RITSA: Toward a Retrieval-Augmented Generation System for

Intelligent Transportation Systems Architecture

Afef Awadid

, André Meyer-Vitali

, Dominik Vereno

and Maxence Gagnant

Technological Research Institute SystemX, Palaiseau, France

German Research Center for Artificial Intelligence (DFKI), Kaiserslautern, Germany

Josef Ressel Centre for Dependable System-of-Systems Engineering, Salzburg, Austria

Keywords: Intelligent Transportation Systems (ITS), ITS Architecture Design and Modeling,

Large Language Model (LLM), Retrieval-Augmented Generation (RAG), Modeling Assistant,

ITS Reference Architectures.

Abstract: Intelligent Transportation Systems (ITS) have significantly transformed the transportation domain by

addressing critical challenges such as traffic safety, cost, and energy efficiency. However, the increasing

complexity of ITS—arising from the extensive range of applications and technologies they encompass—has

made their architectural design modeling time-consuming and challenging, particularly for modelers lacking

specialized expertise. Recent advancements in the literature suggest that large language model (LLM)-based

modeling assistants offer a promising solution to mitigate these challenges. In this context, this paper

introduces the RAG for Intelligent Transportation Systems Architecture (RITSA) project, which seeks to

develop a retrieval-augmented generation (RAG) system to support ITS designers/ modelers throughout the

architecture design process.

1 INTRODUCTION

Intelligent Transportation Systems (ITS), such as

autonomous vehicles, are defined as “those systems

utilizing synergistic technologies and systems

engineering concepts to develop and improve

transportation systems of all kinds” (Giesecke et al.,

2016). The emergence of ITS has significantly

transformed transportation, addressing critical

challenges such as cost efficiency, traffic safety,

speed, and user comfort (Waqar et al., 2023), (Elassy

et al., 2024). However, the growing complexity of

ITS stemming from the extensive range of

applications and services they encompass (Damaj et

al., 2022) has rendered their architectural design

increasingly intricate.

To manage this complexity effectively, Model-

Based Systems Engineering (MBSE) has been widely

recognized as a pivotal approach (Friedenthal et al.,

2014). MBSE leverages models to define the

composition and interfaces of a system’s architectural

layers, including functional, logical, and physical

layers (Haomin et al., 2021). These system models

serve as central artifacts in the systems engineering

process, providing comprehensive representations of

the system and its environment. They incorporate

multiple views to support activities such as planning,

requirements specification, architecture definition,

design, analysis, verification, and validation

(INCOSE, 2023).

Nevertheless, building these models is time-

consuming and challenging, particularly for modelers

who lack expertise in the field of intelligent

transportation. In this context, intelligent modeling

assistants present a promising solution to address this

issue. Against this backdrop, recent studies have

investigated the potential of large language models

(LLMs), such as OpenAI's GPT-4, in providing

support for modeling tasks (Combemale et al., 2023).

LLMs can assist in model development, particularly

in an 'acceleration' mode, where an initial model

already exists (Barke et al., 2023). In such cases,

LLMs contribute by extending the existing model

through the integration of new entities or

functionalities, or by enriching specific elements with

additional characteristics. Thus, the primary role of

LLMs in this context is to enhance and complete pre-

existing models.

However, it is well established that LLMs can

‘hallucinate’ outputs—that is, generate false or

466

Awadid, A., Meyer-Vitali, A., Vereno, D. and Gagnant, M.

RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture.

DOI: 10.5220/0013443300003896

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

In Proceedings of the 13th International Conference on Model-Based Software and Systems Engineering (MODELSWARD 2025), pages 466-473

ISBN: 978-989-758-729-0; ISSN: 2184-4348

misleading content—particularly when applied

outside their domain of expertise or when addressing

complex or ambiguous topics (Aslam and Nisar,

2024). In safety-critical domains such as intelligent

transportation systems, the consequences of

inaccurate outputs can be significant. This highlights

the importance of fine-tuning LLMs and grounding

their responses using information derived from

reference architectures specifically designed for

intelligent transportation, such as ARC-IT

(Architecture Reference for Cooperative and

Intelligent Transportation) (ARC-IT Version 9.3,

2025). This underscores the importance of developing

a Retrieval-Augmented Generation (RAG) system to

specialize large language models (LLMs) by

incorporating information retrieved from domain-

specific sources in the field of intelligent

transportation. RAG is an AI framework designed to

retrieve facts from an external knowledge base,

enabling LLMs to be grounded in accurate, up-to-date

information while providing users with insight into

their generative processes (Martineau, 2023).

With this in mind, this paper provides an overview

of our ongoing research project, RITSA (Retrieval-

Augmented Generation for Intelligent Transportation

Systems).

The remainder of the paper is organized as

follows: Section 2 discusses the theoretical

background and reviews related work. Section 3

outlines the RITSA project, detailing its motivations,

objectives, and solution architecture. Finally, Section

4 concludes the paper and outlines future work.

2 BACKGROUND

2.1 ITS and Reference Architectures

As previously noted, Intelligent Transportation

Systems (ITS) are defined as 'systems utilizing

synergistic technologies and systems engineering

concepts to develop and improve transportation

systems of all kinds' (Giesecke et al., 2016). ITS

integrate advanced technologies, data analytics, and

communication systems to enhance the efficiency,

safety, and environmental sustainability of

transportation networks (Khalid et al., 2019). By

leveraging real-time data, sensor networks, and

intelligent algorithms, ITS aim to alleviate traffic

congestion, reduce travel times, enhance safety, and

minimize environmental impacts (Barba et al., 2012),

(Khalid et al., 2019).

ITS encompass a wide range of applications, from

traffic management and control to autonomous

vehicles, all aimed at enhancing mobility experiences

and addressing challenges associated with

urbanization. Among the key components of ITS that

contribute to improving transportation efficiency,

safety, and sustainability are Vehicular Adhoc

Networks, Intelligent Traffic Lights, Virtual Traffic

Lights, and Mobility Prediction (Elassy et al., 2024).

Given the complexity of ITS, designing their

architectures poses significant challenges,

particularly for designers unfamiliar with such

systems. A reliable approach to address this issue is

to leverage existing domain-specific reference

architectures. A reference architecture represents a

shared and agreed system description, used by a

community of interest as a guide for the development

and evolution of systems (ISO/IEC/IEEE 42020,

2019). It is a key outcome of the domain design

process (ISO/IEC 26552, 2019), encapsulating

knowledge on designing system architectures within

a specific application domain (Nakagawa et al.,

2011). By incorporating proven architectural

elements, reference architectures help mitigate risks

and enhance reliability (Martínez-Fernández et al.,

2015).

A prominent reference architecture in the field of

intelligent transportation systems (ITS) is ARC-IT

(Architecture Reference for Cooperative and

Intelligent Transportation). ARC-IT provides a

common framework for planning, defining, and

integrating (ARC-IT Version 9.3, 2025), offering a

common language for describing the architecture of

these systems. It is structured around four key

viewpoints: enterprise, functional, physical, and

communication. The scope of ARC-IT is delineated

by a set of ITS services, which refer to transportation

services enabled by intelligent transportation

systems. ARC-IT employs the concept of Service

Packages to define the portions of the architecture

necessary to implement a specific service. These

Service Packages encompass elements from each of

the four architectural views required to describe the

service comprehensively. The areas of Services

include Commercial Vehicle Operations, Public

Transportation, Vehicle Safety, Weather, Public

Safety, Traffic Management, Traveler Information,

Data Management, Sustainable Travel, Parking

Management, Maintenance and Construction, and

Support (Kotsi et al., 2020).

Building on this foundation, an effective approach

to assist in the architectural design of ITS is to

leverage a RAG framework. This effectiveness arises

from the framework's ability to combine the

generative capabilities of large language models

(LLMs) for producing comprehensive responses to

RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture

467

needs with the accuracy and reliability of reference

architectures, which offer proven solutions tailored to

the domain.

2.2 LLMs and RAG

Unlike traditional AI systems, large language models

(LLMs) are designed to process and interpret

unstructured data—such as natural language, images,

and complex sensor data—in a manner that closely

resembles human cognition (Katz et al., 2024), (Liu

et al., 2024), (Liu, 2024). LLMs are extensively

utilized across various subfields of natural language

processing (NLP), addressing diverse tasks by

conditioning the models on a few examples (few-shot

learning) or on instructions describing the task (zero-

shot learning). This method of guiding the language

model is known as 'prompting,' and the development

of effective prompts, whether manually or

automatically, has emerged as a prominent area of

research in NLP (Kojima et al., 2022).

These capabilities enable LLMs to process the

vast and diverse data inputs characteristic of modern

transportation systems, enhancing their adaptability

and responsiveness to the dynamic challenges

inherent in transportation. Furthermore, the

integration of LLMs into ITS represents a significant

advancement, facilitating the development of smarter

and more responsive transportation systems that are

better equipped to address future demands (Wandelt

et al., 2024).

However, it is widely acknowledged that LLMs

can generate hallucinated content—outputs that

contradict established factual knowledge—when they

lack sufficient information (Mondal et al., 2025).

Such inaccuracies can have severe consequences,

especially in safety-critical systems like Intelligent

Transportation Systems (ITS). Safety-critical systems

are defined as those whose failure can result in

significant harm, including the loss of human life

(Awadid et al., 2024).

To address this limitation of LLMs, one solution

is to ground them in existing proven references, such

as domain-specific reference architectures. This can

be achieved through Retrieval-Augmented

Generation (RAG), a methodology that integrates the

generative capabilities of LLMs with information

retrieval (Lewis et al., 2020). RAG allows the model

to dynamically access and incorporate relevant

external information during the generation process

(Xia et al., 2024). This approach has been recognized

as a promising strategy, not only for mitigating

hallucination but also for enhancing the domain-

specific expertise and temporal relevance of LLMs

(Chen et al., 2024), (Zhao et al., 2024). By doing so,

RAG improves the controllability and interpretability

of model outputs, making them more reliable and

aligned with the application domain.

2.3 Related Work

The immense potential of LLMs in the field of

transportation has sparked significant interest, not

only among the general public but also within the

research community, as evidenced by a growing body

of literature. One prominent area of research focuses

on the applications of LLMs in Intelligent

Transportation Systems (ITS), with a particular

emphasis on Autonomous Driving (AD).

Autonomous Driving refers to 'the technology and

systems that enable a vehicle to perceive its

environment and make decisions independently,

using artificial intelligence, computer vision, and

sensor technologies to ensure safe driving' (Li et al.,

2024).

The integration of autonomous driving (AD) and

large language models (LLMs) holds significant

promise, with the potential to enhance both user

experience (e.g., through in-vehicle voice assistants

and driving-related decision-making) and vehicle

performance (e.g., by enabling complex

environmental perception, intelligent anomaly

detection, and optimized battery management) (Lei et

al., 2023), (Singh, 2023), (Yang et al., 2023).

From the user perspective, large language models

(LLMs) offer significant potential to enhance

interactions with autonomous vehicles (AVs),

including providing feedback on driving quality,

incorporating personal preferences to influence

driving behavior, and retrieving contextual route

information such as traffic conditions or weather

updates (Chen et al., 2023), (Du et al., 2023), (Xu et

al., 2024). These capabilities are particularly valuable

in scenarios where an autonomous driving (AD)

system encounters uncertainties or requires user input

to support decision-making. In such cases, LLMs

could facilitate comprehensive dialogues with users,

addressing safety, comfort, or route modification

concerns in real time. Currently, communication

between users and vehicles remains rudimentary,

typically limited to keyword-based commands or a

restricted set of predefined use cases. LLMs promise

to enable more natural and intuitive interactions with

broader functionality, potentially exceeding the

original design intent of the vehicle manufacturer or

operator. However, this expanded interaction scope

raises potential security and safety concerns, as users

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

468

may inadvertently interfere with the AD system in

unintended ways.

From the vehicle's perspective, integrating large

language models (LLMs) holds significant promise

for enhancing the performance and safety of

autonomous driving (AD) systems (Wang et al.,

2023). As outlined in (Fu et al., 2024), AD systems

aiming to emulate human driving require three key

abilities: reasoning, interpretation, and memorization.

To explore the feasibility of employing LLMs in AD

scenarios, a closed-loop system was developed,

demonstrating the exceptional reasoning capabilities

of LLMs in addressing long-tail scenarios. In a

similar vein, (Cui et al., 2023) introduced a decision-

making framework called DriveLLM, which

integrates AD stacks with LLMs to enable

commonsense reasoning in decision-making

processes. DriveLLM’s cyber-physical feedback

system allowed it to iteratively learn from mistakes,

thereby improving performance. Additionally, (Wen

et al., 2023) proposed the DiLu framework,

leveraging the emergent abilities of LLMs and real-

world datasets to enhance AD tasks. Complementing

this, (Sha et al., 2023) developed cognitive pathways

and algorithms designed to bridge LLM-based

reasoning with actionable driving commands. Their

approach outperformed baseline methods in single-

vehicle tasks and proved effective for managing

complex behaviors, including multi-vehicle

coordination.

(Deng et al., 2023) proposed an end-to-end test

generation framework designed to automatically

construct test scenarios within an autonomous driving

simulator. The framework leverages ChatGPT-4 to

extract key information from traffic rules, enabling

the generation of targeted test cases. Experimental

evaluations across diverse autonomous driving

systems, traffic regulations, and road maps

demonstrated the framework’s effectiveness in

identifying rule violations and uncovering known

issues. To address information barriers arising from

system and module heterogeneity, (Tian et al., 2023)

introduced a Transformer-based unified framework

called VistaGPT. This framework integrates Modular

Federations of Vehicular (M-FoV) Transformers with

Automated Autonomous (AutoAuto) Driving

Systems to enable seamless interoperability.

Building on the preceding analysis of related

work, existing studies have predominantly focused on

the potential applications of LLMs and augmented

LLMs in Intelligent Transportation Systems (ITS)

from two primary perspectives: enhancing user

experience and improving vehicle performance.

While these perspectives are invaluable, they leave a

critical gap in understanding and leveraging LLMs to

support ITS architecture designers and modelers.

Specifically, the role of LLMs in facilitating the

design, evaluation, and evolution of ITS architectures

has been largely unexplored. Addressing this gap

forms a core motivation for our research project,

RITSA (Retrieval Augmented Generation for ITS),

which aims to pioneer the integration of augmented-

LLMs (RAG) into the ITS architecture design

process, thereby extending their potential beyond user

and vehicle-focused applications.

3 RITSA PROJECT:

AN OVERVIEW

3.1 Motivations for RITSA Project

The motivations behind our research project RITSA

(RAG for Intelligent Transportation Systems

Architecture) are twofold.

3.1.1 Expanding Beyond the ITS end-User

Perspective

The existing body of research on LLMs and

augmented LLMs in Intelligent Transportation

Systems (ITS) demonstrates a growing interest in

leveraging these technologies to enhance the user

experience and optimize vehicle performance. These

studies underscore the transformative potential of

LLM-driven applications in addressing end-user

needs, such as improving in-vehicle communication,

personalizing driving behaviors, and supporting real-

time navigation. However, a critical gap remains:

current research predominantly focuses on the ITS

end-user perspective, overlooking other essential

roles, particularly that of ITS architecture designers

and modelers.

From a systems engineering standpoint, the early

stages of the ITS lifecycle—encompassing

operational analysis, system specification, and

architectural design—are fundamental to ensuring the

system's efficiency, reliability, and adaptability.

Despite their importance, these phases remain largely

unexplored in the context of LLMs and Retrieval-

Augmented Generation (RAG) frameworks.

Addressing this gap constitutes a primary motivation

for the RITSA project. By integrating RAG into the

ITS architecture design process, RITSA seeks to

pioneer a new frontier where LLM-based tools

empower designers and modelers, broadening the

applications of these technologies to encompass not

RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture

469

only end-user and vehicle-centric solutions but also

system-level design and modeling.

3.1.2 Bridging the Gap in Utilizing

Reference Architectures

The concept of "reference architecture" is well-

established and highly valued within the

standardization community, where it is defined as “a

shared and agreed reference system description used

by a community of interest to achieve its business

purposes. It is typically generic and instantiated as

system architectures specific to individual business

purposes” (ISO/IEC/IEEE 42010, 2022). Despite its

recognized significance, feedback from joint research

and industry projects conducted by our research

institute —such as TAM (Trusted Autonomous

Mobility)

, SCA (Secure Cooperative Autonomous

Systems)

, and RTI (Resilience of Intelligent

Transport)

reveals that ITS reference architectures

are often underutilized.

A key factor behind this underutilization is the

lack of intuitive and interactive approaches for

instantiating and tailoring these architectures in real-

world projects. This challenge highlights the need for

tools that make reference architectures more

accessible and user-friendly. The RITSA project

addresses this gap by leveraging RAG

methodologies, which promise to facilitate more

natural and intuitive user interactions. By doing so,

RITSA aims to empower ITS stakeholders to

effectively exploit the full potential of reference

architectures, meeting the practical needs of their

users and enhancing their application in the ITS

domain.

3.2 A High-Level Overview of the RAG

Framework for RITSA

This section presents the high-level architecture of the

RAG framework for the RITSA project, focusing on

its role in assisting ITS designers during the early

phases of the ITS development lifecycle. The RAG

framework is designed to enable ITS designers to

express their needs in terms of the desired ITS

service, such as cooperative autonomous driving,

traffic management, or vehicle-to-infrastructure

communication. Based on this input, the framework

retrieves and synthesizes relevant architecture

elements to generate tailored architecture designs for

the specified ITS service. The high-level overview of

the RAG framework for the RITSA project is given

in Figure 1.

Figure 1: High-level overview of the RAG framework for the RITSA project.

https://www.irt-systemx.fr/projets/tam/

https://www.irt-systemx.fr/en/projets/sca/

https://www.irt-systemx.fr/en/projets/rti/

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

470

The framework presented in Figure 1 is described as

follows:

1. ITS designer interface and prompt input: The

framework begins with an intuitive user

interface (UI), where ITS designers can

express their needs as natural language

prompts. Specifically, the designer articulates

their query in terms of the desired ITS service

that requires architectural design. For

instance, the designer might request the

architecture for a cooperative traffic signal

system or a secure vehicle-to-infrastructure

communication module. In this context, it is

noteworthy that designers are not required to

possess deep technical expertise in ITS

architecture modeling, as prompts serve as the

primary mechanism for initiating the retrieval

and generation of architecture-specific

elements.

2. Query processing via the API: The designer’s

prompt is transferred to the backend through

the application programming interface (API),

which acts as a communication bridge

between the UI and the processing

components. The API translates the natural

language prompt into formats understandable

by the retrieval and generation algorithms. It

ensures compatibility between the user-facing

interface and the underlying algorithms.

3. Information retrieval algorithm: The

Information retrieval algorithm is the first

processing step. It identifies and extracts

relevant information from reference

architecture such as the ARC-IT (Architecture

Reference for Cooperative and Intelligent

Transportation) framework. It serves to match

the designer’s query to the appropriate ITS

service-related content to ensure that retrieved

elements are relevant to the specific ITS

service described in the designer’s prompt.

4. LLM utilization algorithm: Once the relevant

information is retrieved, it is processed by the

LLM Utilization Algorithm, which leverages

advanced Large Language Models (LLMs)

like GPT-4 and BERT. It synthesizes the

retrieved information with the context of the

original designer’s query, and then generates a

comprehensive response that outlines the

architecture design elements for the specified

ITS service.

5. Generated response: The result of the LLM

processing is a Generated Response, which is

tailored to the designer's query. This response

is passed back through the API to the UI,

where the ITS designer can access it. The

output is delivered as a generated architecture

design, comprising a detailed description of

the architecture elements for the specified ITS

service, an explanation of how the elements fit

together to support the desired service, and

references to the ITS reference architecture for

traceability and verification.

4 CONCLUSIONS AND

FUTURE WORK

Due to the inherent complexity and evolving nature

of Intelligent Transportation Systems (ITS), their

architectural design and modeling remain challenging

and time-consuming, particularly for modelers with

limited expertise in these systems. This highlights the

need for intelligent assistants to support the ITS

architecture design and modeling process. In response

to this need, and to address the research gap in this

area, this paper introduces the RAG for Intelligent

Transportation Systems Architecture (RITSA)

project. RITSA aims to assist ITS designers and

modelers during the architectural design process by

leveraging a Retrieval-Augmented Generation

(RAG) framework. By integrating the generative

capabilities of large language models (LLMs) with

information retrieval from proven ITS reference

architectures such as ARC-IT, RITSA seeks to

pioneer a new frontier in ITS design. This approach

broadens the applications of LLM-based tools beyond

end-user and vehicle-centric solutions to encompass

system-level design and modeling. Through this

innovative integration, RITSA empowers ITS

stakeholders to effectively exploit the full potential of

reference architectures, addressing practical needs

and enhancing their applicability within the ITS

domain.

This paper provided a comprehensive overview of

the RITSA project, covering its background—

including Intelligent Transportation Systems (ITS),

reference architectures, large language models

(LLMs), Retrieval-Augmented Generation (RAG),

and related existing work—along with its motivations

and the high-level architecture of the underlying RAG

framework. The project is driven by two key

motivations: (1) extending the application of LLM-

based technologies beyond the ITS end-user

perspective to encompass the ITS designer and

modeler perspective; and (2) addressing the

underutilization of ITS reference architectures by

providing intuitive and interactive means for their

RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture

471

instantiation and tailoring. Building on the high-level

RAG framework architecture presented in this paper,

the immediate next steps for the RITSA project will

focus on two main objectives: (1) implementing the

RAG system and (2) evaluating its performance based

on well-defined criteria.

ACKNOWLEDGEMENTS

The RITSA project is a collaborative research

initiative between the French Technological Research

Institute SystemX, the German Research Center for

Artificial Intelligence (DFKI), and the Austrian Josef

Ressel Centre for Dependable System-of-Systems

Engineering.

REFERENCES

ARC-IT Version 9.3 (2025). Available at https://www.arc-

it.net/.

Aslam, M. S., & Nisar, S. (2024). Ethical Considerations

for Artificial Intelligence Tools in Academic Research

and Manuscript Preparation: A Web Content Analysis.

In Digital Transformation in Higher Education, Part B:

Cases, Examples and Good Practices (pp. 155-196).

Emerald Publishing Limited.

Awadid, A., Robert, B., & Langlois, B. (2024, February).

MBSE to Support Engineering of Trustworthy AI-

Based Critical Systems. In 12th International

Conference on Model-Based Software and Systems

Engineering.

Barba, C. T., Mateos, M. A., Soto, P. R., Mezher, A. M., &

Igartua, M. A. (2012, June). Smart city for VANETs

using warning messages, traffic statistics and intelligent

traffic lights. In 2012 IEEE intelligent vehicles

symposium (pp. 902-907). IEEE.

Barke, S., James, M. B., & Polikarpova, N. (2023).

Grounded copilot: How programmers interact with

code-generating models. Proceedings of the ACM on

Programming Languages, 7(OOPSLA1), 85-111.

Chen, H., Yuan, K., Huang, Y., Guo, L., Wang, Y., & Chen,

J. (2023). Feedback is all you need: from ChatGPT to

autonomous driving. Science China Information

Sciences, 66(6), 1-3.

Chen, J., Lin, H., Han, X., & Sun, L. (2024, March).

Benchmarking large language models in retrieval-

augmented generation. In Proceedings of the AAAI

Conference on Artificial Intelligence (Vol. 38, No. 16,

pp. 17754-17762).

Combemale, B., Gray, J., & Rumpe, B. (2023). ChatGPT in

software modeling. Software and Systems Modeling,

22(3), 777-779.

Cui, Y., Huang, S., Zhong, J., Liu, Z., Wang, Y., Sun, C., ...

& Khajepour, A. (2023). Drivellm: Charting the path

toward full autonomous driving with large language

models. IEEE Transactions on Intelligent Vehicles.

Damaj, I., Al Khatib, S. K., Naous, T., Lawand, W.,

Abdelrazzak, Z. Z., & Mouftah, H. T. (2022).

Intelligent transportation systems: A survey on modern

hardware devices for the era of machine learning.

Journal of King Saud University-Computer and

Information Sciences, 34(8), 5921-5942.

Deng, Y., Yao, J., Tu, Z., Zheng, X., Zhang, M., & Zhang,

T. (2023). Target: Traffic rule-based test generation for

autonomous driving systems. arXiv preprint

arXiv:2305.06018.

Du, H., Teng, S., Chen, H., Ma, J., Wang, X., Gou, C., ... &

Wang, F. Y. (2023). Chat with chatgpt on intelligent

vehicles: An ieee tiv perspective. IEEE Transactions on

Intelligent Vehicles, 8(3), 2020-2026.

Elassy, M., Al-Hattab, M., Takruri, M., & Badawi, S.

(2024). Intelligent transportation systems for

sustainable smart cities. Transportation Engineering,

100252.

Elassy, M., Al-Hattab, M., Takruri, M., & Badawi, S.

(2024). Intelligent transportation systems for

sustainable smart cities. Transportation Engineering,

100252.

Friedenthal, S., Moore, A., & Steiner, R. (2014). A practical

guide to SysML: the systems modeling language.

Morgan Kaufmann.

Fu, D., Li, X., Wen, L., Dou, M., Cai, P., Shi, B., & Qiao,

Y. (2024). Drive like a human: Rethinking autonomous

driving with large language models. In Proceedings of

the IEEE/CVF Winter Conference on Applications of

Computer Vision (pp. 910-919).

Giesecke, R., Surakka, T., & Hakonen, M. (2016, April).

Conceptualizing mobility as a service. In 2016 eleventh

international conference on Ecological Vehicles and

Renewable Energies (EVER) (pp. 1-11). IEEE.

Haomin, W., Haomin, L., Xinai, Z., & Miao, Y. (2021,

April). High-Frequency Data Link System Architecture

Design based on SysML Modeling Language. In

Journal of Physics: Conference Series (Vol. 1884, No.

1, p. 012014). IOP Publishing.

INCOSE (Ed.). (2023). INCOSE systems engineering

handbook. John Wiley & Sons.

ISO/IEC 26552:2019: Software and systems engineering —

Tools and methods for product line architecture design.

Int. Organ. Stand. Geneva, Switzerland (2019).

ISO/IEC/IEEE 42010:2022: Software, Systems and

Enterprise Architecture Description. ISO.

https://www.iso.org/standard/74393.html

ISO/IEC/IEEE 42020:2019:2019-07.: Software, systems

and enterprise — Architecture processes. Int. Organ.

Stand. Geneva, Switzerland (2019).

Katz, D. M., Bommarito, M. J., Gao, S., & Arredondo, P.

(2024). Gpt-4 passes the bar exam. Philosophical

Transactions of the Royal Society A, 382(2270),

20230254.

Khalid, T., Khan, A. N., Ali, M., Adeel, A., ur Rehman

Khan, A., & Shuja, J. (2019). A fog-based security

framework for intelligent traffic light control system.

Multimedia Tools and Applications, 78, 24595-24615.

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

472

Kojima, T., Gu, S. S., Reid, M., Matsuo, Y., & Iwasawa, Y.

(2022). Large language models are zero-shot reasoners.

Advances in neural information processing systems, 35,

22199-22213.

Kotsi, A., Lusco, T., Mitsakis, E., & Sill, S. (2020).

Harmonization and interoperability of C-ITS

Architectures in Europe and USA. arXiv preprint

arXiv:2010.08405.

Lei, L., Zhang, H., & Yang, S. X. (2023). ChatGPT in

connected and autonomous vehicles: benefits and

challenges. Intell. Robot, 3(2), 145-148.

Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V.,

Goyal, N., ... & Kiela, D. (2020). Retrieval-augmented

generation for knowledge-intensive nlp tasks.

Advances in Neural Information Processing Systems,

33, 9459-9474.

Li, M., Liu, X., Chen, S., Yang, L., Du, Q., Han, Z., &

Wang, J. (2024). MST-YOLO: Small Object Detection

Model for Autonomous Driving. Sensors, 24(22), 7347.

Liu, X., Zheng, Y., Du, Z., Ding, M., Qian, Y., Yang, Z., &

Tang, J. (2024). GPT understands, too. AI Open, 5, 208-

215.

Liu, Y. (2024). Large language models for air

transportation: A critical review. Journal of the Air

Transport Research Society, 100024.

Martineau, K., Explainable, A. I., & Generative, A. I.

(2023). What is retrieval-augmented generation?. IBM

Research Blog, 22.

Martínez-Fernández, S., Ayala, C. P., Franch, X.,

Nakagawa, E. Y.: A Survey on the Benefits and

Drawbacks of AUTOSAR. In: Proceedings of the First

International Workshop on Automotive Software

Architecture, pp. 19-26 (2015).

Mondal, H., Mondal, S., & Behera, J. K. (2025). Artificial

intelligence in academic writing: Insights from journal

publishers’ guidelines. Perspectives in Clinical

Research, 16(1), 56-57.

Nakagawa, E. Y., Antonino, P. O., Becker, M.: Reference

architecture and product line architecture: A subtle but

critical difference. In: European Conference on

Software Architecture. Springer, Berlin, Heidelberg,

pp. 207-211 (2011).

Sha, H., Mu, Y., Jiang, Y., Chen, L., Xu, C., Luo, P., ... &

Ding, M. (2023). Languagempc: Large language

models as decision makers for autonomous driving.

arXiv preprint arXiv:2310.03026.

Singh, G. (2023). Leveraging chatgpt for real-time

decision-making in autonomous systems.

Eduzone: International Peer Reviewed/Refereed

Multidisciplinary Journal, 12(2), 101-106.

Tian, Y., Li, X., Zhang, H., Zhao, C., Li, B., Wang, X., &

Wang, F. Y. (2023). VistaGPT: Generative parallel

transformers for vehicles with intelligent systems for

transport automation. IEEE Transactions on Intelligent

Vehicles.

Wandelt, S., Zheng, C., Wang, S., Liu, Y., & Sun, X.

(2024). Large Language Models for Intelligent

Transportation: A Review of the State of the Art and

Challenges. Applied Sciences, 14(17), 7455.

Wang, Y., Jiao, R., Zhan, S. S., Lang, C., Huang, C., Wang,

Z., ... & Zhu, Q. (2023). Empowering autonomous

driving with large language models: A safety

perspective. arXiv preprint arXiv:2312.00812.

Waqar, A., Alshehri, A. H., Alanazi, F., Alotaibi, S., &

Almujibah, H. R. (2023). Evaluation of challenges to

the adoption of intelligent transportation system for

urban smart mobility. Research in Transportation

Business & Management, 51, 101060.

Wen, L., Fu, D., Li, X., Cai, X., Ma, T., Cai, P., ... & Qiao,

Y. (2023). Dilu: A knowledge-driven approach to

autonomous driving with large language models. arXiv

preprint arXiv:2309.16292.

Xia, Y., Jazdi, N., & Weyrich, M. (2024, September).

Enhance FMEA with Large Language Models for

Assisted Risk Management in Technical Processes and

Products. In 2024 IEEE 29th International Conference

on Emerging Technologies and Factory Automation

(ETFA) (pp. 1-4). IEEE.

Xu, Z., Zhang, Y., Xie, E., Zhao, Z., Guo, Y., Wong, K. Y.

K., ... & Zhao, H. (2024). Drivegpt4: Interpretable end-

to-end autonomous driving via large language model.

IEEE Robotics and Automation Letters.

Yang, Z., Jia, X., Li, H., & Yan, J. (2023). A survey of large

language models for autonomous driving. arXiv

preprint arXiv:2311.01043.

Zhao, S., Yang, Y., Wang, Z., He, Z., Qiu, L. K., & Qiu, L.

(2024). Retrieval Augmented Generation (RAG) and

Beyond: A Comprehensive Survey on How to Make

your LLMs use External Data More Wisely. arXiv

preprint arXiv:2409.14924.

RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture

473