes publish and/or subscribe
B. Power Estimation
to topics to respectively send and/or receive these messages.
Services operate on the request/response model; hence, one The power usage of a DC brushless motor may fluctuate
node offers a service that another node invokes using the based on several factors, including the motor’s dimensions,
service name. The node creating the service is the server, operational voltage, and load circumstances. In essence, the
while the nodes calling the service are the clients. Topics power consumption of a DC brushless motor correlates
are suitable for managing continuous data streams, while directly with the mechanical power it delivers. Put simply,
services are best utilised for blocking calls, making them the greater the load the motor must propel, the more power
ideal for remote procedure calls that terminate quickly. it will consume.

402
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
 TABLE I
Jetson Nano (Ubuntu 22.04LTS) Arduino Pro Mini
(baremetal code) W EIGHT BREAKDOWN OF THE VARIOUS UAV PLATFORM COMPONENTS .
ROS2 Humble
IMU UWB I2C Quantity Component Description Weight (g)
Ultrasonic
ROS2 control ROS2 control 1 F450 frame 270
sensors
node node 4 Proppeler 56
Stereo Camera Lidar 4 Brushless DC Motor (1000KV) 208
ROS2 control ROS2 control 4 Electronic speed controller 92
node node 1 11.1V 4400mAh 3SP1 LiPo battery 340
Ultrasonic Safety switch 1 Pixhawk P4 flight controller 16
ROS 2 control ROS2 control 1 NVIDIA Jetson Nano (MYiR Z-Turn) 142 (52)
node node 2 DC-to-DC converter 10
ESC Buzzer 1 RC receiver 6
ROS2 control ROS2 control 4 Ultrasonic sensors 34
node node
1 Arduino Pro Mini board 2
1 UWB Pozyx tag 42
1 ZED 2 stereo camera 160
μXRCE-DDS 2.4/5GHz 1 Hokuyo UURG005 138
agent Wi-Fi Total weight 1516 (1426)

UART

Here, Eb represents the battery energy, measured in watt-

Pixhawk PX4 (v1.14)
hours (Wh), while Phe is the electrical power necessary for
2.4GHz
μXRCE-DDS the UAV to hover. Additionally, Eb can be expressed in
client relation to the battery’s specific energy as
Ground station
optional
Eb = Espec mb nb fDOD (6)
IMU μORB ESC RC control
topic μORB topic μORB topic via
training UAV ROS
where Espec represents the battery’s specific energy, mea-
Buzzer Safety SW
port 2 control sured in Wh/kg, with typical values ranging between 50.7
μORB topic μORB topic node
and 220 Wh/kg for LiPo batteries [19]. In particular the
Espec can be defined as

Fig. 3. ROS2 based communication architecture. Cv

Espec = (7)
mb
C represents the nominal battery capacity provided by the
The power consumption (Watts) of a DC brushless motor manufacturer, v is the nominal voltage between the leads,
can be determined using the formula mb denotes the mass of the battery, nb indicates the battery
efficiency, accounting for heat losses, and fDOD represents
P ower = T orque ⇤ Speed/9.55 (4) the battery’s depth of discharge (DOD) [19].
To prolong battery life, it is recommended not to discharge
Here, torque (Nm) represents the force applied to the motor batteries beyond an 80% depth of discharge [22]. The power
shaft, while speed denotes the motor’s rotational speed in required for hover, Phe , is determined using momentum
revolutions per minute (RPM)4 . theory [23], calculated as follows (assuming an 80% motor
It’s important to recognise that the power consumption efficiency as specified in the datasheet)
of a DC brushless motor can also be influenced by the
3/2
motor’s efficiency, which may vary based on factors such WT o
Ph e =( p )/0.8 (8)
as the motor’s design and construction quality, the materials f 2⇢⇡Nr rprop
utilised, and the operating conditions.
LiPo batteries are popular choice for small UAVs given its where, Nr represents the number of rotors with a radius of
high energy density and high current discharge capabilities rprop , ⇢ is the air density, and f denotes the figure of merit
[20]. As the battery constitutes a significant portion of the (propeller efficiency), which typically lies between between
UAV’s weight, choosing the appropriate one significantly 0.5 and 0.7 [23]. In this work, a value of f = 0.6 is used.
affects flight duration. For the quadcopter platform, the aim From Eq.(5), (6) and (8), WT O from Table I, ⇢ =
is a flight duration of 10 to 15 minutes, with the battery 1.225kg/m3 , rprop = 0.127m and nb = 95% we get a
discharged to 80% depth. Estimating the UAV’s endurance, preliminary estimated endurance of near 8 min. For a more
[21], involves calculating the battery energy and the power accurate estimation of the UAV’s endurance, the Peukert’s
needed for hovering model, as described in [22][24], could be employed, although
Eb this falls beyond the scope of the current study.
E= (5) Table II illustrates the power consumption of the essen-
Ph e
tial parts comprising the quadcopter. The Pozyx tag was
4 Speed in RPM is divided by 60/(2⇡) or 9.55 to convert it to radians intentionally not included in the table, as it is powered by
per second (rad/s). a replaceable CR2450 (3 V, 620 mAh) battery, offering an

403
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.
 TABLE II
R EFERENCES
E XPECTED MAXIMUM POWER DRAW OF THE UAV COMPONENTS .
[1] S. A. H. Mohsan, N. Q. H. Othman, Y. Li, M. H. Alsharif, and
Component Voltage (V) Current (A) Power (W) M. A. Khan, “Unmanned aerial vehicles (uavs): Practical aspects,
BLDC (1000KV) 11.1 13 144.3 applications, open challenges, security issues, and future trends,”
Pixhawk P4 (USB) inc. Intelligent Service Robotics, vol. 16, no. 1, pp. 109–137, 2023.
safety switch and buzzer 5 0.18 0.9 [2] V. Pritzl, M. Vrba, P. Štĕpán, and M. Saska, “Cooperative navigation
Jetson Nano 5 1-3 5-15 and guidance of a micro-scale aerial vehicle by an accompanying uav
MYiR Z-Turn FPGA 5 1.6 8 using 3d lidar relative localization,” in 2022 International Conference
RC receiver 5 0.03 1.5 on Unmanned Aircraft Systems (ICUAS), 2022, pp. 526–535.
Ultrasonic sensor 5 0.006 0.03 [3] J. Sandino, F. Vanegas, F. Maire, P. Caccetta, C. Sanderson, and
Arduino Pro Mini 5 0.016 0.08 F. Gonzalez, “Uav framework for autonomous onboard navigation
and people/object detection in cluttered indoor environments,” Remote
LiDAR 5 0.5 2.5
Sensing, vol. 12, no. 20, p. 3386, 2020.
ZED 2 stereo camera 5 0.38 1.9
[4] L. Hadjiloizou, K. M. Deliparaschos, E. Makridis, and T. Charalam-
bous, “Onboard Real-Time Multi-Sensor Pose Estimation for Indoor
Quadrotor Navigation with Intermittent Communication,” in 2022
autonomy of nearly 5 years according to the manufacturer’s IEEE Globecom Workshops (GC Wkshps), Dec. 2022, pp. 154–159.
[5] L. Hadjiloizou, E. Makridis, T. Charalambous, and K. M. Deli-
specifications. It’s worth noting that powering the Pixhawk paraschos, “Maximum Correntropy Criterion Kalman Filter for Indoor
PX4 FCU via USB significantly reduces power consumption Quadrotor Navigation under Intermittent Measurements,” in 2023 31st
compared to powering it from the battery through the ESC, Mediterranean Conference on Control and Automation (MED), June
2023, pp. 170–175.
primarily due to the lower efficiency regulator in the ESC. [6] NVIDIA, “Jetson Nano Developer Kit,” 2024. [Online]. Available:
The complete multirotor UAV setup is shown in Fig.4. https://developer.nvidia.com/embedded/jetson-nano-developer-kit
[7] MYIR, “Z-turn Board,” 2024. [Online]. Available: https://www.
myirtech.com/list.asp?id=502
[8] C. Kallis, K. M. Deliparaschos, G. P. Moustris, A. Georgiou, and
T. Charalambous, “Incremental 2D Delaunay triangulation core im-
plementation on FPGA for surface reconstruction via high-level syn-
thesis,” in 2017 22nd IEEE International Conference on Emerging
Technologies and Factory Automation (ETFA), Sept. 2017, pp. 1–4.
[9] Stereolabs, “ZED 2 - AI Stereo Camera,” 2024. [Online]. Available:
https://www.stereolabs.com/products/zed-2
[10] L. Zheng, P. Zhang, J. Tan, and F. Li, “The obstacle detection method
of uav based on 2d lidar,” IEEE Access, vol. PP, pp. 1–1, 11 2019.
[11] L. Kneip, F. Tache, G. Caprari, and R. Siegwart, “Characterization of
the compact Hokuyo URG-04LX 2D laser range scanner,” in 2009
IEEE International Conference on Robotics and Automation, May
2009, pp. 1447–1454.
[12] J. Krejsa and S. Vechet, “The evaluation of hokuyo urg-04lx-ug01
Fig. 4. Mutlirotor UAV implementation.
laser range finder data,” in Eng. Mech. 2017, Svratka, CZ, May 2017.
[13] Pozyx Labs, “Pozyx Labs - Accurate Positioning,” 2024. [Online].
Available: https://www.pozyx.io
V. CONCLUSIONS AND FUTURE DIRECTIONS [14] F. Kendoul, D. Lara, I. Fantoni, and R. Lozano, “Nonlinear Control for
Systems with Bounded Inputs: Real-Time Embedded Control Applied
We present a specialised hardware platform addressing to UAVs,” in IEEE Conf. on Dec. and Con., 2006, pp. 5888–5893.
the challenges of indoor navigation research with unmanned [15] H. Voos, “Nonlinear Control of a Quadrotor micro-UAV using
Feedback-Linearization,” in IEEE International Conference on Mecha-
aerial vehicles (UAVs). While UAVs have become increas- tronics, 2009, pp. 1–6.
ingly prevalent in research, particularly in aerial navigation, [16] F. Sabatino, “Quadrotor Control: Modeling, Nonlinear Control Design,
many proposed methodologies depend heavily on simula- and Simulation,” Master’s thesis, KTH Royal Inst. of Tech., 2015.
[17] S. Bouabdallah, “Design and Control of Quadrotors with Application
tions, which may not accurately reflect real-world conditions. to Autonomous Flying,” Ecole Polytechnique Fédérale de Lausanne
Our work introduces a low-cost modular UAV hardware plat- (EPFL), Tech. Rep., 2007.
form specifically designed to enable research and develop- [18] eProsima, “eProsima Micro XRCE-DDS Agent,” 2024. [Online].
Available: https://micro-xrce-dds.docs.eprosima.com/en/stable/agent.
ment in aerial indoor navigation. By prioritising modularity, html
the platform offers researchers the flexibility to customise [19] M. Gatti, “Complete Preliminary Design Methodology for Electric
and scale their experiments to suit the unique requirements of Multirotor,” Journal of Aerospace Engineering, vol. 30, May 2017.
[20] Y. N. Saravanakumar, M. T. H. Sultan, F. S. Shahar, W. Giernacki,
indoor environments while meeting reasonable payload con- A. Lukaszewicz, M. Nowakowski, A. Holovatyy, and S. Stepien,
straints. This approach not only speeds up the development “Power Sources for Unmanned Aerial Vehicles: A State-of-the Art,”
process but also enables the evaluation of UAV navigation Applied Sciences, vol. 13, no. 21, p. 11932, Jan. 2023.
[21] A. Kirenga, H.-I. Lee, and A. Zolotas, “Unmanned aerial system
precision and reliability in real-world indoor scenarios. concept design for rail yard monitoring,” in AIAA SCITECH 2023
The proposed UAV platform facilitates research and devel- Forum, 2023, p. 1733.
opment, with the goal of conducting real-world experiments [22] M. Biczyski, R. Sehab, J. Whidborne, G. Krebs, and P. Luk, “Mul-
tirotor Sizing Methodology with Flight Time Estimation,” Journal of
in complex indoor environments. The authors intend to utilise Advanced Transportation, vol. 2020, pp. 1–14, Jan. 2020.
the UAV platform for performing real scenario experiments, [23] L. Bauersfeld and D. Scaramuzza, “Range, Endurance, and Optimal
building upon their prior research [4][5] that investigated the Speed Estimates for Multicopters,” IEEE Robotics and Automation
Letters, vol. 7, no. 2, pp. 2953–2960, Apr. 2022.
stability conditions of the classical Kalman filter (KF) and [24] M.-h. Hwang, H.-R. Cha, and S. Y. Jung, “Practical endurance esti-
Maximum Correntropy Criterion Kalman filter (MCC-KF) mation for minimizing energy consumption of multirotor unmanned
algorithms in scenarios with packet drops. aerial vehicles,” Energies, vol. 11, no. 9, p. 2221, 2018.

404
Authorized licensed use limited to: ETH BIBLIOTHEK ZURICH. Downloaded on December 22,2024 at 05:22:26 UTC from IEEE Xplore. Restrictions apply.