Conceptual Approaches to Identify the Hazardous Scenarios in Safety

Analysis for Automated Driving Systems

Marzana Khatun

1 a

, Florence Wagner

2 b

, Rolf Jung

and Michael Glass

Kempten University of Applied Sciences, Bahnhofstrasse 61, Kempten, Germany

Institute for Driver Assistance and Connected Mobility, Benningen, Germany

Embedded Systems/ Real-Time Systems, University of Ulm, Ulm, Germany

{marzana.khatun, ﬂorence.wagner, rolf.jung}@hs-kempten.de, michael.glass@uni-ulm.de

Keywords:

Functional Safety, Safety of the Intended Functionality, Hazard Analysis and Risk Assessment, Automated

Driving System, Machine Learning.

Abstract:

To ensure safety of the road users is one of the major challenges in highly automated driving. The technologies

applied in semi or fully-automated vehicles that are safer than human drivers compromise functionalities and

human comfort. A comprehensive understanding of the use of complex driving systems and the Operational

Design Domain (ODD) is essential for the effective deployment and safe operation of Automated Driving

Systems (ADSs). Hazard analysis is a foundation of various safety engineering methods, which include Func-

tional Safety (FuSa) and Safety Of The Intended Functionality (SOTIF). The scenario-based analysis offers

signiﬁcant advantages in the safety analysis of automated vehicles but poses inherent difﬁculties in identify-

ing unknown-hazardous scenarios. The work presented in this paper deals with the conceptual approaches of

hazard scenario identiﬁcation. Moreover, discusses the incorporation of Machine Learning (ML) in Hazard

Analysis and Risk Assessment (HARA) for vehicles equipped with ADSs. Furthermore, this paper can serve

as foundation support for research inquiries related to ADSs validation and safety assessment.

1 INTRODUCTION

The complexity in Automated Driving Systems

(ADSs) has brought safety aspects to the forefront of

research efforts (Tu and Sun, 2023). According to

Society of Automotive Engineers (SAE) J3016:2021

standard, automated or autonomous vehicles are cate-

gorized into six automation levels from level 0 (No

Driving Automation) to level 5 (Full Driving Au-

tomation). First three driving automation levels (0-

2) require a human driver who is at all times su-

pervising the support features and handling the driv-

ing for safety reasons. Level 3 higher are taking

over the driving task while level 3 can detect the

limits of its application range and request the driver

be in control, level 4 and 5 are not requiring some-

one to take over driving. This paper focuses on

driving automation level 4 (High Driving Automa-

tion) and/or level 5 (SAE, 2021). Current Functional

Safety (FuSa) and Safety Of The Intended Func-

tionality (SOTIF) related standards are focusing on

https://orcid.org/0000-0002-3839-1575

https://orcid.org/0009-0009-2515-1116

the automation level 2 where humans are responsi-

ble to ensure driving safety. Standards such as ISO

26262:2018 (ISO26262, 2018) for road vehicle FuSa

and ISO 21448:2022 (ISO21448, 2022) for road ve-

hicle SOTIF consider performance insufﬁciencies of

the complex components used in an ADS. More-

over, the upcoming ISO/PAS 8800 standard is a valu-

able addition that enhances safety in Artiﬁcial Intelli-

gence (AI)/Machine Learning (ML), providing essen-

tial guidelines as well (ISO/PAS8800, 2024).

Fully-automated vehicles are intended to drive ef-

ﬁciently from one point to another. According to a

study, the uses of automated driver-assistance systems

in Europe could reduce the number of accident by

about 15% by 2030 (Deichmann and Steiner, 2023).

ADS-equipped vehicles can add additional value to-

gether with human comfort for automotive industry

and could generate between $300 billion and $400

billion in the passenger car market by 2035, according

to McKinsey analysis (Deichmann and Steiner, 2023).

According to (Feldmann, 2023), Feldmann mentions

that the adoption of safety measures is hindered by

slow progress, mainly due to lack of resources. From

this motivation, the research questions addressing in

Khatun, M., Wagner, F., Jung, R. and Glass, M.

Conceptual Approaches to Identify the Hazardous Scenarios in Safety Analysis for Automated Driving Systems.

DOI: 10.5220/0013250900003890

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

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 1, pages 549-555

ISBN: 978-989-758-737-5; ISSN: 2184-433X

549

this paper with respect to safety analysis of ADS-

equipped vehicles are:

1. What methodologies can be employed to identify

unknown-hazardous scenarios and subsequently

reduce unknown-hazardous concerns in the con-

text of SOTIF-related scenario evaluation?

2. What steps can be involved to perform scenario-

based hazard analysis, including scenario simula-

tion?

3. Can continuous improvement be achieved in haz-

ard analysis and risk assessment to effectively

support the veriﬁcation and validation of auto-

mated driving systems?

4. What strategies or methodologies can be em-

ployed to ensure ongoing enhancement in this

critical aspect of automated vehicle development?

This paper focuses on the scenario-based safety

analysis in terms of hazardous scenario identiﬁcation

and support to reduce the unknown-hazardous scenar-

ios. Unknown-hazardous scenarios are described as

Area 3 based on the scenario category division from

SOTIF standard. Here, the scenarios are classiﬁed

as known-not hazardous (Area 1), known-hazardous

(Area 2), unknown-hazardous (Area 3) and unknown-

not hazardous (Area 4). In relation to ADS safety, the

scenarios assigned to Area 3 are considered to be the

most critical. Scenario modeling and simulation play

a vital role in investigating and understanding the crit-

ical aspects of ADS safety. The contributions of this

paper are listed below:

• Propose conceptual approaches to incorporate a

new technology in hazardous scenario identiﬁca-

tion.

• Description of safety-related scenarios to support

scenario-based hazard analysis and risk assess-

ment for automated vehicles.

• Illustrate incorporation of ML and determinis-

tic approaches in scenario-based Hazard Analysis

and Risk Assessment (HARA) applicable in au-

tomation level 4 or higher. Looking at level 4 and

5 of driving automation the driving features will

be in control and will not require a human driver

to take over.

The outline of this paper is as follows, Section

2 introduces normative deﬁnitions of terms used in

safety and cybersecurity domain. Section 3 presents

the proposed approaches to identify hazardous scenar-

ios and risk assessment. Finally, Section 4 provides a

redundant summary of the discussion and offers an

outlook on the future perspectives of the study.

2 LITERATURE STUDY

2.1 Normative Terms and Deﬁnitions

According to SAE J3016:2021 (SAE, 2021), ADS is

the ”hardware and software that are collectively ca-

pable of performing the entire dynamic driving task

on a sustained basis, regardless of whether it is lim-

ited to a speciﬁc ODD; this term is used speciﬁcally

to describe a Level 3, 4, or 5 driving automation sys-

tem.” The ODD is described as the ”speciﬁc condi-

tions under which a given driving automation system

is designed to function” (ISO21448, 2022). The ODD

is not limited to, environmental, geographical, and

time-of-day restrictions, and/or the requisite presence

or absence of certain trafﬁc or roadway characteris-

tics (SAE, 2021). According to ISO 26262-1:2018

(ISO26262, 2018), SAFETY is the ”absence of unrea-

sonable risk”. FUNCTIONAL SAFETY is deﬁned as an

”absence of unreasonable risk due to hazards caused

by malfunctioning behavior of electrical and/or elec-

tronic systems”. A HAZARD is deﬁned as a ”potential

source of harm caused by malfunctioning behavior of

the item”, and a HAZARDOUS EVENT is deﬁned as

a ”combination of a hazard and an operational situ-

ation”. The HARM indicates the ”physical injury or

damage to the health of persons” (ISO26262, 2018).

According to ISO 21448:2022, a SCENARIO is de-

scribed as ”a temporal relationship between several

scenes in a sequence of scenes, with goals and values

within a speciﬁed situation, inﬂuenced by actions and

events” (ISO21448, 2022). The HAZARDOUS SCE-

NARIO is a ”scenario when harm occurs unless pre-

vented by an entity other than the ADS” (ISO34502,

2022).

2.2 Methods and Technologies

As stated in 26262-1:2018 (ISO26262, 2018), a

FUNCTIONAL CONCEPT is described as ”speciﬁca-

tion of the intended functions and their interactions

necessary to achieve the desired behavior. The func-

tional concept developed during the concept phase de-

scribes a set of speciﬁcation of the functional safety

requirements, with associated information, their allo-

cation to elements within the architecture, and their

interaction necessary to achieve the safety goals”.

HARA is a ”method to identify and categorize haz-

ardous events of items and to specify safety goals and

automotive safety integrity levels related to the pre-

vention or mitigation of the associated hazards in or-

der to avoid unreasonable risk” (ISO26262, 2018).

The growing need for large-scale coverage of sce-

narios in autonomous vehicles is driving the adop-

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

550

tion of technologies such as ML models. These tech-

nologies are used to support scenario-based analysis,

identifying safety-critical parameters, scenario sim-

ulation and reduction in testing and optimize time

efﬁciency. In (Grigorescu and Macesanu, 2020),

GRIGORESCU, TRASNEA, COCIAS AND MACE-

SANU present a survey of the current state-of-the-art

on deep learning technologies used in autonomous

driving. Simulation-based HARA is suggested by

(Oakes, 2021) but focused on fault injection not in

hazardous scenario identiﬁcation. Furthermore, a

concept of hazard scenario identiﬁcation is presented

in (Khatun and Glaß, 2023) based on simulated sce-

nario but risk assessment was not present in detail.

Scenario-based HARA supports to understand and

realize the hazardous events and is widely applied in

automated vehicle’s safety assessment (Madala and

Solmaz, 2023; Khatun and Jung, 2020). To in-

vestigate the hazard and risk methods like, System-

Theoretic Process Analysis (STPA) and dynamic haz-

ard approach are applied in ADS-equipped vehicles

(Chen and Zhao, 2020; Schwalb, 2021). A priori

safety risk assessments of ADS-equipped vehicles

road tests are of great importance as they allow quan-

tiﬁcation of the level of risk in different road scenarios

and provide guidance for such vehicles road tests (Tu

and Sun, 2023),(TR5469, 2024). AI technologies can

not only help to recognize, research and to deﬁne haz-

ardous scenarios, but they can also accelerate func-

tional safety analyses. Although validating AI system

safety poses signiﬁcant challenges, it can be managed

effectively through rigorous conﬁrmation measures.

A scenario-based HARA approch is presented

here with the support of ML model to deﬁne the in-

puts that are the key to establish the base scenario

framework in HARA, applicable for L3 and/or higher

level of automation (L4/L5.)

3 PROPOSED HAZARD

ANALYSIS APPROACHES

3.1 Scenario-Based Hazard Analysis

While scenario-based analysis has many advantages

by means of detail description and reducing misun-

derstanding in autonomous safety analysis, it also has

some shortcomings. The time and effort required

for scenario-based HARA is very high. One of the

biggest challenges is the number of scenarios that

need to be considered for risk assessment. A further

obstacle arises from the limited applicability of real-

world scenarios to automation level 3 and/or higher,

leading to the emergence of novel hazard scenarios

that cannot be effectively derived from real-world sce-

narios. In such cases, the inclusion of simulation-

based scenarios is essential for comprehensive safety

analyses.

Simulation-based approaches have gained signiﬁ-

cant traction for identifying unknown-hazardous sce-

narios (Zhang and Felbinger, 2023). ISO 21448:2022

introduces scenario categories divided into areas 1

to 4, based on the attributes known/unknown and

hazardous/non-hazardous. These categories are used

to facilitate the preparation of the safety concept and

the deﬁnition of hazardous events. The hazard analy-

sis begins with the known hazardous scenarios (Area

2). Simulation-based and knowledge-based investiga-

tions based on safety-related parameters can be used

to reduce the scope of Area 3, as shown by the ar-

rows in Fig. 1. In cases where simulation-based sce-

nario analysis begins with known-hazardous scenar-

ios, the utilization of parameter variation techniques

aids in the identiﬁcation of unknown-hazardous sce-

narios and thus leads to a reduction in the scope of

Area 3 (See Fig. 1).

Figure 1: Scenario Categories Visualization with Areas.

3.2 Proposed Concepts

The conceptual approaches proposed in this take into

account the different representations of real-world

crash scenarios, virtual scenarios, and pre-crash sce-

narios. To generate scenarios methods like, tree dia-

gram, formal rule-based and model-based approaches

can be applied for safety analyses (Khan and Vijay,

2023; Faysal, 2022). For scenario-simulation tools

like, CarMaker (IPG-Automotive, ), MATLAB (Mat-

lab, ), Simulation of Urban Mobility (SUMO) (Li,

Conceptual Approaches to Identify the Hazardous Scenarios in Safety Analysis for Automated Driving Systems

551

2023), Unity3D (Gang, 2016), CARLA Simulator

(Horel, 2022) are utilized in various stages of veriﬁca-

tion and validation. To generate and identify critical

scenarios, SOTIF-related hazard analysis and risk as-

sessment (HARA) includes several areas to consider

(Kon

e and G

eronimi, 2023; Yang et al., 2023; Sana

and Raahemifar, 2023; Kramer et al., 2020). These

include ODD, environmental conditions, and com-

plex system features leading to limitations in human-

performed HARA due to the heightened volume of

scenarios.

The different representations of these scenarios

cause difﬁculties in analyzing of hazards for auto-

mated driving systems. To aid in mitigating chal-

lenges, a generic ﬂow diagram is depicted in Fig.

2. The ﬂow diagram in Fig. 2 shows the possibili-

ties how a scenario database can be created and sim-

ulated for identifying unknown-hazardous scenarios

and continuously update the scenario database.

Figure 2: Proposed Flow Diagram for Scenario-based

HARA.

In the following two concepts, both determinis-

tic (yellow color) and ML model (light blue color)

approaches are integrated within the context of haz-

ardous scenario identiﬁcation and risk assessment for

ADS-equipped vehicles. The concepts include the use

of Hazard and Operability Analysis (HAZOP) key-

words displayed as dark green color (see in Fig. 3 and

Fig. 4) and characteristics that are used in describing

hazardous events (light green color), as shown in Fig.

3 and Fig. 5.

Concept 1. The proposed concept 1 is divided into

two segments. In the ﬁrst segment, the hazardous

events are compiled based on the characteristics of the

scenario descriptions. Finite number of Hazardous

Events (HE) are presented in Fig. 3 from 1 to n

(HE1, HE2, ..., HEn). Then, hazardous events are

combined with the predeﬁned HAZOP keywords by

applying an ML technique. The characteristics men-

tioned in Fig. 3 are operating mode, operational sit-

uation and vehicle functions. The deterministic ap-

proach is marked by a yellow block and can be car-

ried out by a knowledge-based method for describ-

ing hazard events. The number of hazard events can

range from a ﬁnite number between 1 and n (see Fig.

3). The hazardous events database can be employed

to identify previously unknown-hazardous scenarios,

thereby contributing to the reduction of Area 3.

Figure 3: Hazardous Scenario Identiﬁcation - Concept 1

(Segment 1).

In the second segment of concept 1, hazard events

combined with HAZOP keywords are used to formu-

late Accident Scenarios (ACs). Each HAZOP key-

word is associated with individual hazardous events,

to deﬁne multiple sets of accident scenarios (red

color) as shown in Fig. 4. A ML model can be ap-

plied to determine the Severity (S), Exposure (E) and

Controllability (C). Based on these three parameters,

Automotive Safety Integrity Level (ASIL) can be de-

ﬁned using risk graph or risk matrix for risk assess-

ment of ADS-equipped vehicles.

An exemplary case: The operating mode is driv-

ing an ADS-equipped vehicle in an operating sce-

nario in which the vehicle performs a lane change

from the right to the left lane on a straight highway

in sunny weather. To perform the lane change, the

ADS-equipped vehicle uses the steering, acceleration,

and braking functions. There are other road users

on the road that are passed by an ADS-equipped ve-

hicle while it is performing a lane change. A haz-

ardous event can be described as follows: The cam-

era sensor fails to detect the lane marking while an

ADS-equipped vehicle is performing a lane change.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

552

Figure 4: Hazardous Scenario Risk Assessment - Concept

1 (Segment 2).

Some of the traditional HAZOP keywords are no or

not, late, after. Based on the HAZOP keywords,

ACs can be described as follows: A rear-end colli-

sion occurs. Accident database can be used as mixed

datasets. Mixed datasets can be used to identify the

parameters (S, E, C) using deep neural networks, such

as convolutional neural networks and recurrent neural

networks. Finally, a deterministic method can be ap-

plied to determine ASIL based on ISO 26262:2018.

Concept 2. Concept 2 is divided into two segments.

First, hazardous events are considered as database for

ML. Hazardous events are classiﬁed into several sets

as shown in Fig. 5 The HE classiﬁcation sets are pre-

sented in Fig. 5 from 1 to k (set 1, set 2, ..., set k).

Figure 5: Hazardous Scenario Identiﬁcation - Concept 2

(Segment 1).

Afterward, the sets of hazardous events are com-

bined with the HAZOP keywords to describe the ACs.

Several sets of ﬁnite ACs are shown in Fig. 6 that are

considered based on the hazardous events with HA-

ZOP keywords. The accident database can be con-

sidered as a mixed dataset, containing both numerical

and textual values.

Figure 6: Hazardous Scenario Risk Assessment - Concept

2 (Segment 2).

An exemplary case: publicly available hazardous

events from real-world scenarios are used together

with simulated scenarios and considered as the base

for ML models, which can be used to classify haz-

ardous events. Each of these classiﬁcation blocks can

be associated with HAZOP keywords to create ACs.

An AC can be expressed as follows: Accident on the

highway, as vehicles equipped with ADS show an un-

intended behavior and injure a human. The ML model

should provide the parameters (S, E, C), and later

ASIL can be determined based on the deterministic

approach.

4 CONCLUSION

The work showcases the conceptual approaches how

ML models can be applied for the scenario-based

HARA with respect to FuSa and SOTIF aspects. Two

conceptual approaches are presented which gives the

fundamental of incorporating ML in safety analyses

by means of HARA. This paper presents the basics

of HARA through the interplay of ML and determin-

istic approaches. This integration provides valuable

insights for safety mechanisms in hazardous scenar-

ios. The proposed approach not only improves the

depth of risk assessment, but also enables real-time

adjustments and scenario updates as new data be-

comes available, making the risk management pro-

cess more dynamic and responsive. Ultimately, the

use of AI in scenario-based HARA supports a more

proactive security strategy that responds to emerging

threats in complex systems such as autonomous vehi-

cles and other AI-driven applications.

Conceptual Approaches to Identify the Hazardous Scenarios in Safety Analysis for Automated Driving Systems

553

In future, simulation and publicly available real-

world scenario database will be considered for safety

analysis. To perform veriﬁcation of the scenario-

based HARA for ADS-equipped vehicle, safe human

intervention needs to be ensured. The safety pro-

cesses should include all possible safety-critical hu-

man interventions and provide safety measures to re-

duce the risk. To support safety analysis, ML tech-

niques can be applied in HARA with acceptable ver-

iﬁcation and validation processes that include new

test strategies and methods. To validate the scenario-

based HARA for ADS-equipped vehicles, it is essen-

tial to incorporate a human safety intervention pro-

cess. This process is aimed at evaluating the HARA

results in view of the ML applications to maintain the

integrity of the assessment.

ACKNOWLEDGEMENTS

This research work is funded by the Institute for

Driver Assistance and Connected Mobility (IFM),

Junkersstraße 1A, 87734 Benningen, Germany.

REFERENCES

Chen, L., J. J. and Zhao, T. (2020). A novel hazard analysis

and risk assessment approach for road vehicle func-

tional safety through integrating stpa with fmea. Ap-

plied Sciences 10, no. 21: 7400.

Deichmann, J., E. E. H.-K. H. R.-K. M. and Steiner, F.

(2023). Autonomous driving’s future: Convenient and

connected. McKinsey Center for Future Mobility,

McKinsey & Company.

Faysal, J. A. (2022). Formal rule-based scenarios for the de-

sign of safe autonomous vehicles. modeling and sim-

ulation. HAL open Science, Universi

e C

ote d’Azur.

Feldmann, A. (2023). The industry adoption of innovative

safety-analysis approaches for autonomous vehicles.

KTH Industrial Engineering and Management, master

thesis edition.

Gang, S.-M., C. H.-W. K. D.-R. C. Y.-J. (2016). A study on

the construction of the unity 3d engine based on the

webgis system for the hydrological and water hazard

information display. Procedia Engineering, vol 154,

PP. 138–145.

Grigorescu, S., T. B. C.-T. and Macesanu, G. (2020). A

Survey of Deep Learning Techniques for Autonomous

Driving. Elektrobit Automotive and the Robotics

Vision and Control Laboratory (ROVIS Lab), De-

partment of Automation and Information Technology,

Transilvania University of Brasov, Romania.

Horel, J.-B., L. C.-M. L. M.-R. M. L.-e. a. (2022). Using

formal conformance testing to generate scenarios for

autonomous vehicles. In Automation and Test in Eu-

rope - Autonomous Systems Design.

IPG-Automotive. Carmaker: Software tool. https://ipg-

automotive.com/de/anwendungen/autonomes-

fahren/.

ISO21448 (2022). ISO 21448: Road vehicles - Safety of the

intended functionality. ISO Standard, 1 edition.

ISO26262 (2018). ISO 26262: Road vehicles - Functional

Safety: Part 1 to Part 13. ISO Standard, 2 edition.

ISO34502 (2022). ISO 34502: Road vehicles — Test

scenarios for automated driving systems — Scenario

based safety evaluation framework. ISO Standard, 1

edition.

ISO/PAS8800 (2024). ISO PAS 8800: Road vehicles -

Safety and artiﬁcial intelligence. ISO Standard/Under

development, 1 edition.

Khan, J. and Vijay, C. (2023). 27th international technical

conference on the enhanced safety of vehicles (esv),

yokohama, japan. In alidation of safety of the intended

functionality for autonomous driving systems.

Khatun, M., G. M. and Jung, R. (2020). Scenario-based

extended hara incorporating functional safety & sotif

for autonomous driving. In 30th European Safety and

Reliability Conference.

Khatun, M., J. R. and Glaß, M. (2023). Scenario-based col-

lision detection using machine learning for highly au-

tomated driving systems. In In Journal Systems Sci-

ence & Control Engineering, vol. 11, no. 1. Taylor &

Francis.

Kon

e, T. F., B. E.-L. E. M.-F. and G

eronimi, S. (2023).

An approach to guide the search for potentially haz-

ardous scenarios for autonomous vehicle safety vali-

dation. Applied Sciences, vol. 13, no. 11, p. 6717.

Kramer, B., Neurohr, C., B

uker, M., B

ode, E., Fr

anzle, M.,

and Damm, W. (2020). Identiﬁcation and quantiﬁca-

tion of hazardous scenarios for automated driving. In

Model-Based Safety and Assessment.

Li, X., T. S. L.-B. D. X.-N. X. W.-F.-Y. (2023). Ieee trans-

actions on intelligent vehicles, vol. 8, no. 5. In Let-

ters: Advanced Scenario Generation for Calibration

and Veriﬁcation of Autonomous Vehicles.

Madala, K. and Solmaz, M. (2023). Scenario-based risk

quantiﬁcation approach for assuring safety in au-

tonomous vehicles. SAE Technical Paper 2023-01-

0584.

Matlab. Matlab: Software tool, automatic sce-

nario generation. https://www.mathworks.com/help/

driving/ug/automatic-scenario-generation.html.

Oakes, B. J., M. M.-M. S. V.-V. H. D.-J. (2021). Machine

learning-based fault injection for hazard analysis and

risk assessment. In Computer Safety, Reliability, and

Security. Springer International Publishing.

SAE (2021). SAE J3016: Taxonomy and Deﬁnitions for

Terms Related to Driving Automation Systems for On-

Road Motor Vehicles. SAE Standards.

Sana, F., A. N. L. and Raahemifar, K. (2023). Autonomous

vehicle decision-making and control in complex and

unconventional scenarios—a review. Machines, vol.

11, no. 7, p. 676.

Schwalb, E. (2021). Analysis of hazards for autonomous

driving. ASME Journal of Autonomous Vehicles Sys-

tems,vol.1, no. 2: 021003.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

554

TR5469, I. (2024). Iso/iec tr 5469: Artiﬁcial intelligence

— functional safety and ai systems. Technical report,

ISO/IEC.

Tu, H., W. M. L.-H. and Sun, L. (2023). Safety risk assess-

ment for autonomous vehicle road testing. In Trafﬁc

Injury Prevention. Taylor & Francis.

Yang, Y., Kujanp

a, K., Babadi, I. A., Pajarinen, J., and

Ilin, A. (2023). Suicidal pedestrian: Generation of

safety-critical scenarios for autonomous vehicles. In

2023 IEEE 26th International Conference on Intel-

ligent Transportation Systems (ITSC), pages 1983–

1988.

Zhang, X., T. J. T.-K. T. M.-J. S. M. G.-R. M. T. X. G. M. W.

F. F. M. N. N. M. and Felbinger, H. (2023). Ieee trans.

softw. eng., vol. 46, no. 3, ieee press, pp. 991–1026.

In Finding Critical Scenarios for Automated Driving

Systems: A Systematic Mapping Study.

Conceptual Approaches to Identify the Hazardous Scenarios in Safety Analysis for Automated Driving Systems

555