Advancements and Challenges in Intelligent Driving

Technology: From Evolution to Future Prospects

Xuenuoyang Zong

Computer Science and Technology, Wuxi Taihu University, Wuxi, Jiangsu, 214000, China

Keywords: Intelligent Driving Technology, Computer Technology, Artificial Intelligence, Machine Learning.

Abstract: This paper provides an overview of the current challenges and advances in intelligent driving technology,

focusing on the rapid developments in the fields of computer technology, artificial intelligence, and machine

learning that are driving the development of self-driving cars. The emergence of advanced processors, sensors,

and data processing capabilities has greatly enhanced the performance of intelligent driving systems. This

paper also explores the benefits of smart driving technologies in improving road safety, reducing traffic

congestion and optimising transport efficiency. It highlights global efforts to promote smart driving through

supportive policy and regulatory frameworks, and provides support for automotive companies to access

growth opportunities in the field. Further, the article traces the evolution of autonomous driving technology

between its early conception in 1969 and actual testing in the late 20th century, categorised into different

levels of automation. The article also examines some of the self-driving cars available in the market and their

related technologies such as connectivity solutions, machine learning processors, and sensor fusion

technologies, and provides insights into various applications such as semantic segmentation, target detection,

and obstacle avoidance based on deep learning approaches in automated navigation scenarios, which are

important in enhancing driving safety and navigation functions. By discussing the limitations of intelligent

driving technology and future prospects including complex road conditions, legislative issues, liability and

interpretability challenges, this paper highlights the need for continued research and development in this area

and summarises areas where future research and progress may be involved.

1 INTRODUCTION

Intelligent driving technology refers to the use of

advanced computer technology and perception

systems thereby enabling autonomous vehicle

navigation and driving. With the rapid advancement

of computer technology, artificial intelligence and

machine learning, a variety of high-performance

advanced processors, sensors emerge one after

another, supplemented by large-scale data processing

and analysis capabilities, so that intelligent driving

technology has been greatly developed. Secondly,

with the continuous improvement of the overall

economic strength of the nation and the demand for

people's daily travelling means of transport, the

number of private cars has increased dramatically.

Intelligent driving technology can reduce the risk of

accidents caused by human driving errors by

monitoring real-time road conditions, and indirectly

https://orcid.org/0009-0000-9309-8972

optimise road traffic to reduce traffic jams. Nowadays,

countries are also trying to promote the development

of smart driving, and have introduced many welfare

policies and regulatory frameworks that facilitate the

development of car companies. This is an opportunity

for automakers to gain national support.

The concept of self-driving technology was

introduced in 1969, and the earliest practical tests date

back to the end of the 20th century, when the Navlab

project at Carnegie Mellon University in the US first

realised self-driving vehicles on city roads in 1995.

The 2016 U.S. Policy Guidelines for Self-Driving

Vehicles categorizes self-driving technology into six

levels as one of the standards to assess the level of

autonomous driving technology. Currently, the

applied self-driving cars in the market belong to the

partially automated stage, incorporating a very wide

range of technologies. For example, Qualcomm

Technologies can provide connectivity for

Zong, X.

Advancements and Challenges in Intelligent Driving Technology: From Evolution to Future Prospects.

DOI: 10.5220/0012958100004508

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

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 517-521

ISBN: 978-989-758-713-9

517

autonomous driving with modems and Snapdragon

processors that enable vehicles to interconnect and

continuously learn from each other. It is also provided

to perform heterogeneous processing1 for machine

learning, human-computer interaction, intelligent

safety and sensor sensing. Based on the prevalence of

sensors in vehicles, Global Navigation Satellite

System (GNSS) solutions have been proposed to

combine image sensors and inertial sensors in

vehicles as the primary means of determining road

conditions, which is particularly important for self-

driving cars at Level 3 (Herrmann, 2018). Bimbraw

et al. reviews the evolution of self-driving car

technology from radio-controlled vehicles to basic

electronic-guided systems in modern automobiles to

fully self-driving cars of the future. Autonomous

driving technology has gone through several stages of

development, including vision-guided, network-

guided, and lane-keeping systems. Some of the semi-

autonomous driving functions found in contemporary

vehicles include adaptive cruise control, lane-keeping

assistance, and automatic emergency braking, are

based on these systems. In the future, self-driving cars

will enable safe and comfortable transportation, but

there are also challenges such as interaction with

human-driven vehicles, software reliability and

preventing terrorists and criminals from using self-

driving cars (Bimbraw, 2015). Yurtsever et al.

presents the classification of automated driving

systems, interfaces, communications, end-to-end

driving, and driving behavior assessment, and carries

out the views of several car companies on the

classification of automated driving and their

respective standards for automated driving

technology (Yurtsever, 2020).

The rest of this paper is organized as follows:

Section 2 provides a description of related methods in

this domain Section 3 provides a discussion about the

current limitations and future prospects. The

conclusions of this work are provided in Section 4.

2 METHOD

2.1 Semantic Segmentation

Manikandan et al. describes the remarkable progress

of deep learning in the field of semantic image

segmentation and illustrates its wide application in

the realm of autonomous driving. The field of

autonomous driving covers mediated perception

paradigm, behavioral reflection paradigm, and

intermediate paradigm. Among them, the

intermediate paradigm utilizes semantic

segmentation modules to provide affordance metrics

for autonomous driving scenarios to guide final

driving decisions. This research delves into the

significance of semantic segmentation in autonomous

driving, reviews related research, and proposes

directions for future research. Deep semantic

segmentation methods are mainly divided into four

categories: classical methods, convolutional neural

networks, structural models, and spatio-temporal

models. The deep semantic segmentation problem in

autonomous driving is relatively simple, and the

model building can be simplified by many a priori

constraints. Future research can focus on how to

estimate depth and semantics simultaneously in the

network to solve the problems of computational

constraints, the need for large labeled datasets, and

complex outputs faced by autonomous driving

technology, and solutions such as multi-task learning,

end-to-end learning, and modular end-to-end learning

are proposed. The paper delves into the field of deep

learning-based autonomous driving, focusing on the

application of fully convolutional networks and their

variants in semantic segmentation, as well as the role

of structural models and spatio-temporal

characteristics in the segmentation issue

(Manikandan, 2023).

Another study mentions that in July 2016, a Tesla

ModelS was involved in a fatal accident while in

autopilot mode, the first known fatal accident

involving a self-driving vehicle. The researchers used

a deep learning network to detect hidden parts of the

vehicle to help prevent accidents. The study proposes

a target detection based on deep learning and image

restoration method called DID-Alliance (DIDA),

which effectively detects the hidden parts of large

vehicles through a six-step process. The experimental

results verify the effectiveness of the algorithm,

which provides a new approach for the safety of

autonomous driving systems and is supported by the

National Natural Science Foundation of China and

other organizations (Siam, 2017).

2.2 Object Detection

Fang et al. introduces a deep reinforcement learning

(DRL)-based navigation system for self-driving

vehicles, which utilizes an RGB-D camera to acquire

depth information, an adaptive obstacle avoidance

algorithm and a DRL decision-making algorithm to

achieve lane detection, object detection and

navigation functions. The system is realized with

ArduinoUNO controller, MPU9250IMU sensor and

IntelRealsenseD435 depth camera. The experimental

results indicate that the D3DQN algorithm performs

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

518

better in detecting road obstacles and pedestrians, and

successfully navigates an autonomous vehicle using

RGB-D inputs to reach a maximum speed of 30km/h

in the city. Meanwhile, the study also explores the use

of convolutional neural networks for pedestrian

detection and navigation in real-time environments,

which provides new ideas and methods for the field

of future autonomous driving (Fang, 2003).

Aiming at the target detection problem of self-

driving vehicles, this article proposes an improved

model CSPDarknet45_G based on YOLOv4. The

model introduces stochastic regularization and GELU

activation function, which enhances the model's

nonlinearity and generalization ability. The

experimental results are superior to LeakyReLU and

Mish by the inclusion of Database Generation (DBG)

modules. The enhanced ResNet and CSP architecture

is constructed upon the DBG modules, leading to

improved model performance. CSPDarknet45_G

comprises five CSP_ResBock_Body blocks, with

each CSP block containing N Resunit residual

modules. The initial AnchorBox size is determined

using the K-Means++ algorithm to enhance model

stability. Experimental results demonstrate that the

upgraded YOLOv4 model exhibits enhancements

across all performance metrics and excels in various

categories (Xu, 2019).

2.3 Deep Learning–Based Obstacle

Avoid

This paper proposes an approach called IVERSE,

which aims to improve driving safety and is divided

into active and passive systems. The article focuses

on the impact of driver information processing,

including visual sensors, knowledge, expectations

and other factors on driving behavior. To detect

changes in the driving environment, the study

proposes a computational model-based approach and

develops a change detection system based on this

model. This system consists of three components:

perceptual, perceptual and conceptual, and is capable

of dealing with a wide range of environmental

changes, such as shadows, precipitation, reflections

and lane markings. In addition, the study introduces a

computational model of a functional analyzer based

on cognitive processing to improve the model's

ability to handle complex tasks (Yu, 2021).

Ramos et al. proposes a deep learning-based

method for detecting small unexpected obstacles on

roads, which overcomes the limitations of traditional

methods by learning the shape, size, and appearance

context information of objects through deep

convolutional neural networks. Combining the deep

learning detection method with a stereo image fusion

system improves the detection accuracy and reduces

the false positive rate. Concentrating on identifying

small, common, and unforeseen obstacles in driving

situations, a stereo camera-based setup is used to

focus on generic obstacles in 3D space. A 3D obstacle

representation was generated by two main detection

methods, UON and FPHT. Experimental results show

that the OR fusion and probabilistic fusion algorithms

exhibit the best detection performance over a range of

distances, while UON-Stixels and FPHT-Stixels

perform well at longer distances. Future plans include

enhancing the system's robustness by increasing the

amount of training data and integrating the

probabilistic fusion scheme directly into the learning

system (Ramos, 2017).

3 DISCUSSIONS

As shown in Figure 1 (Muhammad, 2020), this is the

Highway Traffic Safety Administration in the United

States has proposed a formal five-level classification

system for autonomous driving. As early as 1920,

foreign countries, especially the United Kingdom, the

United States and Germany, have begun to carry out

research on driverless cars, and has made great

progress in feasibility and practicality. In Europe,

especially Belgium, France, Italy and other countries

have already planned to adopt driverless cars to

operate the transport system, and Germany even

allows experimental self-driving cars to go on the

road. This shows that the supporting infrastructure for

autonomous driving in these countries is relatively

complete and the scope of application of autonomous

driving is very wide. However, although the market

for autonomous driving in China is expanding and the

pace of commercialization is accelerating, it is still at

Level 2. China's road conditions are very complex,

and autonomous driving in different terrains and

different environments will have very differentiated

performance. This leads to a significant reduction in

the general usability of autonomous driving models.

For the time being, most autonomous driving tests

and applications are conducted on terrains like plains

that do not have particularly harsh environments. For

example, assisted driving, as one of the categories of

intelligent driving, most people will use it only on

uncongested straight highway sections, and it will be

relatively useless in terms of congested urban

commuting, as there are many emergencies on urban

roads, such as pedestrians running red lights, and

assisted systems similar to cruise control can lead to

traffic accidents instead. During the day, the sensor’s

Advancements and Challenges in Intelligent Driving Technology: From Evolution to Future Prospects

519

Figure 1: Automatic driving technology classification and evaluation standards (Photo/Picture credit: Original).

field of view is better, and the accuracy of the self-

driving system in the recognition and decision-

making process becomes higher. But at night, when

other conditions on the same road have not changed

significantly, the difference between the data read by

the sensors and that of the daytime is significant, it

reflects the applicability of the autonomous driving

model is not enough. Some advanced solutions

should be considered in the future (Csurka 2017; Qiu,

2022).

In addition, the legislation on intelligent driving is

not sound enough. Laws and regulations are too

lagging behind. Intelligent driving technology is

developing at a fast pace, so existing traffic

regulations are often unable to cover and regulate the

problems and challenges posed by the brand-new

technology. Before self-driving vehicles can enter the

market, they need to undergo real road tests. However,

most of the current road test standards are still for

conventional vehicles, and the relevant road test

standards and specifications for self-driving vehicles

have yet to be supplemented. The most important

point, which is also directly related to the level of

autonomous driving, is the issue of liability when

people drive autonomous vehicles. If an accident

occurs during the process of autonomous driving,

how to define the attribution of responsibility is

currently a major controversial problem. At present,

especially in China, there are still very big flaws in

the self-driving technology. The interpretability of

autonomous driving technology is also one of the

current challenges. Even if the automatic driving

system finally makes a decision, but people cannot

understand its internal decision-making process and

logic, and its own interpretation of the decision is not

necessarily reasonable. And the lack of explanation of

the decision-making process of the self-driving

technology model may affect people’s trust in the

self-driving technology, as such decisions may be

directly related to the lives of the drivers in the vehicle.

4 CONCLUSIONS

In this work, an overview of intelligent driving was

provided. This paper considered the harsh

environment that may be faced in the process of

automatic driving, observed the decision-making

mode and decision-making ability of deep learning

and sensor technology in the automatic driving

system for current vehicle problems, and then

analyzed the impact of algorithms such as neural

network, convolutional neural network and recurrent

neural network used by them on the decision results.

Based on discussions, there are still huge challenges

in the field of intelligent driving, for example, the

safety of the people in the car cannot be guaranteed

during the autonomous driving process. The

automatic driving system is not enough to deal with

complex road conditions, and cannot make logical

explanations for its own decisions, and its

interpretability is not high, and people cannot fully

trust it. The focus has been placed more on the

advancements and results from the algorithmic side

of academia rather than on the product side of

industrialized applications. Therefore, future research

aims to incorporate these elements into the system to

provide a more comprehensive overview.

REFERENCES

Bimbraw, K. 2015. 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

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

520

International Conference on Informatics in Control,

Automation and Robotics (ICINCO). IEEE, 1: 191-198.

Csurka, G. 2017. Domain adaptation for visual

applications: A comprehensive survey. arXiv preprint

arXiv:1702.05374.

Fang, C.Y., Chen, S.W., & Fuh, C.S. 2003. Automatic

change detection of driving environments in a vision-

based driver assistance system. IEEE Transactions on

Neural Networks, 14(3): 646-657.

Herrmann, A., Brenner, W., & Stadler, R. 2018.

Autonomous driving: How the driverless revolution will

change the world. Emerald Publishing Limited.

Manikandan, N.S., Kaliyaperumal, G., & Wang, Y. 2023.

Ad hoc-obstacle avoidance-based navigation system

using Deep Reinforcement Learning for self-driving

vehicles. IEEE Access.

Muhammad, K., Ullah, A., Lloret, J., et al. 2020. Deep

learning for safe autonomous driving: Current

challenges and future directions. IEEE Transactions on

Intelligent Transportation Systems, 22(7): 4316-4336.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation

training. Biomedical Signal Processing and Control, 72,

103323.

Ramos, S., Gehrig, S., Pinggera, P., et al. 2017. Detecting

unexpected obstacles for self-driving cars: Fusing deep

learning and geometric modeling. In 2017 IEEE

Intelligent Vehicles Symposium (IV). IEEE: 1025-1032.

Siam, M., Elkerdawy, S., Jagersand, M., et al. 2017. Deep

semantic segmentation for automated driving:

Taxonomy, roadmap and challenges. In 2017 IEEE 20th

International Conference on Intelligent Transportation

Systems (ITSC). IEEE: 1-8.

Xu, Y., Wang, H., Liu, X., et al. 2019. Learning to see the

hidden part of the vehicle in the autopilot scene.

Electronics, 8(3): 331.

Yurtsever, E., Lambert, J., Carballo, A., et al. 2020. A

survey of autonomous driving: Common practices and

emerging technologies. IEEE Access, 8: 58443-58469.

Yu, W., & Ren, P. 2021. Vehicle and pedestrian target

detection in auto driving scene. In Journal of Physics:

Conference Series. IOP Publishing, 2132(1): 012013.

Advancements and Challenges in Intelligent Driving Technology: From Evolution to Future Prospects

521