The Evolution of Autonomous Driving Technology and Its Ethical

Challenges: A Pedestrian-First Perspective

Mohan Wang

Faculty of Computing and Data Science, Boston University, 15 N Beacon St, Unit 428, Allston, U.S.A.

Keywords: Machine Learning and Deep Learning, Sensor Integration, Data Collection and Processing, Pedestrian Safety,

Ethical Frameworks.

Abstract: This essay examines the progression of autonomous driving technology, explores its core technical

components such as data collection, sensor integration, and real-time decision-making. By integrating

cameras, LiDAR, and radar with sophisticated algorithms, modern autonomous vehicles demonstrate

markedly enhanced perception and navigation capabilities. However, widespread adoption also raises

profound ethical questions. Central to these debates is the dilemma of whether passenger or pedestrian safety

should take precedence, especially when accidents pose life-threatening risks. Drawing on utilitarian,

deontological, and social contract ethical frameworks, the essay contends that protecting pedestrians

represents not only a logical policy choice but also a moral imperative. The discussion underscores the

importance of maintaining the inherent dignity of all individuals and ensuring legal and societal acceptance

for future implementation. By balancing technological development with robust ethical considerations,

autonomous vehicles can move toward broader acceptance and enduring success. Ultimately, the evolution of

autonomous driving will be shaped by both technological progress and the ethical principles that guide its

adoption. As innovation continues, ongoing discussions and thoughtful decision-making will play a crucial

role in shaping its impact on society.

1 INTRODUCTION

In an era of rapid development in artificial

intelligence, autonomous driving technology has

emerged as an innovative advancement, gradually

entering the public eye and becoming a new industry

that major automobile manufacturers are actively

investing in. Through the use of sensors, artificial

intelligence algorithms, data fusion, and high-

performance computing, autonomous driving

technology enables vehicles to perceive their

surroundings, plan routes, and operate safely with

little or no human intervention. In its early stages, this

technology primarily relied on relatively simple

environmental sensing methods and rule-based

algorithms, and its functionality was often limited to

specific environments or structured roads (Chen et al.,

2022; Dhanasingaraja et al., 2014). However, with the

emergence of new technologies, such as using high-

precision sensors, lidar, and cameras to collect data in

real time, autonomous vehicles are now able to

https://orcid.org/0009-0006-6113-0804

continuously gather information about their

surroundings. In addition, the integration of machine

learning, deep learning, and path-planning algorithms

(including path-based decision-making) enables swift

and accurate recognition of, and response to, complex

road conditions and rapidly changing external

environments (Gerdes & Thornton, 2015; Bimbraw,

2015). As autonomous driving technology begins to

satisfy most travel needs, merely following traffic

regulations is no longer sufficient to ensure its

widespread application. Rather, possessing the ability

to carry out independent analysis, risk assessment,

and optimal decision-making represents the greatest

challenge for this technology (Liang, Liu, & Wang,

2024; Martinho, Silva, & Cunha, 2021; Kabir et al.,

2025). When accidents endanger the safety of

pedestrians, passengers, other vehicles, the

environment, surrounding property, or even the

autonomous vehicle itself, people are forced to ask:

by what standard does the system decide whose

interests, or even whose life, should take priority? It

306

Wang, M.

The Evolution of Autonomous Driving Technology and Its Ethical Challenges: A Pedestrian-First Perspective.

DOI: 10.5220/0013688500004670

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Data Science and Engineering (ICDSE 2025), pages 306-310

ISBN: 978-989-758-765-8

is precisely this profound ethical consideration that

underscores the complexity of autonomous driving

technology. Against this backdrop, drawing on

utilitarianism, deontological ethics, and social

contract theory, this essay proposes that the safety of

the pedestrian should take priority.

2 HISTORY AND

DEVELOPMENT

Tracing back to the mid-20th century, some

automobile manufacturers had already begun

experimenting with sensing cables, radar, and other

methods to achieve simple automated driving

functions. In the 1920s, the first radio-controlled car

was designed, opening the door to the development of

autonomous vehicles. Over the following decades,

vehicles with similar electronic guidance systems

gradually moved toward the ideal of full automation.

"1980s saw vision guided autonomous vehicles,

which was a major milestone in technology and till

date we use similar or modified forms of vision and

radio guided technologies" (Bimbraw, 2015).

Entering the 21st century, major companies such as

Tesla, Google, Baidu, and other general automobile

manufacturers began conducting autonomous driving

tests on public roads, accelerating the

commercialization of this technology and gradually

bringing it to the public and the market. "While there

were about 31 million machines with some level of

automation in operation around the world in 2019,

that number is predicted to rise to 54 million by 2024"

(Ignatious, Karthikeyan, & Kumar, 2022). Over the

span of five years, this increase of 23 million vehicles

fully illustrates the rapid growth of autonomous

vehicle market penetration, indicating an expanding

consumer demand. Moreover, the positive market

conditions have generated highly favorable social

impacts. "Although the market dropped by roughly

3% in 2020 because of the economic slowdown

induced by the Covid-19 epidemic, the market is

expected to rise by about 60% between 2020 and

2023" (Ignatious et al., 2022). This comparison of

data shows that the prospects for the autonomous

vehicle industry remain very promising. Even in the

face of global economic challenges, the industry

continues to demonstrate remarkable competitiveness

and strong market demand, suggesting that its

potential is immeasurable. Furthermore, the post-

pandemic economic recovery may introduce a new

group of consumers to autonomous vehicle

technology, thereby spurring further technological

innovation and creating a positive cycle of

development.

3 DATA COLLECTION AND

ANALYSIS

3.1 Data Collection

After examining the history of autonomous driving

technology, this paper will discuss how autonomous

driving technology acquires, processes, and utilizes

data to achieve safe and efficient driving. The first

step in autonomous driving is data collection; only

after gathering data does the process move on to

analysis, computation, and other steps. As the

foundation of this technology, the architecture of

autonomous driving is generally described from two

perspectives: a technical perspective and a functional

perspective. These respectively refer to the hardware

and software layers of the technology, as well as the

processes required for analysis and decision-making.

Tools familiar to the public, such as cameras and

radar, can be classified as sensors, which detect

events or changes in the environment and convert

them into measurable data. Sensors are often

categorized by their transmission range into three

types: short-range, medium-range, and long-range.

Achieving near-perfect detection typically requires

the combined use of multiple sensors. Based on their

operating principles, sensors can be divided into two

major categories: internal state sensors (such as

IMUs, encoders, and GNSS receivers), which

measure forces, wheel loads, and vehicle orientation,

and external state sensors (such as cameras, radar, and

LiDAR), which collect information about the

surrounding environment. Passive sensors (e.g.,

cameras) gather existing energy from the

environment, while active sensors (e.g., LiDAR and

radar) emit signals and measure their reflections.

Three of the most common and widely adopted

external environment sensors are cameras, LiDAR,

and radar. Cameras, as one of the most extensively

applied technologies for environmental observation,

are used not only in autonomous vehicles but also in

everyday cars—for example, in rearview systems,

360-degree surround-view systems, and dashcams.

The operating principle of a camera is: "A camera

produces crisp images of the surrounding by detecting

lights emitted from the surroundings on a

photosensitive surface (image plane) using a camera

lens (placed in front of the sensor)" (Ignatious et al.,

2022). This technology boasts low cost and can

The Evolution of Autonomous Driving Technology and Its Ethical Challenges: A Pedestrian-First Perspective

307

recognize both static and dynamic obstacles

simultaneously, although the data acquired may

sometimes experience relatively higher latency. The

second technology is LiDAR, first developed in the

1960s, which estimates the distance between the

reflected objects and the sensor by emitting infrared

or laser pulses. Its advantages include the ability to

conduct detection and estimation in three dimensions,

as well as providing intensity information about the

reflected objects. Radar operates on principles similar

to those of LiDAR, emitting electromagnetic waves

toward a target area and determining the target’s

relative speed and position by receiving the reflected

waves. It’s easy to see that combining the functions

of these three types of sensors creates a solid

framework for detecting obstacles and gathering

related information. By integrating sensor data with

advanced software and computing systems,

autonomous driving technology can reduce reliance

on human input while enhancing overall efficiency

and safety. As a prerequisite and foundation for all

subsequent calculations and processes, data

collection, along with robust and reliable

environmental evaluation and detection, warrants

significant attention. After all, one of the highest

potential risk factors to others on the road is the

presence of other vehicles. The ability to accurately

and promptly perceive and identify the surrounding

environment is critical to ensuring the safety of

passengers, pedestrians, and the vehicles themselves.

Only by predicting potential danger can the system

respond accordingly based on given instructions.

Equipping an autonomous vehicle with a

comprehensive data-capturing system can greatly

reduce safety risks, earn greater public trust, and

ultimately expand its marketing potential.

3.2 Target Detection

After discussing the data collection techniques in

autonomous vehicles, another critical aspect is object

detection. Vision-based object detection can

generally be categorized into three types: traditional

techniques, machine learning, and deep learning. This

paper focuses on the roles of machine learning and

deep learning in autonomous driving. As a highly

popular topic in artificial intelligence and computer

science in recent years, machine learning primarily

operates by using data and algorithms to split data

into training and testing sets, thereby emulating

human learning processes and gradually reaching

minimal error through continual learning and

training. When applied to autonomous vehicles,

machine learning typically involves two key steps:

first, inputting and processing images to obtain a

region of interest (ROI), and then transforming and

encoding the extracted images, converting high-

dimensional image-space data into lower-

dimensional data, all while continuously refining this

process through training and optimization.

First, during the feature extraction step, various

methods, such as Histogram of Oriented Gradients

(HOG), Haar-like descriptors, Local Binary Patterns

(LBP), Gabor filters, and Speeded-Up Robust

Features (SURF), are commonly used to capture

recognizable vehicle traits. These approaches help

identify consistent patterns that remain reliable when

the vehicle’s orientation or model changes. Some

techniques, like the Deformable Part Model (DPM),

further refine HOG features to deal with more

complex shapes. In the classification stage,

algorithms like AdaBoost, K-Nearest Neighbors

(KNN), Naive Bayes, Support Vector Machines

(SVM), and Decision Trees are often employed. Each

classifier must balance how well it fits its training

data with how effectively it adapts to new inputs.

Ensemble learning, which merges multiple

classifiers, can enhance overall detection

performance. A key challenge is the high

computational cost of searching the entire image for

potential vehicles. To address this, many systems

focus on areas of interest, such as identifying shadows

or other cues, before running feature extraction and

classification. This targeted approach can

significantly reduce processing time while

maintaining reliable detection.

Another object detection algorithm is deep

learning. Because machine learning's reliance on

feature extractors and classifiers can limit a model's

capabilities in certain scenarios, the rise of

convolutional neural networks brought significant

attention to deep learning algorithms, making them

more suitable for object detection methods in the field

of computer vision. Typically, deep learning-based

object detection methods are divided into object

detection-based approaches and segmentation-based

approaches. First, the object detection-based

approach can be categorized into three types: anchor-

based detectors, anchor-free detectors, and end-to-

end detectors. The key difference between anchor-

based and anchor-free detectors is that anchor-based

detectors predefine bounding boxes to detect objects

by partitioning and categorizing vehicles in proposed

regions, then predicting the vehicle's center and

bounding box. Anchor-free detectors directly predict

the object's center point and then cluster them into a

single entity to obtain the bounding box. Some

models derived from YOLO play a crucial role in

ICDSE 2025 - The International Conference on Data Science and Engineering

308

both anchor-based and anchor-free detectors, while

others like FSAF and FCOS use different detection

methods. End-to-end detectors can be viewed as a

derivative and evolution of anchor-based detectors,

but they operate more directly. They do not require

complex preprocessing or postprocessing and only

need to analyze the input image to determine the

target's category and location. The second approach

is the segmentation-based method. Compared to

simpler target-level detection, semantic segmentation

assigns a category to every pixel in an image, giving

a more precise representation of vehicles’ positions

and shapes—crucial for autonomous driving.

Traditionally, vehicle segmentation has followed a

region-based approach that mirrors two-stage object

detectors: first, candidate areas are proposed, and then

a classifier labels the pixels within those regions.

Models like DeepMask, SharpMask, MultipathNet,

and Mask R-CNN exemplify this process by refining

region proposals before creating segmentation masks.

Although this can produce high-quality results, it also

increases computation time, making real-time

deployment more challenging. Consequently,

ongoing research aims to balance accuracy with faster

processing, ensuring that segmentation can meet the

demands of autonomous systems on the road.

4 ETHICAL PROBLEMS

4.1 Utilitarianism

After analyzing the basic operation of autonomous

vehicles, the next important factor is the ethical

problem arising with it. Given the significant

potential risk and direct involvement of human life in

both situations, the most contentious issue is whether

pedestrian or passenger safety should come first. "An

ethical theory that is based on the principle that our

policies and laws should be such that they produce the

greatest good (happiness) for the greatest number of

people" (Tavani, 2016), which also implies to

minimize the total harms, is how Tavani describes it

from a utilitarian standpoint. Pedestrians are usually

the most vulnerable individuals in an accident. They

are much more likely to suffer severe injuries or die

without any kind of protection, such as seat belts,

airbags, or a car frame, and the level of damage they

endure is greater than that experienced by someone

inside a car. Pedestrians are therefore the most

vulnerable population in these situations, with the

greatest risk of suffering serious injury. Prioritizing

pedestrian safety reduces total suffering by

preventing fatalities, according to the utilitarian

ethical framework.

Additionally, rule utilitarianism calls for the

analysis of broader societal ramifications and rule-

based requirements, both of which are strongly

backed by the field of legal accountability. For

example, Massachusetts law clearly states that cars

have an obligation to stop for pedestrians using a

crosswalk in order to safeguard them. Therefore,

adhering to these regulatory requirements is

consistent with building autonomous cars using the

"pedestrians-first" approach. The concern that robots

may prioritize passenger safety over the protection of

innocent pedestrians is a common source of public

anxiety about autonomous driving. Since there are

generally more pedestrians than passengers on the

road, putting pedestrian safety first would greatly

allay public concerns and skepticism. It would

increase the long-term potential for autonomous

vehicle development, improve traffic flow efficiency,

and save public resources, which are essential to

maximizing overall benefits.

4.2 Deontology

Next, Tavani then discusses Kant's categorical

imperative from the perspective of rule deontology,

saying: "Adhere always to that maxim or principle (or

rule) that guarantees that all individuals will be

treated as ends-in-themselves and never merely as a

means to an end" (Tavani, 2016). Kant's other version

of the categorical imperative, which states, "Act

always on that maxim or principle (or rule) that can

be universally blinding, without exception, for all

human beings" (Tavani, 2016), fully addresses the

idea that an action is just if it respects each person's

autonomy and treats them as an end in and of

themselves rather than as a means. In the event that

an autonomous vehicle is configured to put the safety

of its passengers first, the risk is unintentionally

transferred to pedestrians, who never agreed to take

that risk. The fundamental tenet of rule deontology is

violated when potential injustice is created.

A similar example may be found in commercial

aviation, where planes frequently fly for large

portions of the journey on a quasi-autopilot system.

In fact, accidents caused by this technology have

happened in the past. It is completely legitimate for

someone to choose a slower or less convenient form

of transportation over an autonomous system in order

to lessen personal danger. Going back to autonomous

cars, people who believe the manufacturer's safety

assurances can travel more conveniently, but they

also have to accept the potential for increased risk that

The Evolution of Autonomous Driving Technology and Its Ethical Challenges: A Pedestrian-First Perspective

309

comes with new technology. If manufacturers allow a

"passenger-first" policy, people who trust the

technology will benefit from its ease while

simultaneously shifting the risk of the system on

those who are less sure and have avoided it. Since

each person should be seen as an independent

creature with inherent dignity, this effectively gives

the technology a kind of supremacy, breaking the

deontological necessity that prohibits using others

just as a means. Therefore, rather than shifting risks

on uncooperative bystanders or exploiting the safety

of others for profit, the people can only force

automakers to improve the technology itself by

prioritizing pedestrian safety.

5 CONCLUSIONS

In summary, autonomous driving technology has

advanced from early rule-based trials to complex

systems that combine deep learning, machine

learning, and high-precision sensors. Data collection

and processing, which are crucial for perception and

decision-making in real time while driving, are at the

heart of this evolution. With the help of sophisticated

algorithms, cameras, LiDAR, and radar combine to

provide cars the ability to recognize and react to

challenging driving situations more accurately than

ever before. However, ethical issues inevitably arise

as the technology advances and becomes more well-

known. The crucial issue is how to put safety first,

especially for vulnerable pedestrians and the

occupants of the car, without violating people's rights

or placing unwarranted risk on onlookers. Using

ethical frameworks like utilitarianism and Kant's rule

deontology, this essay emphasizes how prioritizing

pedestrian safety maintains both the inherent dignity

of every person and more general societal norms and

legal requirements. By doing this, autonomous

driving may keep moving forward toward broader

acceptance and financial success while preserving a

fair balance between the welfare of the general

population and technological growth.

REFERENCES

Bimbraw, K. 2015, July. Autonomous cars: Past, present

and future a review of the developments in the last

century, the present scenario and the expected future of

autonomous vehicle technology. In 2015 12th

International Conference on Informatics in Control,

Automation and Robotics (ICINCO) (Vol. 1, pp. 191-

198). IEEE.

Chen, R., Hu, J., & Xu, W. 2022. An RRT-Dijkstra-based

path planning strategy for autonomous vehicles.

Applied Sciences, 12(23), 11982.

Dhanasingaraja, R., Kalaimagal, S., & Muralidharan, G.

2014. Autonomous vehicle navigation and mapping

system. International Journal of Innovative Research in

Science, Engineering and Technology, 3(8), 15468–

15474.

Gerdes, J. C., & Thornton, S. M. 2015. Implementable

ethics for autonomous vehicles. In M. Maurer, J. C.

Gerdes, B. Lenz, & H. Winner (Eds.), Autonomes

Fahren (pp. 87–102). Springer Vieweg.

Ignatious, H. A., Karthikeyan, P., & Kumar, R. 2022. An

overview of sensors in autonomous vehicles. Procedia

Computer Science, 198, 736–741.

Kabir, M. M., Kaur, M., & Kassim, A. A. 2025. Terrain

detection and segmentation for autonomous vehicle

navigation: A state-of-the-art systematic review.

Information Fusion, 113, 102644.

Liang, L., Liu, Y., & Wang, Y. 2024. Vehicle detection

algorithms for autonomous driving: A review. Sensors,

24(10), 3088.

Martinho, A., Silva, J. A., & Cunha, J. F. 2021. Ethical

issues in focus by the autonomous vehicles industry.

Transport Reviews, 41(5), 556–577.

Tavani, H. T. 2016. Ethics and technology: Controversies,

questions, and strategies for ethical computing (5th

ed.). Wiley.

ICDSE 2025 - The International Conference on Data Science and Engineering

310