1

Project’s Preface

3 Contributors
3 About The Team
3 Our Vision
3 Team Members

4 Electrical & Control System

4 Regenerative Break System
5 Adaptive Cruise Control System
6 Overall Control System Structure

8 Mechanical System
8 Mechanical Design
9 Vehicle Modeling
9 Hardware Criteria

11
Artificial intelligence System
12 Computer Vision
13 Natural Language Processing
13 The Main AI Assitant

15 Mobile’s Application & IOT

12 Idea of the Application
15 App’s Design
15 The Packages Used
16 The Communication with Microcontroller

17 Business & Investment 19 Main Aspects

17 Business Model 19 Terminologies
17 Cost Analysis 20 Criteria
2
 Contributors
About The Team Our Vision
Adhering to recent developments
AUM Team in technology, our team has shifted
focus from combustion Engines to
Established in 2009, AUMotorsports electric vehicles which introduce
is the official Formula Racing Team fewer greenhouse gasses and
Of Alexandria University. harmful emissions to The
environment.
Our passion is motorsports, and we
aspire to advance the field through Our methods are now inspired by
innovation and environment Rriendly operations.
diligence ,implementing our We have implemented the use of
engineering background and utilizing new technologies such as Adaptive
sound research. Cruise Control and Regenerative
Braking System as an initiative
We have participated in various calling for energy conservation in
competitions across the Whole the automotive
Global. industry.
FSAE Germany 2010. Team Members
FSAE Germany 2012.
Mechanical System:
Shell Eco-marathon 2015 | Mohamed Osama [CEO]
Philippines, Asia
Control System:
FSAE 2016 Italy. Kirollous Mahrous [Chairman]
Shell Eco-marathon 2016 | IOT & Communication System:
Philippines, Asia Mohamed Refaat [BMS Head]
FSAE 2017 Italy.
Artificial Intelligence System:
Shell Eco-marathon 2017 | Omar Medhat Aly
Singapore, Asia
Shell Eco-marathon 2022 |
France, Europe 3
 Electrical &
Control System
Cruise control systems are widely applied in the automobile
industry.

They help drivers maintain a constant speed for a prolonged

amount of time. Recently, there has been a shift towards
completely automated systems and , hence, the introduction of
the Adaptive Cruise Control System (ACCS), which, in a sense, is
an advanced version of the regular cruise control system.

Regenerative Break to greatly improve overall

System electrical performance of EVs
In our system:
We’ll talk briefly about • Battery capacity = 100 kwh
Regenerative Braking System • Our system recovers 10% of
We tried to recover the power total capacity (10 kwh)
loss in the braking in the friction We will save 10% of the
torque by replacing it with turned recharging cost (in best condition)
the motor into generator by *Recording to the data analysis of
adipate the road to decelerate the average driven distance per
instead of braking. month in USA = 1931
Km/month.
It can solve the range anxiety from
consumers by increasing • In our case the 100kwh
provides us with 615 Km.
driving range by 8-10%
1931km/month requires 3.2
charge cycle.so, our system
In addition, a certain commercial
competitiveness in the future leads to 2.88 charge
market The regenerative braking cycle/month
system (RBS) provides an
effective way 4
to greatly improve overall
Adaptive Cruise
Distance & Speed Control
Control System The system should be able to
maintain a constant relative speed
Idea to the lead vehicle;however, if
The system introduced by our there is no lead vehicle(the road is
team is primarily implemented in open), the speed values should
electric vehicles coupled with a converge with the desired
Regenerative Brake System (RBS). values as in a regular cruise control
system.
These two systems are key for The transient response is of utmost
energy conservation in an importance:transitions between
electric vehicle. control schemes should be smooth
The ACCS would manipulate the and safe;otherwise,the system
velocity of the vehicle with would become energy inefficient
respect to its surroundings while and quite uncomfortable for the
also monitoring the distance to passenger.
adjacent vehicles on the road.
Passenger Comfort
The RBS would then utilize the Excessive acceleration and
inertia of the vehicle to charge the deceleration should be avoided for
battery. the comfort of the passenger.
This method is more efficient The jerk(the time derivative of the
than using a mechanical caliper to acceleration) is used to monitor the
brake as it would just dissipate acceleration and the
energy as heat. deceleration of the vehicle. It
should be kept to a small value.
Some criteria are considered
while implementing the ACCS:
stability, distance and speed
Safety
control, passenger comfort, and In emergency situations, the
safety vehicle should be able to come to a
complete stop, even if it might
be uncomfortable for the
Stability passenger.
The system should yield results
converging to the desired results.
The desired results are
acquired as inputs from the user. 5
Overall Control System Structure

Fig.1. Illustration of the Control System

The system consists of the driver, the observer, the upper level control,
and the lower level control.

The driver can turn the system on or off. The driver also provides the
system's control parameters:

- the desired velocity( the velocity of the vehicle when there is no

lead vehicle in sight)

- the desired headway time( the time after which the host vehicle
and the lead vehicle will collide when the lead vehicle makes a
sudden stop and the host vehicle maintains its relative velocity).

- The desired relative distance can be derived from the headway

time.
6
The observer unit is responsible for making sound estimates for
unmeasurable states such as vehicle mass, the aerodynamic coefficient,
and the road slope form utilizing feedback data such as angular velocity
and acceleration of each wheel.

The sensors are responsible for collecting data about the lead vehicle then
this information is transmitted to the upper level controller ,which decides
the desired states of the host vehicle and its acceleration accordingly.
Then, the upper level controller sends control signals to the lower level
controllers( DC motors controller, brakes controller, transmission
controller, etc.).

The performance of the system is measured and then introduced to the

controller as
feedback to ensure that the system overcomes disturbance and works as
predicted.

7
 Mechanical System
The design and production of
Mechanical Design electric cars require careful
consideration of material choices
(EVs).

An EV's weight, strength,

durability, and general performance
can all be significantly impacted by
the materials used in its
construction.

Fig.2. Side View Right

While stronger materials increase

the car's durability and safety, a
lighter vehicle has better efficiency
and a longer driving range.
In order to reduce carbon emissions
and promote sustainable
transportation, the materials
utilized must also be eco-friendly Fig.3. Side View Left
and sustainable.
To make the production and
purchase of EVs more affordable
for customers, the materials must
also be economically feasible.
Overall, as material choice
influences the performance,
sustainability, and price of the
vehicle, it is crucial for the success
of EVs.

Fig.4. Plan View

Our EV’s prototype is
environmentally friendly and aligns By analyzing the forces on the
with Egypt's 2030 vision. The car's vehicle in the free body diagram in
base is composed of acrylic figure 2.1 and by applying
material, which helps to further Newton’s second law of motion,
reduce weight while keeping high we can easily derive the following
strength, while The body of the car formula
is made of lightweight and durable
fiberglass.
This material combination not only
lightens the car's overall weight
but also improves its performance where m is the mass of the vehicle,
helping to promote sustainable x is the longitudinal position of the
transportation in line with the 2030 vehicle, Fxf and Fxr are
vision. the tire forces at front and rear
This electric vehicle concept stands tires, Faero is the aerodynamic
out as an innovative answer for the resistance, Rxf and Rxr are the
transportation needs of the future rolling resistances at front and rear
thanks to the utilization of these tires, and theta is the slope angle of
eco-friendly materials. the road.

We tend to consider using natural

fibers in the next prototypes Hardware Criteria
versions to enhance the
sustainability and eco-friendliness The Body of the Car
of the vehicles. Natural fibers offer The choice of material for the body
a range of benefits over traditional and the chassis of the vehicle is of
synthetic fibers, including lower utmost importance.
environmental impact, improved
sustainability, and better durability. The body plays a critical role in
power consumption of the vehicle.
Vehicle Modeling To this end, the body of our
prototype is made of fiberglass
which is a lightweight
material and ,thus, serves in
conserving the energy used by the
vehicle.

Fig.5. Vehicle Model

9
Prototype Preface
After we have discussed the theory of operation and control of our model,
it’s time to discuss the
prototype created.We would like to note that this prototype has been
made specifically for this competition.
The prototype is in fact a smaller model of our already fabricated vehicle.

Fig.6. Our Life Sized Vehicle

Electrical Components
Four DC motor (one at each wheel)

Two ln98 (H-bridge)

Battery cell lithium 12V

Power Line
Protection Fuse 3A
Raspberry Pi
Raspberry Pi Camera
Raspberry Pi Power Bank
ESP32 10
 Artificial Intelligence
System
In This Section We’re going to discuss the Software Part of
the System [Programming Codes] .
The Programming Codes written by the Developer is actually
the Brain of any Robot and Sometimes we can control the Whole
Progress or almost the Whole Progress with Just Lines of Codes ,
so that Indicates that the more the Perfection or the more the
accurate and Meaningful the Code , The More the System
becomes more Intelligent [Robot’s Brain].
Along the History the Scientists have defined that with a
Terminology of “Machine Learning (ML)” which’s extension of
AI [Artificial Intelligence] but that time making Predictions
according to supposed training modes and the Type of Data
[ Categorical, Numeric, …etc] , Briefly there is three Types of
Machine Learning which are:
-- Supervised Learning (e.g. Linear Regression)
-- Un-Supervised Learning (e.g. Clustering)
-- Reinforcement Learning (e.g. Markov’s Decision Process)
The Story didn’t end here , as the Scientists have came with
the “Deep Learning (DL)” which is the Next level of Machine
Learning and we’re going to use it mainly in our Robot ,so each
Application based on the DL.
Nowadays All Robot’s -or to be more clear- The Intelligent
Robot’s are based on DL and ML full Systems, To make it Talk
you can use the “Natural Language Processing” Techniques, to
make it See you can use the “Computer Vision” Techniques , to
make it do specific stuff and you are going to train it yourself ,
so you can use the “Reinforcement Learning” Techniques.
11
Computer Vision (CV) Model that was entirely Custom
from Training Phase to Dataset as
This Part is for The Image in that time we’ve get Dataset of
Processing Tasks which’s 877 Images and started Annotating
specifically focused on Detection, them all so at the end we got 1754
For that we have used two Models. Images half of them Masks
[ Bounding Boxes on the Target] ,
The First one is the Standard then all of that get trained of a
YOLOv5 [pre-trained] Model that’s MLOPs platform called Roboflow ,
Normally got 80 different Classes that’s also performed Data
- based on coco128 Dataset that’s Augmentation and Preprocessing
also pre-trained Dataset- like Phases on the Images, after that
[Person, Laptop, Phone, sink, we exported all of it in the form of
Toaster, ..etc] “.xmal” files that contains the
Locations of masks on each
That’s so simple in usage , as you corresponding Image, Then
just have to get access to the Entering that Data on a Custom
main YOLO repository from Github YOLO Code made also by
or either download or Clone the Roboflow, then it got trained and
Repository by GIT then start received Accuracy of 95.7% on our
calling the Model [in our Case Data , Finally exporting our Model
we’ve used PyTorch package to in the extension of “.pt” file and
load the Model but you can use it’s ready to use. That Model
Keras], then by using OpenCV mainly is used to detect only Four
package open Video Stream from Classes [Traffic Light, Stop Sign,
the used Camera and then it’ll Speed Limit Sign, Cross-Walk
start creating Bounding Boxes on Sign].
the detected classes that it sees . We actually got a problem in that,
as it detect these classes perfectly
Actually we face some Overfitting but it also detect another objects
in Detection despite the very high that seems to be similar to the
Validation Accuracy of 96% and main classes and that’s also a
that’s fixable by re training the Problem , we tend to fix it by
same Class but using more Dataset proving the Custom Model with
for training and Anotating [as the More classes and that requires
Dataset should be splitted into mainly more Dataset.
Images and their Masks]. That was
the First Model we used , now But In General everything in the
talking about the Second YOLO desired Situation is pretty Good.
12
Natural Language IMBD reviews in the form of “.csv”
file , our target was “review”
Processing (NLP) column of this file so we’ve
performed some Data Analysis on
Now Talking about one of the Most it and at the end exported that
Challenging and in the Same time column in the form of “.txt”
very Useful Topic in our Robot and file ,that was the input of our
that also takes apart in the Coming Model.
Section which is the NLP. That’s Now setting our Thresholds like
actually a Simulation of Human’s Batch Size [that’s number of
Conversations but at that time it’s samples processed before the
Manually generated according to model is updated], Number of
the Provided Data for Training. Epochs , then performing Data
The Idea of NLP in our case is to Preprocessing like One Hot
make to generate Text if the Input Encoding as our data are not
was something New to the Sparse , Encoding the whole
Assistant or something that you Dataset to be able to read by the
didn’t especially provide an Model , converting Characters to
Answer to it and in the Same time Integers and vice versa.
to be kinda Logic and accurate so The Main Model was Two layers
the best way to do this is to use LSTM , one Dropout Layer of 30%
the RNNs “ Recurrent Neural rate [that’s actually prevents
Networks” which actually got overfitting during training] and so,
extension of LSTMs “Long Short Our Model produced 955,010
Term Memory” Layers which able trainable Parameters.
to generate whole Text Sentence
not just a Word as it can hold After we found that the Model
information Longer . failed we’ve chosen to do
That’s actually what we did but something different that will be
unfortunately the Model didn’t discussed in the Following Section.
work well, it has successfully
trained for exactly [4 Hours : 37
Minutes : 49 Seconds ] and we The Main AI Assistant
have successfully exported it as
Now We’re going to talk about the
“.h5” extension but once we tried
Main Assistant where all of what
to use it , an Error occurred and
we did should exist here in that.
seems like the Model got
That took about 2 Months to
Corrupted .
produce that DEMO with that
Describing what we did , the First
Features , so we’re going to
thing we’ve got the Dataset from 13
 The Assistant can make some
discuss what it can do as much as
Human things like telling you Jokes
we could.
and Facts, How Did we do that?
We actually make something like a
Calling up from the last Section
Database in form of “.json” file
our problem we faced after the
excpecting to hear from the User
Failed Model , The Solution we
the desired Words making the
decided to do is getting a Pre-
assistant understand that , that
trained Model to generate Text
word have been called and that
instead of that one that passed
means you need to say a joke or a
away. We found Hugging Face
fact according to the Input [ User’s
Platform that’s actually a
Input], It generates random Jokes
Magnificent Platform where
every time you call it normally and
Companies and Skilled Data
also random Fact from Wikipedia
Scientists post there Models in
every time it’s called.
various Fields.
The Model that we choose is
The Assistant repeat what you
“GPT-2 small “
said again after you say it just as a
Debugging step to make sure it has
heard you correctly and then it
flows through that text as it’s
converted to multiple Strings and
begin checking which feature it
should call at the moment.
Fig.7.GPT-2 Models Versions
One of it’s Features is telling you a
That’s a pre-trained Model on a Complete Weather Forecast
very large corpus of English data in according to your City with a
a self-supervised fashion. By the unique ID and that progress is
way we’ve activated Early actually called “Web Scrapping
Stopping while Calling that Model “ where we extract information
just to prevent Damaging the from something called an “API”
Accuracy. [Application Programming
Interface], the API used in our case
Ok Pretty good , now the Text was Open Weather Map API.
Generation Model is Ready to use
and has been successfully added to
the Assistant. Now an Important
Question what does that Assistant
can do? Fig.8.Open Weather API 14
We’ve also added Emotion Detection Model that predicts your Emotions
according to what you said. We’ve created two Models of that, One of
them is self made by normal Machine Learning Methods using “Sklearn”
Package but it’s Limitations were that the Prediction isn’t that Accurate,
so We’ve searched and found another Model [pre-trained] and it’s called
“NRCLex” [National Research Council Canada’s Emotion Lexicon] that’s a
MIT-approved PyPI project by Mark M. Bailey which predicts the
sentiments and emotion of a given text, That’s also got a problem that
when it can’t detect the Emotion it returns None so once it returns None
the Assistant will switch to use the Old Model [you can just say that’s a
plan B as it’s better than nothing].

The Assistant got the ability to switch between voices [ Male, Female,
Indian-Accent] as that’s what the Speech Recognition Package that
perfom that provides.

Also you can store some Orders in the Assistant's Memory and then you
got the Ability to add more and to remove one of them , the Assistant will
automatically repeat the name of the Order or the Task.

The Name our Assistant is Tornado

Mobile’s Application
Once you open it up you will find a
Idea of the Application Button at the Home Page which will
The Main Purpose or the Idea of take you directly to a Page that
creating and Building that Application Display the Values once you clicked it
is to get a Live Video Stream from the on. The App also got the Live Stream
Robot to the Mobile Phone and also Page and a Side Edge contains
display the Values of the Sensors exist Information about Team Members.

App’s Design The Packages Used

The Application contains mainly three We’ve totally used about Seven
Pages provided with a Curved Packages and in the following
Navigation Bar to facilitate browsing , lines, 15
we’re going to discuss and explain and smooth animation. It has a rich
all of them briefly. We’ll begin set of features, and it is
with cupertino_icons Package. completely customizable and
extendable.
cupertino_icons Package | This
is an asset of a Github repository Kdgaugeview | KDGaugeView is a
containing the default set of icon simple and customizable gauge
assets used by Flutter. control for Android
curved_navigation_bar | A
Flutter package for easy The Communication
implementation of curved
navigation bar.Also it’s the main
with Microcontroller
package in our App as it’s used to The Communication between the App
connect all pages together. and the Robots is considered to be in
flutter_vlc_player | we used three phases.
this package to show live
Firstly webview package but it wasn’t
stream in app.
that efficient.
flutter_native_splash | When
your app is opened, there is a Secondly flutter_bluetooth_serial
brief time while the native app package but we failed to send
loads Flutter. By default, during command to raspberry pi used in this
this time, the native app displays Project
a white splash screen. This
package automatically generates Thirdly we’ve tried to use firebase
iOS, Android, and Web-native which is an app development platform
code for customizing this native that helps you build and grow apps .
splash screen background color and we would install it on raspberry pi
and splash image. Supports dark but unfortunately we didn’t have
mode, full screen, and platform- enough time to search about it and
specific options install it on our app and our robots
flutter_launcher_icons | A
command-line Tool which
simplifies the task of updating
your Flutter app's launcher icon.
syncfusion_flutter_charts |
Create various types of cartesian,
circular and spark charts with
seamless interaction,
responsiveness, 16
 Business &
Investment
Business Model

Cost Analysis
COMING SOON..
17
 Main Aspects
Terminologies
Robotics | is an interdisciplinary Servo Motor | is a rotary Machine Learning | s a field of inquiry
branch of computer science and actuator or linear actuator that devoted to understanding and building
engineering. allows for precise control of methods that 'learn', that is, methods
Geared Motor | is a component angular or linear position, velocity that leverage data to improve
whose mechanism adjusts the and acceleration. performance on some set of tasks.
speed of the motor, leading them Motor Driver | acts as an interface Deep Learning | is a subset of machine
to operate at a certain speed between the motors and the learning, which is essentially a neural
IMU | Inertial Measurement Unit control circuits. network with three or more layers.
Working & Applications. Pandas | is a software library Natural Language Processing | Natural
Lithium Ion Battery | A lithium-ion written for the Python language processing (NLP) refers to
or Li-ion battery is a type of programming language for data the branch of computer science—and
rechargeable battery which uses manipulation and analysis. more specifically, the branch
the reversible reduction of lithium Matplotlib | is a plotting library for of artificial intelligence or AI—
ions to store energy. the Python programming language concerned with giving computers the
PID | PID means proportional, and its numerical mathematics ability to understand text and spoken
integral and the derivative is a extension NumPy. words in much the same way human
controller used to control the Seaborn | is a Python data beings can.
processes by adjusting the visualization library based on Computer Vision | is a field of artificial
manipulated variable to keep matplotlib. intelligence (AI) enabling computers to
process variable at a set-point. NumPy | is a library for the Python derive information from images,
Infrared Sensor | An infrared programming language, adding videos and other inputs.
(IR) sensor is an electronic device support for large, multi- YOLO | an algorithm that uses neural
that measures and detects dimensional arrays and matrices, networks to provide real-time object
infrared radiation in its along with a large collection of detection.
surrounding environment. high-level mathematical functions CNN | a kind of network architecture
Ultrasonic Sensor | is an electronic to operate on these arrays. for deep learning algorithms and is
device that measures the distance PyTorch | open source machine specifically used for image recognition
of a target object by emitting learning framework that and tasks that involve the processing
ultrasonic sound waves, and accelerates the path from research of pixel data.
converts the reflected sound into prototyping to production RNN |A recurrent neural
an electrical signal. deployment.. network (RNN) is a class of artificial
Microcontroller | is a small Sklearn | is a free software neural networks where connections
computer on a single VLSI machine learning library for the between nodes can create a cycle,
integrated circuit chip. Python programming language. It allowing output from some nodes to
LIDAR Sensor | a type of laser features various classification, affect subsequent input to the same
distance sensor that measures the regression and clustering nodes.
range, or depth from a surface. algorithms including support- LSTM | Long short-term
Autonomous Car | is a vehicle vector machines,..etc. memory (LSTM) is an artificial neural
capable of sensing its environment DataFrames | is a data structure network used in the fields of artificial
and operating without human that organizes data into a 2- intelligence and deep learning.
involvement. dimensional table of rows and Speech Recognition | is an
Arduino UNO | is a microcontroller columns, much like a spreadsheet. interdisciplinary sub-field of computer
board based on the ATmega328P Keras | is an open-source software science and computational linguistics
(datasheet). library that provides a Python that develops methodologies and
Atmega328P | is a high interface for artificial neural technologies that enable the
performance yet low power networks. recognition and translation of spoken
consumption 8-bit AVR OpenCV | a library of programming language into text by computers with
microcontroller that's able functions mainly aimed at real-time the main benefit of search-ability.
to achieve the most single clock computer vision. API | An application programming
cycle execution of 131 powerful Web Scraping | interface (API) is a way for two or
instructions. Scraping a web page involves more computer programs to
fetching it and extracting from it. communicate with each other.
18
Criteria

Font | Radian
Main Titles Font Size | 45
Sub-Titles Font Size | 26
Page’s Text Font Size | 18
Article’s Title Font Size | 22
Article’s Text Font Size | 12-16

♾ The First Word in each text, Titles and Sub-Titles are

always set to be BOLD
♾ Important Titles/ Names/ Terminologies found in any
Text will be colored by Orange
♾ The Normal Text will be in White
♾ The Questions & Figure Numbers are in Dark Yellow
♾ Terminology Title will be in Orange
♾ After each section a Line is add under the Section which
splits the Sections between each other

19
20