Porfolio-Đã G P

BỘ GIÁO DỤC VÀ ĐÀO TẠO
TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT

THÀNH PHỐ HỒ CHÍ MINH
KHOA: CƠ KHÍ CHẾ TẠO MÁY
----------

PORFOLIO
KHOA: CHÍNH TRỊ VÀ LUẬT

TRƯỜNG ĐẠI HỌC
Instructor: D. SƯ PHẠM
Trần KỸĐảm
Ngọc THUẬT
Implementer
No
BÁO CÁO PROJECT CUỐI KỲStudent ID
Full name
1 Phạm Hoàng Phúc 21146295

TIỂU LUẬN
TƯ TƯỞNG HỒ CHÍ MINH
Thủ Đức, tháng 11 năm 2023

1.
2.
----------

PORFOLIO

TRƯỜNG ĐẠI HỌC
Instructor: D. SƯ PHẠM
Trần KỸĐảm
Ngọc THUẬT
Implementer
No
BÁO CÁO PROJECT CUỐI KỲStudent ID
Full name
1 Phạm Hoàng Phúc 21146295

TIỂU LUẬN
TƯ TƯỞNG HỒ CHÍ MINH
Thủ Đức, tháng 11 năm 2023

TRƯỜNG ĐẠI HỌC1SƯ PHẠM KỸ THUẬT
TABLE OF CONTENT
I. Free Paragraph..............................................................................................................3
II. Interview .....................................................................................................................4
III. Title and abstract .......................................................................................................5
IV. Paragraph ................................................................................................................ 12
V. Content Presentation ................................................................................................ 16
VI. Research Paper ....................................................................................................... 18
2
I. Free paragraph
3
II. Interview
I appreciate KUKA firm giving me a chance to interview on such a wonderful
day like today. My name is Pham Hoang Phuc, I’m 20 years old. Mechatronics is my
area of study at Ho Chi Minh City University of Technology and Education. I have
gained a lot of soft skills and in-depth knowledge about the topic. I typically have
knowledge of manipulators and the capacity to comprehend machine components. I also
particularly like the things your firm produces because it is a company that specializes
in manipulators. I'm so passionate about your business because of this. My weakness
appears to be my lack of work experience. My strengths include self-assurance,
determination, and a high level of learning. My kindness, lack of professional experience,
and rather slender frame are also my drawbacks. My idol is Thomas Edison. Then, I'd
want to discuss my family. My family consists of 4 persons, including my parents and
brother. In addition, I really love basketball. My life's objective is to make at least as
much money as I lost on school and money I borrowed from my family. Overall, hiring
me to fill a vacancy in your organization won't be a mistake because I won't let you
down with my working abilities.
4
III. Title and abstract (4 + 1)
3.1 Application of the sales smart service of stationery of the company PHP
Stationery is one of the places that attracts a large number of students, young
people and even the elderly to shop for necessary items in life. Typically, products of
PHP stationery service will include erasers, student notebooks, notepads, blue ink pens,
red ink pens,...Due to the Covid-19 pandemic, which is ravaging the Thu Duc City
region in particular and Ho Chi Minh City in general, there is a growing need for school
supplies. However, since people are unable to leave their homes, online purchasing
services have been more and more popular. Thus, the PHP online stationery business
was established with the goal of meeting the demands of students in the Thu Duc City
area who are shopping for school supplies, particularly Ho Chi Minh University of
Technology and Education. Through the creation of a website that promotes the items
offered by PHP firm on all university sites on Facebook, it has drawn in a significant
number of buyers from students’s universities throughout Thu Duc city. The PHP online
shopping service has, however, run across certain operational issues, such as the
website's propensity to fail under heavy traffic loads. As a result, PHP has gathered a lot
of feedback from users who have made purchases in order to enhance the website that
PHP runs. Conclusion, adult and student alike have benefited greatly from the
availability of many essential school supplies thanks in large part to the PHP’ s online
stationery services.
5
3.2 The Development and Production of Washing Automatic Hand Robot using
Arduino Board Controller
6
7
3.3 The Development of The Smart Controlling Cars
8
3.4 The Development and Production of Cleaning Robot PHP
9
3.5 Fault Diagnosis of AC Servo Motor with Current Signals Based on Wavelet
Decomposition and Template Matching Methods.
This paper presents a diagnosis technique to detect and identify faults in AC servo
motors. The first phase of stator currents among three phases is digitized and stored in
the time domain. Wavelet transform is employed to convert the signals onto time
frequency domain because the time domain based approach is not suitable for detecting
state features of the current signals. Pre-processing algorithms that includes a kind of
mean-filtering, synchronization with Hilbert transform and difference are consecutively
performed to the raw signal to determinate features. Wavelet decomposition is applied
to the difference values by the optimally selected mother wavelet and the features are
calculated from the transformed signals. The extracted features are compared with the
motor fault templates for the template matching method. The results based on real data
show that the proposed approach is very useful to extract features of the signals for fault
diagnosis.
Comment:
 Purpose: A diagnosis technique to detect and identify faults in AC servo motors.
 Method:
 The first phase of stator currents among three phases is digitized and
stored in the time domain.
10
 Wavelet transform is employed to convert the signals onto time frequency
domain because the time domain based approach is not suitable for
detecting state features of the current signals.
 Pre-processing algorithms that includes a kind of mean-filtering,
synchronization with Hilbert transform and difference are consecutively
performed to the raw signal to determinate features.
 Wavelet decomposition is applied to the difference values by the
optimally selected mother wavelet and the features are calculated from the
transformed signals.
 The extracted features are compared with the motor fault templates for the
template matching method.
 Result: Based on real data show that the proposed approach is very useful to
extract features of the signals for fault diagnosis.
 Conclusion: Thank to the methods in order to analyze and identify faults in AC
servo motors.
11
IV. Paragraph (4 + 1)
4.1 The Benefits of Robotics in Our Life
Robotics' invention is the key to a new era of cutting-edge science and technology
that meets the demand for automation by removing workers from hazardous professions
and giving them control, rationality and mental effort greater. Thus, it is impossible to
dispute the advantages of robots in daily life. The first, Robotics ensure safety for
workers. In particular, everyone working on the Bosch company’ manufacturing lines
must handle and be in close proximity to large gear, as well as areas with high
temperatures and plenty of sharp things that could cut them experiencing significant
workplace accidents; therefore, by delegating risky tasks to robots, needless risks can
be removed, ensuring worker safety. Next, Robotics handle all jobs quickly, efficiently
and accurately. For instance, the manipulators robots in Bosch's manufacturing line are
precisely preprogrammed, which helps the robot accelerate, concentrate on its task, and
consistently achieve high production process export efficiency. Finally, Robots raise the
standard of products. In particular, the Kuka Robot's exact programming makes it easier
for the manufacturing and installation cycle to run smoothly, which lowers the number
of defective items and produces today's flawless products extra. In conclusion, Robots
today bring many benefits to people to improve their lives, as well as promote socio-
economic development.
4.2 The Application of Recent Technology 4.0
The manufacturing industry's intelligentization of production and management
procedures is known as “Technology 4.0”. Every industry, every nation, and every
production, management and governance system in the globe are changing as a result of
this revolution. The first, the application of Technology 4.0 in Education. For instance,
in 2020, as a result of the Covid-19 pandemic, schools have shifted to using online
teaching techniques, facilitated by sophisticated technology 4.0 that allows pupils to
readily access the internet and expand their knowledge on it. Next, the application of
Technology 4.0 in Medicine. During the recent Covid -19 epidemic, 4.0 technology has
supported productivity in medical declaration, helping people not have to go to the
hospital but still be diagnosed by a team of specialists 24/7. Finally, the application of
12
Technology 4.0 in Agriculture. The application of cloud computing aims to provide high
quality output products and food safety and hygiene for consumers. In conclusion, with
a supply chain built on an intelligent model, technology 4.0 enables businesses to be
more adaptable in the production process that results in the development and
improvement of people's lives as well as the manufacture of high-quality goods at
competitive rates, saving time and money.
4.3 The Benefits of Learning English
Although mastering English communication offers three incredible advantages
that few people are aware of, English has long been a widely used language for
communication worldwide. People from across the world are able to connect together,
better understand one another, and grow together steadily and powerfully in all spheres
of society as a result of learning English. First of all, studying English increases our
global travel and cultural awareness. Typically, I have used to visiting a lot of locations
in Europe. Because I could converse with the locals in English, I was able to get a lot of
knowledge and insight into their cultures from the sites I visited. Next, Learning English
also brings benefits of increasing connection - increasing relationships. For instance,
when Hung traveled to Manchester, England, he helped Rachelle Juliette with her year-
end school project, which led to Hung becoming friends with her. His fluency in English
allows him to listen and comprehend with ease, which helps him assist her with her
schoolwork. Finally, acquiring knowledge of English will also boost your self-
assurance. Because we already speak English, we can readily make contact and interact
with foreigners when we are out on the street and no longer feel awkward when we
encounter them. In conclusion, learning English make us many benefits in our life, we
feel comfortable conversing and interacting with strangers, listening to them speak, and
even reading documents written in different languages.
4.4 What Exactly Is Artificial Intelligence?
Within computer science, artificial intelligence is a broad discipline. It focuses
on creating intelligent machines that can function without requiring human input. These
days, we can imagine a machine that can function in accordance with our requirements
without requiring instructions. Artificial Intelligence has applications in surveillance,
13
cell phones, video games, and automobiles. Artificial intelligence is used by numerous
well-known social media sites, like Facebook, Google, and many more. AI is used in
Google and other websites' recommendations. Artificial intelligence is demonstrated via
virtual aid programs such as Google aid and Siri on iPhones. The artificial intelligence
products Alexa and Amazon Echo, which are made by Amazon, are another well-known
example. We have encountered autonomous vehicles. They do employ artificial
intelligence as a guiding principle.
4.5 The Proportion of the Population Aged 65 and over between 1940 and 2040 in
USA, Sweden and Japan
The United States, Sweden, and Japan are the three countries represented in the
line graph, which shows the percentage of senior people (those over 65) over a century
starting in 1940 and extending to 2040. Despite having a lower percentage of the elderly
population in their early 40s, Japan will have a higher percentage of older people in the
future—one-fourth of the world's population will be older people in 2040, according to
the graph.
According to the figure, in 1940, the average proportion of elderly persons in each
of the three countries was between 5 and 10 percent. Japan had the lowest percentage of
elderly people, while the United States had the greatest. The United States' aging
population increased gradually until it hit 15% in 2000. By 2040, it's predicted to have
increased to about 25%. In a similar vein, the proportion of elderly persons in Sweden
14
increased from roughly 7% in 1940 to 15% in 2000 and is expected to reach 25% by
2040.
Japan, on the other hand, has the oldest population, which was less than 5% until
2005 and is predicted to rise to the maximum percentage of 30% by 2040. The graph
clearly reveals that the ratio of aged population will increase in 2040 and almost one of
every four citizens in these three countries would be over 65 years old in 2040.
4.6 The Indispensable Role of Technology in Modern Life and National Development
Scientific innovations or technology is the most amazing thing that has ever
happened to humankind. In this modern world, technology is no more a luxury; it has
become a necessity. Life without technology is next to impossible since humans have
become dependent on technological advancements. For every human activity, we need
a machine, and then there comes technology. It has made all our lives easier and more
convenient and has also made transportation, education, labour, etc., more accessible.
Technological breakthroughs change from time to time in today’s fast-paced, ever-
changing world. However, in today’s world, technology plays a critical role in
strengthening a country’s economy and people’s lives.
Comment:
 Purpose:
 The passage emphasizes the significance of scientific innovations and
technology for humanity.
 It focuses on explaining why technology has become an indispensable part
of modern life.
 Content:
 Describes the human dependence on technology and machinery in daily
activities.
 Highlights the benefits of technology, making life easier and more
convenient, and improving accessibility in areas such as transportation,
education, and labor.
15
 Method:
 Uses clear and understandable language to convey the author's opinions
and perspectives.
 Provides specific examples, such as transportation, education, and labor,
to illustrate how technology has changed our approach to these areas.
 Conclusion:
 Concludes by emphasizing the critical role of technology in strengthening
a country's economy and enhancing the lives of its citizens.
V. Content Presentation
Biofeedback for human-robot interaction in the context of

General purpose
collaborative assembly
1. The aim of this paper is to systematically investigate potential

applications of biofeedback for human-robot interaction (HRI) in
the context of collaborative assembly.
2. Within this paper, bodily functions are not to be reported back
Detail purpose
to the human, but made accessible to a machine.
3. For the derivation of the models, different approaches for
potential interaction systems for collaborative assembly are
collected.
1. Emphasis is placed on the interaction of different modalities of

biofeedback. Considering the determined parameters and
influencing factors, new biosignal-based interaction models are
to be derived that compensate for weaknesses of existing systems
Content
and enable safe and error-free collaboration between humans and
machines.
2. The biological data shall help to improve the interaction
between human and robot.
16
3. 13 interaction models for the task types of information
exchange, transport, precision and workload reduction could be
derived.
1. The integration of biofeedback

2. A distinction is made between electroencephalogrphy (EEG)
biofeedback, electrocardiography (ECG) biofeedback,
electromyography (EMG) biofeedback, electrodermal activity
(EDA) biofeedback, skin temperature biofeedback,
plethysmography biofeedback, respiration feedback and
Method
biofeedback of internal organs.
3.
Information exchange
Transport
Precision
Workload reduction
1.
In this way, the robot should have access to information from its
human counterpart.
It enables the robot to interpret the state of the corresponding
employee, which offers enormous potential in a wide variety of
Result + applications.
Discussion + From an economic point of view, the optimized exchange of
Comment information increases the efficiency of HRC, thus making the use
of robots useful in smaller series sizes as well as in unstructured
production environments with changing products and processes.
2.
Plethysmography biofeedback and biofeedback of internal organs
are excluded.
17
In addition to the common biofeedback measurements, the
recording of the eyes will be examined in more detail.
In this context, three wellknown methods for recording eye
movements, eye positions, eyelid closures, and pupillary
responses will be presented: the electro – oculogram (EOG), the
mechano – oculogram (MOG) and infrared camera systems.
3.
If the model under consideration can completely fulfill the sub –
objective, the highest rating of 3 is given.
To avoid subjective influences, all main objectives are weighted
equally.
Therefore, each of the four main objectives receives an influence
of 25%.
1. Biofeedback promotes a more autonomous, flexible and

adaptive behavior of the robot.
2. The development of the models is based on the thesis of
psychophysiology that conclusions can be drawn about the inner
– psychological state of a person by means of physiological
Conclusion measurands. The disciplines of physiology and psychology are
inferentially linked.
3. Accordingly, attention is paid to use no more than three
beneficial sensors. The suitability of the resulting possible sensor
combinations is to be quantified by a utility analysis.
VI. Research Paper
18
Proceedings of the 17th World Congress
The International Federation of Automatic Control
Seoul, Korea, July 6-11, 2008
Fault Diagnosis of AC Servo Motor with Current Signals Based on

Wavelet Decomposition and Template Matching Methods
Yountae Kim, Hyeon Bae, Sungshin Kim* and George Vachtsevanos**
*
School of Electrical and Computer Engineering, Pusan National University
Busan, Korea (Tel: 82-51-510-2374; e-mail:{dream0561, baehyeon, sskim}@pusan.ac.kr)
**
School of Electrical and Computer Engineering, Intelligent Control Systems Laboratory
Georgia Institute of Technology, Atlanta, GA 30332, USA (e-mail: gjv@ece.gatech.edu)
Abstract: This paper presents a diagnosis technique to detect and identify faults in AC servo motors. The
first phase of stator currents among three phases is digitized and stored in the time domain. Wavelet
transform is employed to convert the signals onto time-frequency domain because the time domain based
approach is not suitable for detecting state features of the current signals. Pre-processing algorithms that
includes a kind of mean-filtering, synchronization with Hilbert transform and difference are consecutively
performed to the raw signal to determinate features. Wavelet decomposition is applied to the difference
values by the optimally selected mother wavelet and the features are calculated from the transformed
signals. The extracted features are compared with the motor fault templates for the template matching
method. The results based on real data show that the proposed approach is very useful to extract features of
the signals for fault diagnosis.
1. INTRODUCTION The diagnostic methods to identify the above faults may

involve several different types of fields of science and
The most popular way of converting electrical energy to technology [1-3]. Several methods are applied to detect the
mechanical energy is an induction motor that plays an faults in induction motors as the following:
important role in modern industrial plants. The risk of motor
failure can be remarkably reduced if normal service • Electromagnetic field monitoring
conditions can be arranged in advance. In other words, one • Temperature measurements
may avoid costly expensive downtime of plant by proper
time scheduling of motor replacement or repair if warning of • Radio frequency (RF) emissions monitoring
impeding failure can be obtained in advance. In recent years, • Noise and vibration monitoring
fault diagnosis has become a challenging topic for many
electric machine researchers. The major faults of electrical • Acoustic noise measurements
machines can be broadly classified as follows [1]: • Motor current signature analysis (MCSA)
• Broken rotor bar or cracked rotor end-rings In this research, we use several data mining techniques based
on time-series for fault diagnosis. First, we introduce
• Static and/or dynamic air-gap irregularities synchronizing method that is very important issue in
• Bent shaft (akin to dynamic eccentricity) digitalized time-series measurement equipments. Wavelet
transform is a method for time varying or non-stationary
• Shorted rotor field winding signal analysis, and uses a new description of spectral
• Bearing and gearbox failure decomposition via the scaling concept. Wavelet theory
provides a unified framework for a number of techniques,
Faults in electric machines produce one of more of the
which have been developed for various signals processing
following symptoms:
applications. One of its feature is multi-resolution signal
• Unbalanced air-gap voltages and line currents analysis with a vigorous function of both time and frequency
localization. This method is effective for stationary signal
• Increase losses and reduction in efficiency processing as well as non-stationary signal processing.
• Excessive heating Mallat’s pyramidal algorithm based on convolutions with
quadratic mirror filters is a fast method similar to FFT for
We will treat four major faults those are cracked bearing, signal decomposition of the original signal in an orthonormal
broken rotor, bent rotor shaft, bearing wear or misalignment. wavelet basis or as a decomposition of the signal in a set of
These four faults are mechanically related faults. These faults independent frequency bands. The independence is due to the
are more frequently happened than others. orthogonality of the wavelet function [4]. But, the
978-3-902661-00-5/08/$20.00 © 2008 IFAC 7239 10.3182/20080706-5-KR-1001.2785

17th IFAC World Congress (IFAC'08)
performance is changed according to selection of mother eccentricity when the position of the minimal radial air-gap
wavelet. Unfortunately, selection of suitable mother wavelet length is fixed in the space. This caused by the ovality of the
is very difficult. To overcome this problem, we construct a stator core or by the incorrect positioning of the rotor or
non-supervising method. It supports the selection of the most stator at the commissioning stage. In case of dynamic
suitable mother wavelet for fault diagnosis. eccentricity, the center of rotor is not at the center of rotation,
so the position o minimum air-gap rotates with the rotor. This
2. FAULTS AND APPROACH METHODS FOR maybe caused In case of dynamic eccentricity, the center of
DIAGNOSIS rotor is not at the center of rotation, so the position o
minimum air-gap rotates with the rotor. This caused by a bent
2.1 Bearing faults rotor shaft by a bent rotor shaft, bearing wear or
misalignment, mechanical resonance at critical speed, etc. In
practice an air-gap eccentricity of up to 10% is permissible.
Though almost 4~50% of all motor failures are bearing
Both static and dynamic eccentricities tend to exist in
related, very little has been reported in the literature
practice. Using MCSA the equation describing the frequency
regarding bearing related fault detection techniques. Bearing
components of interest is:
faults might manifest themselves as rotor asymmetry faults
from the category of eccentricity related faults [5]. The ⎡ (1 − s) ⎤
vibration frequency of the fault is as follows, f ⎢(kR ± nd ) ± v⎥ (4)
⎣ p ⎦
f1[ Hz ] = ( f r / N ) f r [1 − bd cos( β ) / d p ] (1)
where nd=0 in case of static eccentricity, and nd=1, 2, 3, … in
where fr is the rotational frequency, N is the number of balls, case if dynamic eccentricity f is fundamental supply
bd and dp are the ball diameter and ball pitch diameter frequency, R is the number of rotor slots, s is slip, p is the
respectively, and β is the contact angle of the ball. The number of pole pairs, k is any integer and υ is the order of the
following equation includes the vibration frequency and stator time harmonics.
current spectrum [6].
Other equations are also presented in the literature as low
f bng =| f e ± m. f v | (2) frequency components for mixed eccentricity [7]. As it is
obvious, sometimes, different faults produce nearly the same
where m=1, 2, 3, … for the vibration harmonic contributions, frequency components or behave like healthy machine, which
fe is electrical power supply frequency and fv is one of the make the diagnosis impossible. This is the reason why new
bearing characteristic vibration frequency. techniques must also be considered to reach a unique policy
for distinguishing among faults. Park’s vector based upon
Artificial intelligence or neural networks have been voltage and current has been proposed to detect the motor
researched to detect bearing related faults on line. And also fault.
adaptive, statistical time-frequency methods are studying to
find bearing faults. 3. SYNCHRONIZATION WITH HILBERT TRANSFORM
2.2 Rotor bar faults 3.1 Hilbert transform
Rotor failures now account for 5-10% of total induction The Hilbert transform of the signal x(t) is defined to be the
motor failures. Broken rotor bars give rise to a sequence of signal whose frequency components are all phase shifted by -
side bands given by: π/2 radians. The resulting signal is denoted
f b = (1 ± 2ks) f , k = 1, 2, 3, K (3) xˆ (t ) = H {x(t )} (5)
where f is the supply frequency and s is the slip. Frequency xˆ (t ) is produced by passing x(t) through a filter with transfer
domain analysis and parameter estimation techniques have function[9].
been widely used to detect this type of faults.
H ( f ) = − j sgn( f ) (6)
In practice, the current side bands around fundamental may
exist even when the machine is healthy [7]. Also rotor The magnitude and phase of H(f) are
asymmetry, resulting from rotor ellipticity, misalignment of
the shaft with the cage, magnetic anisotropy, etc. shows up at H( f ) =1 (7)
the same frequency components as the broken bars [8].
Therefore other features of this fault need to be investigated. π (8)
∠H ( f ) = − sgn( f )
2
2.3 Eccentricity faults
This phase shift is very useful to synchronize each measured
signals.
Eccentricity faults are the condition of unequal air-gap
between the stator and rotor. It is called static air-gap
7240
xˆ (t )
2 2
x(t ) H ( x(t )) : phase shift term +∞ Ψ ( w) 0 Ψ ( w)
∫0 w
dw = ∫
−∞ w
dw = Cψ < +∞ (11)
x(t ) : source
When Wf (b, a j ) is known only for a < a0 to recover f , we
Fig. 1 Hilbert transform.
need a complement of information corresponding to
Wf (b, a j ) for a < a0 . This is obtained by introducing a
3.2 Synchronization
scaling function φ that is an aggregation of wavelet at scales
We use Hilbert transform to align phase of each measured larger than 1. ψˆ ( w) and ϕˆ ( w) are Fourier transforms of ψ(t)
signal. We can get the phase of the original signal by using and ϕ(t) respectively. ψ(t) is a band pass filter, and ϕ(t) is a
arctangent from a created phase shift term in result of Hilbert low-pass filter. Taking positive frequency into account
transform and the original signal. We align each signal to ϕˆ ( w) has information in [0, π] and ψˆ ( w) in [π, 2π]. Therefore
zero degree, as shown in Fig. 2. they both have complete signal information without any
redundancy. Decomposition of the signal in [0, π] using
Find -90° phase point using Hilbert transform
Hilbert Transform Phase Plane of Two Current Signal
Mallat’s wavelet algorithm gives:
1 Real Signal 1
Transformed Signal
0.8 0.8
0.6 h( n) = 2 − j ϕ ( 2 −1 t )ϕ (t − n)
(12)
0.6
0.4 0.4 , j = 0,1, 2, K

g ( n) = 2 − jψ ( 2 −1 t )ϕ (t − n)
Imaginary Number
Current [m A]
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4 Wavelet decomposition does not involve the signal in [π, 2π].
-0.6 -0.6
-0.8
In order to decompose the signal in whole frequency band,
-0.8
-1 -1 wavelet packet can be used for this purpose. After

0 1 2 3
Times [s]
4 5 6 -1 -0.5 0
Real Number
0.5 1
decomposition for l times, we will get 2l frequency bands
each with the same bandwidth. That is:
Shifting
Synchronized Signal ⎡ (i − 1) f n (i ) f n ⎤ (13)
10
⎢ 2 , , i = 1, 2, K, 2 j
8
Same Phase Point ⎣ 2 ⎥⎦
6
where fn is the Nyquist Frequency, in the ith frequency band.

4
2
C urrent [m A ]
0
Wavelet packet de-composes the signal into one low-pass
-2 filter h(n) and 2l-1 band-pass filters g(n), provides diagnosis
-4 information in 2l frequency bands.
-6
-8
Functions h(n) and g(n) can be obtained by inner product of
-10
0 20 40 60
Sample [n]
80 100 120 ψ(t) and ϕ(t).
Fig. 2 Synchronization procedure of the raw data. h( n) = 2 − j ϕ ( 2 −1 t )ϕ (t − n)

, t ∈ R, n ∈ Z (14)
g ( n) = 2 − jψ ( 2 −1 t )ϕ (t − n)
4. WAVELET DECOMPOSITION
Wavelet transform was applied to extract fault features from A j ( n) = ∑k h( k − 2n) A j −1
, n = 1, 2, 3, K (15)
the difference signals. A wavelet is a function ψ belonging to D j ( n) = ∑ g g ( k − 2n) A j −1
L2(R) with a zero average. It is normalized and centered in
the in the neighborhood of t=0. A family of time-frequency where A0(k) is the original signal and Aj is the low frequency
atoms is obtained by scaling ψ by aj and translating is by b approximation at the resolution j. Dj is called high frequency
[10-13]: detail signal. After de-composition of j time, we can obtain
one approximation signal Aj and D1, D2, …, Dj detail signals.
−j
⎛t −b⎞ (9)
ψ a ,b = a 2 ψ ⎜ j ⎟
⎝ a ⎠ Wavelet packet decomposition is:
These atoms also remain normalized. The wavelet transform x2 n (t ) − 2 ∑ h(k ) xn (2t − k )
of f belonging to L2(R) at the time b and scale aj is: k (16)
x2 n+1 (t ) − 2 ∑ g (k ) xn (2t − k )
+∞ 1 t −b
Wf (b, a j ) = f , ϕb ,a j = ∫ f (t ) ϕ*( )dt (10) k
−∞
a j aj
where x1(t) is the original signal. Comparing (16) with (14),
A real wavelet transform is complete and maintains energy we can find that Aj in (14) is decomposed but also Dj in (14)
conservation as long as the wavelet satisfies a weak is decomposed in (16).
admissibility condition which is:
7241
Wavelet and wavelet packet decompose the original signal of 16,384 sampling numbers, 3kHz maximum frequency, and
that is non-stationary or stationary into independent 2.1333 measuring time.
frequency bands with multi-resolution. We assume that most
of fault characteristic is embedded in high frequency band Input: Stator current
and apply wavelet decomposition to find a characteristic of
each fault in stator current. It can produce approximation 5.1
coefficient (cA) and detail coefficient (cD) in first step Mean filter
decomposition. Fig. 3 shows characteristics of discrete
wavelet transform. The detail coefficient means high
frequency of source signal. We use this detail coefficient as a Synchronize
feature of each fault. The produced coefficients depend on a
kind of mother wavelet. Because of that, selection of a
suitable mother wavelet is also important. Find a difference
between ‘input current’
Pre-processed current signal and ‘healthy’ current
5.2
Fault templates 5.3

1. BF template Wavelet
2. BR template decomposition
3. MA template
4. UB template
‘db1’mother wavelet Correlation with input
and fault templates
Find the best matching

cALow Frequency cDHigh Frequency rate
Downscale 1/2
Diagnosis result
Approx. coef. Detail. coef.
Fig. 4 Flow chart of diagnosis processes. (BF: bearing fault,

BR: broken rotor bar, MA: misalignment, UB: unbalance)
Applied fault types in this study are broken rotor, faulted

bearing, bowed rotor, unbalance, and static and dynamic
eccentricity case. If the wavelet decomposition is
Fig. 3 Characteristics of DWT (ex: bearing fault current). implemented in the fault detection of induction motors, the
unsynchronized current phase problem should influent the
5. FAULTS DIAGNOSIS AND EXPERIMENTAL detection results much. The result of wavelet decomposition
RESULTS has time element. If target signals are not synchronized as
shown in Fig. 5, the unexpected results will appear in the
In this chapter, we propose an algorithm to diagnose and wavelet decomposition.
exactly explain each state. The proposed method was tested
and verified by Matlab software. In Fig 4, this flow chart Therefore the signals are re-sampled by synchronizing signals
shows whole processes of our proposed diagnosis. First, a with phase 0 by using Hilbert transform. And the average
measured unknown coming signal has pre-processing stage value divided by one cycle signal is calculated to reduce the
same as making templates. And this is mathematically noise of original signals which is named by mean filtering as
compared with 4 templates by correlation. shown in Fig. 6. A fault current has two elements those are a
healthy current element and fault current element. The fault
current element was caused from illegal condition of fault
5.1 Current signals and data pre-processing
motor which is kind of electromagnetic pulse. We can get
these faults element from difference between averaged fault
Motor rating applied in this paper is dependent on the signal and healthy signal. In Fig. 7, the template is produced
electricity conditions. The rated voltage, speed, and by difference between averaged fault signal and healthy
horsepower are 220V, 3450rpm, and 5kW, respectively. And signal. Theses produced templates have high frequency terms
the implemented motor specification includes the number of that relate with characteristics of each fault. In this reason, we
slot, the number of pole, slip, etc. The specification of used apply wavelet decomposition and extract the detail term. It is
motor is 34 numbers of slots, 4 numbers of poles, and 24 considered to improve the performance and introduced in
numbers of rotor bars. The slip is determined by calculating next section.
an actual motor speed and a rated speed. The specification of
measured input current signals under this condition consists
7242
5.2 Fault templates
We make four fault templates consist of broken rotor bar,

bearing fault, misalignment, and unbalance cases to apply a
template matching method for diagnosis. Each fault case has
40 test sets and averages these sets to make fault templates.
These processes are shown in Fig 8.
Input:
Faulted current signal
(known)
Mean filter
Fig. 5 Original current signals.

Synchronize
Find a difference
between ‘input current’
and ‘healthy’ current
No
Iteration = 40 ?
Yes
Take an average Make fault

templates
Fig. 6 Data synchronizing and mean filtering results. Fig. 8 Difference results of 4 faults.
Input 5.3 Applying Wavelet Decomposition and Finding a Suitable
(Filtered, Synchronized) Averaged healthy current
10
Synchronized Signal
10
Synchronized Signal Wavelet for Mother Diagnosis
8 8
6 6 We apply wavelet decomposition in produced templates. We

4
2
4
2
use equation (16) for decomposition of each template then
C urrent [m A ]
C urrent [m A ]
0 0 produce approximation coefficient and detail coefficient in

-2 -2
first step decomposition. The detail coefficient means high
frequency of source signal. These become new fault
-4 -4
-6 -6
-8 -8 templates and these are used for a template matching method.

-10
0 20 40 60
Sample [n]
80 100 120
-10
0 20 40 60
Sample [n]
80 100 120 A correlation is used to find out matching rate. But, before
this process, we need to find most suitable mother wavelet
for the motor diagnosis. We seek out best suitable mother
Difference: Bearing fault Difference: Broken rotor bar wavelet in various mother wavelet sets investigated in Table
0.2 0.5 1. We try a template matching and find out diagnosis results
in each mother wavelet at 40 times (totally 200 samples data,
0.1
0
each fault 40 samples). The criterion for evaluation is used
0 hit rate of diagnosis. And we find that the Daubechies
wavelet of window size 1 (same as Harr wavelet) is a best
-0.1 -0.5
suitable mother wavelet for our diagnosis as shown in Table
20 40 60 80 100 120 20 40 60 80 100 120 2. The diagnosis result of Daubechies wavelet has about
Difference: Misalignment Difference: Unbalance
97.5% accuracy rate. Especially, misalignment and unbalance
0.4 0.5 cases have 100% accuracy rate.
0.2
6. CONCLUSIONS
0 0
The proposed method is bases upon several time-series data
-0.2
mining techniques. The stator currents are measured by the
-0.4 -0.5 current meters and stored in the time series data. Pre-
20 40 60 80 100 120 20 40 60 80 100 120
processing methods are applied to treat the signals, because
the intact time series data is not suitable to represent the
Fig. 7 Difference results of 4 faults.
7243
current signals. After preprocessing of the signals, the [12] I. Daubechies (1992), Ten Lectures on Wavelets,
features have to be extracted by the time series data mining Philadelphia, PA: SIAM.
methods that include synchronization and wavelet analysis. [13] Y. Meyer (1993), Wavelets: Algorithms and
The discovered features are constructed to the motor faults Applications, Philadelphia, PA: SIAM.
templates for the template matching method. The wavelet
analysis is possible method to detect faults of induction
Table 1. Mother wavelets for DWT.
motors, but it needs to find most suitable mother wavelet for
the motor diagnosis. Then we know that Daubechies wavelet Wavelets Window size
is the most suitable mother wavelet for diagnosis of motor. It
Daubechies 1-15
can increase the diagnosis accuracy rate of about 97.5%.
Symlets 2~15
Coiflets 1~5
ACKNOWLEDGEMENT
Meyer -
This work was supported by “Research Center for Logistics
Information Technology (LIT)” hosted by the Ministry of
Education & Human Resources Development in Korea. Table 2. A list of results of diagnosis test sets in various
mother wavelets (Each 40 times).
REFERENCES
Wavelets BF BR MA UB Total
[1] P. Vas (1993), Parameter Estimation, Condition Haar 37 39 40 40 156
Monitoring, and Diagnosis of Electrical Machines,
coif1 31 38 20 33 122
Clarendron Press, Oxford.
[2] G. B. Kliman and J. Stein (August 1990), Induction db2 28 38 19 34 119
motor fault detection via passive current monitoring, sym2 28 38 19 34 119
International Conference in Electrical Machines, db3 25 40 18 28 111
Cambridge, MA, pp. 13-17. sym3 25 40 18 28 111
[3] G.K. Singh and SAS AlKazzaz (2003), Induction db4 23 37 16 25 101
machine drive condition monitoring and diagnostic db5 25 36 12 26 99
research—a survey, Electric Power Systems Research,
sym7 26 32 15 26 99
vol. 64, pp. 145–158.
sym11 25 31 13 24 93
[4] K. Abbaszadeh, J. Milimonfared, M. Haji, and H. A.
Toliyat (2001), Broken Bar Detection In Induction Motor db10 25 31 11 24 91
via Wavelet Transformation, IECON’01: The 27th db11 19 29 16 27 91
Annual Conference of the IEEE Industrial Electronics db12 22 33 10 26 91
Society, pp. 95-99. sym4 25 37 6 22 90
[5] Masoud Haji and Hamid A. Toliyat (30 Sept.-4 Oct. sym15 24 31 12 23 90
2001), Pattern Recognition-A Technique for Induction coif3 25 34 7 24 90
Machines Rotor Fault Detection Eccentricity and Broken
dmey 25 31 9 25 90
Bar Fault, Conference Record of the 2001 IEEE Industry
Applications Conference, vol. 3, pp. 1572-1578. db9 25 33 8 23 89
[6] Jafar Zarei and Javad Poshtan (2007), Bearing fault sym5 23 35 7 24 89
detection using wavelet packet transform of induction coif2 24 36 6 23 89
motor stator current, Tribology International, vol. 40, sym14 25 32 8 22 87
Issue 5, pp. 763-769. db6 25 34 4 23 86
[7] S. Nandi, H. A. Toliyat (1999), Condition Monitoring sym12 25 32 7 22 86
and Fault Diagnosis of Electrical Machines – A Review, coif4 25 33 6 22 86
IEEE Industry Applications Conference, vol. 1, pp. 197-
sym6 24 32 7 21 85
204.
[8] B. Yazici, G. B. Kliman (March/April 1999), An coif5 25 32 7 21 85
Adaptive Statistical Time-Frequency Method for db7 24 34 5 21 84
Detection of Broken Bars and Bearing Faults in Motors db13 25 31 4 22 82
Using Stator Current, IEEE Trans. On Industry Appl., sym8 24 33 4 21 82
vol. 35, no. 2, pp. 442-452. db14 25 30 5 21 81
[9] B. Sklar (2001), Digital Communications Fundamentals db15 25 31 5 20 81
and Applications. 2nd edition, Prentice-Hall, 2001.
sym10 25 31 5 20 81
[10] C. Gasquet, P. Witomski (1998), Fourier Analysis and
db8 23 33 5 19 80
Applications: Filtering, Numerical Computation,
Wavelets, Springer Verlag, New York. sym9 23 32 6 19 80
[11] S. Mallat (1998), A Wavelet Tour of Signal Processing, sym13 25 31 5 18 79
San Diego, CA: Academic. *BF: bearing fault, BR: broken rotor bar, MA: misalignment,
UB: unbalance.
7244
Available online at www.sciencedirect.com
ScienceDirect
Procedia CIRP 118 (2023) 952–957
16th CIRP Conference on Intelligent Computation in Manufacturing Engineering, CIRP ICME ‘22, Italy
Biofeedback for human-robot interaction in the context of collaborative

assembly
Patrick Rückerta,*, Hannah Wallmeiera, Kirsten Trachta
a
Bremen Institute for Mechanical Engineering, University of Bremen, Badgasteiner Str. 1, 28359 Bremen
* Corresponding author. Tel.: +49-421-218-64838; E-mail address: rueckert@bime.de
Abstract
In a variety of applications in the medical as well as in the technical sector, biofeedback refers to endogenous data of the user. In this paper, the
potential application fields of biofeedback specifically for human-robot interaction in the context of collaborative assembly are investigated.
The worker's endogenous signals are to be captured and made available to the robot. This allows an optimization of the cooperation between
humans and robots. The aim is to reduce behavioral adaptation on the part of the employee. Several models for different task types of
collaborative robots are developed, showing the diversity of possible applications. The application scenario of robot adaptation based on stress
detection is exemplified as a use case.
© 2023 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 16th CIRP Conference on Intelligent Computation in Manufacturing Engineering
Keywords: Human-robot interaction; Assembly; Biofeedback
1. Introduction each employee over the entire work period. It is therefore not
possible to exclude the possibility of the employee being either
Human-robot collaboration (HRC) enables the reduction of overchallenged or underchallenged.
the workload of employees, making optimal use of the strengths One possible solution for optimizing HRC systems is the
of humans and robots in a given case. Thus, the demographic integration of biofeedback. The body's own signals are recorded
change, the shortage of skilled workers and the increasing and made available to the user. [4] In this way, the robot should
desire for variability can be managed by HRC. [1] HRC have access to information from its human counterpart. It
applications are continuously being developed and are no enables the robot to interpret the state of the corresponding
longer just a research concept, but are gradually forming a employee, which offers enormous potential in a wide variety of
keystone in the industrial workflow [2]. applications. Thus, biofeedback can increase awareness within
In this context, the communication interfaces between workplaces. Work-related complaints can be reduced as well as
humans and robots represent the foundation for user accidents, making it possible to ensure the safety and health of
acceptance. The effectiveness of communication significantly employees. [5] From an economic point of view, the optimized
influences the efficiency of the overall system. [3] An intuitive exchange of information increases the efficiency of HRC, thus
operation and user-friendliness are therefore particularly making the use of robots useful in smaller series sizes as well
important in collaborative use cases. Until now, in many cases as in unstructured production environments with changing
the human adapts to the patterns of the robot. Behavior products and processes [6].
characteristics, communication interfaces and communication The aim of this paper is to systematically investigate
protocols must therefore be known and executed correctly. By potential applications of biofeedback for human-robot
default, the robot has no information about the employee's state interaction (HRI) in the context of collaborative assembly. With
of mind. Accordingly, the robot behaves in the same way for this background, existing applications of biofeedback in the
2212-8271 © 2023 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 16th CIRP Conference on Intelligent Computation in Manufacturing Engineering
10.1016/j.procir.2023.06.164
Patrick Rückert et al. / Procedia CIRP 118 (2023) 952–957 953
context of robotics in the manufacturing, medical, and In the context of this study, only non-invasive methods are
consumer domains are examined. Emphasis is placed on the considered in more detail. Plethysmography biofeedback and
interaction of different modalities of biofeedback. Based on the biofeedback of internal organs are excluded. In addition to the
existing approaches and considering the determined parameters common biofeedback measurements, the recording of the eyes
and influencing factors, new biosignal-based interaction will be examined in more detail. In this context, three well-
models are to be derived that compensate for weaknesses of known methods for recording eye movements, eye positions,
existing systems and enable safe and error-free collaboration eyelid closures, and pupillary responses will be presented: the
between humans and machines. Primarily, it is assumed that electro-oculogram (EOG), the mechano-oculogram (MOG),
biofeedback promotes a more autonomous, flexible and and infrared camera systems (Table 1).
adaptive behavior of the robot. The development of the models is based on the thesis of
psychophysiology that conclusions can be drawn about the
2. Biofeedback inner-psychological state of a person by means of physiological
measurands. The disciplines of physiology and psychology are
The term biofeedback originated in clinical psychology to inferentially linked. [8]
influence disorders caused by malfunctions of the biological
system. Biofeedback is understood as a treatment in which 3. Applications of biofeedback in production
physical functions of the patient are continuously fed back in a
visual, acoustic or tactile way. The feedback of physiological In production, biofeedback sensors are used for a diverse range
processes improves the willful self-control of bodily functions. of applications. Based on a literature review, the following
In this way, the patient can learn to influence the functions, application areas can be identified: control, analysis &
which can reinforce positive changes. [4,7] assessment, stress detection, and emotion detection. In Table 2,
Within this paper, bodily functions are not to be reported the subsequent examples are assigned to the respective
back to the human, but made accessible to a machine. The application areas.
original therapy approach of biofeedback is therefore not
pursued. Instead, the biological data shall help to improve the Table 2. Literature review on applications of biofeedback in production.
interaction between human and robot. The robot is supposed to Application Example
act in accordance with the operator's performance capabilities. area
Depending on the biological process, special methods are Control Proof of concept of EOG-controlled industrial robots [9].
available to measure the biofeedback. These are usually used in EMG-controlled microgripper [10]
combination. A distinction is made between Myoelectric-controlled robotic glove for grasping
electroencephalography (EEG) biofeedback, assistance [11]
electrocardiography (ECG) biofeedback, electromyography EEG-controlled robots (via mental tasks, external stimuli,
(EMG) biofeedback, electrodermal activity (EDA) or hybrid systems) [12,13,14]
biofeedback, skin temperature biofeedback, plethysmography Online position control via EEG [15].
biofeedback, respiration feedback and biofeedback of internal
Analysis & Myo wristbands for action recognition [16].
organs. [7] Assessment EMG based stress analysis for the evaluation of assistance
Table 1. Biofeedback methods considered in this paper. systems [17].
Biofeedback methods Measured body function EMG-based assessment of ergonomic examination

methods [18].
EEG biofeedback Electrophysiological processes of the brain
Stress Real-time stress detection based on ECG signals for
ECG biofeedback Heart rate detection human-machine system design [19].
Heart rhythm Evidence of possible stress reduction by robot in HRI
EMG biofeedback Muscle activity [20].
EDA biofeedback Sweat gland activity Evaluating the acceptance of different movement patterns
of the robot by recording the comfort of the employee
Skin temperature Skin temperature [21].
biofeedback
Predicting mental states of the employee in HRI based on
Blood pressure the employee's skin temperature [22].
Respiration biofeedback Respiration rate
Emotional Emotional communication in HRI by decreasing Heart
Respiration quality detection frequency [23].
Respiration pressure Determination of mental states in HRI via EEG analyses
EOG Eye movement [24]
MOG Eyelid closure Analysis of the quality of an HRI using facial

myographies [25].
Infrared camera systems Eye movement
Facial expression recognition based on EMG evaluations
Eye position [26].
Eyelid closure Investigation of emotional impact in teleoperation tasks
Pupil reaction using EDA sensors [27].
Deriving motion rule for robots based on EDA data [28].
954 Patrick Rückert et al. / Procedia CIRP 118 (2023) 952–957
3.1. Application of robot control 3.3. Application of Stress detection
Different sensors can be useful for controlling robots. The consideration of mental stress, in contrast to physical
Aungsakul et al. investigated various features of the EOG and stress, has not been taken into account for a long time. Stress
were able to prove that industrial robots can be controlled by as a physical and psychological reaction of the organism to
eye movement. [9] Furthermore, a control of the robot strain is commonly known and can increase the probability of
synchronously to its own processes is possible. Through EMG heart attacks. [30] Accordingly, the ability to recognize a
sensors, muscle movements can be transferred to the robot. In dangerous stress level is crucial for healthy work. One way to
this context, Jain et al. developed an artificial muscle finger to detect mental workload is by measuring the ECG. There are a
control a microgripper. The gripper is controlled by the variety of low-cost, non-invasive, and portable ECG devices
myoelectric signals picked up from the human index finger. available [31]. In this way, the physical condition of the subject
[10] Cheon et al. presented a myoelectric-controlled robotic can be examined remotely. [32] Therefore, integration into
glove for assisting forceful grasping. Five EMG sensors are production is not a problem. At a higher stress level, the cardiac
attached to the proximal forearm region for this purpose. [11] load increases, so that the period and the shape of the ECG
Additionally, there have been initial research efforts to use changes. It is mainly the R-R interval that is considered.
EEG data to control industrial robots in collaborative assembly However, beyond that, other characteristics can be illuminated
applications. This is expected to enable intuitive and redundant to properly classify the workload despite the variability
human-machine communication and closer human-robot between subjects. Qu et al. included the different parameters of
interaction. Based on this, it is expected to increase the ECG in the stress classification to incorporate the results
productivity in assembly. [14] The basic prerequisite for such into the development of a human-machine system. [19] Dang
applications is the development of EEG caps that enable and Tapus investigated the effects of the presence of robots on
smooth and reliable EEG signal acquisition. In this way, use in the stress perception of individuals. The cardiovascular signals
real-world environments can be considered. EEG controls are prove that a robot can indeed have a positive effect on stress
distinguished according to whether an external stimulus is reduction. [20] However, this largely depends on the robot's
needed for command generation. Hybrid systems represent a movement pattern. Zanchettin et al. investigated the effects of
combination of both control options and can thus exploit robot movements on the subject's acceptance. For this purpose,
synergies. [29] they recorded myoelectric and electrodermal signals in addition
Control via external stimuli may be more useful in the to the ECG. The results show that a human-like movement of
production environment. Middendorf et al. therefore the robot increases social acceptance. The stress levels of the
implemented a brain-computer interface that converts visually subjects could be reduced in this way. [21] He et al. referred to
evoked responses into control commands for the operation of the skin temperature of the face to obtain information about the
devices [12]. Minati et al. developed a multimodal control psychological state of the employee. Compared to the
approach in which EOG, EMG, and EEG data as well as head presented methods, skin temperature feedback has an immense
movements are used [13]. In addition to robot control, ECG can advantage, as the temperature measurement does not
be used for online position control of a robotic arm. Deviations necessarily require physical contact of the sensor. [22]
of a known pattern, in this case an erroneous final position of
the robot, can be observed in the EEG by changing the
frequency patterns. As a result, a correction can be made to the 3.4. Application of Emotion detection
robot position. [15]
Motions of employees are especially crucial in human-robot
interactions, as they can influence collaboration. [34] Due to
3.2. Application of analyses & assessment this, a wide variety of physiological signals of the employee are
picked up during the interaction. EEG, ECG, EDA, EOG,
Biofeedback is often used in production to analyze and EMG, and skin temperature measurement can all contribute to
assess processes. Based on recorded sensor data, movements emotion detection. [21] Maruyama et al. used heart rate for
can be identified to further predict the outcome of actions. Al- emotion detection. Heart rate, as a result of signals from the
Amin et al. developed a concept based on the recording of sympathetic nervous system, provides information about the
EMG data using Myo wristbands. [17] Furthermore, in the subject's affective state. [23] Toichoa Eyam et al. accessed
context of process optimization, physical strain at the EEG data for emotion detection in human-robot interaction.
workplace is often of interest. To record the load, EMG The acquired neural signals can be examined using machine
electrodes are attached to the stressed muscles in order to learning algorithms to identify emotion detection patterns. The
record the changes in tension. Graham et al. evaluated the robot can then adapt its parameter settings to the employee's
effectiveness of an assistance system based on the recorded emotional state and in this way become more responsive to the
data sets. [18] Mathiassen et al. evaluated the validity of employee [24]. Besides EEG, EMG is another possibility for
ergonomic examination procedures. [19]. emotion detection. With the help of facial myographs, reactions
to visual and acoustic signals can be. Kulić and Croft
investigated the quality of human-robot interaction using facial
myographs [25]. Elevated EMG signals in the face are thereby
associated with unpleasant emotions Further, EMG data are
used for the identification of facial expressions. In this process, 4. Robot adaptation based on stress detection
a broad variety of facial signals must be derived. The analysis
represents a complicated area of biomedical signal processing. 4.1. User story of the scenario
[26] Alternatively, EDA can be recorded. EDA biofeedback is
often used for emotion detection due to the simplicity of the Based on the utility analysis, the application scenario of
measurement setup. Advantageously, the EDA sensor can be robot adaptation based on stress detection is exemplified as an
used as a wrist sensor in the form of a watch. This allows the use case. This presents the task type of workload reduction.
wearer freedom of movement. [27] Hanajima et al. were able This model aims at adapting the robot behavior to the stress
to derive a motion rule for human-friendly robots, using perceived by the employee. Speed reduction seems to be an
electrodermal properties. Therefore, the speed of the robot is appropriate measure. Hanajima et al. proved in their work that
reduced when it is close to the human. [28] the reduction of robot speed correlates with decreasing stress
levels. [28].
4. Models for biofeedback in human-robot interaction If the employee is exposed to stress this results in physical
reactions. Heart rate and electrodermal activity increase due to
For the derivation of the models, different approaches for the increased mental workload. In addition, a growing tension
potential interaction systems for collaborative assembly are follows, so that the myoelectric potentials rise. This can be
collected. From the analysis of the presented fields of measured, for example, in the shoulder area. On average,
application in production, as well as medicine and other fields, breathing becomes more high-frequency and deeper. Fig. 1
13 interaction models for the task types of information shows the user story for the interaction model.
exchange, transport, precision and workload reduction could be
derived. Often, multiple biofeedback sensors can be used for Interaction model: robot adaptation based on stress detection
the same application. On the one hand, to meet the redundancy

criterion and to consider the interaction of different types of
biofeedback, it is suggested to use at least two biofeedback
sensors in one interaction system. On the other hand, the
number of sensors should be limited in order to reduce the costs
and to improve the user friendliness. Accordingly, attention is
paid to use no more than three beneficial sensors. The
suitability of the resulting possible sensor combinations is to be User Story: „As a user, I want the robot to adapt to my stress levels so that my
quantified by a utility analysis. mental health is protected.
The evaluation criteria of the utility value analysis are Conversation:
derived from economic, functional, safety and operating target ▪ Adaptation of the robot to measured body data.
Feedback on measured body data
values. To avoid subjective influences, all main objectives are
▪
▪ Constant behavior of the robot unless a request to do so has been received
weighted equally. Therefore, each of the four main objectives ▪ Speed can be manually reset
receives an influence of 25%. This share is in turn divided Acceptance Criteria
equally among the sub-objectives The fulfillment of the sub- ▪ Biofeedback is continuously recorded and evaluated
objectives is evaluated on a scale from 0 to 3. If the model ▪ Visual feedback of body data via LED bars
under consideration can completely fulfill the sub-objective, ▪ Individual determination of stress thresholds
▪ Robot slows down or stops when stress thresholds are exceeded
the highest rating of 3 is given. Table 3 shows the applications ▪ Reset of speed by pressing a button
with the maximum utility value for the respective application
field. Table 3. Four applications with highest utility value. Fig. 1. User story of the use case scenario
Table 3. Interaction models with maximum utility value for the respective
application field. 4.2. Use case diagram of the scenario
Task Type Nr Interaction model Utility Modality
value The corresponding use case diagram is shown in Fig. 2. The
biological data of the user are continuously recorded (use case
Information 3 Hazard detection 3.0 HF, EDA
exchange
"Record user input"). If both signals of the selected model rise
EDA, Breath into critical ranges, the control command to change the robot
HF, Breath speed is transmitted ("Reduce robot speed" use case). In case
Transport 5 Robot control by 2.13 IRCS, EMG of excessive changes, the speed can be set to 0. If only one
eye movement parameter increases, the robot behavior remains constant. The
Precision 7 Object handling 2.75 EMG, IRCS redundant acquisition is intended to avoid a superfluous
Workload 8 Robot adaptation 3.0 HF, EDA increase in cycle time. The data is continuously made available
reduction based on stress HF, Breath
to the operator via the LED bars (use case "Visual output"). The
detection robot speed can be reset by pressing a button if the stress is not
HF, EMG
caused by the work situation (use case "Reset robot speed").
EDA, Breath
EDA, EMG
Breath, EMG
956 Patrick Rückert et al. / Procedia CIRP 118 (2023) 952–957
4.4. Measurement methodology
In order to adapt the threshold values of the measurement

system to test persons, their individual body values must be
determined in advance. The necessary methodical procedure is
shown in Fig. 4. First, a stress test must be performed. This
results in the individual stress-induced heart rate and EDA data,
as well as the evaluation of the mental state over time. The body
values must be compared with the mental state. Based on these
data, the respective regression analysis of the heart rate and
EDA values on the psychological state is performed. In this
way, each level of psychological well-being (from 0 to 10) can
Fig. 2. Use Case diagram for robot adaptation based on stress detection. be assigned a corresponding heart rate and EDA value. In
addition, coupling of the stress-induced heart rate and EDA
4.3. Measurement system values is possible. Subsequently, different ranges of stress
experience are defined. Depending on the range definitions,
When designing the measurement system, wireless data different behavioral levels of the robot are defined. For a
transmission is crucial in order to maintain the operator's reasonable robot adaptation, the heart rate and EDA related
freedom of movement. In addition, the operator should have stress threshold should serve as the limit of the lowest
access to his own body data in order to understand possible adaptation level. This is calculated using the resting values. A
changes in the robot's speed. This can be done with the help of mental state of level 10 should always result in a robot stop.
LED bars. If stress values are not caused by the work situation,
the employee is free to reset the robot's speed by pressing a
button. The measurement would be equally possible on the
finger, wrist or earlobe. EMG sensors pick up myoelectric
signals at the trapezius muscle to detect increasing tension in
the shoulder area. EDA is measured by sensors on the index
and middle fingers. Various respiratory parameters can be
registered via respiratory belts. Depending on which model
variant is selected, corresponding sensors are omitted.
The sensors required for recording the individual
biofeedback data must be bundled on the test person to allow
flexible and convenient handling. For this purpose, an
wristband is made which can be easily attached to the forearm
(Fig. 3). The wristband is variably adjustable so that it can be
individually adapted by different test subjects. By lining the
wristband with foam, the risk of injury is significantly reduced.
In addition, a pleasant wearing comfort is achieved. Fig. 4. Methodology for measurement of individual stress levels.
In addition to the heart rate and EDA sensor, the EMG
sensor, the control unit (microcontroller), a 9V battery, the 4. Summary and Outlook
button, the room temperature and humidity sensor and the LED
bars are also located on the wristband. The results of the literature analysis confirm the assumption
that biofeedback can contribute to an improvement of human-
robot interaction in the context of collaborative assembly.
Based on a detailed investigation of existing biofeedback
applications from a wide variety of fields, new biosignal-based
interaction models were derived taking into account the
parameters and influencing factors identified. In doing so, the
diversity of biofeedback applications was underlined by
modeling in the different task areas of collaborative robots.
Thirteen model ideas were evaluated using utility analysis. Five
of the ideas were further specified and elaborated. Each model
can address weaknesses of existing collaborative systems and
enable safer and more user-friendly collaboration between
humans and machines. The use of user stories, use case
diagrams, and event-driven process chains ensure the
feasibility of the models. Practical implementation of one of the
models, "Robot adaptation based on stress detection", further
Fig. 3. Measurement system for stress related workload. confirms the feasibility.
In further investigations, the reproducibility of the body [13] Middendorf, M; McMillan, G; et al. (2000): “Brain-computer interfaces
values under constant conditions has to be considered. If this based on the steady-state visual-evoked response”. In: IEEE transactions
on rehabilitation engineering a publication of the IEEE Engineering in
cannot be confirmed, further features must be used to Medicine and Biology Society, 8 (2), p. 211-214.
distinguish between stress, strain and danger responses. In this [14] Minati, L; Yoshimura, N; et al.. (2016): “Hybrid Control of a Vision-
respect, the slope of the EDA curve can be considered. In a Guided Robot Arm by EOG, EMG, EEG Biosignals and Head Movement
danger situation, the skin conductance drops drastically in the Acquired via a Consumer-Grade Wearable Device”. In: IEEE Access, 4, p.
short term. Stress, on the other hand, leads to a steady slow 9528-9541.
[15] Bhattacharyya, S; Konar, A; et al. (2017): “Motor imagery and error
decrease. In addition, the addition of further biofeedback related potential induced position control of a robotic arm”. In: IEEE/CAA
sensors is possible. The lid closure activity shows significant Journal of Automatica Sinica, 4 (4), p. 639-650.
differences between the scenarios and is therefore [16] Al-Amin, M; Tao, et al. (2019): “Action Recognition in Manufacturing
recommended. Assembly using Multimodal Sensor Fusion”. In: Procedia Manufacturing,
Furthermore, the quantification of the influence of 39, p. 158-167.
[17] Graham, R.B; Agnew, M.J; et al. (2009): “Effectiveness of an on-body
individuality seems to be useful. For this, a statistical lifting aid at reducing low back physical demands during an automotive
experimental design with multiple subjects is necessary. For assembly task: assessment of EMG response and user acceptability”. In:
this, approval by the Ethics Board should be obtained Applied ergonomics, 40 (5), p. 936-942.
beforehand. The developed method for thresholding is based [18] Mathiassen, S.E; Burdorf, A; et al. (2001): “Statistical power and
on the assumption that the body values of further test persons measurement allocation in ergonomic intervention studies assessing upper
trapezius EMG amplitude A case study of assembly work”. In: Journal of
behave similarly to those recorded. If the responses of body Electromyography and Kinesiology, 2002 (12), p. 45-57.
values to stress and danger situations differ significantly from [19] Qu, H; Gao, X; et al. (2021): “Classification of mental workload based on
those measured, it will be necessary to repeatedly verify that multiple features of ECG signals”. In: Informatics in Medicine Unlocked,
classification is possible based on EDA and heart rate data 24, p. 100575.
alone. [20] Dang, T.-H.-H; Tapus, A. (2015): “Stress Game: The Role of Motivational
Robotic Assistance in Reducing User’s Task Stress”. In: International
Journal of Social Robotics, 7 (2), p. 227-240.
References [21] Zanchettin, A.M; Bascetta, L; et al. (2013): “Acceptability of robotic
manipulators in shared working environments through human-like
[1] Kuhlenkötter, B; Thomas, et al. (2014): Mensch-Roboter-Kollaboration - redundancy resolution”. In: Applied ergonomics, 44 (6), p. 982-989.
Von der industriellen Produktion bis zum Anwendungsgebiet [22] He, C; Mahfouf, M; A. et al. (2018): Facial Temperature Markers for
Rehabilitation, p. 253. Mental Stress Assessment in Human-Machine Interface (HMI) Control
[2] Zeising, P; Brending, et al (2015): Sichere Mensch-Roboter- Kollaboration System. Band 2: SCITEPRESS – Science and Technology Publications.
durch Prädiktion. Oldenburg: Oldenbourg Wissenschaftsverlag, p.271. [23] Maruyama, T; Makikawa, M; et al. (2007): ECG Measurement Using
[3] Müller, R; Franke, J; Heinrich, D; Kuhlenkötter, B; Raatz, A; Verl, A. Capacitive Coupling Electrodes for Man-Machine Emotional
(2019): Handbuch Mensch-Roboter-Kollaboration. München: Hanser, p Communication.
354. [24] Toichoa Eyam, A; Mohammed, W.M; et al. (2021): “Emotion-Driven
[4] Rief, W; Birbaumer, N. (2011): “Grundsätzliches zu Biofeedback”. In: Analysis and Control of Human-Robot Interactions in Collaborative
Rief, W; Birbaumer, N.-P; Bernius, P. (eds.) Biofeedback. Grundlagen, Applications”. In: Sensors (Basel, Switzerland), 21 (14).
Indikationen, Kommunikation, Vorgehen ; mit 48 Tabellen, 3., vollst. [25] Kulić, D; Croft, E. (2007): “Physiological and subjective responses to
überarb. und erw. Aufl. Stuttgart: Schattauer, p. 1-7. articulated robot motion”. In: Robotica, 25 (1), p. 13-27
[5] Pirker-Binder, I. (2016): Prävention von Erschöpfung in der Arbeitswelt. [26] Hamedi, M; Salleh, S.-H; et al. (2013): “EMG-based facial gesture
Betriebliches Gesundheitsmanagement, interdisziplinäre Konzepte, recognition through versatile elliptic basis function neural network”. In:
Biofeedback. Biomedical engineering online, 12.
[6] Naumann, M; Dietz, T; et al. (2014): “Mensch-Maschine-Interaktion”. In: [27] Yang, E; Dorneich, M.C. (2017): “The Emotional, Cognitive,
Bauernhansl, T; Hompel, M. ten; Vogel-Heuser, B. (eds.) Industrie 4.0 in Physiological, and Performance Effects of Variable Time Delay in
Produktion, Automatisierung und Logistik. Anwendung, Technologien, Robotic Teleoperation”. In: International Journal of Social Robotics, 9
Migration. Wiesbaden: Springer Vieweg, p. 509-523. (4), p. 491-508.
[7] Linden, M; Hautzinger, M. (2015): Verhaltenstherapiemanual. Mit 23 [28] Hanajima, N; Goto, T; et al. (2005): A motion rule for human-friendly
Tabellen. 8., vollst. überarb. Aufl. Berlin: Springer,. p.97. robots based on electrodermal activity investigations and its application to
[8] Vossel, G; Zimmer, H; et al. (1998): Psychophysiologie. Stuttgart: mobile robot.
Kohlhammer., p. 11. [29] Aljalal, M; Ibrahim, et al. (2020): “Comprehensive review on brain-
[9] Aungsakul, S; Phinyomark, et al. (2012): “Evaluating Feature Extraction controlled mobile robots and robotic arms based on
Methods of Electrooculography (EOG) Signal for Human-Computer electroencephalography signals”. In: Intelligent Service Robotics, 13 (4),
Interface”. In: Procedia Engineering, 32, p. 246-252. p. 539-563.
[10] Jain, R.K; Datta, S; et al. (2013): “Design and control of an IPMC [30] Wagner-Link, A. (2013): Stress. Wie Sie Stressoren erkennen und
artificial muscle finger for micro gripper using EMG signal”. In: Belastungen besser bewältigen können. 23. Aufl.
Mechatronics, 23 (3), p. 381-394 [31] Faruk, N; Abdulkarim, A; et al. (2021): “A comprehensive survey on
[11] Codispoti, M; Bradley, M.M; et al. (2001): “Affective reactions to briefly low-cost ECG acquisition systems: Advances on design specifications,
presented pictures”. In: Psychophysiology, 38 (3), p. 474-478. challenges and future direction”. In: Biocybernetics and Biomedical
[12] Mohammed, A; Wang, L. (2020): “Advanced Human-Robot Engineering, 41 (2), p. 474-502
Collaborative Assembly Using Electroencephalogram Signals of Human [32] Rahman, M.M; Rimon, M.A.H; et al. (2019): Affordable Smart ECG
Brains”. In: Procedia CIRP, 93, p. 1200-1205. Monitoring Using Arduino & Bluetooth Module.
ScienceDirect
IFAC PapersOnLine 54-4 (2021) 177–182
Detection
Detection of
of Spatiotemporal
Spatiotemporal Interferences
Interferences
Detection
of Road
Detection of
of Spatiotemporal
Traffic with IoT-Connected
Spatiotemporal Interferences
Sensor
Interferences
of
of Road
Road Traffic
Detection of with IoT-Connected
Spatiotemporal
Traffic with IoT-Connected Sensor
Interferences
Sensor
of Road Traffic Beacons
with IoT-Connected
of Road Traffic with IoT-Connected Sensor
Beacons
Beacons Sensor
Beacons
Beacons
Olaf Czogalla ∗
René Schönrock ∗∗
Olaf Czogalla ∗∗∗ René Schönrock ∗∗ ∗∗
Olaf Czogalla ∗ René Schönrock ∗∗ ∗∗
∗ Olaf Czogalla René Schönrock
∗ Institut für AutomationOlaf Czogalla und Kommunikation,
∗
René Schönrock Magdeburg,
∗∗ ST 39106
∗ Institut für Automation und Kommunikation, Magdeburg, ST 39106
∗
Institut für Germany
Automation (e-mail:
und olaf.czogalla@ifak.eu).
Kommunikation, Magdeburg, ST 39106
∗ Germany (e-mail:
∗∗Institut für Automation und olaf.czogalla@ifak.eu).
Kommunikation, Magdeburg,
∗ Institut für Automation
Institut für Germany (e-mail:
Automation und Kommunikation,
und Kommunikation,
olaf.czogalla@ifak.eu).Magdeburg, ST
Magdeburg, ST 39106
ST 39106
∗∗
∗∗ Institut für Automation
Germany und
(e-mail: Kommunikation,
olaf.czogalla@ifak.eu).Magdeburg, ST 39106
39106
∗∗
Institut fürGermany
Automation
Germany (e-mail:
und
(e-mail: rene.schoenrock@ifak.eu).
Kommunikation,
olaf.czogalla@ifak.eu).Magdeburg, ST 39106
∗∗ Germany (e-mail: rene.schoenrock@ifak.eu).
∗∗ Institut fürGermanyAutomation und Kommunikation, Magdeburg, ST 39106
(e-mail: rene.schoenrock@ifak.eu).
Institut fürGermany
Automation und Kommunikation,
(e-mail: rene.schoenrock@ifak.eu). Magdeburg, ST 39106
Germany (e-mail: rene.schoenrock@ifak.eu).
Abstract: Efficient collection and real-time provision of information on roadworks with impact
Abstract: Efficient collection and real-time provision of information on roadworks with impact
on traffic flow
Abstract: are essential
Efficient collectionto to
and reliable
real-time and effectiveoftraffic and transport management for
on traffic flow
Abstract: are essential
Efficient collection and reliable
real-time andprovision
provision
information
effectiveoftraffic
information
on roadworks
and transport
on roadworks
with impact
management
with impact for
urban
on
Abstract: and
traffic regional
flow are
Efficient road networks.
essential
collection to
and In
reliable order
real-time and to enable
effective
provision road
trafficoperators
and to impose
transport efficient
management traffic
for
urban
on and flow
traffic regionalare road networks.
essential to In order
reliable andto enable of roadinformation
operators on
to roadworks
impose with impact
efficient traffic
management
urban
on and flow
traffic
management
measures,
regionalare
measures, road comprehensive
networks.
essential to
comprehensive reliable amounts
In order and
amounts toeffective
of dataroad
enable
effective
of data
traffic
on both
traffic
on both
and
and
transport
traffic
operators demand
to impose
transport
traffic demand
management
and interferences
efficient
management
and
for
traffic
interferences for
urban
of the roadand regional
infrastructureroad networks.
are required. In order
Roadworksto enable road
management operators
systems to impose
are used efficient
for planningtraffic
of
management
urban
of the and
road measures,
regional
infrastructureroad comprehensive
networks.
are required. In amounts
order
Roadworksto of data
enable on both
road
management traffic
operators
systems demand
to impose
are usedand interferences
efficient
for planningtraffic
of
management
construction measures,
work zones,comprehensive
whereas exact amounts
real-timeof data on both
information on traffic
opening, demand and interferences
spatial-timely progress
of the
managementroad
construction infrastructure
measures, are required.
comprehensive
work zones, whereas Roadworks
amounts of management
data
exact real-time information on both systems
traffic
on systems are
demand
opening,are usedand
spatial-timelyfor planning
interferences
progress of
of
andthecompletion
road infrastructure
construction work of roadworks are required.
are only Roadworks
to obtain by management
connected sensors to beused for planning
deployed directlyof
of
andthecompletion
construction work ofzones,
road infrastructure roadworks
zones,
whereas
whereas areexact
are required.exact
real-time
to obtaininformation
only Roadworks
real-time management
by connected
information
on systems
on
opening,
sensors to
opening,
spatial-timely
are beuseddeployed
spatial-timely
progress
for planningdirectly
progress of
in the
and work zone.
completion
construction work To
ofTo thatwhereas
roadworks
zones, end,are anexact
innovative
only to obtain
real-timesensor-based
by connected
information construction
on sensors to
opening, barricade
be deployed
spatial-timely light progress
–directly
in the
in
and the work
completion zone. of as that
roadworks end, an innovative
are only to obtainsensor-based
by connected construction
sensors barricade light – in the
following
in
and the work denoted
completion zone.ofTo sensor
that
roadworks end, beacon
an only
are – has
innovative been
to obtain developed
sensor-based
by connected which
constructionallowsto
sensors
be geo-referencing
the
barricade
deployed directly
light –directly
in theof
following
in the work denoted
zone. as
To sensor
that end, beacon – has
an innovative been developed
sensor-based which
constructionallowsto be geo-referencing
the
barricade deployed
light – sensor
in theof
road
in the works
following work in
denotedreal
zone. time
as
To with
sensor
that end,sub-meter
beacon
an – accuracy.
has
innovative been The combined
developed
sensor-based which use
construction of multiple
allows the
barricade connected
geo-referencing
light – in of
the
road worksdenoted
following in real time with sub-meter
as sensor beacon accuracy. The combined useallows
of multiple connected sensor
lightsworks
road
following allows the
inthe
denoted extent
realextent
time of the
with
as sensor work –site
sub-meter
beacon
has tobeen
accuracy. developed
be exactly
The combined which
mapped. The
useThe the geo-referencing
construction
of multiple connected site related
sensorof
lights
road allows
works in real time of the
with work –site
sub-meter has tobeen
accuracy. developed
be exactly
The combined which
mapped. use allows
of the geo-referencing
construction
multiple connected site related
sensorof
information
lights
road allowsinis
works isthe
provided
real extent
time to
with an
of antheurban
work data
sub-meter data
site platform forcombined
to be exactly
accuracy. The further
mapped. processing
useThe
of with connected
construction
multiple software systems
site systems
related
sensor
information
lights allows the provided to
extent of and urban
the coordination. platform for further processing
work site to be exactly mapped. The construction site related with software
for roadworks
information
lights allowsis information
provided
extent to of antheurban
work data site platform for further processing with software site systems
for roadworks
information isthe
information
provided to and
an coordination.
urban data to be exactly
platform mapped.
for further The construction
processing with software related
systems
Copyright
for roadworks
information © 2021
is The Authors.
information
provided to and
an This is an open
coordination.
urban data access
platform article
for under
further the CC BY-NC-ND
processing with license systems
software
for roadworks information and Datacoordination.
(https://creativecommons.org
Keywords:
for roadworks
Keywords:
Internet of Things,
information
Internet of Things, /licenses/by-nc-nd/4.0/)
and aquisition, Traffic control
coordination.
Data aquisition, Traffic control
Keywords: Internet of Things, Data aquisition, Traffic control
Keywords: Internet of Things, Data aquisition, Traffic control
Keywords: 1. INTRODUCTION
Internet of Things, Data aquisition, Traffic countries.
controlIn the U.S. connected safety equipment has been
1. INTRODUCTION countries. In the U.S. connected safety equipment has been
1. INTRODUCTION developed In
countries. andtheintroduced
U.S. connected by commercial
safety equipment manufacturers
has been
1. INTRODUCTION developed In
countries. andtheintroduced
U.S. connected by commercial
safety by manufacturers
equipment has been
Roadwork construction 1. sites
INTRODUCTIONon highways and urban roads and service
developed
countries. providers
and
In the introduced
U.S. such
connected as
by shown
commercial
safety Sheckler (2020).
manufacturers
equipment has been
and service
Roadwork construction sites on highways and urban roads developed and introduced by commercial manufacturers providers such as shown by Sheckler (2020).
have a major
Roadwork impact on theon
construction urban traffic and transport
roads and The service
product line
providers named suchiConeas shown includes specialized
by Sheckler ap-
(2020).
have a major
Roadwork impact onsites
construction theon
sites
highways
urban
highwaystraffic and
andand urban
urbantransport developed
The service
and
roads plications product and introduced
line
providers named such by
iCone commercial
as shown includes
by manufacturers
specialized
Sheckler ap-
(2020).
system.
have a
Roadwork The
major knowledge
impact
construction onabout
the
sites onspatial
urban
highways extent
traffic and and
and urbantemporal
transport
roads The
and product
service ofproviders
connected
line named suchdevices
iCone
as that
shownincludesareSheckler
by customized
specialized for
ap-
(2020).
system.
have The
a major knowledge
impact about
on the urbanspatial extent
traffic and
and for temporal
transport plications
The product of connected
linesuch
named devices that
iConebarrels,includesare customized
specialized for
ap-
timeframe
system.
have a The
majorofknowledge
the
impactconstruction
onabout
the site istraffic
spatial
urban essential
extent and
and temporal
transport various
deter- plications
The purposes
product of connected
line named as devices
traffic
iCone that
includes traffic
are cones, flag-
customized
specialized for
ap-
timeframe
system. The ofknowledge
the constructionabout site isextent
spatial essential and for
temporal various purposes
deter- plications of such as devices
connected traffic barrels,
that traffic
are cones, flag-
customized for
mining effective
timeframe traffic
ofknowledge
thetraffic and transport
construction management
site isextent
essential for mea- ging
deter- various batons, of arrow
purposes suchtrailers
as trafficand barrels,
hazard lights.
traffic This suite
cones, flag-
system.
mining
timeframe Theof
effectivethe about spatial
and transport
construction site andfor
management
is essential mea- plications
temporal
deter- ging
various batons,
purposes connected
arrow such as devices
trailers traffic that are
and barrels,
hazard customized
lights.
traffic This suite
cones, for
flag-
sures.
mining It
timeframe is the
effective
of therole of
traffic municipal
and
construction authorities
transport
site is to
management
essential authorize,
for mea-
deter- of
gingproducts
various batons,
purposescollects
arrow such and
trailers
as transmits
and
traffic hazardreal-time
barrels, lights.
traffic information
This
cones, suite
flag-
sures. It is the role of municipal authorities to authorize,
mea- ging of products
batons, collects and
arrow trailers transmits
and hazard real-time information
mining effective
coordinate,
sures. and to
Iteffective
is the
traffic
inform and transport
traffic users management
about construction about
of the status
products arrowof
collects the transmits
and roadways into lights.
real-time
This using
the cloud suite
information
mining
coordinate,
sures. androle
It isinthe to
role
of municipal
traffic
inform and
ofmitigate traffic
municipal
authorities
transport
users about
authorities
to
management authorize,
construction
to mea- ging
authorize, about
of batons,
the status
products collects oftrailers
and and hazard
the transmits
roadways into lights.
real-time This using
the cloud suite
information
activities
coordinate,
sures. It is order
and
the to
role to
inform
of their
traffic
municipal obstructing
users about
authorities effects
construction
to authorize, on cellular
about
of products mobile
the status radio
collects of communication.
the
and roadways
transmits In
into France
real-timethe cloudhas been
using
information
activities
coordinate, in and
order to to mitigate
inform their
traffic obstructing
users about effects on about
construction cellularthe mobile
status radio
of communication.
the roadways In France
into the cloud has using
been
traffic quality.
activities in orderHowever,
to in
mitigate practice
their it is
obstructing observedeffects that
on developed
cellular
about a
mobile
the comparable
radio device
communication. to be In applied
France to
has traffic
been
coordinate,
traffic and
quality.
activities in order to toinform
However, traffic
in
mitigate users
practice
their itabout construction
is observed
obstructing effects that
on developed
cellular astatus
mobile of communication.
comparable
radio the roadways
device to into the cloud
beInapplied
France to
has using
traffic
been
the planned
traffic quality.
activities and
in order authorized
However,
to mitigate in construction
practice it
their obstructingis process
observed is
effects not
that cones
developed
on cellular called a
mobile Smartroadcone
comparable device
radio communication. that tois used
be to
Inappliedimprove
applied
France to road
traffic
hastraffic
been
the planned and authorized construction it is process is that
not cones called
developed a Smartroadcone
comparable devicethatto is used
beon to improve road
traffic
aligned
the
quality.
with and
planned
traffic quality.
However,
theHowever,
real
authorized
in practice
roadworks
in progression
construction
practice it is
observed
on site.is that
process
observed The
not worker’s
cones
developed safety
called a when theydevice
Smartroadcone
comparable arethatat torisk
is used
be road
to orto
improve
applied tohighway
road
traffic
aligned
the with the
plannedworksite real
and authorizedroadworks progression
construction on site. The worker’s
cones safety
called when
Smartroadcone they are at
that risk
is on
used road
to or
improve highway
road
authorized
aligned withworksite
the real extent
roadworks may differ from process
progression the actual
onactual
is not
oc- lanes
site.is The cones thas
worker’s is Smartroadcone
safety
called presented
when they in Destrian
are at risk is (2019). toThe
on road ormanufac-
highway
the planned
authorized
aligned
cupied area
authorized
and
withworksite
the
or the
authorized
real extent
roadworks
section-wise
extent
may
may
construction
differ from process
progression
roadwork
differ from
the
onactual
permission
the site. maynot
The
oc-
lanes
oc- worker’s
turer
lanes
thas
claims
thas
is presented
safety
is thatwhenthe
presented they arethat
in Destrian
electronic
in Destrianat risk used
(2019).
device on can
(2019). road improve
The
be
The ormanufac-road
highway
mounted
manufac-
aligned
cupied with
area
authorized the
or the
worksite real roadworks
section-wise
extent progression
roadwork
may differ on
permission
from theaactual site. The
may worker’s
turer
oc- lanes thas safety
claims that
is cone.whenthe
presented they are
electronic
in Destrian at risk
device on road
can
(2019). The beor highway
mounted
manufac-
not be in
cupied area
authorized the permitted
orpermitted timeframe.
the section-wise
worksite extent may roadworkTherefore,
differ frompermission real-time
theaactual may on anythas
oc- turer
lanes roadis
claims that Connectivity
the electronic
presented in Destrian is device
provided
(2019). through
canThebe manufac-Blue-
mounted
not be in
cupied areathe or the timeframe.
section-wise Therefore,
roadwork permission real-time
may on anyclaims
turer road cone. Connectivity
that between
the electronic is device
provided canthrough
bedevices Blue-
mounted
information
not be
cupied in
areathe on
or the
permitted
the exact geographical
timeframe.
section-wise extent
Therefore,
roadwork aof
permission each
real-time
mayof tooth
on
turerany technology
road
claims cone.
that Connectivity
the single
electronic is self-powered
provided
device can through
be mounted in
Blue-
information
not be in theonon the
permitted exact geographical
timeframe. extent
Therefore, of
aof each
real-time of tooth
on any technology
road cone. between
Connectivity single is self-powered
provided devices
through in
Blue-
the authorized
information
not be in theon and
the
permitted active
exact construction
geographical
timeframe. sites
extent
Therefore, in the range
each of close
tooth
on anydistances
technology
road cone. and 4G
between LTE-M
Connectivity single and Narrowband
self-powered
is self-powered
provided Internet
devices
through in
Blue-
the authorized
information and
the active
exact construction
geographical sites
extent inaof real-time
the range
each of close
tooth distances
technology andbetween
4G LTE-M single and Narrowband Internet
devices in
of
thetheir authority
authorized andis of great
active advantage
construction for both
sites in the urban
theeachrange of Things
close
of tooth distances(NB-IoT).
technology and 4G A solar
LTE-M rechargeable
and Narrowband warning light
Internet
information
of
thetheir
authorized
planning
on and
authority
department
the exact
greatgeographical
is active
of advantage for
construction
and construction
public roadfor
extent
both
sites in of
authorities.
the
the urban
range
The of
of Things
close
called distances
SmartBLight andbetween
(NB-IoT). 4G ALTE-M
has
solarsingle
been brought and self-powered
rechargeable
Narrowband
to market
devices
warning Internet
inInternet
in
light
thelight
US
of
thetheir authority
authorized
planning and
department is active
of great
and advantage
public road both
sites in
authorities.the urban
range
The Things
close
called distances(NB-IoT).
SmartBLight and 4G ALTE-M
has solar
been rechargeable
and
brought Narrowband
to warning
market in the US
of their authority
requirements for theis ofsensor
greatdevice
advantageinclude for the the urban of
bothcapabilities thatThings
features (NB-IoT).
likewisehas A solar rechargeable
connectivity to a cloud warning
server. light
The
planning
of department
their authority and device
is ofsensor
great public
advantage roadforauthorities.
bothcapabilities
the urbanThe of called
that SmartBLight
Things
features (NB-IoT).
likewisehas A been brought
solar
connectivity to ato
rechargeable market
warningin thelight
US
requirements
planning
to collect
requirements
for the
department
geographical
for the and device
position
sensor public
by
include
roadaccuracy
such
include
the
authorities.
the to The called
detect
capabilities that SmartBLight
functionality
features includes
likewise been brought
localization
connectivity to atocloud
service, market
cloud
server.
in theThe
operational
server. US
sta-
The
planning
to department
collect geographical
requirements for the and
position
sensor public road
by such
device include authorities.
accuracy
the The
to detect that
capabilities called SmartBLight
functionality
features includes
likewise has been
localization
connectivity brought to ato
service, market
cloud in
operational
server.the US
sta-
The
blocked
to lanes
collectlanes and thesensor
geographical road
positioncurvature.
by such Likewise,
accuracy the detec- that
to detect tus update,
featuresover-the-air firmware maintenance and remote sta-
requirements
blocked
to collect for
andthe
geographical the road
position device
curvature. include
byofsuch Likewise,
accuracy the detec- functionality
the capabilities
to detect tus update,
functionality
includes
likewise localization
over-the-air
includes connectivity
localization
service,
firmware maintenanceto a cloud
service,
operational
server.
and remote
operational The
sta-
tion of
blocked
to opening
collectlanes andand
geographical theclosure
road
position times
curvature.
byofsuchthe road
Likewise,
accuracy construction
the detec-
to detect activation
tus update,
functionality and deactivation
over-the-air
includes firmware
localizationof the warning
maintenance
service, light
and
operational such
remoteas
sta-
tion of opening
blocked lanes andand theclosure
road times
curvature. the road construction
Likewise, the detec- tus activation
update, and deactivation
over-the-air firmwareof the warning light
maintenance and such
remoteas
site are
tion of
blocked measured
opening
lanes andand by
the evaluation
closure
road times
curvature.of the
of the device
road
Likewise, messages
construction
the detec- Selevan
activation
tus update, (2020).
and The above
deactivation
over-the-air described
firmwareof the connected
warning
maintenance lightintelligent
and such
remoteas
site
tion are
of measured
opening and by evaluation
closure times of the
of theroaddeviceconstruction Selevan (2020).
messages activation and The above described
deactivation of the connected
warning lightintelligent
such as
that are
site
tion are measured
of transmittedandby during
evaluationoperation.
of thethe device barricade
Selevan (2020).
messages activation lights haveabove
The
and deactivation to be ofdiscerned
described fromlight
connected
the warning other novel
intelligent
such as
that areopening
site are transmitted
measured closure
by during times
evaluation of
of theroad
operation. deviceconstruction
messages Selevan barricade lights
(2020). haveabove
The to be discerned
described from other
connected novel
intelligent
that
site are
are transmitted
measured by during
evaluationoperation.
of the device messages warning
barricade
Selevan lights
lights
(2020). on the
have
The market
to
above be that
described are
discerned equipped
from
connected with
other GNSS
novel
intelligent
that are transmitted duringOF operation. warning
barricade lights
lights on the
have market that are
to besynchronization equipped
discerned from the with
other GNSS
novel
2. STATE THE ART receiver lights
warning
barricade for the
lights reason
on have to of
the market that are equipped with blinking
GNSS
that are transmitted 2. STATEduringOF operation.
THE ART receiver lights
warning for theon reason
the market ofbesynchronization
discerned
that areautomaticfrom the
equipped other
with novel
blinking
GNSS
2. STATE OF THE ART frequency
receiver
warning among
for the multiple
on reason lights
oflights and
synchronization equippedthe setting the
blinking
2. STATE OF THE ART frequencylights
receiver among
for the the market
multiple
reason ofan thatand areautomatic
synchronization with
thesettingGNSSthe
blinking
Developments for digitization
2. STATE OF THEof roadworks
ART safety man- order
frequency of flashing
among strobes
multiple lightsinstalled
and line
automatic of lights
setting such
the
Developments for digitization of roadworks safety man- receiver order of for
frequency the reason
flashing
among strobesofan
multiple synchronization
lightsinstalled
and line ofthe
automatic blinking
lights
setting such
the
agement have been
Developments for progressed inofseveral
digitization roadworks leading industrial
safety man- as presented
order of by
flashing Rea (2018).
strobes an Those
installed are designed
line of to
lights easily
such
agement have been
Developments progressed inofseveral
for digitization roadworks leading man- frequency
industrial
safety as presented
order among
of flashingby Reamultiple
(2018).
strobes lights
an Those andare
installed automatic
designed
line setting
of lights the
to easily
such
agement have been
Developments for progressed inofseveral
digitization roadworks leading industrial
safety man- as presented
order of by Rea
flashing (2018).
strobes an Those
installed are line
designed
of to easily
lights such
agement have been progressed in several leading industrial as presented by Rea (2018). Those are designed to easily
Copyright
agement © 2018been
2405-8963 have
Copyright IFAC
© progressed
2021 The Authors.in several
This isleading
an open industrial as presented
access article under the CC BY-NC-NDby Realicense
(2018).. Those are designed to easily
Copyright © 2018 IFAC
Peer review
Copyright © under responsibility of International Federation of Automatic Control.
2018 IFAC
10.1016/j.ifacol.2021.10.030
178 Olaf Czogalla et al. / IFAC PapersOnLine 54-4 (2021) 177–182
makes use of differential GNSS technology. Therefore, the

developed solution takes advantage of the high precise real-
time positioning service SAPOSthat
R is available for use
all over Germany Jahn (2020). The satellite positioning
service SAPOSisR a joint project of the Working Group of
the Surveying Authorities of the States of the Federal Re-
public of Germany (AdV). SAPOSprovidesR the official
geodetic spatial reference in the form of a Germany-wide
network of GNSS reference stations and correction data by
the international standard RTCM SC-104 Kendal (2010)
(US-Radio Technical Commission for Maritime Services
Special Committee No. 104) that are transmitted to the
mobile GNSS receiver over the internet in cycles of one
second. By this technology, the required real-time position-
ing accuracy of less than one meter is achieved, provided
that reception conditions of both GNSS and correction
data is guaranteed. Within urban settings the reception
conditions are often affected by buildings, trees and other
Fig. 1. Open IoT-Sensor beacon mounted on vertical obstacles that cause multi-path signal propagation. In
panel barricade. The additional weight of the sensor cases of such external surrounding conditions, the required
printed circuit board is within certified tolerance of accuracy cannot be warranted. However, the practical field
the barricade light. Flashing LED sits in focus of the tests have shown that sub-meter precision is achieved in
Fresnel lens lid allowing nightly warning strobe light. 75% of overall measured position fixes.
deploy and create a unique intelligent wireless sequential
operation of lights in forming a clear directional path 4. ARCHITECTURE AND IMPLEMENTATION
for traffic during lane closures, but have no localization
function and cloud connectivity. The electronic components of the IoT-sensor board were
designed to be encapsulated for proper operation of the
3. REQUIREMENTS OF DESIGN sensor in such a way that all specified requirements of
the system are met. Such requirements include shock re-
The deployment of high sensitive electronic components in sistance, theft protection, weather resistance, shock resis-
rough roadwork environment requires a rugged design of tance/impact safety, inconspicuousness and the simplest
the end product. The applied design followed the principle possible handling during use and maintenance. By in-
to integrate sensor components into existing casings of stalling the sensor board in a barricade light housing,
roadworks barrier and warning equipment as shown in it is guaranteed that the hardware and electronics are
Figure 1. Barricade lights are top-mounted on vertical protected from rain and splash water. For the integration
panel barricades and safety barriers. By installing the into existing housing solutions it was established a close
sensor board in a warning light housing, it is guaranteed coordination with respective manufacturers of roadworks
that the hardware/electronics are protected from rain and safety systems. Also, the German Federal Highway Re-
splash water. The warning light housing is screwed onto search Institute as the responsible certification author-
the intended support of the warning beacons with a special ity was consulted during the design and test process, so
lockable key. that existing regulations and granted permissions could
be taken into account. The main components of the IoT-
The casing of the flashing light is waterproof and provides sensor board are shown in the schematic diagram of Figure
a standardized energy supply of 6-volts lantern battery 2. The main microcontroller unit (MCU) contains the
4R25 that provides sufficient energy for the sensor board. LTE (4G) modem and the Narrowband-IoT communica-
In practical roadworks safety and barricading procedures tion module using LTE-M (machine) type communication
a cyclic exchange of worn batteries is ensured. The use standard for wide area network with low power require-
of rechargeable batteries is not yet a common practice ments. For the primary task of positioning is was selected
the roadworks barrier service business. The design of the a single band GNSS locating module that is connected over
sensor board was led by the principle of re-usability and a serial connection to the MCU. The Inertial Measurement
versatility. Re-usability refers to the replacement of the Unit (IMU) includes a temperature sensor and works as
sensor board into recycled or replaced barricade light movement detector and ambient temperature measure-
casings. Versatility refers to the ability to install the ment unit. They were selected based on the requirements
sensor board into casings from various manufacturers. The of the corresponding system specification and are con-
affiliated developer group has established a collaboration nected in between by I2 C-bus. The battery voltage mea-
with leading German manufacturers. A series production surement unit observes the power supply and shuts down
of the sensor beacon is imminently feasible. the device in case of low battery capacity after sending
a warning message. External universal wired and wireless
3.1 Essential precision objectives interfaces exist via USB, Bluetooth for communications
between nearly located sensors, LTE (4G) and GNSS. The
The described application in the area of roadworks safety interfaces meet the requirements regarding the amount of
management spurs high objectives in terms of positioning data to be transmitted, transmission cost efficiency and
accuracy. These requirements led to a sensor design that real-time capability. For transmission of the location data
Olaf Czogalla et al. / IFAC PapersOnLine 54-4 (2021) 177–182 179
as well as for the remote configuration, message protocols the developer’s institute to carry out and evaluate the
are used that have been specially developed for the use of accuracy of positioning. In the test field, an open area
IoT components. with high-precision surveyed reference points was selected,
which allows for a simple and easy evaluation of the
4.1 Sensor based IoT-connectivity positioning results. Realistic scenarios such as positioning
near the walls of houses were also simulated and probed in
The availability of LTE-M as a Narrowband cellular stan- the test field. Subsequently, the field tests were extended
dard for the use of WAN connectivity of the IoT device to urban areas in real street space. In a first scenario a
is essential for both low-cost data transfer and highly number of 10 sensor beacons were deployed in an open
reliable correction data reception as a precondition of high- street space with relatively fair conditions with respect
precision localisation. LTE-M is data transfer standard to GPS reception and cellular connectivity. These implied
that was just recently introduced after roll-out of LTE mostly open sky conditions with minor obstructed field of
in Europe. The advantage of LTE-M over Narrowband- view by trees and buildings as well as sufficient distance to
IoT (NB-IoT) is its comparatively higher data rate and metal fences or walls. In subsequent scenarios, conducted
lower latencies but requires more bandwidth. In an earlier over the period of eight months, also less advantageous
stage of the field test the IoT sensor beacon was operated circumstances were tested in which up tp 50 sensor beacons
with NB-IoT. During that period the connectivity was were placed in narrow streets closer to trees and buildings.
frequently interfered leading to interruptions of data trans- The term of spatiotemporal interference in the context of
fer, more often re-connect attempts and therefore faster this paper is used with respect to interference of road traf-
battery drain. The stability of data transfer increased fic by temporal restrictions of road space due to construc-
significantly by altering the data transfer method to LTE- tion measures. In order or detect the section-by-section
M standard in the IoT-device firmware. progress of roadwork contruction sites, as spatiotemporal
interference with road traffic, identical sensor beacons were
4.2 Server side IoT-Platform software deployed for a moving road construction section.
The core component of the IoT-platform is a server appli-

cation that runs on a server with an Internet connection. 5.2 Battery drain characteristics
The tasks of the IoT-platform consist of receiving data
from the connected sensor beacon, processing it and mak-
ing it available to third parties, such as data platforms The energy supply of the IoT-Sensor beacon is provided
of municipal authorities or commercial vendors and ser- through the 6-volts lantern battery 4R25 as the standard
vice providers of roadwork safety equipment. Appropriate low-cost power source for barricade lights used by various
software modules were developed for this purpose. The manufacturers of roadworks safety equipment. Lantern
data communication with beacons takes place within a batteries are capsulated in a waterproof case and con-
VPN. The position and attribute data are transmitted to tain four 1.5 V cells of size D zinc-carbon batteries. The
the IoT software platform as well as to the urban data battery block provides a capacity of approximately 8000
platform and are displayed at the front-end system of the milli-ampere hours (mAh). The discharge characteristics
IoT platform in a map-based representation, see Figure 3. of zinc–carbon elements are determined by the length and
The browser-based front-end software of the IoT-platform frequency of discharge cycles and the discharge current.
visualizes the positioning data by forming the geographical The zinc-carbon cell features a typical voltage relaxation
extent of transmitted sensor beacon positions. The algo- behaviour. Under load conditions the voltage drops ap-
rithm is mapping the site beacon positions as a polygon. proximately from 0.5 volts up to 0.8 volts depending on
For each sensor beacon that is put into operation the algo- the remaining voltage under no-load conditions. The less
rithm adds a point to the polygon. The algorithm automat- remaining voltage is measured the higher is the voltage
ically assigns the positions to a geometry by evaluating the drop under load conditions. If the voltage falls under 4
probability of belonging to a construction site object. For volts, the power management of the beacon shuts down
this purpose, the spatial relations of the captured beacon power supply, such as shown in Figure 3.
positions are evaluated. The algorithm is implemented on During the practical field tests the frequency of data
the server side. transmission from the sensor beacon was configured with
24 transmissions a day significantly higher than it is
5. RESULTS planned for future regular operation at 2 transmissions a
day. Therefore, a general forecast for duration under real
5.1 Field testing scenarios deployment conditions can be given under the assumptions
of a 12h idle cycle. In order to provide a rule of thumb
The goals of the field tests were to test the developed sen- for remaining beacon run time a laboratory test was
sor beacon hardware together with the associated firmware carried out under near-real conditions with a scheduled
and IoT-platform software under the realistic conditions. transmission cycle of 12 h. After battery replacement the
The important test object was the communication of sensor beacon was in continuous operation for 6 weeks. For
the GNB IoT-sensor with the central IoT-platform (IoT- runtime forecast of the beacon a drop of 0.036 volts per
server) under real conditions. The main questions about day or, respectively 26 days per volt drop was determined.
the availability of the communication channel, its latency Thus, the remaining runtime can be approximated by
and error behavior were tested and evaluated. The first subtracting the actual voltage under load with 4 volts
tests were carried out in a test site at the premises of (shutdown) and multiplying the remainder by 26 days.
Fig. 2. Schematic diagram of sensor printed circuit board.
Fig. 3. Battery drain curve of a single beacon from Oct 1st, 2020 till Dec 20, 2020, blue line indicates recovered voltage
under no load condition, brown line indicates voltage under full load condition, straight lines mark the idle time
interval from last time of transmission until battery replacement.
5.3 Precision analysis the installation of the sensor close to a metal fence causing
magnetic interference for the correction data received by
To analyse the precision of the sensor technology a compre- LTE. The histogram in part B of Figure 4 indicates the
hensive field test under real conditions was conducted over frequency of deviation distances for a bucket width of
a timeframe of eight months. Multiple urban roadwork 15 centimetres. Spectral peaks occur at 1.5 meters, 2.5
construction sites were equipped with sensor beacons to meters and 5 meters in the distribution visibly and indicate
delineate the blocked construction area. In order to obtain sensor movements to positions in a greater distance from
the reference position for single sensor beacons, they were the reference position. If only the red cluster positions
additionally surveyed with a high-precision positioning (n=219) from unmoved sensor beacon are considered for
device: a survey RTK rover. The highly accurate reference precision analysis in comparison of the reference position,
positions were compared with sensor beacon positions over the 75% quantile is 0.56. This means that 75% of the
a long term period of several weeks. The analysis led beacon positions are within a distance of 0.56 meters
to valuable interpretations of the results resulting in a from the reference point, respectively the median, that
number of conclusions. In Figure 4 A is shown an excerpt is 50% of the measured positions, are within a distance
of the field test for a single sensor beacon in the time from of 0.31 meters from the reference point. In result of the
mid-March until end-June 2020 at a road construction analysis the following reasons for significant deviations of
site. During that deployment the sensor must have been collected positioning data to exact reference positions were
replaced by construction personnel several times. Within identified:
the timeframe of 48 days a total of n=2014 positions 1. Poor cellular connectivity, resulting in no or frequently
were collected in an hourly interval. The coloured clusters interrupted connections to the SAPOS server and thus no
indicate the re-positioning of the beacon during the entire correction of the position possible using the correction data
operational time, which can be made visible by a post- This case occurred in connection with the mobile network
animation over the time course of the recorded positions. connection using Narrow Band-IoT during the time of use
The movement took place from the red cluster to the top in the firmware of the sensor beacon. Especially in certain
green, further to black, to right green and finally to the locations connectivity problems occurred frequently over
blue cluster, that are all located in an extent of approxi- the test period, as shown in Figure 5. The high number
mately 10 meters. Each of the identified clusters feature a of green dots which represent uncorrected positions of the
relatively small scattered distribution except of the right quality level Q4 indicate the effect on positioning accuracy.
green cluster. The reason for this wider distribution was
Olaf Czogalla et al. / IFAC PapersOnLine 54-4 (2021) 177–182 181
Fig. 4. A: Moving sensor beacon positions collected in a field test: red = first , top green=second, black=third, right
green=fourth, blue=fifth position, black cross (red cluster)=reference position, B: Histogram of deviations from
reference positions, spectral peaks indicate sensor movements.
However, inaccuracies due to poor cellular connections and location of the beacon falsify the result. The algorithm
a resulting lack of position correction decreased signifi- for the post-processing process is shown in Algorithm 5.4
cantly from the time when LTE-M was used for commu- and was implemented as described below.
nication.
2. Interference due to unfavourable ambient conditions Algorithm 1: Sensor data post-processing
Poor ambient conditions, such as trees, passing vehicles or Input: Pi set of positioning data for sensor i
metal construction fences can lead to significant deviations Output: pi resulting center of gravity
from the reference position that cannot be corrected. for all i do
Especially in the shown example in Figure x it is noted that - restrict data of P osi as current position to
the sensor beacon was located very close to a metal barrier position of last sensor movement;
fence. The image shows here a clear one-sided scattering in - restrict resulting data set to interval of x days;
western direction, away from the metal fence, which was - restrict resulting data set to n data tuples;
located east of the beacon close to the roadside. - calculate the center of gravity of remaining data
position coordinates: pi ;
3. General scattering of the positions In general, it cannot - return result pi ;
be guaranteed that a measured position - even if it has the end
highest quality level - is within the 50 cm radius. Even with
a good line of sight to satellites, a passing vehicle can cause
a position measured with good quality to be distorted and 6. DETECTION OF SPATIOTEMPORAL
thus lie outside the intended radius. Basically, there will VARIATIONS OF ROADWORKS AREA
always be measured positions that are not within the 50 cm
radius, but with a post processing these can be corrected.
Road construction measures with the associated excava-
tions usually are divided into spatially successive con-
5.4 Post processing
struction sections. In the course of processing these road
construction measures, the immediate work areas, which
The post-processing procedure was carried out for several are marked and delimited by the sensor warning lights, are
sensors with data of a period of 2 weeks. Figure 5 shows the successively shifted.
original positions and the positions of one sensor beacon
corrected by post-processing. The yellow circle marks the When the work area changes, the sensor warning lights
50 cm diameter around the respective reference position. are now to be used to automatically detect a shift in the
Q4 position data were filtered out beforehand, they are construction site area. Thus, a change in the marked area
not included in the calculation. Almost 100% accuracy of at the time of the change is detected in real time and
the 50 cm diameter was achieved by post-processing. The transmitted for processing and display on a map-based
decisive factor here is how the data is distributed, and browser application.
any movement of the beacon has not yet been taken into
The installed sensors at the construction barriers present
account.
their position, location with an average radial deviation of
Therefore, the time of the latest movement of the referring 50 cm, activation time and activation end (dismantling of
sensor beacon is saved in the database, to be able to discern the construction site) in an automated, near real-time as
sensor collected positions in the post-processing process. It well as geo-referenced manner. The sensor data supplied
is important to note that all beacon points are included in and processed can thus represent the polygon or geometry,
the calculation, which were sent since the last movement i.e. the location of the individual beacons (correspond-
of the beacon. This prevents that points of a previous ing to the vertices of the polygon) of the construction
Fig. 5. A: Original positions and B: Corrected positions by post-processing of identical sensor beacon data.
project. It is possible to add additional data, parameters, the sensor beacon is provided through non-rechargeable
information (e.g. about the type of construction site or standard batteries that are preferred in barrier lights
traffic information defined within the project) at any time which poses no additional service operations for roadworks
(either directly at the sensor or in the supporting software safety equipment providers. The field tests hav proven
solution or by input frontend). The beacons equipped with the longevity of power supply under real conditions. The
the sensor system automatically assign themselves to a impact of lower temperatures in winter season on battery
construction site. A unique job site ID for this job site durability is subject of long-term tests which results will
is either created directly and sent along or formed in the be available in the near future.
IoT platform. Possibly, the ID can be directly assigned to
an approved construction measure through the interface ACKNOWLEDGEMENTS
to the urban platform by means of the corresponding
construction measure-specific internal authority ID. Research covered in this paper benefited from participa-
tion in the project “GeoNetBake”. The project is funded
The detection of sensor movements is achieved by use of by the German Federal Ministry of Transport and Digital
an internal measurement unit including accelerator sensors Infrastructure (BMVI) under grant number 16DKV20002
for x-y-z axes. For the purpose of movement detection and is co-funded among other projects by the City of
all three acceleleration axes are observed and evaluated. Hamburg. Thanks are expressed for the good cooperation
The acceleration signal is acquired with a sampling rate with the companies horizont group and Nissen Elektrobau
of 25 ms. The vertical signal component is subtracted without which the present research work could not have
by gravity (9.81m/s) and filtered through a first order been carried out so successfully.
low pass filter. Parameterizable thresholds are defined to
detect deviations of the signal that can be characterized REFERENCES
either as movements of the IoT sensor beacon or low- Sheckler, R. (2020). What You Need to
frequent movements caused by the influene of wind. Both Know About Connected Work Zones,
causes are clearly discerned by threshold values that are https://www.iconeproducts.com/news/what-you-
parametrized and can by adjusted in retrospect by over- need-to-know-about-connected-work-zones, ICone
the-air maintenance connection access of the IoT sensor Products LLC, Cato, NY, USA.
beacon. Destrian, Mathieu (2019). Smart cone to protect work-
ers on construction, building sites, roads lane cuts and
7. CONCLUSION railyards, https://intellinium.io/author/mdestrian/, In-
tellinium, Aix-en-Provence, France.
A stable high precision localisation of beacon positions Selevan JR. (2020), Smart B Light, Pi-lit https://www.pi-
requires both the utilization of RTK rover technology at lit.com/smart-light, PI lit R Pi Variables Inc, Tustin,
clear satellites line of sight and interference-free IoT con- CA, USA.
nectivity for correction data transfer. Without clear line Rea, M., Skinner, N., and Bullough, J. (2018) A Novel
of sight multipath reception of satellite signal take place, Barricade Warning Light System Using Wireless Com-
which harms the accuracy of positioning significantly. A munications, SAE Technical Paper 2018-01-5036, 2018,
stable LTE-M connectivity of correction signal reception https://doi.org/10.4271/2018-01-5036.
from the base station is essential for sub-meter accuracy. Jahn, C.-H., Riecken, J., Trautvetter, C., Freitag,
Given these both conditions, a sub-meter positioning accu- M., Kurtenbach, E., Fabian, G., Dick, H.-G.: Quo
racy of less than 50 cm can be achieved with approximately vadis SAPOS? – Zukünftige Entwicklungen des Posi-
75% of measured positions. Further, an important conclu- tionierungsdienstes der Landesvermessung, in DVW-
sions from the field trials it is resumed that the preferred Schriftenreihe Bd. 87.
design of fully integrated barrier light casing has proven Kendal, Furgerson: RTCM/Special Committee 104, Cur-
to withstand rugged operation conditions in roadwork rent Status and Focus. International Symposium on
zones. It is a requirement that handling procedures of GNSS, Space-based and Ground-based Augmentation
service personnel for battery replacement are maintained Systems and Applications, November 29-30, 2010, Brus-
as simple as possible. Therefore, the energy supply for sels, Belgium.
ScienceDirect
IFAC PapersOnLine 56-1 (2023) 132–137
Exploiting
Exploiting Different
Different Symmetries
Symmetries for
for
Exploiting
Exploiting
Trajectory Different
Different
Tracking Symmetries
Symmetries
Control with for
for
Exploiting
Trajectory
Exploiting
Trajectory Different
Tracking
Different
Tracking Symmetries
Control
Symmetries
Control with
with for
for
Trajectory Tracking
Application to Control
Quadrotors
with

Trajectory
Trajectory Tracking
Application to
Tracking
Application to Control
Quadrotors
Control with
Quadrotors with

Application
Application to
to Quadrotors
Quadrotors
Application
Matthew ∗
Hampsey Pieterto van
Quadrotors
Goor Tarek Hamel ∗∗
∗
Matthew Hampsey ∗ Pieter van Goor ∗ Tarek Hamel ∗∗
∗ ∗ ∗ ∗∗
Matthew
Matthew Hampsey
Hampsey Robert
∗ Pieter Mahony
van Goor ∗ ∗ Tarek
Tarek Hamel Hamel ∗∗
∗ Pieter
Robert van
Mahony Goor ∗ ∗ ∗∗
Matthew Hampsey
Matthew HampseyRobert ∗ Pieter
Robert van
Mahony
PieterMahonyvan GoorGoor ∗ ∗ Tarek Hamel ∗∗
Tarek Hamel
∗
∗ Robert Mahony ∗
∗ Systems Theory and Robotics
Robert Group,
Mahony School of Engineering,
∗ Systems
Australian Theory
National and Robotics
University, Group,
Australia School
(e-mail: of Engineering,
{first name}.{last
∗ Systems Theory and Robotics Group, School of Engineering,
∗ Systems
Australian Theory
University, Group,
Australia School
(e-mail: of Engineering,
{first name}.{last
∗ Systems
Australian Theory
National and Robotics Group, School of Engineering,
{first
Systems
Australian TheoryUniversity,
University, Australia
name}@anu.edu.au).
Group,
Australia
name}@anu.edu.au).
(e-mail:
School
(e-mail: of name}.{last
Engineering,
{first name}.{last
Australian
∗∗
I3S
Australian National
(University
National University,
Côte d’Azur,
University, Australia
name}@anu.edu.au).
CNRS,
AustraliaSophia(e-mail:
(e-mail: {first
Antipolis)
{first name}.{last
and
name}.{last Insitut
∗∗
∗∗ I3S (University Côte name}@anu.edu.au).
d’Azur, CNRS,
name}@anu.edu.au). Sophia Antipolis) and Insitut
∗∗ I3S (University
Universitaire deCôte
France,d’Azur,FranceCNRS, Sophia
(e-mail:
name}@anu.edu.au). Antipolis)
thamel@i3s.unice.fr). and Insitut
I3S (University Côte d’Azur, CNRS, Sophia
∗∗ Universitaire de France, France (e-mail: thamel@i3s.unice.fr). Antipolis) and Insitut
∗∗ I3S (University Côte d’Azur, CNRS, Sophia Antipolis)
Universitaire
I3S (Universityde
Universitaire de France,
Côte
France,d’Azur,France
France (e-mail:
CNRS, Sophia
(e-mail: Antipolis) and
thamel@i3s.unice.fr).
thamel@i3s.unice.fr). and Insitut
Insitut
Universitaire
Universitaire de de France,
France, FranceFrance (e-mail:
(e-mail: thamel@i3s.unice.fr).
thamel@i3s.unice.fr).
Abstract: High
Abstract: High performance
performance trajectory
trajectory tracking
tracking control
control of of quadrotor
quadrotor vehicles vehicles is is an an important
important
challenge in
Abstract:
Abstract:
challenge inHigh
aerialperformance
High
aerial robotics. Symmetry
performance
robotics. Symmetry
trajectory
trajectory is tracking
is a fundamental
fundamental
tracking
a control
control property
of
of quadrotor
quadrotor
property of physical
of physical
vehicles
vehicles systems
is
is
systems an
an and offers
important
important
and offers
Abstract:
the potential High
to performance
provide a tooltrajectory
to design tracking control
high-performance of quadrotor
control vehicles
algorithms. is an
We important
propose
challenge
Abstract:
challenge
the potential
challenge
in
in
in
aerial
High
aerial
to robotics.
performance
robotics.
provide
that a
Symmetry
trajectory
Symmetry
tool to is
is
design a fundamental
tracking control
a high-performance
fundamental property
of quadrotor
propertycontrol of
of physical
vehicles
physical
algorithms. systems
is
systems an
We and
important
and
propose offers
offers aa
design
the potential
challenge
the
designpotentialin aerial
methodology
to
aerial
to
methodology
robotics.
provide
robotics.
provide
that a
a Symmetry
takes
tool any
to
Symmetry
tool
takes to
any
is
is aa symmetry,
given
design
design
given
fundamental
symmetry,
high-performance
fundamental
high-performance property
linearises
property
linearises the
control
control
the
of physical
associated
algorithms.
of associated
physical
algorithms. systems
error
systems
error We
We inand
propose
and
propose
in a
offers
a single
single
offersaa
the
set of
design
the
designpotential
of
potential to
to provide
coordinates,
methodology
methodology and
thatuses
provide
that a
uses tool
takes
atakes LQR
tool to
to design
any
any high-performance
to obtain
given symmetry,
design
given obtain aalinearises
symmetry,
high-performance linearises control
high performance
performance algorithms.
the associated
control
the associated control;
algorithms. error
error We
We aninpropose
in singleaa
approach
aaa single
propose
set
design coordinates,
methodology and
that takes LQRany design
given to
symmetry, high
linearises the associated control; error an in approach
we
set
set
we term
designof
of
term Equivariant
coordinates,
methodology
coordinates,
Equivariant and
andRegulator
that uses
takes
uses
Regulator design.
LQR
LQR design
anydesign
design. givenWe
We show
to that
obtain
symmetry,
to
showobtain
that a quadrotor
high
linearises
aaquadrotor
high vehicles
performance
the
performance
vehicles admit
associated control;
control;
admit several
error
several an
an a single
different
approach
inapproachsingle
different
set
we
set
we of
term
of
term coordinates,
symmetries: the
Equivariant
coordinates,
Equivariant and
direct
and uses
product
Regulator
uses
Regulator LQR
LQR
design.design
symmetry,
design.designWe
We to
the
show
to
showobtain
extended
that
obtain
that a high
pose
quadrotor
high
quadrotor performance
symmetry vehicles
performance
vehicles and control;
admitthe
control;
admit pose
several
several an
and
an approach
velocity
different
approach
different
symmetries:
we term the direct product symmetry, the extended pose symmetry and the pose and velocity
we term Equivariant
symmetry,
symmetries:
symmetries:
symmetry, and
the
Equivariant
the
and show
direct
direct
show
Regulator
that
product
Regulator
product
that each
design.
eachsymmetry,
symmetry
design.
symmetry,
symmetry
We
Wethe show
can
show
thecan
that
be
extended
that
extended
be used
used
quadrotor
poseto symmetry
quadrotor
poseto define
symmetry
define
vehicles
aaapose
vehicles global
and
and
global
admit
admitthe
theerror. several
pose
several
pose
error. We
and
and
We
different
compare
velocity
different
velocity
compare
symmetries:
the linearised
symmetry,
symmetries:
symmetry, the
and
the
and direct
systems
show
direct
show product
via
that
product
that each symmetry,
simulation
each symmetry
symmetry,
symmetry and the
find
thecan
canextended
that
be
extended
be the
used
used pose
poseto
to symmetry
extended
define
symmetry
define a and
global
and
globaland the
the pose
pose
error.
pose
error. and
and
We
and
We velocity
compare
velocity
compare
the linearised
symmetry,
symmetries
the linearised and systems
show
outperform
systems via
that
the
via simulation
each
direct symmetry
simulation and
product
and find
can that
be
symmetry
find that the
used
in
the theextended
to define
presence
extended a pose
global
of
pose and
large
and pose
error.
disturbances.
pose and
We
and velocity
compare This
velocity
symmetry,
the linearised
symmetries and show that
systems
outperform via
the each
direct symmetry
simulation and
product can that
find be used
symmetry the
in to
the define apose
extended
presence global
of and
large error.
pose
disturbances.We
and compare
velocity
This
the
the linearised
suggests
symmetries that
linearised
symmetries systems
choices
outperform
systemsofvia
outperform ofvia
the simulation
equivariant
the direct
simulation
direct and
and
product
product find
group that
symmetry
andgroup affine
find that
symmetry the
in
the
in the
theextended
symmetries
presence
extended
presence pose
have
of
pose
of and
improved
large pose
disturbances.
and disturbances.
large pose and and velocity
linearisation This
velocity
This
suggests
symmetries that choices
outperform equivariant
the direct and affine symmetries have improved linearisation
error.
suggests
symmetries
suggests
error. that
that choices
outperform
choices of
of the direct product
equivariant
equivariant and
product
and symmetry
group
symmetry
group affine
affine in the
the presence
insymmetries
presencehave
symmetries of
of large
have improved
large
improved disturbances.
disturbances.linearisation
linearisation This
This
suggests
error. that choices of equivariant and group affine symmetries have improved linearisation
Copyright
suggests
error.
error. ©
that 2023 The
choices Authors. This
of equivariant is an open
and access
group article
affine under
symmetries the CC BY-NC-ND
have improved license linearisation
error.
(https://creativecommons.org
Keywords:
Keywords: Tracking, UAVs,
Tracking, UAVs,/licenses/by-nc-nd/4.0/)
Application of
Application of nonlinear
nonlinear analysis
analysis and and design,
design, Flying Flying Robots, Robots,
Guidance
Keywords:
Keywords: navigation
Tracking,
Tracking, and
UAVs, control
Application of nonlinear analysis and design, Flying Robots,
Guidance
Keywords:
Guidance Tracking, UAVs,
Keywords:navigation
Guidance navigation
Tracking,
navigation
and
UAVs,
and
UAVs,
and
Application
Application of
control
control
Application
control of nonlinear
of nonlinear analysis
nonlinear analysis and
analysis and design,
and design, Flying
design, Flying Robots,
Flying Robots,
Robots,
Guidance
Guidance navigation
navigation and
and control
control
1. INTRODUCTION
1. tion (Lee et al. (2010)). Using aa Lie group structure also
1. INTRODUCTION
1. INTRODUCTION
INTRODUCTION
tion
allows
tion
tion
(Lee
(Lee
(Leeone et
to
et
et
al.
work
al.
al.
(2010)).
in
(2010)).
(2010)). local Using
logarithmic
Using
Using a
a
Lie
Lie
Lie
group
error
group
group
structure
coordinates
structure
structure
also
in
also
also
1. INTRODUCTION
INTRODUCTION allows
ation (Lee
single
allows one to
et
chart.
one to work
al.In
work in
(2010)).
Farrell
in local
local et logarithmic
Usingal. aa Lie
(2019),
logarithmic theerror
group
error coordinates
structure
authors in
also
consider
coordinates in
Autonomous
Autonomous drones
drones1. have
have steadily
steadily grown
grown in
in popularity
popularity tion
allows
a (Lee
single one et
to
chart. al.
work
In (2010)).
in
Farrell local Using
et logarithmic
al. (2019), Lie group
theerror structure
coordinates
authors also
consider in
over past
Autonomous decades,
drones withhavemodernsteadilyapplications
grown in in a diverse
popularity allows
the
a single
single
allows
a one
quadrotor
one to
chart.
to
chart. work
In
work
Insystem in
Farrell
in
Farrell local
placed
local et logarithmic
al. on thethe
et logarithmic
al. (2019),
(2019), theerror
direct
error coordinates
authorsproduct
consider
coordinates
authors consider in
Lie
in
Autonomous
over past drones
decades, withhavemodernsteadily grown
applications in inpopularity
a diverse the
a quadrotor
single chart. Insystem
Farrell placed
et al. on
(2019), the the direct
authorsproduct
considerLie
Autonomous
array
over of
past drones
domains,
decades, have
including
with modernsteadily
photography, grown
applications in popularity
cinematography,
in a diverse the
group
a quadrotor
single
the SO(3)
chart.
quadrotor ×In R
system 3
Farrell
system × R 3
placed
. The on
3et al. (2019),
placed on the
corresponding
the the direct
authors
direct product
local
consider
product Lie
loga-
Lie
Autonomous
over
array past
of domains,drones
decades,
domains, withhave
including modernsteadily
photography, grown
applications in inpopularity
a diverse group SO(3) × R 3
3 × areR . The corresponding local loga-
over
over
array past
surveying,
array of
past
of
surveying,
decades,
inspection,
decades,
domains, with
with
including
inspection,
modern
delivery,
including modern
delivery, and
photography,racing.cinematography,
applications
applications
photography,
and racing.
in aa diverse
Quadrotors
cinematography,
in
cinematography,
Quadrotors are the
diverse
are
rithmic
group
the
group
rithmic
quadrotor
coordinates
SO(3)
quadrotor
SO(3)
coordinates
×
×
system
R
system
R 3 ×
3 × are
R
R
placed
3used
3 . The
placed
.. The
3used
toon
toon the
the aadirect
track
corresponding
corresponding
track direct product
trajectory local
product
trajectory local with Lie
Liea
loga-
loga-
with a
array
a of
surveying, domains,
particularly including
compelling
inspection, photography,
research
delivery, and and
racing. cinematography,
development
Quadrotors plat-
are group
standard
rithmic
group
rithmic SO(3)
SO(3)LQR
coordinates
coordinates×
× R
R 3 ×
formulation.are
×
are R
R 3used
.
used The
The to
to corresponding
track
corresponding
track a
a trajectory
trajectory local
local loga-
with
loga-
with a
a
array
a of
surveying, domains,
particularly including
inspection,
compelling delivery,photography,
researchand racing.
and cinematography,
Quadrotors
development are
plat- standard
rithmic LQR
coordinates formulation.are used to track a trajectory with a
asurveying,
form owing
particularly
surveying,
aform
particularlyinspection,
to mechanical
compelling
inspection,
compelling delivery,
delivery,
researchand
simplicity
researchand andracing.
and and
racing. Quadrotors
versatile
development
Quadrotors
development are
flight
plat-
plat- standard
rithmic
are standard LQR
coordinates
LQR formulation.
formulation.are used to track a trajectory with a
a owing
particularly
dynamics.
form owing A to
tokeymechanical
compelling
goal
mechanical in simplicity
research
quadrotor
simplicity and and versatile
development
control
and is tracking
versatile flight
plat-
flight a In parallel
standard LQR to the control
formulation. work, significant progress has
a
formparticularly
owing
dynamics. to compelling
Atokey
keymechanical
goal research
simplicity
inpresence
quadrotor and development
and
control versatile
is tracking plat-
flight a beenIn parallel
standardmade in
In parallel
parallel LQR to the
to exploiting control
formulation.
controlsymmetry
the control work, significant
work, significant for observer
significant progress has
and filter
progress has
form
desiredowing
dynamics. trajectory
A mechanical
in the
goal thein simplicity
quadrotor and
and versatile
of external
external
control disturbances
is trackingflight
flightaa In been made to
in the
exploiting work,
symmetry for progress has
form
desiredowing
dynamics.
dynamics.
and state
desired
A
A
tokey
trajectory
key
mechanical
goal
in
goal
perturbations.
trajectory in the
in
in
simplicity
quadrotor
presence
quadrotor
This
presence of
problem
of
control
control
has
external
versatile
a
is
is tracking
disturbances
tracking
long history
disturbances a In
In
beenparallel
design.
been made
parallel
made to
In Mahony
in
to
in the
exploiting
the
exploiting control
et al. (2008),
control work,
symmetry
work,
symmetry the Lie observer
significant
for
significant
for groupprogress
observer
observer
and
structure
and
progress
and
filter
has
of
filter
has
filter
dynamics.
desired
and state A key goal
trajectory in
perturbations. theinpresence
quadrotor
This of
problem control
external
has ais tracking
disturbances
long history a beendesign.
design. madeIn
In Mahony
in
Mahony exploiting et
et al.
al. (2008),
symmetry
(2008), the
the Lie
for
Lie group
observer
group structure
and
structure of
filter
of
desired
in
and general
state trajectory
robotics in
perturbations. the
and presence
several
This of external
classical
problem has disturbances
control
a long design
history SO(3)
been
design. is
madeIn used in
Mahony to design
exploiting et al.a complimentary
symmetry
(2008), the for
Lie attitude
observer
group filter.
and
structure In
filter
of
desired
and
in state
generaltrajectory
robotics in
perturbations. the
and presence
This
several of
problem external
classical has a disturbances
long
control history
design SO(3)
design.
SO(3) is
In
is used
Mahony
used to
to design
design et al.a
a complimentary
(2008),
complimentary the Lie attitude
group
attitude filter.
structure
filter. In
of
In
and state
methodologies
in general perturbations.
are
robotics available.
and This problem
One
several of the
classical has
mosta long
control history
studied ap-
design Bonnabel
design.
SO(3) In
is et
Mahony
used al.to (2008),
design et al.aan invariant
(2008),
complimentary the observer
Lie group
attitudeis proposed
structure
filter. of
In
and
in state perturbations.
general
methodologies robotics
are and
available.This
several problem
One classical
of the has
mosta long
control history
studieddesign
ap- Bonnabel
SO(3)
by
Bonnabel is
exploiting et
used
et al.
al.to
the(2008),
design
(2008),structureaan
an invariant
complimentary
of the
invariant observer
unit attitudeis
quaternions.
observer is proposed
filter.
The
proposed In
in
in general
proaches
methodologies
general
methodologies is robotics
the
robotics
are and
application
are available.
and
available.several
of
several the
One
One classical
ofLinear
the
classical
of the control
Quadratic
most studied
control
most studieddesign
Reg-
ap-
design
ap- SO(3)
Bonnabel
by is
exploiting used
et al.to
the design
(2008),structureaan complimentary
invariant
of the observer
unit attitudeis
quaternions. filter.
proposed
TheIn
proaches is the
thein application of theof Linear Quadratic Reg- Bonnabel
group et SEal. (2008), an
methodologies
ulator (LQR)
proaches (LQR)
methodologies
proaches
ulator is
is the are
are
in
available.of
trajectory
application
available.
application
trajectory ofOne
error
the
One
the
error of the most
coordinates
Linear
the
Linear
coordinates
most studied
((Anderson
Quadratic
studied
Quadratic
((Anderson
ap- Lie
Reg-
ap-
Reg-
by
by
Lie
exploiting
Bonnabel
exploiting
group et
SE al. (3)
2the
2the
(2008),
(3)
structure
structure
is an invariant
is introduced
introduced
of
of the
invariant
the in unit
in
observer
Barrau
observer
unit
Barrau
andis
quaternions.
quaternions.
and
proposed
is BonnabelThe
proposed
BonnabelThe
proaches
ulator
and is
(LQR)
Moore, the in
2007, application
trajectory
chapter 4. of the
error
Tracking Linear
coordinates Quadratic
Systems)). ((AndersonReg-
However, by
Lie
by
Lie exploiting
(2015) group to
exploiting
group SE
design
SE 2the
(3)
the
(3)a structure
is introduced
filter
structure
is for
introduced of the
inertial
of the in
in unit
Barrau
unit
Barrau quaternions.
navigation. and
quaternions.
and Bonnabel
This
BonnabelThe
work
The
proaches
ulator
and Moore, is
(LQR) the in
2007, application
trajectory
chapter of the
error
4. Tracking
Tracking Linear
coordinates Quadratic
Systems)). ((AndersonReg-
However, (2015)
Lie group to design
SE 2
2 (3)
a filter
is for
introduced inertial in navigation.
Barrau and This work
Bonnabel
ulator
classical
and
ulator (LQR)
Moore, LQR
(LQR) in
2007,is
in trajectory
posed
chapter
trajectory for
4. error
systems
error coordinates
on R m ((Anderson
Systems)).
coordinates whereas
However,
((Anderson the is
Liefurther
(2015)
(2015) group to
to developed
design
SE
design (3)a
a in
filter
is
filter the
for
introduced
for landmark
inertial
inertial in paper
navigation.
Barrau
navigation. and Barrau
This
Bonnabel
This and
work
work
and
and Moore,
classical
Moore,
natural
classical LQR
state
LQR
2007,
2007,is
space
chapter
posed
chapter
is chapter
posed
4.
for Tracking
systems
4. Tracking
of a quadrotor
for Tracking
systems on R
Rm
m
Systems)).
Systems)).
is aon
on smooth
However,
whereas
However,
manifold.
m whereas
the
the
is further
(2015)
Bonnabel
is further to developed
design
(2017),
2
developed aawhich in
filter
in the
for
proves
the landmark
inertial
landmark paper
thatnavigation.
thepaper Barrau
This
local Barrau and
work
logarithmic and
and Moore,
classical
natural LQR
state2007,is
space posed
of a 4.
for systems
quadrotor is a R
Systems)).
smooth However,
whereas
manifold. the (2015)
is further
Bonnabel to design
developed
(2017), filter
which in for
the
provesinertial
landmarkthat navigation.
the paper
local This
Barrau work
logarithmic and
classical
natural LQR
state is
space posed
of a for systems
quadrotor is on
a R m
m whereas
smooth manifold. the is
is further
linearisation
Bonnabel
further
Bonnabel developed
is
(2017),
developed
(2017), exact which
which in
for
in the
group
proves
the
proves landmark
affine
landmarkthat
that the paper
systems.
the local
paper
local Barrau
More
Barrau
logarithmic and
recent
logarithmic and
classical
natural LQR
state
If a dynamical is
space posed
system of a for systems
quadrotor
is quadrotor is
posed on aissmooth on
a R
smooth whereas
manifold.
manifold the
that work linearisation
Bonnabel in
linearisation is
(2017),
this exact
field
is exact which
showsfor
for group
proves
that
group affine
group that
affine systems.
the
affine
systems.local
and More recent
logarithmic
equivariant
More recent
natural
If a
natural state
dynamical
state space
system
space of
of a
is
a posed
quadrotor on a is a smooth
smooth
a smooth manifold.
manifold
manifold. that Bonnabel
linearisation
work in (2017),
this is exact
field which
showsfor proves
group
that group thataffine
affine the local
systems. and logarithmic
More recent
equivariant
admits
If
If a
a a Lie group
dynamical
dynamical system
system symmetry,
is
is posed
posed then
on
on a
a the group
smooth
smooth symmetry
manifold
manifold that
that linearisation
symmetries
work in
linearisation this is
is exact
lead
field toshows
exact for
for group
simplifiedthat
group affine
error
group
affine systems.
dynamics
affine
systems. and andMore recent
improved
equivariant
More recent
admits
If
cana a
be used
admits a Lie
dynamical
Lie togroup
definesymmetry,
system
group is
errorposed
symmetry, then
on a
coordinates
then the
smooth
the group
for symmetry
manifold
control
group symmetry that
design. work
symmetries
work in
in this
this field
lead
field to shows
simplified
shows that
that group
error
group affine
dynamics
affine and
and and equivariant
improved
equivariant
If
cana dynamical
admits
be a Lie
used to system
group
define is
errorposed
symmetry, on a
then
coordinates smooth
the manifold
group
for controlsymmetry that
design. linearisation
symmetries
work
symmetriesin this error
lead
field
lead to
to (Barrau
simplified
shows
simplifiedthat and Bonnabel
error
group
error dynamics
affine
dynamics (2020),
and and
and Ng et al.
improved
equivariant
improved
admits a linearisation error (Barrau and Bonnabel (2020), Ng et al.
This
can be
admits
can
This be a Lie
approach
used
Lieto
used
approach togroup
define
group
define
is
symmetry,
is powerful,
error
symmetry,
error
powerful,
and
and
then
coordinates
then
coordinateswasthe
was
group
theoriginally
for control
group
for control
originally
symmetry
used
design.
symmetry
design.
used
to (2022),
to
symmetries
linearisation
symmetries
linearisation van lead error
Goor
lead
error to
to et simplified
(Barrau
al. (2021)).
simplified
(Barrau and
anderror
Bonnabel
error
Bonnabel dynamics
dynamics and
(2020),
and
(2020), improved
Ng et
improved
Ng et al.
al.
can
can be
derive
This be used
an
used to
to define
almost-globally
approach is
define error
powerful,
error coordinates
stable
and non-linear
coordinateswas for control
attitude
originally
for control design.
used con-
design. to (2022),
linearisation
(2022), van
van Goor
error
Goor et
et al.
(Barrau
al. (2021)).
(2021)).and Bonnabel (2020), Ng et al.
This
derive
This approach
an
approach is
is powerful,
almost-globally
powerful, and
stable
and was
non-linear
was originally
attitude
originally used
used con-to
to linearisation
(2022),
The van
choice error
Goor
of et
symmetry (Barrau
al. (2021)).
isand Bonnabel
important, (2020),
because Ng
it et al.
deter-
troller
derive
This
derive on
an SO(3)
approach
an is (Meyer
almost-globally
powerful,
almost-globally (1971))
stable
and
stable and
was later
non-linear
non-linear a
originallyquaternion-
attitude
attitudeused con-
con-to (2022),
The van
choice Goor
of et
symmetry al. (2021)).
is important, because it deter-
troller on SO(3) (Meyer (1971)) and later a quaternion- (2022), van Goor et al. (2021)).
derive
troller
based
derive
troller
based
an
on
an
on almost-globally
SO(3)
attitude (Meyer
controller
almost-globally
SO(3)
attitude (Meyer
controller
stable
(1971))
(Wie
stableand
(1971))
(Wie
non-linear
and later
Barba
non-linear
and
and later
Barba
a
aattitude
quaternion-
(1985)).
attitude
quaternion-
(1985)). con- mines
con-
This
This
The
The
mines
The
the non-linear
choice
choice
the
choice
of
of symmetry
non-linearsymmetry error is
error
dynamics
is important,
important,
dynamics
and,because
and,
to a largeit
because
to aa large it extent,
deter-
deter-
extent,
troller
approach
based
troller on SO(3)
was
attitude
on SO(3) also (Meyer
more
controller
(Meyer (1971))
recently
(Wie
(1971)) and
applied
and
and later
Barba to
later a quaternion-
aerial
(1985)).
a robotics
quaternion-This the
mines
The
minesquality
the
choice
the ofof
non-linear
of
non-linearsymmetry
the subsequent
symmetry error
error is
is important,
linearisation.
dynamics
important,
dynamics and,
and, because
toIn
because
to a large
largeit
recent
it deter-
work,
extent,
deter-
extent,
based
approach attitude
was controller
also more (Wie
recently and
appliedBarba to (1985)).
aerial This
robotics the
mines
Cohen
the quality
the
quality et of the
non-linear
al.of (2020)
the subsequent
error
use
subsequent the linearisation.
dynamics
SE (3)
linearisation. and,
symmetry toIn
Ina recent
large
for
recent work,
extent,
(local)
work,
based
approach attitude
by utilising
based was
attitude controller
also
a also more
symmetry
controller (Wie
recently
on(WieSO(3) and
applied
and Barba
R3 ×
× Barba R(1985)).
3 to aerial
3
3 to design
(1985)). This
robotics
This mines
a Cohen
the the
quality et non-linear
al.of the
(2020) errorthe
subsequent
use dynamics
linearisation.
SE 2 and, toIna large
recent extent,
work,
approach
by utilising wasa more
symmetry recently
on SO(3) applied
× R to
× aerial
R to robotics
design a the quality
control
Cohen etof a
al.of the
quadrotor
(2020) subsequent
use system:
the SE 2 (3)
linearisation.
this
(3) symmetry
is
symmetry In
the for
recent
first
for (local)
work,
time
(local)a
approach
nonlinear
by utilising
approach was
wasa also
cascaded more
symmetry
also more recently
controller
on to
SO(3)
recently applied
×
track
applied R 3 to
× aerial
R
to R
3
attitude
aerialto robotics
and
designposi-
robotics a the
Cohenquality
control etof al.
aof the
(2020)
quadrotor subsequent
use the
system: linearisation.
SE 2 (3)
this symmetry
is In
the recent
for
first work,
(local)
time a
by
by utilising
nonlinear
utilising a
a symmetry
cascaded
symmetry on
controller
on SO(3)
to
SO(3) ×
track
× R
R
3
×
× R
3
3 toand
3attitude
to design
designposi- a
a Cohen
symmetry
control etof al.
other
a (2020)
than
quadrotor use
SO(3) the
system: SE
(or a2
2 (3)
direct
this symmetry
isproduct
the for
of
first (local)
SO(3))
time a
nonlinear cascaded
by utilisingcascaded
nonlinear a symmetry controller
on SO(3)
controller to track
to track 3attitude
× R attitude
× R toand
3 and
designposi-
posi- Cohen
a control
symmetry etof al.
a
other (2020)
quadrotor
than use
SO(3) the
system: SE
(or a (3)
this
2 direct symmetry
is the
product for
first
of (local)
time
SO(3)) a
nonlinear cascaded
This research controller to track attitude and posi- control
was
symmetry
control
symmetryusedof offoraa quadrotor
other
other quadrotor
than
quadrotor
than SO(3) system:
control.
SO(3) system:(or
(or this
aa direct
this
direct is
is the
product
the
product first
of
first
of time
SO(3))
time
SO(3)) a
a
nonlinear
This research was
cascaded supported
wasGrant controller
supported
by the
to Australian
track
by the Australian
Research
attitude
Researchand Council
posi-
Council
was
symmetry
was used
used for
other
for quadrotor
than
quadrotor control.
SO(3)
control. (or a direct product of SO(3))

through
This Discovery
research was DP210102607
supported by the “ Exploiting
Australian the
Research Symmetry
Council symmetry
was
In used
this other
for
paper than
quadrotor
we SO(3)
proposecontrol. (or
the a direct
‘Equivariantproduct of SO(3))
Regulator’
This research
through wasGrant
Discovery supported by the Australian
DP210102607 “ Exploiting Research
thethe Council
Symmetry was
In used
this for
for quadrotor control.
apaper we propose the ‘Equivariant Regulator’

of Spatial
This
through Awareness
research for 21st
wasGrant
Discovery supported Century
by theAutomation”
DP210102607 Australian
“ and
Research Franco-
Council was used quadrotor control.

through
of This Discovery
research
Spatial wasGrant
Awareness DP210102607
supported
for Research
21st byProject
Century “ Exploiting
Exploiting
theAutomation”
Australian the
thethe
Research
and
Symmetry
Symmetry
Council
Franco- (EqR),
In
In this
this general
paper
paper we
we design propose
propose methodology
the
the for trajectory
‘Equivariant
‘Equivariant track-
Regulator’
Regulator’
Australian
through
of
of Spatial International
Discovery
Awareness
SpatialDiscovery
Awareness Grant
for DP210102607
21st Century
for Research
21st Century “ “Advancing
Exploiting
Automation”
Automation” and
Autonomy
the
and Symmetry
the
the Franco-
Franco-
for (EqR),
In
ing this
(EqR),on a
a general
paper
symmetry
general we design
design propose
groups. methodology
We the
methodology show for
‘Equivariant
thatfor a trajectory
trajectory track-
Regulator’
trajectory on a
track-
through
Australian Grant
International DP210102607Project “ Exploiting
“Advancing the Symmetry
Autonomy for In this
(EqR), aapaper
general wedesign propose the
methodology ‘Equivariant
for Regulator’
Unmanned
of
Australian
of
Robotic Systems”
Spatial Awareness
Australian for Research
International
International
Spatial Awareness
Unmanned Robotic for Research
(IRP
21st Century
21st Century
Systems” (IRP
ARS).
Automation”
Project
Project “Advancing
“Advancing
Automation”
ARS).
and
and
the Franco-
Autonomy
Autonomy
the Franco- for
for ing
(EqR),
ing on
on symmetry
general
symmetry groups.
design
groups. We
methodology
We show
show that
thatfor aatrajectory
trajectory
trajectory
trajectory
track-
on
track-
on a
a
Australian
Unmanned International
Robotic Research
Systems” (IRP Project
ARS). “Advancing Autonomy for (EqR),
ing on a general
symmetry design
groups. methodology
We show thatfor atrajectory
trajectory track-
on a
Unmanned Robotic Systems”
Unmanned Robotic Systems” (IRP ARS).
(IRP ARS).
Australian International Research Project “Advancing Autonomy for ing
ing on
on symmetry
symmetry groups.
groups. We
We show
show that
that a
a trajectory
trajectory on
on a
a
2405-8963 Copyright
Unmanned © 2023 The
Robotic Systems” (IRP ARS).This is an open access article under the CC BY-NC-ND license.
Authors.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2023.02.023
Matthew Hampsey et al. / IFAC PapersOnLine 56-1 (2023) 132–137 133
manifold equipped with a free and transitive group action ture. This gives a natural identification between elements
can be lifted onto the corresponding Lie group. We show of G and G. The notation ξ X is used to indicate
that if the system is equivariant, we can formulate a global identified elements ξ ∈ G and X ∈ G. A group torsor
error for which the error dynamics can be written in terms naturally inherits a free and transitive left action from
of only the error state and transformed system inputs. We the group G induced by applying left translation. That is,
derive the general form of the linearisation in logarithmic given arbitrary ξ ∈ G , Y ∈ G, with ξ X ∈ G, the group
coordinates. Exploiting the generic design methodology action φ : G × G → G is defined by φ(Y, ξ) := ζ Y X.
proposed, we consider the same control algorithm imple- If multiple Lie groups share the same underlying manifold
mented for different choices of Lie group symmetry, and G , then this torsor admits multiple symmetries. We give
compare the resulting tracking performance in simulation. three examples of this for the state-space of the quadrotor
later in this paper.
2. PRELIMINARIES
2.1 Notation 3. PROBLEM FORMULATION
For a thorough introduction to smooth manifolds and Lie The following development draws on recent work by the
group theory, the authors recommend Lee (2012) or Tu authors (Mahony et al. (2022), Mahony et al. (2020)).
(2010). However, in this paper we develop a left-handed symmetry
for the control problem, following the lead of Cohen et al.
Let M be a smooth manifold. For arbitrary ξ ∈ M , the (2020) and Lee et al. (2010), rather than the right-handed
tangent space of M at ξ is denoted by Tξ M . For another symmetry used for the observer problem. Several results
smooth manifold N and a smooth mapping h : M → N , that we use (without proof) in the present development
Dξ|ζ h(ξ) :Tζ M → Th(ζ)N are straightforward analogies of known results for right
handed symmetry.
δ → Dξ|ζ h(ξ)[δ]
denotes the differential of h(ξ) evaluated at ξ = ζ in the 3.1 Symmetry and Equivariant Systems
direction δ ∈ Tζ M . The notation Dh will also be used
when the argument and base point are implied. The space Let M be a smooth manifold and the input space L a finite-
of smooth vector fields on M is denoted with X(M ). dimensional vector space. Consider the dynamical system
ξ˙ = fu (ξ), ξ(0) = ξ0 (1)
Let G be a Lie group. The identity element is denoted I.
If G is an m-dimensional Lie group, TX G is a real vector where ξ ∈ M , u ∈ L and f : L → X(M ) written u → fu
space of dimension m (Lee (2012)). The Lie algebra g is is affine. Given a Lie group G acting on the state-space
isomorphic to Rm . The“wedge” operator (·)∧ : Rm → g is M , a lift (Mahony et al. (2020)) is a smooth function
a linear isomorphism from the classical Rm vector space Λ : M × L → g satisfying
to the abstract Lie-algebra g. The inverse “vee” operator Dφξ Λ(ξ, u) = fu (ξ) (2)
(·)∨ : g → Rm is defined such that (v ∧ )∨ = v for all v ∈ ˚
Rm . Given X ∈ G define the left translation LX : G → G where φ : G × M → M is the group action. Let ξ be a
by LX (Z) = XZ. For U ∈ g, the left translation of U by X fixed point in M , termed the origin. The lifted system is
is defined by DLX U ∈ TX G. Similarly, RX Z = ZX and defined to be
DRX U ∈ TX G is termed the right translation of U by X. Ẋ = DRX Λ(φξ̊ (X), u), φ(X(0), ˚
ξ) = ξ(0) (3)
For a matrix Lie group DLX U = XU and DRX U = U X.
for all X ∈ G, u ∈ L. A trajectory of the lifted system
For fixed X ∈ G, the adjoint map, AdX : g → g, is defined projects down to a trajectory of the original system via
by φ(X(t), ˚ξ) = ξ(t) (Mahony et al. (2020)). Given a suitable
AdX (U ) = DLX DRX −1 U group action ψ : G × L → L, a lift Λ(ξ, u) is said to be
For a matrix Lie group, it is clear that AdX (U ) = XU X −1 . equivariant if it satisfies 1
AdX Λ(ξ, u) = Λ(φX (ξ), ψ(X, u)), (4)
Given a Lie algebra g and a function F : g → g, F ∨ is the
corresponding mapping F ∨ : Rm → Rm defined by for all X ∈ G, ξ ∈ M and u ∈ L.
F ∨ (x) := F (x∧ )∨ In this paper, we are concerned with the particular case
Given a Lie algebra g and a function H : Rm → g, the where M = G , a Lie-group torsor. Given that G is a group
notation H ∨ : Rm → Rm is defined by H ∨ (x) := H(x)∨ . torsor, there is a unique element I ∈ G that is a natural
Using analogous notation, log local coordinates around the choice for the reference or origin point, ˚
ξ I. Choosing
identity on G are denoted log∨ : G → Rm , log∨ (X) := the origin to be the identity element of G allows us to
(log(X))∨ . directly identify elements of G and G since G X =
XI = φX (I) = X ∈ G . The natural left action induced by
Let M be a smooth manifold and G a Lie group. A left identification of G with G can be used to define the lift
action is a function φ : G × M → M satisfying φ(I, ξ) = ξ Λ(X, u) by
and φ(Y, φ(X, ξ)) = φ(Y X, ξ) for all ξ ∈ M and X, Y ∈ G. Λ(X, u) = DRX −1 f (X, u)
For a fixed X ∈ G, the partial map φX : M → M maps
ξ → φ(X, ξ). Similarly, for a fixed ξ ∈ M , the partial map where for f (X, u), X ∈ G is taken in the torsor, while
φξ : G → M maps X → φ(X, ξ). A group action is called for Λ(X, U ), X ∈ G is taken in the group. Clearly,
free if for all ξ ∈ M , φ(X, ξ) = ξ only if X = I. A group f (X, u) = DRX Λ(X, u) and Λ is a lift.
action is called transitive if for each pair ξ, ζ ∈ M , there sdd In this paper, we consider three different group struc-
exists a X ∈ G such that φ(X, ξ) = ζ. A homogeneous tures on the same group torsor. In each case the group
space is a manifold M that admits a transitive group action multiplication is different and the system lift Λ will be
φ : G × M → M . The Lie group G is referred to as a different, even though the underlying dynamics f (ξ, u) are
symmetry of M .
1 For right-handed symmetry (observer design) the lifted system
If G is a Lie group, the torsor G is the underlying manifold was DLX Λ and the equivariant lift condition is for AdX −1 Λ(ξ, u) =
of G stripped of its group multiplication and group struc- Λ(φX (ξ), ψ(X, u)) (Mahony et al. (2022)).
134 Matthew Hampsey et al. / IFAC PapersOnLine 56-1 (2023) 132–137
the same. Each lifted system still projects to the same Note that ψX ≡ I implies that the error dynamics are
dynamics on the torsor via the different group multiplica- group affine, but that the converse does not hold in
tions. general.
3.2 Trajectory tracking on the symmetry group

4.1 Error dynamics in local coordinates
For the system (1), an initial condition ξ(0) = ξ0 along
with an input trajectory u(t) for t ∈ [0, T ] determines The equivariant regulator is based on linearising local
a system trajectory, an integral curve ξ : [0, T ] → M coordinates for the error dyanmics. The local coordinate
chart considered is the logarithm log∨ : G → Rm that
satisfying ξ˙ = fu (ξ) for all t ∈ [0, T ]. Given an initial maps a local neighbourhood of the identity in G to g
condition ξd (0) = ξd0 , and desired input ud (t) for t ∈ [0, T ], and then to Rm . Define notation for the local coordinates
the desired trajectory ξd (t) is an integral curve ξd : [0, T ] → ε := log∨ (E).
M satisfying ξ˙d = fud (ξd ) for all t ∈ [0, T ]. Typically, ξd (t) Proposition 4.3. The first order dynamics of ε are given
represents a known generated trajectory. by
The tracking task is to choose a u(t) that steers the ε̇ = A(t)ε + B(t)ũ + O (ε, ũ2 ), (8a)
system trajectory ξ(t) to ξd (t). In order to formulate where
the equivariant regulator we identify a lifted trajectory ∨
Xd ∈ G such that φ(Xd , ˚ ξ) = LXd ˚ξ = ξd . For a general A(t) = AdX −1 ◦DE|Xd Λ(E, ud ) ◦ DLXd , (8b)
d
˚
origin ξ the different group multiplications lead to different ∨
B(t) = AdX −1 ◦Du|ud Λ(Xd , u) . (8c)
lifted trajectories. However, by choosing ˚ ξ = I then d
LXd (˚ξ) = Xd for all group structures and we can consider If f is equivariant, then the system matrices can be sim-
a single lifted system trajectory that is identified with the plified to
desired trajectory Xd ξd . This correspondence is only ∨
true for free group actions and for the particular choice of A(t) = DE|I Λ(E, ψ(Xd−1 , ud )) , (8d)
∨
origin ˚
ξ I. B(t) = Du|ψ(X −1 ,ud ) Λ(I, u)Du|ud ψ(Xd−1 , u) . (8e)
d
Define the equivariant error
E := Xd−1 X, (5) If f is group affine, then
∨
noting that E = I if and only if X = Xd . The tracking task A(t) = DE|I Λ(E, ud ) , (8f)
is to drive E → I. The equivariant error will depend on ∨
the different group structures chosen and leads to different B(t) = Du|ud Λ(I, u) . (8g)
closed-loop responses.
The proof can be found in Appendix ??.
4. EQUIVARIANT REGULATOR
Let M be a manifold and let L be a finite-dimensional 5. EXAMPLE
vector space. We consider the dynamical system (1). Let
ξ(t) be the system trajectory and let ξd (t) be a desired In this section we consider trajectory tracking control for
trajectory to be tracked. Let G be a free symmetry on a quadrotor vehicle. The quadrotor is modelled as a rigid
M and let Λ be a system lift (2) for f . Then, since the body, with state ξ = (R, x, v) ∈ SO (3) × R3 × R3 . The
action is free, there exist corresponding trajectories X(t) = matrix R denotes the orientation of the body frame with
respect to the inertial frame, and x and v respectively
φ−1
ξ̊
ξ(t), Xd (t) = φ−1
ξ̊
ξd (t). Recall the definition (5) of the denote the position and translational velocity of the body
equivariant error E. For simplicity of presentation, we use frame with respect to the inertial frame, expressed in the
the group elements X, Xd ∈ G as the arguments for Λ via inertial frame.
the identification between M and G.
Proposition 4.1. The time derivative of E is given by The input u = (Ω, T ) ∈ L = R4 consists of the body-fixed
angular velocity Ω ∈ R3 and the combined rotor thrust
Ė = DRE AdX −1 [Λ(Xd E, u) − Λ(Xd , ud )]. (6) T ∈ R. There are low-level angular rate and rotor speed
d
Proof. control loops that we do not model in the present paper.
Ė = DLX −1 Ẋ − DRX DLX −1 DRX −1 Ẋd The skew symmetric operator (·)× : R3 → so(3) (skew-
d d d
symmetric matrices) is defined such that a× b = a×b for all
= DLX −1 DRX DRX −1 DRXd [Λ(X, u) − Λ(Xd , ud )]
d d a, b ∈ R3 . The quadrotor dynamics, ξ˙ = fu (ξ) = (Ṙ, ẋ, v̇)
= DRE AdX −1 [Λ(X, u) − Λ(Xd , ud )].
d
for f : L → X(SO (3) × R3 × R3 ) are given by (Cohen et al.
Substituting X = Xd E gives the required result. (2020))
If the lift Λ : M × L → g is equivariant, there exists a Ṙ = RΩ× , (9a)

group action ψ : G × L → L such that AdY Λ(X, u) = ẋ = v, (9b)
Λ(Y X, ψ(Y, u)). In this case, Ė can be written (analogous T
to Mahony et al. (2020)) v̇ = − Re3 + ge3 − RDR v, (9c)
m
Ė = DRE [Λ(E, ψ(Xd−1 , u)) − Λ(I, ψ(Xd−1 , ud ))] (7) where the matrix D denotes rotor drag, m denotes the
mass of the quadrotor and g denotes the acceleration due
Definition 4.2. If the error dynamics have the form to gravity.
Ė = DRE [Λ(E, u) − Λ(I, ud )],
For this paper, we shall assume that wind speed is neg-
then the system is said to be group affine (Barrau and ligible. We follow the common approach taken in the
Bonnabel (2017)). literature (Mahony et al. (2012); Kai et al. (2017)) and
model aerodynamic drag as a linear drag in the rotor plane Pose and velocity symmetry: The pose and velocity
(Mahony et al., 2012, Eq. 10). That is, symmetry treats the attitude and position as a pose with
an SE(3) (special Euclidean group) symmetry and uses
D = c1 (I − e3 e
3 ). (10)
a direct product R3 symmetry for velocity. The group
Define a new input, termed the compensated thrust (Kai product is given by
et al. (2017)) by (Q1 , p1 , γ1 )(Q2 , p2 , γ2 ) = (Q1 Q2 , Q1 p2 + p1 , γ1 + γ2 ).
1
T̄ = T − c1 e
3 R v. (11) The group action on SO (3) × R3 × R3 is given by
m
Setting the input u = Ω, T̄ ∈ L = R4 , the resulting φ((Q, p, γ), (R, x, v)) = (QR, Qx + p, v + γ).
dynamics are Since the velocity is treated separately, it is possible
Ṙ = RΩ× , (12a) to model either the inertial velocity or the body-frame
ẋ = v, (12b) velocity. The system model (12) used an inertial velocity.
However, setting vb = R v and replacing (12c) by v̇b =
v̇ = −T̄ Re3 + ge3 − c1 v. (12c) 1
−Ω× vb − m T̄ e3 + gR e3 − c1 vb the resulting system is
This input transformation significantly simplfies the tra- written in terms of a body velocity. That is, the new system
jectory generation (§6.1) and the symmetry analysis that state is written ξ = (R, x, vb ) where v = Rvb . This change
follows. allows us to find a finite-dimensional equivariant extension
The manifold SO (3) × R3 × R3 serves as the torsor for the for the system model that was not possible for the standard
following three distinct Lie group structures. model with the pose and velocity symmetry.
Direct product symmetry: The first realisation consid- The corresponding lift Λ : (SE(3) × R3 ) × L → se(3) × R3
ered uses the direct product group structure SO(3) × R3 × is given by
R3 where SO(3) is the special orthogonal group and R3 is ΛR = RΩ× R , (15a)
a group under addition. The group product is given by
(Q1 , p1 , γ1 )(Q2 , p2 , γ2 ) = (Q1 Q2 , p1 + p2 , γ1 + γ2 ). Λx = −RΩ× R x + Rvb , (15b)
The Lie algebra associated with SO(3) × R3 × R3 is Λvb = −Ω× vb − T̄ e3 + R ge3 − c1 vb , (15c)
so(3) × R3 × R3 . The group action on SO (3) × R3 × R3 where Λ((R, x, v), u) = (ΛR , Λx , Λvb ) ∈ se(3) × R3 . The
is derived from identifying the system coordinates with resulting system can be shown to be equivariant with an
group elements: φ((Q, p, γ), (R, x, v)) = (QR, p + x, γ + v). appropriate velocity extension (see Appendix ??). The
This action is free and transitive. system, however, is not group affine. The equivariant error
Comparing (12) to the definition (3), the system lift Λ : is given by E = (Rd R, Rd (x − xd ), v − vd ).
(SO(3) × R3 × R3 ) × L → (so(3) × R3 × R3 ) is given by
6. SIMULATION
ΛR = RΩ× R , (13a)
Λx = v, (13b) 6.1 Trajectory Generation
Λv = −T̄ Re3 + ge3 − c1 v, (13c) The quadrotor system (12) is differentially flat (Mellinger
and Kumar (2011)). We choose the flat outputs to be the
where Λ((R, x, v), u) = (ΛR , Λx , Λv ). The induced equiv-
ariant error is position xd ∈ R3 and the heading αd ∈ [0, 2π).
E = (Rd R, x − xd , v − vd ). Given (xd (t), αd (t)), vd is given directly by ẋd . Then,
Extended Pose symmetry: The extended pose symmetry from the quadrotor dynamics (12), Rd e3 and T̄d can be
SE2 (3) integrates a special Euclidean symmetry on posi- computed via
tion and velocity (Barrau and Bonnabel (2015)). Identify- −ẍd + ge3 − c1 vd
ing the system coordinates with group elements, the group Rd e 3 = and T̄d = −ẍd +ge3 −c1 vd .
−ẍd + ge3 − c1 vd
action on SO (3) × R3 × R3 is given by
Defining θ = cos−1 e 3 Rd e3 , µ = e3 × (Rd e3 ), the full
φ((Q, p, γ), (R, x, v)) = (QR, Qx + p, Qv + γ).
attitude Rd can be computed by Rd = exp(αd e3 ) exp(θµ).
The corresponding system lift Λ : SE2 (3) × L → se2 (3) is The angular velocity Ωd can be computed via the time
given by
derivative of the attitude, Ω× −1
d = Rd Ṙd . Note that in
ΛR = RΩ× R , (14a) contrast to Farrell et al. (2019) and Cohen et al. (2020),
×
Λx = −RΩ R x + v, (14b) there is no requirement to neglect drag forces or compute
an iterative solution to compute Rd due to the thrust
Λv = −RΩ× R v − T̄ Re3 + ge3 − c1 v, (14c) modification (11).
where Λ((R, x, v), u) = (ΛR , Λx , Λv ) ∈ se2 (3). The equiv- 6.2 Symmetry group linearisations of the quadrotor system
ariant error is defined by
E = (Rd R, Rd (x − xd ), Rd (v − vd )). Applying the methodology outlined in Section 4 to the
quadrotor systems discussed in Section 5, we obtain the
By computing (Λ(E, u) − Λ(I, ud ))E, the lifted quadrotor following linearisations (in local logarithmic coordinates).
system on extended pose SE2 (3) can be shown to be
In all cases, ε = (rε , xε , vε ) and ũ = Ω − Ωd , T̄ − T̄d .
group affine (Definition 4.2). This also marks an important
distinction between the development here and the develop- Direct product symmetry (SO(3) × R3 × R3 ):
ment in Cohen et al. (2020), in which the quadrotor system  
model on SE2 (3) is not group affine. It is interesting to −Ω× 0 0
 0
d
0 I  I 0
note that the direct product lifted system (5.0.1) is not ε̇ = 
T̄d ε + 0 0 ũ. (16)
group affine: in this sense the extended pose lifted system Rd e× 0 −c 1 I 0 −R d e 3
3
is more structured. m
136 Matthew Hampsey et al. / IFAC PapersOnLine 56-1 (2023) 132–137
||log(RE )||2 [rad]
||log(RE )||2 [rad]

2 2 100
100
||∆Ω||2 [rad/s]
||∆Ω||2 [rad/s]
80
80
1 1
60 60
0 40 0 40
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
3 20 20
3
||∆x||2 [m]
||∆x||2 [m]
0 0
2
0.0 0.5 1.0 1.5 2 0.0 0.5 1.0 1.5
1 1
6 SO(3) × R × R
3 3
SO(3) × R × R3
3
0 SE2 (3) 0 15 SE2 (3)

0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
||∆T̄ ||2 [N] SE(3) × R3 SE(3) × R3
||∆T̄ ||2 [N]

4 15
10 10
||∆v||2 [m/s]
||∆v||2 [m/s]
10
2
5 5
5
0 0 0 0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
time (s) time (s) time (s) time (s)
Fig. 1. Median squared norm of state error regulating Fig. 2. Median squared norm of state error tracking Lis-
hover over time for 2000 simulations. Convergence is sajous curve over time for 2000 simulations. Conver-
obvious after π2 seconds, so only results for t ∈ [0, π2 ] gence is obvious after π2 seconds, so only results for
are shown. Shaded area shows 5th to 95th percentile t ∈ [0, π2 ] are shown. Shaded area shows 5th to 95th
of mean squared error. percentile of mean squared error.
Extended pose symmetry (SE2 (3)):
  SO(3) × R3 × R3
−Ω× d 0 0 SE2 (3)
 0 −Ω×  I 0 0.08
I SE(3) × R3
ε̇ =  d ε + 0 0 ũ. (17)
T̄d × 0 −e3
e3 0 −Ω×
d − c 1 I 0.06
m
RMSE
Pose and velocity symmetry (SE(3) × R3 ):

  0.04
−Ω× d 0 0 I 0
 × ×  0 0
ε̇ = −vbd −Ωd I ε+ ũ. 0.02
g(Rd e3 ) 0 −Ω× − c 1 I vb×d −e3
d
(18)
0.00
∆T̄ [N] ∆x [m] ∆v [m/s] ∆Ω [rad/s] log(RE ) [rad]
6.3 Simulation Results
A simple case study is presented to compare the relative Fig. 3. RMSE tracking Lissajous curve over time for
performance of the EqR design using different symmetry 5000 simulations in the presence of process noise. Bar
groups. The trajectories are designed and tracked by shows median and error bars represent 20th and 80th
simulating the quadcopter and controller dynamics in percentile.
Python. The quadrotor states are assumed to be known 1
exactly at all times. The drag constant c1 = 0.25 is used xd (t) = (cos(2t), sin(4t), 2), αd (t) = 0, t ∈ [0, π]. (19)
and the discrete-time LQR parameters are as follows: rε 2
state cost matrix 1.0I3 ; xε state cost matrix 2.0I3 ; v state This trajectory has the property that is non-trim, and
so will result in time-varying A and B matrices for all
cost matrix 0.1I3 ; Ω̃ input cost matrix 0.5I3 ; and T̄˜ scalar symmetries. The initial orientation, position and velocity
input cost 0.5. are randomly perturbed in each component by sampling
6.4 Point regulation and transient response from Gaussian distributions of means 0 and standard
deviations 0.8, 0.6 and 0.3, respectively. The results of 2000
The goal of this simulation is to investigate the transient Monte Carlo trials are shown in Figure 2 and discussed in
responses of the different controllers when tracking a hover § 6.7.
point trajectory, defined by
xd (t) = (0, 0, 0) and αd (t) = 0, for t ∈ [0, π]. 6.6 Asymptotic tracking performance
This trajectory has the property that the linearisations The goal of this simulation is to investigate the asymptotic
(16-18) at the goal trajectory are identical. Thus, the only responses of the different controllers. The trajectory to
differences in response will be due to transient effects. be tracked is the Lissajous curve (19). The initial states
The initial orientation, position and velocity are randomly are exactly the initial desired states, but the system
perturbed in each component by sampling from Gaussian derivatives are perturbed in each component at every time
distributions of means 0 and standard deviations 0.8, 0.6 step by sampling from Gaussian distributions of mean 0
and 0.3, respectively. The results of 2000 Monte Carlo and standard deviation 0.1. The distribution of RMSE
trials are shown in Figure 1 and discussed in § 6.7. over 5000 Monte Carlo trials are shown in Figure 3 and
discussed in § 6.7.
6.5 Lissajous trajectory and transient response
6.7 Discussion
The goal of this simulation is to investigate the transient
responses of the different controllers when tracking a There are two important factors to analyse in Figure 1:
Lissajous trajectory, defined by the median response and the outlier performance. The
median response of each controller is very similar, although Barrau, A. and Bonnabel, S. (2017). The invariant ex-
there is a higher error seen in the transient response of tended kalman filter as a stable observer. IEEE Trans-
the direct product controller in the position component. actions on Automatic Control, 62(4), 1797–1812. doi:
There is a significantly wider range of performance of 10.1109/TAC.2016.2594085.
the direct product controller compared to the extended Barrau, A. and Bonnabel, S. (2020). Extended kalman fil-
pose and pose and velocity controllers across all state and tering with nonlinear equality constraints: A geometric
input components; that is, the direct product controller approach. IEEE Transactions on Automatic Control,
performs poorly more frequently. Importantly, the perfor- 65(6), 2325–2338. doi:10.1109/TAC.2019.2929112.
mance discrepancy is not simply due to the controllers Bonnabel, S., Martin, P., and Rouchon, P. (2008).
expending more energy; indeed, the direct product symme- Symmetry-preserving observers. IEEE Transactions
try appears to consistently expend more energy for worse on Automatic Control, 53(11), 2514–2526. doi:
performance. The results for tracking the Lissajous curve
with different initial conditions in Figure 2 are almost 10.1109/TAC.2008.2006929.
identical to the hover regulation results in Figure 1. This Cohen, M.R., Abdulrahim, K., and Forbes, J.R. (2020).
indicates that the observed performance of the controllers Finite-Horizon LQR Control of Quadrotors on SE2 (3).
is not trajectory-specific. Figure 3 reveals no discernible IEEE Robotics and Automation Letters, 5(4), 5748–
difference in asymptotic tracking performance between 5755. doi:10.1109/LRA.2020.3010214.
symmetries. Farrell, M., Jackson, J., Nielsen, J., Bidstrup, C., and
McLain, T. (2019). Error-State LQR Control of a
These simulations, along with the fact that the hover tra- Multirotor UAV. In 2019 International Conference on
jectory results in identical linearised dynamics and error Unmanned Aircraft Systems (ICUAS), 704–711. doi:
definitions, indicate that the salient differences must be
due to the particular symmetry’s Lie algebraic represen- 10.1109/ICUAS.2019.8798359.
tation of the system state, and the associated higher- Kai, J.M., Allibert, G., Hua, M.D., and Hamel,
order system dynamics. Previous research has shown that T. (2017). Nonlinear feedback control of
choices of equivariant or group affine symmetries lead to quadrotors exploiting first-order drag effects.
lower linearisation error (Barrau and Bonnabel (2020), Ng IFAC-PapersOnLine, 50(1), 8189–8195. doi:
et al. (2022)). A possibility for the improved performance https://doi.org/10.1016/j.ifacol.2017.08.1267. 20th
seen here is that the group affine structure of the extended IFAC World Congress.
pose and the equivariant structure of the pose and ve- Lee, J.M. (2012). Introduction to Smooth Manifolds.
locity symmetry have reduced linearisation error. In this Graduate Texts in Mathematics. Springer.
way, identifying specific properties of a given symmetry Lee, T., Leok, M., and McClamroch, N.H. (2010). Geomet-
reveals insight into the underlying geometry of the system ric tracking control of a quadrotor UAV on SE(3). In
and provides motivation for choice of symmetry prior to 49th IEEE Conference on Decision and Control (CDC),
implementation. 5420–5425. doi:10.1109/CDC.2010.5717652.
Mahony, R., Hamel, T., and Pflimlin, J.M. (2008). Non-
The improved performance of the extended pose symmetry linear complementary filters on the special orthogonal
compared to the direct product symmetry is consistent group. IEEE Transactions on Automatic Control, 53(5),
with the results of Cohen et al. (2020). This, and the
improved performance of the pose and velocity symmetry 1203–1218. doi:10.1109/TAC.2008.923738.
compared to the direct product symmetry suggests that Mahony, R., Hamel, T., and Trumpf, J. (2020).
the most important component of rigid body symmetries is Equivariant Systems Theory and Observer Design.
the coupling of the rotational and positional components of arXiv:2006.08276 [cs, eess].
the state. The quadrotor dynamics are otherwise invariant Mahony, R., Kumar, V., and Corke, P. (2012). Multirotor
with respect to position, so this coupling may positively aerial vehicles: Modeling, estimation, and control of
impact the LQR Riccati equation solutions (for these quadrotor. IEEE Robotics Automation Magazine, 19(3),
trajectories and sets of LQR gains). 20–32. doi:10.1109/MRA.2012.2206474.
Mahony, R., van Goor, P., and Hamel, T. (2022). Ob-
server design for nonlinear systems with equivariance.
7. CONCLUSION Annual Review of Control, Robotics, and Autonomous
Systems, 5(1), 221–252. doi:10.1146/annurev-control-
This paper presented the Equivariant Regulator: a general 061520-010324.
methodology for trajectory tracking on manifolds with a Mellinger, D. and Kumar, V. (2011). Minimum snap tra-
free symmetry. We discussed several special cases of sym- jectory generation and control for quadrotors. In 2011
metry, and analysed how these special cases simplify error IEEE International Conference on Robotics and Au-
trajectory linearisation. We presented the sample case of tomation, 2520–2525. doi:10.1109/ICRA.2011.5980409.
the quadrotor, generated sample trajectories and used the
methodology to derive error trajectory linearisations and Meyer, G. (1971). Design and global analysis of spacecraft
LQR controllers to track these trajectories in the presence attitude control systems.
of perturbations. We showed that equivariant and group Ng, Y., Van Goor, P., Hamel, T., and Mahony, R. (2022).
affine symmetries result in improved performance in the Equivariant observers for second order systems on ma-
presence of initial perturbations. trix lie groups. IEEE Transactions on Automatic Con-
trol, 1–1. doi:10.1109/TAC.2022.3173926.
Tu, L. (2010). An Introduction to Manifolds. Universitext.
REFERENCES Springer New York.
van Goor, P., Hamel, T., and Mahony, R. (2021). Equiv-
Anderson, B. and Moore, J. (2007). Optimal Control: Lin- ariant filter (eqf).
ear Quadratic Methods. Dover Books on Engineering.
Dover Publications. Wie, B. and Barba, P.M. (1985). Quaternion feedback
Barrau, A. and Bonnabel, S. (2015). Invariant filtering for spacecraft large angle maneuvers. Journal of Guid-
for pose ekf-slam aided by an imu. In 2015 54th IEEE ance, Control, and Dynamics, 8(3), 360–365. doi:
Conference on Decision and Control (CDC), 2133–2138. 10.2514/3.19988.
doi:10.1109/CDC.2015.7402522.
Memories - Materials, Devices, Circuits and Systems 4 (2023) 100030
Contents lists available at ScienceDirect
Memories - Materials, Devices, Circuits and Systems

journal homepage: www.elsevier.com/locate/memori
Robust PID controllers tuning based on the beetle antennae search algorithm
Spyridon D. Mourtas a,b , Chrysostomos Kasimis c , Vasilios N. Katsikis a ,∗
a
Department of Economics, Division of Mathematics-Informatics and Statistics-Econometrics, National and Kapodistrian University of Athens,
Sofokleous 1 Street, 10559 Athens, Greece
b
Laboratory ‘‘Hybrid Methods of Modelling and Optimization in Complex Systems’’, Siberian Federal University, Prosp. Svobodny 79, 660041 Krasnoyarsk, Russia
c
Department of Physics, Electronics Laboratory, University of Patras, Patras, GR-26504, Greece
ARTICLE INFO ABSTRACT

Keywords: The core components of both traditional and contemporary control systems are the proportional–integral–
Metaheuristics derivative (PID) control systems, which have established themselves as standards for technical and industrial
Beetle Antennae Search applications. Therefore, the tuning of the PID controllers is of high importance. Utilizing optimization
Optimization
algorithms to reduce the mean square error of the controller’s output is one approach of tuning PID controllers.
Feedback control systems
In this paper, an appropriately modified metaheuristic optimization algorithm dubbed beetle antennae search
PID controller
(BAS) is employed for robust tuning of PID controllers. The findings of three simulated experiments on
stabilizing feedback control systems show that BAS produces comparable or higher performance than three
other well-known optimization algorithms while only consuming a tenth of their time.
1. Introduction However, unlike the approach used in [11], which involves mini-
mizing the integral of time multiplied by the absolute value of error
Since its introduction decades ago, proportional–integral–derivative (ITAE), our approach for fine-tuning PID controllers involves minimiz-
(PID) controllers have been effectively employed in process-controlled ing the mean square error (MSE) of the closed-loop control system sim-
domains of industry including metallurgy, power, machinery, and light ulation similar to [10]. As a result, the BAS algorithm is appropriately
industries [1]. Two of the key factors for the PID controller’s continued modified to address this specific problem of tuning PID controllers. The
application are its ease of analysis and design, along with its ease findings of three simulated experiments on stabilizing feedback control
of implementation. A PID controller employs feedback to establish a systems show that BAS produces comparable or higher performance
control loop, making it simple to grasp and adapt to a wide range of than three other well-known optimization algorithms while only con-
control systems [2]. However, PID controllers have drawbacks. The PID
suming a tenth of their time. The effectiveness of BAS in fine-tuning
control system’s performance is tuned by adjusting the PID parameters,
PID controllers is therefore confirmed once more.
particularly Kp , Ki , and Kd . The controlled system’s performance will
The following is a list of the major contributions of our research in
be subpar or perhaps unstable as a result of improper tuning [3].
this paper.
Tuning the PID controller for accurate and reliable closed-loop con-
(1) A fine-tuning PID controllers approach that involves minimizing
trol turns into an NP-Hard problem, just like all non-convex optimiza-
tion problems [4]. As a result, a variety of heuristic and metaheuristic the MSE of the closed-loop control system simulation is presented.
algorithms have been used in recent years to tune PID controllers, (2) An appropriately modified BAS algorithm for tuning PID con-
including whale optimization [5], particle filtering optimization [6], trollers is introduced.
fruit fly [7], invasive weed optimization [8], bee [9], differential (3) Three simulated experiments on stabilizing feedback control
evolution (DE) and genetic algorithm (GA) [10]. Such as in [11], the systems are carried out with the intention of comparing the perfor-
beetle antennae search (BAS) algorithm is used in this research to mance of BAS against three other well-known optimization algorithms
tune PID controllers. It is important to note that BAS is competent in fine-tuning PID controllers.
of efficient global optimization, that it has recently gained significant The following hierarchy serves as the foundation for the paper’s
traction in a number of scientific fields, including machine learning overall organization of sections. Section 2 presents preliminary in-
[12,13], robotics [14], engineering [15], and finance [16], and that formation on PID-based feedback control systems and also describes
it has undergone a number of modifications, including binary [17] and the problem of minimizing the MSE of the closed-loop control system
semi-integer [18] versions, to better address various problems. simulation. Section 3 introduces the modified BAS algorithm for robust
∗ Corresponding author.
E-mail addresses: spirmour@econ.uoa.gr (S.D. Mourtas), chrkasim@upatras.gr (C. Kasimis), vaskatsikis@econ.uoa.gr (V.N. Katsikis).
https://doi.org/10.1016/j.memori.2023.100030
Received 9 December 2022; Received in revised form 3 January 2023; Accepted 10 February 2023
2773-0646/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
S.D. Mourtas, C. Kasimis and V.N. Katsikis Memories - Materials, Devices, Circuits and Systems 4 (2023) 100030
Fig. 1. Structure of PID-based control system.
tuning of PID controllers. Section 4 presents three simulated experi- where 𝐷(𝑠), 𝑅(𝑠), respectively, signify the Laplace transforms of the
ments on stabilizing feedback control systems, where the performance input and output signals and
of BAS against three other well-known optimization algorithms in fine- ∑𝑛−1 𝑖
tuning PID controllers is investigated. The final remarks are included 𝛽𝑖 𝑠
𝑄(𝑠) = ∑0 𝑛 , (2.4)
in Section 5. 𝑖
0 𝑖𝑠
𝛼
2. Problem formulation where 𝑄(𝑠) signifies the transfer function of the plant and its coeffi-
cients 𝛼𝑖 , 𝛽𝑖 ∈ R.
The two basic types of control loops are open loop control and Further, to change the transfer function 𝑅(𝑠) from the s-domain to
closed loop control. The controller’s response is dependent on process the z-domain the zero-order-hold (ZOE) technique [21] is utilized. It
feedback in the context of the process variable’s (PrV) value in closed is worth mentioning that the ZOE technique is used to transform the
loop control, whereas in open loop control, the controller’s response is discrete-time signal 𝑞(𝑡𝑘 ) generated by the controller in (2.2) into a
independent of the PrV. In a closed loop controller, a feedback loop
piecewise continuous-time constant signal. Particularly, assuming the
ensures that the controller manages the PrV to be equal to the input
next z-transfer function:
of reference. In other words, a closed-loop controller eliminates the
shortcomings of an open-loop controller by using feedback to control 𝛽1 𝑧 + 𝛽0 𝑦(𝑧)
𝑅(𝑧) = = , (2.5)
the states or outputs of a dynamical system. Because of this, closed loop 𝑧2 + 𝛼1 𝑧 + 𝛼0 𝑞(𝑧)
controllers are often referred to as feedback controllers [19].
its difference equation may be expressed as below [20]:
Since the 1920s, practically all analogue control systems have used
the PID controller, which is a conventional feedback controller design. 𝑦(𝑡𝑘 ) = −𝛼1 𝑦(𝑡𝑘−1 ) − 𝛼0 𝑦(𝑡𝑘−2 ) + 𝛽1 𝑞(𝑡𝑘−1 ) + 𝛽0 𝑞(𝑡𝑘−2 ). (2.6)
Both its theoretical understanding and practical use derive from that
era. When there is a discrepancy between a desired setpoint and a In our approach for fine-tuning PID controllers, the tracking errors
measured PrV, a PID controller calculates an error value 𝑒(𝑡) and cor- of the PID controller can be quantified by MSE after 𝑦(𝑡𝑘 ) is retrieved
rects it utilizing proportional, integral, and derivative elements. Keep in from (2.6). Note that the MSE is a measure of the quality of an
mind that these three elements use the error signal to create the control estimator. It is always a positive number that gets smaller as the error
signal. Presume that 𝑞(𝑡) is the control signal supplied to the system, 𝑦(𝑡)
gets closer to zero since it is derived from the square of the Euclidean
is the measured output, 𝑟(𝑡) is the target output, and 𝑒(𝑡) = 𝑟(𝑡) − 𝑦(𝑡) is
distance. Therefore, for 𝑘 in number samples, the MSE of the PID
the tracking error, a PID controller’s general form is as follows:
controller’s tracking errors is computed as follows:
𝑡
d𝑒(𝑡)
𝑞(𝑡) = Kp 𝑒(𝑡) + Ki 𝑒(𝜏)d𝜏 + Kd , (2.1)
1 ∑(
𝑘
∫0 d𝜏 )2
MSE = 𝑟(𝑡𝑗 ) − 𝑦(𝑡𝑗 ) . (2.7)
where Kp , Ki , Kd ∈ R+ , respectively, signify the coefficients for the 𝑘 𝑗=1
0
proportional, integral, and derivative components. Note that R+ 0
signi-
With this approach, closed-loop stability can be attained more quickly
fies non-negative real numbers. The appropriate closed loop dynamics
can be achieved by adjusting the values of these three coefficients, while the phase margin of the tuned PID loop can be decreased. The
frequently iteratively by tuning and without specific understanding of previously mentioned discretized method for computing the MSE of
a plant model. the closed-loop control system simulation is further described as an
In this paper, the MSE of the closed-loop control system simulation objective function in Alg. 1.
is minimized to fine-tuning PID controllers. This can be achieved after
discretizing the complete feedback control system process. The goal Algorithm 1 Discretized process for computing the MSE of the
is to produce all of the simulation outputs for the closed-loop control closed-loop control system simulation.
system before calculating the MSE of the tracking errors. As a result, Input: The vector 𝑥 = [Kp , Ki , Kd ], the s-transfer function 𝑅(𝑠), the
the discrete-time equivalent of the PID controller, rather than the
output 𝑟(𝑡), the period’s end 𝑇 and the sampling period 𝛥𝑡.
continuous-time equivalent (2.1), is used [20]:
( ) 1: procedure Obj_Function(𝑥, 𝑅(𝑠), 𝑟(𝑡), 𝑇 , 𝛥𝑡)
K 2: Transform the 𝑅(𝑠) from s-domain to z-domain, and set 𝑡 = 2𝛥𝑡
𝑞(𝑡𝑘 ) =𝑞(𝑡𝑘−1 ) + Kp + Ki 𝛥𝑡 + d 𝑒(𝑡𝑘 )
𝛥𝑡 3: while 𝑡 ≤ 𝑇 do
( ) (2.2)
2Kd K 4: Set 𝑞(𝑡) inline with (2.2) using the components of 𝑥
− Kp + 𝑒(𝑡𝑘−1 ) + d 𝑒(𝑡𝑘−2 ),
𝛥𝑡 𝛥𝑡 5: Set 𝑦(𝑡) inline with (2.6), and 𝑒(𝑡) = 𝑟(𝑡) − 𝑦(𝑡)
where 𝑘 signifies the index of sample and 𝛥𝑡 signifies the sampling 6: 𝑡 ← 𝑡 + 𝛥𝑡
period. 7: end while
Assume the plant model of Fig. 1, which is a single-input and single- 8: Set MSE inline with (2.7)
output (SISO) control system and usually operates in continuous-time. 9: end procedure
The next s-transfer function can be used to represent this plant model: Output: The MSE of the closed-loop control system simulation.
𝑅(𝑠) = 𝑄(𝑠)𝐷(𝑠), (2.3)
2
3. Tuning PID controllers through BAS
This paper’s main focus is on the computational techniques utilized

to quickly and accurately address the suggested challenge of fine-tuning
PID controllers. As a result, we employ the memetic metaheuristic
algorithm BAS, whose primary characteristic is its short running time.
According to [22], the strength of the odor beetles perceive on the
antennas determines how they use their two antennas to find food. This
behavior is imitated by the BAS algorithm’s optimal solution finder,
and this approach makes it possible to use novel optimization methods
(see [23–25]).
The challenge of fine-tuning PID controllers may be composed in
the next non-convex optimization problem:
minimize 𝑓 (𝑥) (3.1)

subject to 𝟎 ≤ 𝑥 ≤ UB ⋅ 𝟏 (3.2)
where the objective function 𝑓 (𝑥) corresponds to the MSE output of

Alg. 1, UB ∈ R+ 0
signifies the upper bound value, 𝟎, 𝟏 ∈ R𝑛 signify a
vector of zeros and ones, respectively, and 𝑥 ∈ R𝑛 signifies the unknown
variable. By imitating the beetle’s behavior, we attempt to solve the
(3.1)–(3.2) optimization problem.
Suppose that the beetle’s position at the 𝑖th time moment is repre-
sented by the vector 𝑥𝑖 with 𝑖 = 1, 2, … , and that the objective function
𝑓 (𝑥) represents the odor concentration at the position 𝑥, with the source
point of the odor being linked with the minimum value of 𝑓 (𝑥). The
model of searching behavior is thus defined as follows by the beetle’s
haphazard search path:
𝜆
ℎ= , (3.3)
𝜖 + ‖𝜆‖
where 𝜖 = 2−52 and 𝜆 ∈ R𝑛 signifies a vector of 𝑛 random elements. To
mimic the searching behaviors of the beetle’s antennae, the right (𝑥𝑅 ) Fig. 2. The modified BAS algorithm for addressing the optimization problem of
and left (𝑥𝐿 ) antennae are formulated as follows: (3.1)–(3.2).
𝑥𝑅 = 𝑥𝑖 + 𝜁𝑖 ℎ, 𝑥𝐿 = 𝑥𝑖 − 𝜁𝑖 ℎ. (3.4) Table 1
Feedback control systems configuration.
where 𝜁𝑖 signifies the sensing width of the antennae that correlates Plant 𝛽0 𝛼2 𝛼1 𝛼0
to capacity to exploit at the 𝑖th time moment. Further, consider the
1 1 1 10 20
candidate optimal solution (𝑥𝐶 ), 2 3 5 25 50
3 2 10 100 10
𝑥𝐶 = |𝑥𝑖 + 𝜉𝑖 𝜁𝑖 sign(𝑓 (𝑥𝐿 ) − 𝑓 (𝑥𝑅 ))|, (3.5)
where 𝜉𝑖 signifies a size step which takes into consideration the pace
of convergence following an increase in 𝑖 during the search, and sign(⋅) 4. Simulated experiments
signifies a sign function. It is important to note that 𝑥𝐶 ≥ 𝟎. However,
in order to guarantee that 𝑥𝐶 ≤ UB, we set all the elements of 𝑥𝐶 that This section compares and contrasts BAS’ performance with those of
are greater than the UB equal to the UB. That is, for every element 𝑗 state-of-the-art metaheuristic algorithms such as slime mould algorithm
of 𝑥𝐶 , we set: (SMA) of [26], DE of [27] and GA of MATLAB in addressing the
{ 𝑗 problem of (3.1)–(3.2) for fine-tuning PID controllers. A few parameters
𝑥𝐶 , 𝑥𝑗𝐶 ≤ UB
𝑥𝑗𝐶 = . (3.6) and symbols must also be explained, and some more information must
UB, 𝑥𝑗𝐶 > UB be provided as a prelude to the upcoming tests. Three plant models
Then, the detecting behavior is expressed as below: based on (2.4) are specifically taken into consideration, one for each
{ simulation, and their parameters are shown in Table 1. 𝛥𝑡 = 0.01,
𝑥𝐶 , 𝑓 (𝑥𝐶 ) ≤ 𝑓 (𝑥𝑖 )
𝑥𝑖+1 = . (3.7) 𝑇 = 10, and 𝑟(𝑡) = 1 have been set for each simulation, while UB = 104
𝑥𝑖 , 𝑓 (𝑥𝐶 ) > 𝑓 (𝑥𝑖 ) and 𝑖max = 103 have been set in all metaheuristics. In SMA, DE, and GA,
Finally, the update rules of 𝜁 and 𝜉 are presented below: the population size has also been set at 50. For benchmarking reasons,
the simulations also include the results of the MATLAB function pid-
𝜁𝑖+1 = 0.991𝜁𝑖 + 0.001, 𝜉𝑖+1 = 0.991𝜉𝑖 . (3.8) tune with default settings. Finally, the simulation results are shown in
Fig. 3 and the time consumption (TC) of the algorithms is presented in
It is important to mention that the initial conditions of the aforemen- Table 2.
tioned approach are the following: The simulation results for the plants of Table 1 are shown in Fig. 3.
𝑥0 = 𝟎, 𝜁0 = 𝜉0 = UB∕2, (3.9) Particularly, the parameters Kp , Ki , and Kd convergence during BAS
iterations for the plants 1–3 are shown in Figs. 3(a)–3(c), respectively.
while we have set 𝑛 = 3 for fine-tuning PID controllers. The modified There, it can be seen that the variable 𝑥, which is made up of the
BAS algorithm for addressing the (3.1)–(3.2) optimization problem is parameters Kp , Ki , and Kd , converged to the ideal outcome before the
presented in the diagram of Fig. 2. 600th iteration. The convergence of the objective function in Alg. 1
3
Fig. 3. (3(a)–3(c)): The parameters Kp , Ki , Kd convergence during BAS for the plants 1–3. (3(d)–3(f)): The objective function convergence during BAS for the plants 1–3. (3(g)–3(i)):
The minimum MSE produced by SMA, DE, GA and BAS for the plants 1–3. (3(j)–3(l)): Results from simulating the fine-tuned PID utilizing pidtune, SMA, DE, GA and BAS in
the plants 1–3.
during BAS iterations for the plants 1–3 is shown in Figs. 3(d)–3(f), terms of closed-loop stability and phase margin, and SMA has the worst
respectively. There, it can be seen that the objective function converged results in terms of closed-loop stability. Additionally, GA performs
to the minimum MSE before the 600th iteration. The minimum MSE similarly to pidtune while both achieve the poorest phase margin
produced by SMA, DE, GA and BAS for the plants 1–3 is shown in results. In plant 2 of Fig. 3(k), the PID controller tuned with BAS has
Figs. 3(g)–3(i), respectively, where it can be seen that BAS always gave achieved the quickest closed-loop stability and the minimum phase
the best results. Additionally, among all the plants examined, GA gave margin, whereas DE has the second best results in terms of closed-loop
the poorest results, while DE gave the second-best results. stability and phase margin, and GA has the worst results in terms of
The results from simulating the fine-tuned PID utilizing pidtune, closed-loop stability. Also, SMA performs similarly to pidtune while
SMA, DE, GA and BAS in the plants 1–3 are shown in Figs. 3(j)– both achieve the worst phase margin results. In plant 3 of Fig. 3(l),
3(l), respectively. Particularly, the PID controller tuned with BAS has BAS performs similarly to DE. That is, the PID controllers tuned with
achieved the quickest closed-loop stability and the minimum phase BAS and DE have achieved the quickest closed-loop stability and the
margin in plant 1 of Fig. 3(j), whereas DE has the second best results in minimum phase margin. Additionally, SMA has the second best results
4
Table 2 [5] E.S. Ghith, F.A.A. Tolba, Real-time implementation of tuning PID controller
Algorithms average TC. based on whale optimization algorithm for micro-robotics system, in: 2022
Algorithm SMA DE GA BAS 14th International Conference on Computer and Automation Engineering, ICCAE,
2022, http://dx.doi.org/10.1109/iccae55086.2022.9762448.
Average TC 135 s 140 s 130 s 11 s
[6] D.A. Plaza-Guingla, R.M. Idrovo, A.J. Valencia, C.S. Lopez, Enhancing the
performance of the particle filtering optimization algorithm for the tuning of
PID controllers, in: Proceedings of the 2017 the 5th International Conference on
Control, Mechatronics and Automation, ICCMA 2017, Association for Computing
in terms of closed-loop stability and phase margin, GA has the third Machinery, New York, NY, USA, 2017, pp. 95–99, http://dx.doi.org/10.1145/
best and pidtune has the worst. 3149827.3149850.
The algorithms average TC is presented in Table 2. In particular, the [7] X. Liu, Y. Shi, J. Xu, Parameters tuning approach for proportion integration
differentiation controller of magnetorheological fluids brake based on improved
SMA, DE, GA, and BAS average TCs are generated by running all three
fruit fly optimization algorithm, Symmetry 9 (7) (2017) 109, http://dx.doi.org/
simulations 100 times. The TCs indicate that BAS requires only 11 s to 10.3390/sym9070109, URL https://www.mdpi.com/2073-8994/9/7/109.
perform 103 iterations, while SMA, DE, and GA require 135, 140, and [8] R.K. Mandava, P.R. Vundavilli, [Tuning of pid] controller parameters of a
130 s, respectively. As a result, DE is the slowest algorithm, whereas biped robot using IWO algorithm, in: Proceedings of the 2018 4th International
BAS is the fastest because it only needs a tenth of the time of SMA, DE, Conference on Mechatronics and Robotics Engineering, ICMRE 2018, Association
for Computing Machinery, New York, NY, USA, 2018, pp. 90–94, http://dx.doi.
and GA.
org/10.1145/3191477.3191504.
In summary, PID controllers can be fine-tuned through the (3.1)– [9] M.A. Şen, V. Bakırcıoğlu, M. Kalyoncu, Performances comparison of the bees
(3.2) optimization problem using SMA, DE, GA, and BAS. To support algorithm and genetic algorithm for PID controller tuning, in: Proceedings of the
the preceding conclusions, simulation results for plants 1–3 on stabiliz- 5th International Conference on Mechatronics and Control Engineering, ICMCE
’16, Association for Computing Machinery, New York, NY, USA, 2016, pp.
ing feedback control systems show that BAS produces comparable or
126–130, http://dx.doi.org/10.1145/3036932.3036951.
higher performance than SMA, DE, and GA while consuming one-tenth [10] M.S. Saad, H. Jamaluddin, I.Z.M. Darus, Implementation of PID controller tuning
the time. using differential evolution and genetic algorithms, Int. J. Innovative Comput.
Inf. Control 8 (11) (2012) 7761–7779.
[11] Y. Fan, J. Shao, G. Sun, Optimized PID controller based on beetle antennae
5. Conclusion
search algorithm for electro-hydraulic position servo control system, Sensors 19
(12) (2019) 2727.
This paper introduced a modified BAS algorithm for tuning PID con- [12] T.E. Simos, V.N. Katsikis, S.D. Mourtas, Multi-input bio-inspired weights and
trollers by reducing the MSE of the controller’s output. Three simulated structure determination neuronet with applications in European central bank
publications, Math. Comput. Simulation 193 (2022) 451–465, http://dx.doi.org/
experiments on stabilizing feedback control systems were carried out
10.1016/j.matcom.2021.11.007.
with the intention of comparing the performance of BAS against the [13] T.E. Simos, S.D. Mourtas, V.N. Katsikis, Time-varying black-litterman portfolio
well-known optimization algorithms SMA, DE, and GA in fine-tuning optimization using a bio-inspired approach and neuronets, Appl. Soft Comput.
PID controllers. According to the simulation results, BAS produces 112 (2021) 107767, http://dx.doi.org/10.1016/j.asoc.2021.107767.
[14] Y. Cheng, C. Li, S. Li, Z. Li, Motion planning of redundant manipulator with vari-
comparable or higher performance than SMA, DE, and GA while only
able joint velocity limit based on beetle antennae search algorithm, IEEE Access
consuming one-tenth of their time. The fact that the proposed modified 8 (2020) 138788–138799, http://dx.doi.org/10.1109/ACCESS.2020.3012564.
BAS algorithm can only be applied to one type of optimization problems [15] X. Li, H. Jiang, M. Niu, R. Wang, An enhanced selective ensemble deep learning
is a limitation. Further study may improve the performance of the method for rolling bearing fault diagnosis with beetle antennae search algorithm,
proposed modified BAS algorithm or perhaps change its architecture Mech. Syst. Signal Process. 142 (2020) 106752.
[16] V.N. Katsikis, S.D. Mourtas, Computational Management, Vol. 18 of Modeling
to allow it to be used for a variety of optimization problems.
and Optimization in Science and Technologies, Springer, Cham, 2021, pp. 305–
323, http://dx.doi.org/10.1007/978-3-030-72929-5_14, Ch. Portfolio insurance
Declaration of competing interest and intelligent algorithms.
[17] S.D. Mourtas, V.N. Katsikis, V-shaped BAS: Applications on large portfolios
selection problem, Comput. Econ. (2021) http://dx.doi.org/10.1007/s10614-
The authors declare that they have no known competing finan- 021-10184-9.
cial interests or personal relationships that could have appeared to [18] V.N. Katsikis, S.D. Mourtas, Diversification of time-varying tangency portfo-
influence the work reported in this paper. lio under nonlinear constraints through semi-integer beetle antennae search
algorithm, AppliedMath 1 (1) (2021) 63–73, http://dx.doi.org/10.3390/
appliedmath1010005.
Data availability [19] J.J. DiStefano, A.R. Stubberud, I.J. Williams, Feedback and Control Systems, third
ed., in: Schaums outline series, McGraw-Hill Education, 2013.
No data was used for the research described in the article. [20] S. Mourtas, V. Katsikis, C. Kasimis, Feedback control systems stabilization using
a bio-inspired neural network, EAI Endorsed Trans. AI Robot. 1 (1) (2022) 1–13,
http://dx.doi.org/10.4108/airo.v1i.17.
Acknowledgments [21] L. Wang, PID Control System Design and Automatic Tuning using
MATLAB/Simulink, Wiley-IEEE Press, 2020, http://dx.doi.org/10.1002/
9781119469414.
This work was supported by the Ministry of Science and Higher
[22] X. Jiang, S. Li, [Bas:] beetle antennae search algorithm for optimization
Education of the Russian Federation (Grant No. 075-15-2022-1121). problems, 2017, arXiv preprint abs/1710.10724. arXiv:1710.10724. URL http:
//arxiv.org/abs/1710.10724.
[23] Z. Zhu, Z. Zhang, W. Man, X. Tong, J. Qiu, F. Li, A new beetle antennae search
References
algorithm for multi-objective energy management in microgrid, in: Proc. 13th
IEEE Conf. Industrial Electronics and Applications, ICIEA, 2018, pp. 1599–1603.
[1] M. Shamsuzzoha (Ed.), PID Control for Industrial Processes, IntechOpen, 2018, [24] Q. Wu, X. Shen, Y. Jin, Z. Chen, S. Li, A.H. Khan, D. Chen, Intelligent beetle
http://dx.doi.org/10.5772/intechopen.69592. antennae search for UAV sensing and avoidance of obstacles, Sensors 19 (2019)
[2] M. Tabish, A. Kalam, A. Zayegh, Robot DC servo motor parameters estimation 1758, http://dx.doi.org/10.3390/s19081758.
in a closed loop using BAT optimisation algorithm, in: 2019 International [25] X. Xu, K. Deng, B. Shen, A beetle antennae search algorithm based on Lévy
Conference on Electrical, Communication, and Computer Engineering, ICECCE, flights and adaptive strategy, Syst. Sci. Control Eng. 8 (2020) 35–47, http:
2019, http://dx.doi.org/10.1109/icecce47252.2019.8940713. //dx.doi.org/10.1080/21642583.2019.1708829.
[3] K. Sharma, D.K. Palwalia, A modified PID control with adaptive fuzzy controller [26] S. Li, H. Chen, M. Wang, A.A. Heidari, S. Mirjalili, Slime mould algorithm: A
applied to DC motor, in: 2017 International Conference on Information, Commu- new method for stochastic optimization, Future Gener. Comput. Syst. 111 (2020)
nication, Instrumentation and Control, ICICIC, 2017, http://dx.doi.org/10.1109/ 300–323, http://dx.doi.org/10.1016/j.future.2020.03.055.
icomicon.2017.8279151. [27] X.S. Yang, Nature-Inspired Optimization Algorithms, first ed., Elsevier insights,
[4] O.A. Somefun, K. Akingbade, F. Dahunsi, The dilemma of PID tuning, Annu. Rev. Elsevier, Amsterdam, 2014, literaturangaben.
Control 52 (2021) 65–74, http://dx.doi.org/10.1016/j.arcontrol.2021.05.002.
5
Computers and Electronics in Agriculture 203 (2022) 107487
Contents lists available at ScienceDirect
Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag
EOdal: An open-source Python package for large-scale agroecological

research using Earth Observation and gridded environmental data
Lukas Valentin Graf a,b ,∗, Gregor Perich a , Helge Aasen a,b
a Crop Science, Institute for Agricultural Science, ETH Zurich, Universitätsstrasse 2, CH-8092 Zürich, Switzerland
b
Earth Observation of Agroecosystems Team, Devision Agroecology and Environment, Agroscope, Reckenholzstrasse 191, CH-8042 Zürich, Switzerland
ARTICLE INFO ABSTRACT
Dataset link: https://doi.org/10.5281/zenodo.7 Earth Observation by means of remote sensing imagery and gridded environmental data opens tremendous
278252 opportunities for systematic capture, quantification and interpretation of plant–environment interactions
Keywords: through space and time. The acquisition, maintenance and processing of these data sources, however, requires
Satellite data a unified software framework for efficient and scalable integrated spatio-temporal analysis taking away the
Python burden of data and file handling from the user. Existing software products either cover only parts of these
Open-source requirements, exhibit a high degree of complexity, or are closed-source, which limits reproducibility of
Earth Observation research. With the open-source Python library EOdal (Earth Observation Data Analysis Library) we propose a
Ecophysiology novel software that enables the development of fully reproducible spatial data science chains through the strict
use of open-source developments. Thanks to its modular design, EOdal enables advanced data warehousing
especially for remote sensing data, sophisticated spatio-temporal analysis and intersection of different data
sources, as well as nearly unlimited expandability through application programming interfaces (APIs).
1. Introduction this finding on an exhaustive review of existing software tools. For

example, the philosophy of the OpenDateCube1 initiative is about
Images from Earth Observation (EO) satellites and in-situ observa- integrated analysis of EO data using standardized interfaces. Installing
tions are of great importance for ecophysiological (Caparros-Santiago the software, however, is complex as setting up a database instance
et al., 2021) and agroecological research (Karthikeyan et al., 2020). is required. The Framework for Operational Radiometric Correction
Such data can be used to determine plant traits and allow mapping for Environmental monitoring (FORCE) (Frantz, 2019) is primarily
of plant growing conditions for larger areas using standardized meth- designed for the creation of Analysis-Ready-Data (ARD) but does not
ods (Weiss et al., 2020). Open-access, high-resolution satellite data provide interfaces for data analysis workflows. Analysis workflows us-
such as from the European Space Agencies’ Sentinel-2 (S2) mission ing standardized interfaces are the main subject of the openEO2 project.
can resolve field heterogeneity and provide site-specific farming mea- openEO, however, focuses on cloud environments. Thus, data sets that
sures operationally. Examples include yield estimates (Marshall et al., are not available on externally operated web platforms are currently
2018; Perich et al., 2022), extraction of phenological metrics (Duarte excluded. This applies particularly to (experimental) research data sets.
et al., 2018), variable irrigation rates (Barker et al., 2018) and site-
Researchers therefore often spend a significant amount of time getting
specific fertilization scheduling (Mittermayer et al., 2022). Remotely
the data into an analysis-ready format. Based on the analysis of more
sensed plant traits and their dynamic development over time can
than 3000 productive machine learning pipelines at Google, Xin et al.
further be augmented with environmental covariates such as climate,
(2021) identified great potential for optimization in the area of data
soil and terrain data, as well as information about farm management
management and pre-processing, which also appears to be true in the
to perform integrated analysis on material and energy fluxes across
EO area.
spatio-temporal scales (Asam et al., 2018).
However, accessing, managing and analyzing EO data is complex For these reasons, we developed the Earth Observation Data Analy-
and often requires solid knowledge of geographic information sci- sis Library (EOdal) as an open-source Python (3.8+) package designed
ence and coding to properly handle large spatial data sets. We base to make EO data analysis tools available to researchers without the
∗ Corresponding author at: Crop Science, Institute for Agricultural Science, ETH Zurich, Universitätsstrasse 2, CH-8092 Zürich, Switzerland.
E-mail address: lukasvalentin.graf@usys.ethz.ch (L.V. Graf).
1
https://www.opendatacube.org/
2
https://openeo.org/
https://doi.org/10.1016/j.compag.2022.107487
Received 16 June 2022; Received in revised form 2 November 2022; Accepted 7 November 2022
Available online 18 November 2022
0168-1699/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
L.V. Graf et al. Computers and Electronics in Agriculture 203 (2022) 107487
Fig. 1. Overview of EOdal and its three layers: The core layer (top) handles different input data sources in a standardized way. The analysis layer (bottom left) builds upon the
core and the processing layer (bottom right) for data maintenance, querying, intersection and integration into user-defined data science pipelines.
need for in-depth knowledge about geoinformation science, coding and are three classes in the EOdal core layer: Bands, RasterCollections and
remote sensing data handling. A key aspect of EOdal is the ability to derived (inherited) sensor-specific classes. The Band class represents
apply spatial data science methods to data from different sources within the base class. In simple terms, a Band refers to a two-dimensional array
a unified framework based on open-source tools. We also intend to pro- referenced in a geo-spatial coordinate reference system. A RasterCollec-
vide researchers with an alternative to proprietary software solutions tion is a collection of zero to 𝑛 Band objects, which can exist in different
such as the widely used Google Earth Engine (Gorelick et al., 2017). spatial reference systems, grid cell sizes (pixel sizes) and spatial extents.
The functionality and structure of EOdal is explained in Section 2 Band objects in a collection are identified by names (e.g., ‘‘blue’’,
of this paper. In Section 3, we present a reproducible use case based on Fig. 1 top right) instead of numeric indices. Bands and RasterCollections
an agricultural research question followed by a discussion in Section 4. can be created from any geo-referenced raster data set understood
by Geospatial Data Abstraction Library (GDAL, e.g., GeoTiff), vector
features (e.g. Shapefile, GeoJSON), and from numerical arrays (Python
2. Description of EOdal libraries, e.g. NumPy, Zarr). Sensor-specific RasterCollections include
all the functionalities and attributes of the RasterCollection class, but
EOdal consists of three layers as shown in Fig. 1. The layers are are tailored to the requirements and capabilities of specific imaging
organized in a triangle to emphasize their inter-dependencies. The core sensors such as, for example, S2, or Landsat. The inheritance-based soft-
layer (Fig. 1, top) provides the Python classes required to perform I/O ware design allows the introduction of further sensors and makes the
operations to read and write geo-spatial data sets in a generic way. It core layer extensible also with regard to upcoming future EO platforms
is also the basis for data warehousing, i.e., the storage of metadata. such as the hyperspectral Copernicus expansion mission CHIME3 and
Class inheritance extends its capabilities to specific EO sensors such as beyond.
S2 Multispectral Imager (e.g., for convenient reading of data organized A further central element of the core layer is the collection of meta-
in the Satellite Archive for Europe structure). (Pre-)processing of these data in a spatio-temporal catalog allowing the filtering of records by
datasets is accomplished in the processing layer (Fig. 1, lower right). data source, time period, and geographic region of interest (ROI). EOdal
Processing steps such as spatial reprojection are often a necessity to supports Spatio-Temporal Asset Catalogs (STAC), which are available
in many cloud environments that provide geo-spatial (satellite) data,
combine different datasets for analysis. The analysis layer (Fig. 1, lower
such as Microsoft Azures Planetary Computer,4 Amazon Web Services
left) enables automatized, reproducible EO data management and is
(AWS) Earth5 and the Copernicus Data and Information Access Services
the backbone for data-driven analysis of geo-spatial data sets and their
(DIAS). For local deployment, EOdal offers the possibility to store
spatio-temporal intersection.
metadata in a PostgreSQL database with the spatial PostGIS extension.
Deployment of EOdal is independent of an Operating System. Fur-
In terms of spatial data models and standards, EOdal supports area
thermore, EOdal can be used on local premises (e.g. for processing
(polygons and multi-polygons) as well as point features following the
research datasets) but also in cloud environments for fast access to
general feature model defined in the ISO 19109 standard (ISO, 2015).
large, freely accessible data such as the global S2 or Landsat archive.
This allows point-based in-situ observations, for example from weather
To enable fast and scalable deployment required for large-scale analysis
stations or ground sensors, to be intersected with EO data and auxiliary
tasks EOdal can be installed into containerized environments (Docker
data sources such as Digital Elevation Models (see Section 3).
containers).
2.1. Core layer 3

https://www.esa.int/Applications/Observing_the_Earth/Copernicus/
Going_hyperspectral_for_CHIME
The data model of the core layer (Fig. 1 top) follows the object- 4
https://planetarycomputer.microsoft.com/
5
oriented programming paradigm and includes different classes. There https://aws.amazon.com/de/earth/
2
Fig. 2. Result of a single Jupyter notebook run using EOdal (v0.0.1) on Microsoft Planetary Computer. S2 false-color infra-red composite of the field parcel before the heavy
rainfall events in June and July 2021 (a) and spatial heterogeneity in winter wheat green-up during spring 2022 (b–c). In addition, cumulative growing degree days (GDD) and
daily precipitation sums derived from a nearby weather station are shown (f). The extent of the flood in 2021 – evident as darker areas in (b–c) – corresponds to a depression
visible in the Digital Elevation Model (e). The flood affected plant growing conditions in spring 2022 as shown in the MSAVI time series (d) of a ‘flooded’ pixel (blue) and a
pixel that was little affected by the floods in 2021 (orange). The dashed lines in (d) correspond to the timing of (a), (b) and (c), respectively, whereas the blue rectangle in (d)
indicates the approximate timing of the flooding in 2021.
2.2. Processing layer 2.4. Applications of EOdal
The processing layer (Fig. 1, lower right) provides functionalities for EOdal has already been used for studies in agroecological re-
EO data preparation and (pre-)processing. Data preparation is usually search. Perich et al. (2022) used EOdal for pixel-based prediction of
necessary to enable the merging datasets from different sensors and crop yield using S2 time series. In Graf et al. (2022), EOdal is used to
platforms. This includes image manipulation methods such as spatial propagate radiometric uncertainty from S2 reflectance factors to phe-
resampling and reprojection from one coordinate system into another nological metrics. Moreover, EOdal drives the EO platform at the Swiss
or masking operations to mask out bad quality observations or land Federal Center of Excellence for Agricultural Research, Agroscope,
cover classes. underpinning its relevance within an operational and governmental
environment.6
2.3. Analysis layer
3. Usage example: Interpreting in-field growth heterogeneity
through time with topographic and meteorological data
The analysis layer (Fig. 1, bottom) is used for the analysis of data
from EO satellites and environmental covariates such as meteorological, We illustrate here the potential of EOdal for integrated spatio-
terrain, soil and land use data. It builds upon the core layer to query temporal analysis over an example ROI from western Switzerland near
and access different data sources and the processing layer to prepare the Lake Neuchâtel (46.98◦ 𝑁, 7.07◦ 𝐸). All results were produced by a single
data analysis-ready. The underlying complex data management such as Jupyter notebook (https://doi.org/10.5281/zenodo.7278252) running
the merging of data from satellite tiles and filling of no-data values is EOdal v0.0.1 on Microsoft Planetary Computer. Due to large-scale
hidden from the user in the processing layer (see Section 2.2). land subsidence, the entire region – a former peat land – is subject
EOdal provides interfaces to widely-used open-source Python li- to increased flood risk (Egli et al., 2020). An agricultural parcel (12
braries (e.g., geopandas, numpy, xarray). This allows users to integrate ha) was affected by flooding during heavy rainfall (360 mm within
their own EO processing workflows and modules (c.f. Section 2.4) 30 days) between mid-June and July 2021, causing flood damage to
to e.g. estimate biochemical, structural or integrated traits such as
chlorophyll, leaf area index, yield or land surface phenology (Fig. 1,
6
lower left). http://www.eoa-team.net/
3
the previously homogeneous green canopy. Fig. 2 shows S2 derived Declaration of competing interest
false-color infra-red images of the parcel in 2021 before the flood
(reddish tones in Fig. 2a). In the spring 2022, differences in canopy One or more of the authors of this paper have disclosed potential or
greenness of the emerging winter wheat are evident (Fig. 2b and c). pertinent conflicts of interest, which may include receipt of payment,
Darker areas, which indicate lower soil cover, can be discerned in either direct or indirect, institutional support, or association with an
the otherwise red-colored canopy. The spatial pattern within the field entity in the biomedical field which may be perceived to have potential
corresponds to the patterns within the DEM (Fig. 2e) and reveals that conflict of interest with this work. For full disclosure statements re-
parts with lower elevation values were more impacted by flooding than fer to https://doi.org/10.1016/j.compag.2022.107487.Lukas Valentin
the rest of the field. Extraction of the Modified Soil Adjusted Vegetation Graf reports financial support was provided by Swiss National Science
Index (MSAVI, Qi et al., 1994) time series from two selected S2 pixels Foundation.
(Fig. 2d) allows to evaluate the impact of the flood in 2021 on the
growth dynamic in 2022 and confirms the findings from the false-color Data availability
S2 images.
As additional usage example, the growth pattern of a field is inves- Code required to rerun the analysis presented in this paper is
tigated in relation to meteorological data, often used to validate and available under: https://doi.org/10.5281/zenodo.7278252.
calibrate ecophysiological growth models. In this case (Fig. 2f), the
cumulative growing degree days (in ◦ C) and cumulative daily rainfall Acknowledgments
amount (in mm) were derived from a nearby weather station (Ins)
operated by the Swiss Federal Office of Meteorology and Climatology The authors thank Achim Walter from the Crop Science group
(MeteoSwiss) calculated from November 1st 2021. at ETH for providing the IT infrastructure and in particular Norbert
Kirchgessner for his support with data storage and implementation.
Furthermore, we thank Alfred Burri for on-site support and contri-
4. Discussion and conclusion
butions to the usage example and Fabio Oriani for valuable com-
ments on the manuscript. The development of EOdal was conducted
The usage example shown in Fig. 2 highlights how EOdal can be within the project ‘PhenomEn’ funded by the Swiss Science Foundation
used to combine different environmental covariates with vegetation (grant number IZCOZ0_198091). GP acknowledges funding through the
dynamics obtained from satellite time series to develop a holistic project ‘DeepField’ of the Swiss Federal Office for Agriculture (BLW).
understanding of plant–environment interactions. This simple example
shows how EOdal lowers the barriers of entry to EO analysis unlocking References
potential for data-based decision-making in agroecology. In addition,
by using Docker and cloud infrastructure, the analysis can be extended Asam, S., Callegari, M., Matiu, M., Fiore, G., De Gregorio, L., Jacob, A., Menzel, A.,
to larger spatial and temporal scales. For the future we envision an Zebisch, M., Notarnicola, C., 2018. Relationship between Spatiotemporal Vari-
ations of Climate, Snow Cover and Plant Phenology over the Alps—An Earth
ecosystem of user modules that build on EOdal, complement themselves Observation-Based Analysis. Remote Sens. 10 (11), 1757. http://dx.doi.org/10.
and are developed and made freely available to the EO community. 3390/rs10111757, URL: http://www.mdpi.com/2072-4292/10/11/1757.
EOdal is in line with recent developments in agriculture that allow Barker, J.B., Heeren, D.M., Neale, C.M.U., Rudnick, D.R., 2018. Evaluation of vari-
able rate irrigation using a remote-sensing-based model. Agricult. Water Manag.
(technically inexperienced) users to run complex analyses: Godara
203, 63–74. http://dx.doi.org/10.1016/j.agwat.2018.02.022, URL: https://www.
et al. (2022) developed a platform called ‘‘AgriMine’’. The platform sciencedirect.com/science/article/pii/S0378377418301082.
enables spatio-temporal analysis of issues in Indian agriculture based Caparros-Santiago, J.A., Rodriguez-Galiano, V., Dash, J., 2021. Land surface phe-
on help-desk calls improving the Indian agricultural extension service. nology as indicator of global terrestrial ecosystem dynamics: A systematic
Improvements in variety testing were enabled by an online platform review. ISPRS J. Photogramm. Remote Sens. 171, 330–347. http://dx.doi.org/
10.1016/j.isprsjprs.2020.11.019, URL: https://linkinghub.elsevier.com/retrieve/pii/
in China that makes processes more efficient and lines up with re- S0924271620303269.
cent incentives in high-throughput phenotyping efforts (Pan et al., Duarte, L., Teodoro, A.C., Monteiro, A.T., Cunha, M., Gonçalves, H., 2018.
2022). Similarly, EOdal can contribute to increased use of EO data in QPhenoMetrics: An open source software application to assess vegetation phe-
agriculture. nology metrics. Comput. Electron. Agric. 148, 82–94. http://dx.doi.org/10.
1016/j.compag.2018.03.007, URL: https://www.sciencedirect.com/science/article/
Although we focus on agriculture (Section 2.4), EOdal is not limited pii/S0168169916302162.
to it. Rather, EOdal can be used in all disciplines that process EO data Egli, M., Gärtner, H., Röösli, C., Seibert, J., Wiesenberg, G.L.B., Wingate, V., 2020.
and require open, reproducible data analysis workflows. Thus, we ex- Landschaftsdynamik im Gebiet des Grossen Mooses - Moorböden, Wassermanage-
ment und landwirtschaftliche Nutzung im Spannungsfeld zwischen Produktivität
pect EOdal to trigger further open-source developments, either through
und Nachhaltigkeit. Schriftenreihe Physische Geographie 68, http://dx.doi.org/
the release of additional Python packages that build on the existing 10.5167/uzh-190633, URL: https://www.zora.uzh.ch/id/eprint/190633/, ISBN:
functionalities, or through its integration into existing data processing 9783906894164 Num Pages: 91 Place: Zürich Publisher: Geographisches Institut
frameworks. Ultimately, EOdal is intended to provide researchers with der Universität Zürich.
Frantz, D., 2019. FORCE—Landsat + sentinel-2 analysis ready data and beyond.
an alternative to proprietary software platforms while also easing the
Remote Sens. 11 (9), 1124. http://dx.doi.org/10.3390/rs11091124, URL: https:
burden of data management for practitioners. For this reason EOdal is //www.mdpi.com/2072-4292/11/9/1124, Number: 9 Publisher: Multidisciplinary
particularly well suitable for educational activities in the field of EO. Digital Publishing Institute.
This supports the transition to reproducible science, benefiting the EO Godara, S., Toshniwal, D., Parsad, R., Bana, R.S., Singh, D., Bedi, J., Jhajhria, A.,
Dabas, J.P.S., Marwaha, S., 2022. AgriMine: A Deep learning integrated spatio-
community as a whole.
temporal analytics framework for diagnosing nationwide agricultural issues using
farmers’ helpline data. Comput. Electron. Agric. 201, 107308. http://dx.doi.org/10.
CRediT authorship contribution statement 1016/j.compag.2022.107308, URL: https://www.sciencedirect.com/science/article/
pii/S0168169922006172.
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017.
Lukas Valentin Graf: Conceptualization, Software, Investigation, Google earth engine: Planetary-scale geospatial analysis for everyone. Remote Sens.
Visualization, Writing – original draft, Writing – review & editing. Gre- Environ. 202, 18–27. http://dx.doi.org/10.1016/j.rse.2017.06.031, URL: https://
www.sciencedirect.com/science/article/pii/S0034425717302900.
gor Perich: Conceptualization, Software, Writing – original draft, Writ-
Graf, L.V., Gorroño, J., Hueni, A., Walter, A., Aasen, H., 2022. Propagating the sentinel-
ing – review & editing. Helge Aasen: Supervision, Writing – original 2 top-of-atmosphere radiometric uncertainty into land surface phenology metrics
draft, Writing – review & editing. using a Monte Carlo framework. (Submitted for Publication).
4
ISO, 2015. ISO 19109:2005 Geographic information — Rules for application schema. Perich, G., Turkoglu, M.O., Graf, L.V., Wegner, J.D., Aasen, H., Walter, A., Liebisch, F.,
URL: https://www.iso.org/standard/39891.html. 2022. Pixel-based crop yield mapping and prediction using spectral indices and
Karthikeyan, L., Chawla, I., Mishra, A.K., 2020. A review of remote sensing applications neural networks on Sentinel-2 time series data. (Submitted for Publication).
in agriculture for food security: Crop growth and yield, irrigation, and crop losses. Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H., Sorooshian, S., 1994. A modified
J. Hydrol. 586, 124905. http://dx.doi.org/10.1016/j.jhydrol.2020.124905, URL: soil adjusted vegetation index. Remote Sens. Environ. 48 (2), 119–126. http://
https://www.sciencedirect.com/science/article/pii/S0022169420303656. dx.doi.org/10.1016/0034-4257(94)90134-1, URL: https://www.sciencedirect.com/
Marshall, M., Tu, K., Brown, J., 2018. Optimizing a remote sensing production efficiency science/article/pii/0034425794901341.
model for macro-scale GPP and yield estimation in agroecosystems. Remote Sens. Weiss, M., Jacob, F., Duveiller, G., 2020. Remote sensing for agricultural applica-
Environ. 217, 258–271. http://dx.doi.org/10.1016/j.rse.2018.08.001, URL: https: tions: A meta-review. Remote Sens. Environ. 236, 111402. http://dx.doi.org/10.
//www.sciencedirect.com/science/article/pii/S0034425718303663. 1016/j.rse.2019.111402, URL: https://www.sciencedirect.com/science/article/pii/
Mittermayer, M., Maidl, F.-X., Nätscher, L., Hülsbergen, K.-J., 2022. Analysis of site- S0034425719304213.
specific N balances in heterogeneous croplands using digital methods. Eur. J. Xin, D., Miao, H., Parameswaran, A., Polyzotis, N., 2021. Production machine learning
Agron. 133, 126442. http://dx.doi.org/10.1016/j.eja.2021.126442, URL: https:// pipelines: Empirical analysis and optimization opportunities. In: Proceedings of
www.sciencedirect.com/science/article/pii/S1161030121002136. the 2021 International Conference on Management of Data. ACM, Virtual Event
Pan, S., Zhao, X., Han, Y., Wang, X., Yang, F., Wang, S., Liu, Z., Zhang, Q., Zhang, Q., China, pp. 2639–2652. http://dx.doi.org/10.1145/3448016.3457566, URL: https:
Wang, K., 2022. Online information platform for the management of national vari- //dl.acm.org/doi/10.1145/3448016.3457566.
ety test of major crops in China: Design, development, and applications. Comput.
Electron. Agric. 201, 107292. http://dx.doi.org/10.1016/j.compag.2022.107292,
URL: https://www.sciencedirect.com/science/article/pii/S0168169922006044.

Porfolio-Đã G P

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Porfolio-Đã G P

Uploaded by

Copyright:

Available Formats

BỘ GIÁO DỤC VÀ ĐÀO TẠO

TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT

BỘ GIÁO DỤC VÀ ĐÀO TẠO

BỘ GIÁO DỤC VÀ ĐÀO TẠO

1 Phạm Hoàng Phúc 21146295

BỘ GIÁO DỤC VÀ ĐÀO TẠO

Thủ Đức, tháng 11 năm 2023

BỘ GIÁO DỤC VÀ ĐÀO TẠO

BỘ GIÁO DỤC VÀ ĐÀO TẠO

1 Phạm Hoàng Phúc 21146295

BỘ GIÁO DỤC VÀ ĐÀO TẠO

Thủ Đức, tháng 11 năm 2023

II. Interview .....................................................................................................................4

III. Title and abstract .......................................................................................................5

IV. Paragraph ................................................................................................................ 12

V. Content Presentation ................................................................................................ 16

VI. Research Paper ....................................................................................................... 18

Biofeedback for human-robot interaction in the context of

1. The aim of this paper is to systematically investigate potential

1. Emphasis is placed on the interaction of different modalities of

1. The integration of biofeedback

1. Biofeedback promotes a more autonomous, flexible and

VI. Research Paper

Fault Diagnosis of AC Servo Motor with Current Signals Based on

1. INTRODUCTION The diagnostic methods to identify the above faults may

978-3-902661-00-5/08/$20.00 © 2008 IFAC 7239 10.3182/20080706-5-KR-1001.2785

2.2 Rotor bar faults 3.1 Hilbert transform

f b = (1 ± 2ks) f , k = 1, 2, 3, K (3) xˆ (t ) = H {x(t )} (5)

0.4 0.4 , j = 0,1, 2, K

-1 -1 wavelet packet can be used for this purpose. After

where fn is the Nyquist Frequency, in the ith frequency band.

Fig. 2 Synchronization procedure of the raw data. h( n) = 2 − j ϕ ( 2 −1 t )ϕ (t − n)

Fault templates 5.3

Find the best matching

Fig. 4 Flow chart of diagnosis processes. (BF: bearing fault,

Applied fault types in this study are broken rotor, faulted

5.2 Fault templates

We make four fault templates consist of broken rotor bar,

Fig. 5 Original current signals.

Take an average Make fault

6 6 We apply wavelet decomposition in produced templates. We

0 0 produce approximation coefficient and detail coefficient in

-8 -8 templates and these are used for a template matching method.

Biofeedback for human-robot interaction in the context of collaborative

* Corresponding author. Tel.: +49-421-218-64838; E-mail address: rueckert@bime.de

Keywords: Human-robot interaction; Assembly; Biofeedback

2212-8271 © 2023 The Authors. Published by Elsevier B.V.

Biofeedback methods Measured body function EMG-based assessment of ergonomic examination

MOG Eyelid closure Analysis of the quality of an HRI using facial

3.1. Application of robot control 3.3. Application of Stress detection

the same application. On the one hand, to meet the redundancy

The evaluation criteria of the utility value analysis are Conversation:

4.4. Measurement methodology

In order to adapt the threshold values of the measurement

makes use of differential GNSS technology. Therefore, the

The core component of the IoT-platform is a server appli-

Fig. 2. Schematic diagram of sensor printed circuit board.

3.2 Trajectory tracking on the symmetry group

If the lift Λ : M × L → g is equivariant, there exists a Ṙ = RΩ× , (9a)

||log(RE )||2 [rad]

||log(RE )||2 [rad]

0 SE2 (3) 0 15 SE2 (3)

||∆T̄ ||2 [N]