Universidad Politécnica de Santa Rosa.

Automotive Systems Engineering

Aguila Jimenez Mario Antonio

PROJECTS FORMULATION AND EVALUATION

Machine learning for vehicle improvements

 1.-Introduction
GENERATE AN IDEA
Machine learning, big data and artificial intelligence (AI) can
offer several significant contributions in the vehicle space.
● Autonomous Driving
● Predictive Maintenance using big data, large amounts of
data related to vehicle performance and health can be
collected and analyzed.
● Improving Road Safety
● Traffic Management
● Cooperative Perception machine learning for connected
autonomous vehicles
● Fuel Economy analyze driving patterns and traffic
conditions to optimize fuel economy and reduce emissions,

2.- Statement of the problem

In the current context of the automotive industry, the adoption

of emerging technologies such as machine learning, big data
and artificial intelligence (AI) has generated significant advances,
offering various contributions to improve efficiency, safety and
sustainability in the field of the vehicles. However, despite these
innovations, challenges and problems persist that require
attention and effective solutions.
Vehicles throughout time have been a fundamental part of the
development of new technologies, leading to autonomous
vehicles through technologies with the potential to enhance
transportation by reducing traffic, improving safety, and
enhancing individual driving experiences.

One of the main challenges is to build an effective system for

making necessary corrections. Developing a system in which
vehicles are connected to a database that allows them to
continuously improve is crucial.

The study is innovative in its application of machine learning

methods in predicting practical field scenarios, encompassing
numerous factors that cannot be considered through
conventional regression methods.

As machine learning techniques become predominant in

autonomous vehicles, especially in their perception subsystem,
we propose the possibility of designing a cooperative perception
system enabled for machine learning for connected autonomous
vehicles.

I believe that the use of big data and machine learning could
yield positive results; however, creating an intelligent system like
this will require many years of work and gaining society's trust in
entrusting their lives to artificial intelligence.
This statement of the problem highlights the need to address
these challenges to fully realize the potential of emerging
technologies in the automotive industry, moving towards a future
of safer, more efficient and sustainable mobility.
3.- Theoretical framework
Background of Machine Learning in Automobiles:

Machine learning serves for the development of programs and

applications that can learn from the data and information they
collect. Machine learning goes hand in hand with Big Data, as it
autonomously processes information based on the received
dataset, continually learning as more data is acquired.

Currently, the combination of neural networks is enhancing the

performance of systems to which they are applied. The
widespread use of digital platforms results in vast amounts of
information, tracing our activities in the use of applications and
web navigation. In this context, the development of Big Data and
its integration into business processes has also contributed to
enhancing algorithms and the advancement of machine
learning.
Machine Learning and Deep Learning capabilities are
continually growing to harness the accessibility provided by
emerging technologies, including wireless technologies.
Furthermore, the future of vehicular networks will revolutionize
both autonomous and conventional vehicles.

Enabling Machine Learning access to internet networks in

vehicles will play a crucial role in constructing the perception
system of autonomous and semi-autonomous vehicles. Despite
significant developments in connected vehicle configurations,
they remain vulnerable to various security measures, such as
sensors, cameras, radars, and other features in vehicles that
share information with onboard sensors and nearby vehicles.

These data can serve various purposes, such as obtaining

information about vehicle kinetics, traffic flow, road and network
conditions, etc. Such data can potentially be used to enhance
the performance of the vehicular ecosystem through adaptive
data-based decision-making. However, it can also be exploited
for malicious objectives. Therefore, ensuring data integrity and
security is of paramount importance to mitigate diverse risks
and attacks on Connected and Autonomous Vehicles (CAV).
Current Applications of Artificial Intelligence in Automobiles:

Perception for an autonomous vehicle is defined as the ability of

a vehicle collecting information and extracting relevant
knowledge from sensor data to develop a contextual
understanding of the environment. Together with localization
and mapping, path planning, decision making, and vehicle
control modules, an autonomous vehicle is able to successfully
navigate itself on roads. Cooperative perception, on the other
hand, enables vehicles to share local perception data with each
other (or infrastructures) through wireless communications. One
of the prime reasons for developing cooperative perception is
the need to maximize the line of sight and field of view of
autonomous vehicles. In addition, it could reduce the uncertainty
in local object detection results and increase perception
accuracy. With increasing situational awareness, cooperative
perception is able to expand vehicles’ field of view, resulting in
safer driving decisions. [1]
Intelligent Transportation Systems (ITS) and computational
systems’ rapid development opened new scientific research in
smart traffic safety with comfort and efficient solutions. Artificial
Intelligence (AI) has been widely used to optimize traditional
data-driven approaches in different research areas [1]. AI-based
on the Vehicle-to-Everything (V2X) system obtains information
from various sources, i.e., car, train, bus, etc., and enables to
increase the realization of drivers and forecast to avoid
accidents. This progression has directed to the opportunity to
understand smart driving, which was built on the idea of copying
real driving comportment, while avoiding human mistakes and
bringing comfortable safety to drivers. Many services have been
invented from crowd and light road traffic to adapting traffic, a
legacy from self-based vehicle systems to the IoV [2]. IoV is
addressed to change the interaction between the vehicles,
roadside stations, on-board stations, and environments to
communicate data and multimedia between various networks.
The motivation of IoV is to be adopted and build the
human-vehicle-roadside onboard IoT Connected services within
the various vehicle and different networks.[2]
Advances in machine learning (ML) have been one of the
biggest innovations of the last decade. Nowadays, ML models are
used extensively in different industrial fields like autonomous
vehicles, medical diagnosis, and robotics to perform various
tasks such as speech recognition, object detection, and motion
planning. Among different ML models,
Deep Neural Networks (DNNs) (LeCun, Bengio, and Hinton 2015)
are well-known and widely used for their powerful representation
learning in high-dimensional data.
The development of safety critical systems relies on stringent
safety methodologies, designs, and analyzes to prevent hazards
at the time of failure. In the automotive field,
ISO26262 and ISO/PAS 21448 are two main safety standards used
to address safety of electrical and electronic components. [3]

CONNECTED AND AUTONOMOUS VEHICLES (CAVS): HISTORY,

INTRODUCTION, AND CHALLENGES
‘’The Society of Automotive Engineers (SAE) has defined a
taxonomy of driving automation that is organized in six levels.
The SAE defined the potential of driving automation at each
level that is described next and depicted in Figure 2. Moreover,
according to a recent scientometric and bibliometric review
article on autonomous vehicles, different naming conventions
have been used over the years to refer to autonomous vehicles.
These names are illustrated in Figure 3; note that the year
denotes the publication year of first paper mentioning the
corresponding name. • Level 0 No automation: all driving tasks
and major systems are controlled by a human driver; • Level 1
Function-specific automation: provides limited driver assistance,
e.g., lateral or longitudinal motion control; • Level 2 Partial
driving automation: at least two primary control functions are
combined to perform an action, e.g., lane keeping assistance
and adaptive cruise control; • Level 3 Conditional driving
automation: enables limited self-driving automation, i.e., allows
the driver to temporarily deviate his attention from driving to
perform another activity but the presence of driver is always
required to retake control within a few seconds; • Level 4 High
driving automation: an automated driving system performs all
dynamic tasks of driving, e.g., monitoring of the environment
and motion control. However, the driver is capable of getting full
control of the vehicle’s safety-critical functions under certain
scenarios; • Level 5 Self-driving automation: an automated
driving system performs all dynamic functions of driving and or
feature thereof is designed to function, including, but not limited
to, driving modes” [18]. ODD refers to the domain of operation
which an autonomous vehicle has to deal with. An ODD
representing an ability to drive in good weather conditions is
quite different from an ODD that embraces all kinds of weather
and lighting conditions. The SAE recommends that ODD should
be monitored at run-time to gauge if the autonomous vehicle is
in a situation that it was designed to safely handle.’’[4]
Challenges and Ethical Considerations

The challenges and ethical considerations associated with

implementing Machine Learning (ML) in automobiles are crucial
to understanding and responsibly addressing the impact of
these technologies on road safety and user privacy. Some of
these challenges and ethical considerations are developed at
length here:
When implementing technologies like artificial intelligence, we
find ourselves faced with decisions involving risk, which imply
ethical situations and cases. Therefore, safety is the utmost
priority in the automotive industry, and machine learning
algorithms must be highly reliable to ensure optimal
performance. The data used in machine learning models
inherently carries biases, which can result in discriminatory
decisions. Efforts should be made to work towards fairness and
equity.

Careful attention to these challenges and ethical considerations

is essential to ensure that the implementation of machine
learning in vehicles is safe, fair, and respectful of fundamental
ethical values. Collaboration between the industry, regulators,
and society at large is key to effectively addressing these
challenges.

In accident situations, the question of responsibility and

decision-making becomes fundamental. Autonomous vehicle
algorithms must be programmed to make quick and safe
decisions in real-time, considering the safety of vehicle
occupants and other road users.

Ethical programming of algorithms is crucial to ensure that

decisions made by an autonomous vehicle are ethical and
socially acceptable. This includes the ability to assess and
minimize harm in emergency situations.

A key ethical principle is harm minimization. Algorithms should

be designed to minimize harm in the event of an accident,
whether by protecting vehicle occupants or making decisions
that reduce the risk of injuries to other people on the road.

4.-Scope
The scope of this project in vehicles would be significant, and the
impact it would have depends on various factors, including the
complexity of the objectives and the amount of resources
available.

The scope of this project could encompass:

1. **Development of Autonomous Driving Algorithms:**

- Designing algorithms for autonomous driving to enhance
vehicle capabilities.
2. **Driver Assistance Systems:**
- Implementing systems to assist and support drivers during
their journey.
3. **Ethics and Safety:**
- Addressing ethical considerations and ensuring the safety of
the implemented technologies.

The focus of this project is on an algorithm that determines the

best user decisions based on big data. This project could find
applications in various contexts to bring about improvements in
processes.
5- Hypothesis:
Given the successful implementation of the machine learning
project in automobiles, it is hypothesized that the following
outcomes will be achieved:

*Improved Vehicle Safety

It is expected that the integration of autonomous driving

algorithms and driver assistance systems will significantly
contribute to reducing traffic accidents, thereby enhancing road
safety.

*Optimization of User Decision-Making

The application of the algorithm focused on data-driven

decision-making should result in the optimization of user
decisions, improving efficiency and the overall driving
experience.

*Impact on Transportation Efficiency

The implementation of artificial intelligence technologies is

anticipated to generate a positive impact on transportation
efficiency, reducing travel times and enhancing traffic flow.

Within the premise that the project is executed successfully,

addressing all practical and ethical challenges appropriately.
6.-Research design:

Developing a 2^2 project with machine learning and artificial

intelligence in autonomous vehicles is highly beneficial for
obtaining important information on how various established
variables affect algorithmic decisions.

1. Identification of Relevant Factors (N):

- MACHINE LEARNING ALGORITHM

- UPDATE FREQUENCY
2. Definition of Factor Levels:

Factor 1: MACHINE LEARNING ALGORITHM

- Level 1: Random Forest
- Level 2: Support Vector Machine (SVM)
Factor 2: Model Update Frequency
- Level 1: Every 1 hour
- Level 2: Every 6 hours
Definition of factors n
2n= factors 2^2=4 factors

3. Construction of Experiment Matrix:

No. Algorithm Type Update Frequency

1 Random Forest Every 1 hour

2 Random Forest Every 6 hour

3 SVM Every 1 hour

4 SVM Every 6 hour

4. Experiment Execution:
Implement each experiment according to the combination of
factors defined in the matrix. For example, execute experiment 1
with Random Forest and updates every 1 hour.
5. Statistical Analysis:
Collect relevant data, such as model accuracy, processing time,
and other performance indicators. Use statistical techniques like
ANOVA to analyze variability and determine the importance of
each factor and its interactions.
6. Conclusions:
Based on the results, we can conclude the optimal algorithm
type and update frequency to enhance the performance of the
machine learning system in vehicles. For instance, we might find
that Random Forest with updates every 1 hour is the most
effective configuration.

This experimental design systematically explores how different

combinations of algorithms and update frequencies affect the
system's performance, providing valuable insights for the
optimization and continuous improvement of the machine
learning system for vehicles.
7- Results Analysis:
Software Used:
- To carry out the results analysis of this project, statistical
software Python with statistical libraries was employed.
Variable of Interest:
- The variable of interest for the analysis is the "Accuracy of the
Machine Learning System for Vehicles," which can be measured
in terms of the model's success rate in predictions on test data.
Reliability and Objectivity:
- Result reliability is ensured through the collection of accurate
data and the execution of controlled experiments. Objectivity is
maintained by adhering to a predefined protocol for
experimentation and analysis, avoiding biases, and ensuring
consistency in result evaluation.
Statistical Hypothesis Testing:
- Statistical hypothesis tests were applied to determine the
RESULTS OF factors and their interactions in the performance of
the machine learning system for vehicles. Some hypothesis tests
that could have been used include:
1. t-Test for Mean Comparison: It could be applied to compare
the average accuracy between different types of algorithms.
2. Analysis of Variance (ANOVA):It would be used to assess if
there are significant differences in model accuracy between
different model update frequencies.

Results Interpretation: - Based on the results of the hypothesis

tests, the relative influence of each factor and their interactions
on the accuracy of the machine learning system is interpreted.
For example, it could be concluded that a specific type of
algorithm has a statistically significant impact on model
accuracy, or that the model update frequency significantly
affects the overall system performance.
Anexos
Factorial Design

A factorial design is a type of experimental design that allows the

study of the effects multiple factors can have on a response.
When conducting an experiment, varying the levels of all factors
simultaneously instead of one at a time allows for the
examination of interactions between the factors.

A complete factorial design is one in which researchers measure

responses for all combinations of factor levels. Minitab offers two
types of complete factorial designs:

1. Complete 2-level factorial designs that only contain factors with

2 levels.
2. General complete factorial designs that include factors with
more than two levels.

In a complete factorial design, all possible levels of each factor

are manipulated in combination with all levels of other factors.
This enables the study of the effect of each factor individually as
well as interactions between factors. Each unique combination of
factor levels is referred to as a "treatment."

This type of design is useful for efficiently identifying the impact

of factors and their interactions on the response variable, as it
allows for the study of multiple factors in a single experiment.
However, as the number of factors increases, the complexity and
number of treatments also increase exponentially, making the
design challenging to handle in practical terms.
The 2^n factorial design is commonly employed in fields such as
engineering, materials science, industrial process research, and
other disciplines where understanding how multiple factors
affect a response or outcome is essential.

ANOVA
Analysis of Variance (ANOVA) for a Single Factor is a statistical
method used to examine differences in the means of three or
more groups.

Typically, ANOVA for a single factor is employed when there is a

single independent variable or factor, and the goal is to
investigate whether variations or different levels of that factor
have a measurable effect on a dependent variable. The central
idea behind ANOVA is to compare variabilities among groups
with variability within groups. If the differences between the
means of the groups are greater than the variabilities within the
groups, it is concluded that there are statistically significant
differences.

ANOVA can be applied in various contexts, including scientific

experiments, clinical studies, data analysis in market research,
among others. There are several types of ANOVA, but one of the
most common is the Single-Factor ANOVA, which is used when
comparing the mean of a variable among three or more groups.

ANOVA for a Single Factor is a test for differences between group

means. It is a statistical method to test the null hypothesis (H0)
that three or more population means are equal against the
alternative hypothesis (Ha) that at least one of the means is
different. Using the formal notation of statistical hypotheses with
k means, we would write:
here µi is the mean of the ith level of the factor.
The t-test

What is a t-test?

A t-test (also known as Student's t-test) is a tool for evaluating the

means of one or two groups through hypothesis testing. A t-test
can be used to determine whether a single group differs from a
known value (a one-sample t-test), whether two groups differ from
each other (independent samples t-test), or if there is a
significant difference in paired measures (a paired samples or
dependent samples t-test).

The t-test is applied when dealing with small samples and the
population standard deviation is unknown. The goal is to assess
whether the observed difference between the means of two
groups is statistically significant or could have occurred by
chance.

The test generates a t-value, which is compared to a critical

value from the Student's t-distribution or is used to calculate a
p-value. If the p-value is less than a pre-established significance
level (commonly 0.05), the null hypothesis is rejected, and it is
concluded that there is sufficient evidence to suggest a
significant difference between the means.
There are different versions of the t-test based on the type of
data and study design. Two of the most common versions are:

1. Independent Samples t-test: Used when comparing the means

of two different groups, and the observations in one group are
unrelated to the observations in the other group.

2. Paired Samples or Dependent Samples t-test: Applied when the

observations in one group are related or paired with the
observations in the other group (e.g., before and after treatment
in the same set of subjects).

Overall, the t-test is a valuable statistical tool for analyzing

differences between two groups and is commonly employed in
scientific research, clinical studies, and other fields where
comparing sample means is necessary.

Yang, Q., Fu, S., Wang, H., & Fang, H. (2021).

Machine-learning-enabled cooperative perception for connected
autonomous vehicles: Challenges and opportunities. IEEE
Network, 35(3), 96-101.[1]
Ali, E. S., Hasan, M. K., Hassan, R., Saeed, R. A., Hassan, M. B., Islam,
S., ... & Bevinakoppa, S. (2021). Machine learning technologies for s
ecure vehicular communication in internet of vehicles: recent
advances and applications. Security and Communication
Networks, 2021, 1-23.[2]
Mohseni, S., Pitale, M., Singh, V., & Wang, Z. (2019). Practical
solutions for machine learning safety in autonomous vehicles.
arXiv preprint arXiv:1912.09630.[3]
Qayyum, A., Usama, M., Qadir, J., & Al-Fuqaha, A. (2020). Securing
connected & autonomous vehicles: Challenges posed by
adversarial machine learning and the way forward. IEEE
Communications Surveys & Tutorials, 22(2), 998-1026.[4]

Glorious Insight. (2022, 23 diciembre). Artificial intelligence in the

automotive industry | Glorious insight | #AI for Automotive
industry [Vídeo]. YouTube.
https://www.youtube.com/watch?v=pHkEhYa4UlI[5]
Diseños factoriales y factoriales fraccionados - MiniTab. (s. f.). (C) Minitab, LLC.
All rights Reserved. 2023.
https://support.minitab.com/es-mx/minitab/21/help-and-how-to/statistical-modeling/
doe/supporting-topics/factorial-and-screening-designs/factorial-and-fractional-factor
ial-designs/#:~:text=Un%20dise%C3%B1o%20factorial%20es%20un,las%20intera
cciones%20entre%20los%20factores[6]
ANOVA de un factor. (s. f.). Introducción a la estadística | JMP.
https://www.jmp.com/es_mx/statistics-knowledge-portal/one-way-anova.html#:~:tex
t=El%20an%C3%A1lisis%20de%20varianza%20(ANOVA,de%20tres%20o%20m
%C3%A1s%20grupos.[7]
La prueba T. (s. f.). Introducción a la estadística | JMP.
https://www.jmp.com/es_mx/statistics-knowledge-portal/t-test.html#:~:text=Una%2
0prueba%20t%20puede%20usarse,de%20muestras%20dependientes%20o%20parea
da).[8]