AI IN AUTONOMOUS VEHICLES

1. Introduction

Artificial intelligence (AI) uses data, computers and technology to

simulate the human mind’s problem-solving and decision-making
abilities (Anjum, 2021). AI can be defined as “the study of agents
that receive precepts from the environment and perform actions”
(Harris, 2022). It is essentially the endeavour of producing
systems with human-like cognitive behaviour such as the ability
to reason, solve a problem, discover meaning and perceive from
past experience and act accordingly. Machine Learning and Deep
Learning approaches are combined in AI (Copeland, 2022). Today,
AI has numerous uses.AI has increasingly gained importance due
to its ability to address various problems in business. AI is
making our daily lives more convenient and efficient. One of the
growing applications of AI is in the field of automotive industry and
self-driving cars are an excellent example of that. (Education,
2022).

Self-driving vehicles, also known as autonomous or driverless

cars, are cars or trucks which do not require human drivers to take
control, for safely operating the vehicles. Such cars are composed
of sensors in software to control, navigate and drive the vehicle.
Self-driving cars are essentially built using artificial intelligence. In
self-driving cars applications of AI can be deployed in conjunction
with advanced technological innovations like GPS, radar, camera,
cloud services and control signals. AI can further enhance users’
experience by adding value features such as blind-spot
monitoring, emergency braking and driver-assist steering
(Dilmegani, 2022). The problems like poor road safety, lesser
independence for the disabled, high costs, less productivity,
traffic congestion, high travel time, environmental pollution
associated with conventional cars can be prevented with self-
driving cars (Benefits of Self-Driving Vehicles, 2018). Today the
vehicles are not just machines they are intelligent, highly
advanced, technological, and innovative machines. The main
motivations behind research on autonomous vehicles (AV) are
safe driving, increase in population and vehicles on the road,
comfortable and stress free driving and effective use of available
resources (Parekh et.al, 2022).
According to Bathla et.al (2021) AI powered applications play a
major role in designing AVs intelligent system especially in
improving the safety standards (2021). The study also
emphasised that in order to implement AI in practical complex
environments, the autonomous system needs to be integrated
with multiple advanced technologies like Internet of Things (IoT),
cloud computing and block chain. IoT lets AVs collect relevant
data automatically. IoT sensors collect road traffic related data at
various traffic signals and AI models use this data to take further
decisions. AI powered Natural Language Processing (NLP) and
speech recognition applications are used to understand the text
and speech instructions in AVs. Safavi et al (2021) discuss the
functioning of sensors in autonomous vehicles. Advanced neural
networks are used to predict the malfunctioning of sensors such
as faulty sensor prediction, identification, and isolation.

1.1. Current Scenario of the Autonomous Vehicles Market

GreyB (2022) discussed the findings of various research reports
and summarized that the self-driving vehicles market has the
potential to be worth $87 billion by 2030. According to him there
will be no Level 4 or Level 5 vehicles in operation by 2030. Level 2
vehicles are expected to account for 92 percent of total market
share, with Level 3 vehicles accounting for the rest. He further
highlighted the findings of Allied Market Research that the market
might grow from $54.23 billion in 2019 at a CAGR of 39.47 % to
reach $556.67 billion in 2026. It also predicts that Europe would
have the highest CAGR of 42.6 % between 2019 and 2026. As
shown in the Figure (1), by 2035, driverless vehicles are expected
to account for 25% of total car sales, with 15% being partially
autonomous and 10% beingfully autonomous compared to 12.4%
in 2025. As per most industryexperts, North America will become
the leading market for self- driving vehicles. Further, the study
states that the United States will be a leader in the autonomous
vehicle market.
Figure 1: The Share of Autonomous Vehicle Sales in the Total
Market(2025 vs 2035)

Source: (GreyB, 2022)

However, McCarthy analysed research surveys and reported that
there are other countries that are better prepared to adopt self-
driving cars than the United States (McCarthy, 2018). As shown
in Figure
(2) the Netherlands has the best level of preparation for
autonomous driving. The survey ranked countries based on 26
variables divided into four categories: policy and regulation,
technology and innovation, infrastructure, and consumer
acceptance. With outstanding infrastructure, a highly supportive
government, and passionate adoption of electric vehicles, the
Netherlands ranked firstin the rating. Singapore ranks second, after
amending its Road Traffic Act to allow self-driving cars to be tested
on its streets. The United States high rank in the index is due to its
intensive testing programme and because it has the largest number
of businesses developing self- driving technology of any country. It
scored well in technology and innovation, but performed poorly in
the infrastructure and in the consumer adoption of autonomous
vehicles.

2. Consumer’s Attitude towards Self-Driving Cars

Buchholz (2020) reported in Statista the results of a survey conducted
in six countries, regarding the consumer perception of the trust and
safety of self-driving cars. The results are shown in Figure (3). Overall
the trust in the safety of autonomous vehicles has increased over the
initial three years, even though the rate of increase slowed. Improving
technology led to increased trust among people of South Korea and
Japan. However, accidents caused by autonomous vehicles, like
Tesla that caused accidents led to doubts about the safety of self-
driven cars. This is reflected in the fluctuating acceptance. The people
of China have been accepting of autonomous vehicles and thus,
the percentage of doubt has decreased over the initial three years.
Technology concerns, however, resurfaced in the country in 2020, as
they did in India.

3. Review of Literature

Ma et al., (2020) observed that AI advancements have sped up

the development and spread of self-driving vehicles in the
transportation sector. AI has become a key component of AVs for
understanding the external environment and for making effective
decision when in motion, thanks to the availability of extensive
data from numerous sensor devices and improved computational
capabilities. It is critical to understand how AI works in AV
systems in order to attain the objective of full automation i.e., self
-driving. Frtunikj (2018) found that data preparation, model
generation, and model delivery are the three primary stages in the
development of deep learning systems for self-driving vehicles.
Vinkhuyzen & Cefkin (2016) observed that Autonomous
Vehicles are trained by computer programs and in case of
uncertainty, they don’t have the capability to judge the
environment like humans. Engineers need to pre define all
possible situations. This can easily be done in games like chess,
where the program defines all the possible ways and rules to
follow, but in the
case of Autonomous Vehicles it is complex as it has to deal with a
real-world environment like humans do.

Faisal et al. (2019) studied and identified that Autonomous

Vehicles can perform five basic operational functions when they
can fully replace human drivers. These functions are localization,
planning, perception, control, and management. These
functions give them an advantage over conventional vehicles in
platooning, eco driving, crash avoidance, lane keeping, lane
changing, parking , and adapting to traffic, signals, pedestrians
and other objects in the environment. Gambino et al. (2020)
studied computers in a Computers are social actors (CASA)
theoretical background for examining human perceptions on
artificial intelligence. CASA states that humans form a social
relationship with machines because they interact with computers
in the same way they interact with humans. Cunneen et al. (2019)
emphasised that both the safety issue and the ethical argument
are concerned with predicting the societal, ethical, and legal (SEL)
effectsof AVs on decisional competence. The disparity, according
to a critical analysis of the two approaches stem from a lack of
engagement at the required meta-level to build insightful
appropriate conceptual models of Autonomous Vehicles’
decisional capacity, as well as a failure to consider the
significant differences in how society and users understand
human and computer decision-making. In reality, the automated
driver actions and human driving judgments is yoked to the basic
subject of AV’s societal, ethical, and legal (SEL) impact. This
emphasizes the importance of questioning how AV driving
decisionsare conceptualized.

3. Discussion
3.1. Scenario of Artificial Intelligence
Between 1940 and 1960, there was a significant interest in the
connection between technological advancements and a system
which can understand the combination of machines and humans.
The word AI was first coined by John McCarthy in 1956. AI has
drastically changed business operations today. Agriculture,
education, automotive, healthcare and finance are some sectors
that have shown a rapid growth. The aim of implementing AI is to
replace technology and not jobs. AI’s impact is not limited to
businesses but voice assistants and smart robots are all around
us. The use of AI based systems like chatbots and Alexa have
changed every area of our lives(Council of Europe, 2022).

3.2. Levels in Autonomous Vehicles

Assis (2018) studied the research report of Society of Automotive
Engineering (SAE), for autonomous vehicles and discussed that
there are six levels of Advanced Driver Assistance Systems
(ADAS). As shown the in the Figure (5) the steps towards a
completelyautonomous car are as follows:

• Level Zero: No automation, and all dynamic driving

activities, such as acceleration or slowing, steering,
 braking, and so on,are performed by humans.

• Level One: Driver assistance via an

acceleration/decelerationor steering system that uses data
about driving circumstances

• Level Two: Partial vehicle automation that includes

automated acceleration, deceleration, and steering

• Level Three: Conditional automation of the driving mode,

which involves effective takeover by an automated driving
system in response to a request from the driver

• Level Four: High automation refers to a vehicle that can

handle all driving operations under specific situations
without the assistance of a human driver

• Level Five: Complete automation refers to a vehicle

that is capable of performing all driving jobs/functions in
all conditions.

Figure 5: Five Stages of Autonomy

Source: (Assis, 2018)

3.3. Applications of AI in Autonomous Vehicles

To build an autonomous vehicle that drive like humans, producers
require to furnish these vehicles with combined sensory and
cognitive functions. Further, they need to be provided with the
operational capabilities that humans use while driving a vehicle
(Gadam, 2019).AI for autonomous vehicles includes three steps: i.
Data collection ii.Path planning iii. Action.

i. Data Collection
As shown in figure (6) automotive vehicles use sensor devices
like lidars, radar, camera to collect data from the vehicle and the
environment. This data includes objects on the road, traffic, road
infrastructure and other vehicles.

Figure 6 : Components of Autonomous Vehicle

Source: The role of artificial intelligence in autonomous vehicles,

2020 (Hristozov A, 2020)

• Ultrasonic: Ultrasonic sensors, also known as SONAR

(SOund NAvigation Ranging), use ultrasound waves to
send and receive signals from nearby objects. Autonomous
Vehicle Technology Report (2020) explains that the
transducer is the main component of the ultrasonic sensor.
It receives a digital signal as output from Engine Control
Units (ECU). Actually, analogue signals that are amplified
and converted as digital signals. This is used for
measuring the position of the objects close to the vehicle
when parking (Jahromi, 2021).

• Cameras:

As shown in Figure (9), the data is collected through high

resolution cameras, depth cameras, and different types of
sensors which are placed on each side of the car (left,
right, front, and rear) to get a 360-degree view of the
environment. Most new cars today already use cameras
for assisted parking. Khayyam et al. (2019) discussed that
computer vision algorithms are used to analyse the signals
from various sensors and cameras. The image produced
from these cameras are transformed from lower to higher
level of information. Computer vision segments classify
 the physical objects and finally reconstructs 2D images
into 3D.

• Data Processing

The figure (15) shows the data processing system in

autonomous cars. In AV the car’s software processes
input data, primarily from sensors, to provide the
necessary signal values for controllers, ensuring that the
car is moving safely in all scenarios. Sensors detect any
road, traffic, vehicles and other obstacles like humans by
collecting all the relevant information about the objects.
Further, these sensors have higher perception than
humans. Extremely sophisticated sensors and extensive
algorithms are required to achieve safe, autonomous driving.
Machine learning techniques and neural networks are used
to classify objects in the video data. It is essential that the
variety of sensors have distinct hardware/ software
modules. This enables concurrent data processing and, as
a result, faster decision-making. Each sensor unit can run
its own AI algorithm and then transmit its findings to other
devices or to the central processing computer.

ii. Path Planning

Path planning for AV is defined as finding the safest, most

convenient, and economically profitable paths from point A to point
B by utilising previous driving experiences, which helps the AI
agent make better decisions in the future. As shown in the figure
(16) multiple-model route planning algorithms for manoeuvring
target tracking are usedto forecast the behaviour of all objects in
the environment and the roadway. The estimate of the trajectory
of each item is estimated based on this, allowing Advanced Driver
Assistance Systems (ADAS) the capability to react quickly. These
algorithms analyse numerous alternative actions for each object at
the same time, then connect them with updated on-road
observations received over wireless network finally, the
programme determines the likelihood of each possible object
motion. After that, high-probability movements are employed to
construct the predicted trajectory. Following the definition of
trajectories, path planning technology decides the most
 appropriate vehicle behaviour (Path Planning for Autonomous
Vehicles (2018).

Figure 16: Path planning

Source: Path Planning for Autonomous Vehicles (2018)

iii. Action

The AV is able to recognise items on the road, steer through traffic,

parking spots, barriers, entertainment, traffic signals, bicycles,
people, working zones, weather patterns, and other cars, by
navigating according to the AI agent’s decisions, and safely arrive
at its destination. AI powered control systems such as brake,
steering and gesture controls, voice recognition, economic fuel,
safety system, eye tracking and other driving monitoring systems
are also being installed in autonomous vehicles. The AV process
loops, which include data gathering, path planning, and action,
will be repeated on a regular basis.
4. Challenges

AVs have become a reality today after several years of

research and development, however, there are still huge
challenges in the design and implementation of autonomous
systems. As the level of autonomy increases in AVs, the
challenges also become tougher. Challenges related to data
perception, localization, planning, control, and prediction must be
addressed. Parekh et.al (2022) discovered that the main
challenge is in the implementation of technology in the complex
real world environment. Further, non-technological factorssuch as
customers’ trust and their behaviour also play a major role in
deciding AV’s future success.
Road and traffic conditions are unpredictable and differ from place
to place. Lanes are not clearly defined, there are potholes,
mountainous and tunnel routes without clear external signals for
direction. Weather Conditions are also unpredictable and
constantly change from area to area. AVs must be able to
function well irrespective of the weather conditions, be it sunny,
rainy, stormy or clear. Its effect on the functionality of the car must
be addressed, as failure or downtime is not an option.Traffic
Conditions keep changing. AVs would be subject to a variety of
traffic circumstances once they are on the road. There would be
other cars, AVs and humans around them. Traffic could be highly
controlled and self-regulated, or it could be erratic and
unregulated when people are breaking rules and driving rashly.
Events or conditions may occur unexpectedly. Even a few
centimetres per minute of mobility matters in congested areas.

i. Radar Interference from different AVs on the road will

hinder navigation of self-driving cars. When there are
several such AVs present on the road together, it will be
a complex and difficult task to differentiate between its
own signal and the signal reflected from another vehicle.
Even ifthere are numerous radio frequencies available for
radar, it is unlikely that this frequency range will be
sufficient for allcars manufactured.

ii. Big Data Analytics is a challenge as poor data

management can significantly stifle innovation. It is
necessary to implement both training and decision-
making systems for large amounts of AV data (Khayyam
et al., 2019).

iii. 3D Map Creation is a complex task. Usually, self-

drivingcar companies feed map data to the car’s system
after conducting a trial run on the road using a
sophisticated machine learning algorithm. However, if
the passenger chooses to visit a site not included in the
navigation system, the self-driving car may not be able to
provide the required guidance. Moreover, even after the
time-consuming 3D maps are created, self-driving cars
may still face difficulty due to changes in the traffic
 signals or construction work. (Tata Elxsi , 2022)

iv. Complex Social Interactions required for driving are still

difficult for self-driving cars. Driving is a highly social
activity that involves complex interactions with other
drivers, bikers, and pedestrians for which humans rely on
their general intelligence. Robots lack this sense.
Drivers
There would be other cars, AVs and humans around
them. Traffic could be highly controlled and self-
regulated, or it could be erratic and unregulated when
people are breaking rules and driving rashly. Events or
conditions may occur unexpectedly. Even a few
centimetres per minute of mobility matters in congested
areas.

v. Radar Interference from different AVs on the road will

hinder navigation of self-driving cars. When there are
several such AVs present on the road together, it will be
a complex and difficult task to differentiate between its
own signal and the signal reflected from another vehicle.
Even ifthere are numerous radio frequencies available for
radar, it is unlikely that this frequency range will be
sufficient for allcars manufactured.

vi. Big Data Analytics is a challenge as poor data

management can significantly stifle innovation. It is
necessary to implement both training and decision-
making systems for large amounts of AV data (Khayyam
et al., 2019).

vii. 3D Map Creation is a complex task. Usually, self-

drivingcar companies feed map data to the car’s system
after conducting a trial run on the road using a
sophisticated machine learning algorithm. However, if
the passenger chooses to visit a site not included in the
navigation system, the self-driving car may not be able to
provide the required guidance. Moreover, even after the
time-consuming 3D maps are created, self-driving cars
may still face difficulty due to changes in the traffic
 signals or construction work. (Tata Elxsi , 2022)

viii. Complex Social Interactions required for driving are still

difficult for self-driving cars. Driving is a highly social
activity that involves complex interactions with other
drivers, bikers, and pedestrians for which humans rely on
their general intelligence. Robots lack this sense.
Drivers face different driving circumstances that include
vision,communication, managing four-way junctions, or a
cop directing cars around a scene of the accident.

ix. Data privacy and cybersecurity is another challenge, as

such cars can be easily hacked and disrupted. For
example, if the car industry tries to build systems that
allow various automobiles to communicate with one
another on the road, like if one receives a message to
slam on the brakes, he should be able to safely trust that
message as that is the very basis of Vehicle-to-Vehicle
communication (V2V). However, safeguarding that
system may be incredibly challenging(Plumer, 2016).

x. Regulations for a completely autonomous system do not

exist. As per the current automobile regulations, the
human driver is liable for any accidents or he takes
over in case of any emergencies. In AVs the software
makes all the decisions and choices, because of this
occupants may not be alert and will generally be in a
relaxed mode. Attention might require sometimes, if they
don’t act on time, it may be too late to get out of the
situation.

xi. Social acceptability related to self-driving cars is a main

topic of discussion today. There have been several
serious accidents involving the current models of
automated cars, including those of Tesla. The
acceptability of such cars concerns not only those
considering buying the car but also others sharing the
same roads. The general public must be included in
decisions about the deployment and use of self-driving
vehicles or such a technology may be
 rejected(McDermid, 2020).

Future Scope

During the covid-19 pandemic period, delivery services

experienced a significant increase as individuals shopped online
for meals, groceries, and other products. The technology of AV has
the potential to alter how items are transported from warehouse
to storefront, meals are delivered from restaurants to our front door,
and packages are delivered from merchants to our mailboxes.
While COVID-19 pandemic proved the importance of AI powered
technological applications. (GreyB, 2022) Even though
automobile companies are manufacturing level 4 self-driving cars,
the production is limited to only a few units, as the public is still
sceptical of purchasing them. However, in the next ten years, the
production will increase and most will opt for self-driving cars, as
the future requirements demand people to buy autonomous
vehicles. In terms of the efficiency of AI-driven automobiles, it
can be anticipated that AVs support zero-emission, with the
majority ofvehicles being electric or hybrid (Dubizzle, 2022). Level 5
autonomous seems impossible as of now. From more than 60
companies that applied for permits in order to test their driverless
cars in California, only six were granted permission to test their
vehicles without drivers on public roads. However, since testing is
currently in the process, it will take more than a decade for this to
come to fruition and adapt to different roads of different cities
(Bellan. R, 2022).

5. Conclusion
The study has given an overview the applications of AI in self-
driving cars. It has also summarised the challenges, public
perception and current scenario of the Autonomous Vehicles
market. It is anticipated that by 2030, autonomous vehicles will
account for 25% of all vehicles on the world’s roadways. Advanced
technology and sensors are used by driverless automobiles to
navigate the road safely. Multiple videocameras and sensors allow
these vehicles to read road signs, recogniseroad boundaries, traffic
lights, and the positions of other vehicles. All of this data is
analysed and processed in a central control system, which
subsequently manages the steering and movement of the car at
a safe distance from other vehicles. Currently, a lot of research
and development (R&D) work is being done in this field, all over
the world. Companies such as Google, Tesla, Mercedes, Ford,
Nissan, Volkswagen, and Hyundai are leading the way by utilising
cutting- edge technology such as artificial intelligence, machine
learning, androbotics engineering. Many start-ups in India are also
exploring self-driving technologies.

References:

Aakash, goel. (2016, October 3). What Tech Will it Take to Put Self-
Driving Cars on the Road? Engineering.Com.
https://www.engineering. com/story/what-tech-will-it-take-to-put-
self-driving-cars-on-the- road

Advantages of Artificial Intelligence in Self Driving Cars | dubizzle. (2022,

April 6). UAE’s Leading Autos Blog | Dubizzle Cars. Retrieved
October 23, 2022, from https://www.dubizzle.com/blog/cars/ai-
self-driving-cars/

All-In-One Drive. (n.d.). Retrieved October 23, 2022, from http://www.

aiodrive.org/sensor.html

Anjum, B., PhD. (2021, December 15). Defining data science, machine
learning, and artificial intelligence - Is it possible to differentiate
between them? | Towards Data Science. Medium. Retrieved
October 23, 2022, from
https://towardsdatascience.com/defining-data-science-machine
- learning-and-artificial-intelligence-95f42a60b57c

Assis, C. (2018, January 7). AI in 2018: Waiting for driverless cars to

becomea reality. MarketWatch. Retrieved October 23, 2022, from
https:// www.marketwatch.com/story/ai-in-2018-waiting-for-
driverless- cars-to-become-a-reality-2017-12-27

Autonomous Vehicle Technology Report. (2020, July 22). Wevolver.

Retrieved October 23, 2022, from
https://www.wevolver.com/article/2020.
autonomous.vehicle.technology.report

Bellan. R (2022, February 10). Fewer-autonomous-vehicle-companies-in-

california-drive-millions-more-miles-in-testing Retrieved October 23,
2022, from https://techcrunch.com/2022/02/10/fewer-autonomous
-vehicle- companies-in-california-drive-millions-more-miles-in-
testing/