Knowledge-Based Approach to Generate Scenarios for Testing

Highly Automated On-Sight Train Operations

Lucas Greiner-Fuchs

1,2

and Martin Cichon

Institute of Vehicle Technology, Nuremberg Institute of Technology, Keßlerplatz 12, Nuremberg, Germany

Institute of Vehicle System Technology, Karlsruhe Institute of Technology, Rintheimer Querallee 2, Karlsruhe, Germany

Keywords: Scenario-Based Testing, Scenario Generation, Automatic Train Operation, Automated Driving System,

Operational Design Domain.

Abstract: Scenario-based test methods are cumulatively used for developing and testing highly automated railway

vehicles, similar to the automotive industry. However, due to significant differences between the two

technologies, existing approaches in the automotive sector cannot be directly applied to railways. Therefore,

it is necessary to develop revised and new processes and methods that are tailored to the rail sector.

The primary step in scenario-based testing is to set up appropriate test scenarios. A significant challenge faced

by the rail industry is the limited availability of measured data from actual railway operations. For this reason,

knowledge-based data sources need to be primarily used and considered in the scenario generation process.

This paper presents a basic approach to define sufficient quantity of test scenarios for highly automated

railway vehicles, using as an example a sensor-supported system for on-sight train operation. The approach

uses the system definition of the automated system as input, includes the operational design domain, and

considers railway-specific data through formalities and knowledge sources. Scenarios are then systematically

derived in three steps: description, combination, and derivation. In the end, a set of testable scenarios is

generated that can be used for virtual and real field testing of automated train operations.

1 INTRODUCTION

The rail system is a high priority in the current focus

of politics and society. The system offers low

greenhouse gas emissions and high energy efficiency

through the use of electromobility, as well as high

levels of traffic safety and capacity with low land

consumption for passenger and freight transport.

Despite the social and global advantages, the

proportion of rail passenger transport performance

has stagnated in recent years, and rail freight transport

has even decreased in Europe. This is primarily

caused by the low economic performance of the rail

system. However, Europe is expected to experience a

significant increase in the rail sector. To manage this

growth, rail transport must increasingly focus on

digitalisation and automation. (Zintel et al., 2023).

The development of highly automated driving

systems (HADS), has a high priority in current

research and industry. In addition to, for example

increasing driving comfort in the automotive sector

and compensating for the lack of personnel in the

railway sector, automation primarily improves the

capacity of transport systems with at least same levels

of safety and reliability. With the establishment of the

SAE J3016 standard (SAE International, 2021) of the

automotive industry in 2014 and the IEC 62290-1

standard (IEC, 2006) from 2006 as the Grade of

Automation (GoA) level in the railway industry, both

have created a detailed description and categorisation

of HADS in terms of their degree of automation.

In addition to the focused development of HADS,

extensive testing of these systems is necessary.

Alternative test approaches, such as scenario-based

testing, are increasingly coming into focus. For

automotive, extensive research projects have already

been launched, such as the Pegasus project family

(Pegasus, 2020; VVM, 2022). Despite early high

levels of automation, such as the first GoA4-System

on the Port Island line in Kobe, Japan in 1981 (Powell

et al., 2016), there are still few established research

approaches or test methods and strategies for

demonstrating the safe operation of Automatic Train

Operation (ATO) Systems, especially for not

restricted and intersection-free rail traffic. The

similarities between the definitions of automation

394

Greiner-Fuchs, L. and Cichon, M.

Knowledge-Based Approach to Generate Scenarios for Testing Highly Automated On-Sight Train Operations.

DOI: 10.5220/0012693500003702

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

In Proceedings of the 10th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2024), pages 394-401

ISBN: 978-989-758-703-0; ISSN: 2184-495X

levels suggest comparable testing approaches. Also,

the grade of complexity of ATO-Systems is

comparable to that of automated road transport. There

are parallels, particularly in the area of on-sight train

operation, where similar sensor systems are used. The

focus is on monitoring the environment and keeping

the track clear. Established approaches and tools of

the automotive industry can serve as inspiration, but

different Use Cases (UC) and system boundaries

require a review and revision as well as a derivation

of new methods for the test process. When it comes

to rail-bounded driving, it is important to consider

additional knowledge sources and processes due to

varying regulations, physical conditions, and UCs.

During test development, a strong focus is placed

on the definition of appropriate scenarios. In this

context, there are scenarios in rail traffic that are very

similar to those in road traffic, such as situations

involving signal detection, passing groups of people

(station entrance, city traffic) or simple speed

changes. However, typical road traffic manoeuvres

and situations such as overtaking, following or mixed

traffic (cars, cyclists, pedestrians, etc.) are only used

to a limited extent in the rail sector or are not possible

due to the system. In contrast, there is a greater focus

on data and radio transmission scenarios or on

monitoring the clearance gauge. Furthermore, the use

of a separate traffic network, interfaces to other traffic

participants (gated and ungated level crossings, entry

and exit of passengers at regularly scheduled stops)

as well as regulations in train operation (automatic

train control, block signalling, train radio) have to be

considered. Rail operations are limited due to the use

of rail-bound vehicles and long braking distances

caused by high masses and low coefficients of friction

in wheel-rail contact. Equipping test vehicles and

demonstrators and carrying out real field tests and

measurements involves greater effort and cost in the

rail sector. Rail vehicles can only be driven by trained

personnel, and the availability of vehicles and track

sections for system testing and data recording is also

a major challenge. This has an impact on the paucity

of real-world measurement data. Vehicle and

environmental data from real drives and situations are

a useful source for creating appropriate scenarios. In

the automotive industry there are many data sets

available, some of which are publicly accessible, due

to extensive endurance campaigns (Guo et al., 2018).

This input is not available in the rail industry, which

is why it is not possible to derive test scenarios

directly from measurement data.

Considering all the reasons mentioned above, this

paper shows the current state of research on a

methodological approach for the knowledge-based

generation of sufficient scenarios for testing ATO. The

focus here is on the aspect of perception in on-sight

train operation, in which the driver has full

responsibility for track monitoring in the non-

automated case. The objective of this paper is to

demonstrate a method comparable to the state of the art

in the automotive industry and to advance the

development and testing of HADS for rail vehicles. At

the beginning some basic information about scenario-

based testing are mentioned. Afterwards the generation

process for relevant railway scenarios is introduced.

2 SCENARIO-BASED TESTING

Due to reasons of economy, clarity, time efficiency,

and organization, conventional test approaches, such

as distance-based testing, are no longer practical.

Instead of executing millions of test kilometres in the

real world, potential eventualities in the application

field of the HADS are described as completely as

possible through different scenarios. These are tested

in a coordinated process first in a virtual environment

and additional partially in the field. (Schuldt, 2017).

The sections below provide basic information on the

term of a scenario and scenario generation methods.

2.1 Scenario Characterization

To gain a better understanding of a scenario, the terms

scene, scenery and situation are described. A scene is

a snapshot of the environment, including the scenery,

dynamic elements, actors, and watchers. The scenery

contains the stationary elements that make up the

fundamental environment of the scene. Finally, the

situation describes the functional information through

the scene, such as interactions and states of different

movable elements, as well as missions or tasks.

Starting with an initial scene, changes described by

the situation and ending with a final scene, a scenario

is a temporal development of scene elements.

Different scenes are combined by actions and events.

(Ulbrich et al., 2015).

The description of a scenario can be presented at

different levels of abstraction, depending on the

specific UC. In the Pegasus project, three scenario

levels were defined based on the work of (Bagschik

et al., 2017). The functional scenario provides the

most abstract description, where the situation is

specified in a semantic form. The logical scenario

then concretises the semantic variables using

parameter spaces. Finally, the concrete scenario is

formed as an instance of the logical scenario using a

variation of these parameters. All variables have

Knowledge-Based Approach to Generate Scenarios for Testing Highly Automated On-Sight Train Operations

395

detailed parameters, and the scenario is fully

described. (Bagschik et al., 2017; Pegasus, 2020).

2.2 Scenario Generation Methods

There are several cross-domain approaches to derive

scenarios. This paper focuses on methods of the

automotive sector because of its thematic proximity

to railways. The survey conducted by (Riedmaier et

al., 2020) provides a general overview of scenario

generation and presents various approaches to derive

scenarios. In particular, data-based and knowledge-

based approaches have proven to be effective for

scenario generation.

To create a set of scenarios from recorded field

data, a sufficient amount of real driving data must be

available. Recording this data requires extensive

endurance projects and data management. The use of

real driving data ensures that all derived scenarios are

realistic and applicable. However, it is important to

note that the scenario set only covers what the data set

includes, and critical situations may be missing.

In the knowledge-based approach, having a solid

foundation of knowledge data is essential. The

information collected from various sources must be

linked appropriately to generate scenarios. One

common method is the use of an ontology (Bagschik et

al., 2018). Other approaches include equivalent or

specially developed combination languages (Fremont

et al., 2018). At (Menzel et al., 2018), a detailed

example of the knowledge-based generation process is

mentioned based on (Bagschik et al., 2018). Semantic

scenarios are generated using an ontology, specified

with the 6-Layer Model presented in (Scholtes et al.,

2021) and then converted into concrete test scenarios

for use in OpenSCENARIO (ASAM e. V., 2024) by

parameterizing the semantic variables. To utilize the

distinct advantages of data-based and knowledge-

based scenario generation, (Hao et al., 2023) propose

an approach that combines both methods.

Based on current research and definitions, a

railway-specific approach for scenario-based testing

is introduced, taking into account the requirements of

the railway system and the challenges of data

generation. The following chapter presents the

developed method and its individual components.

3 GENERATION OF RELEVANT

RAILWAY SCENARIOS

ATO can be used for different fields of application. A

methodical approach for the development of a tool

chain for scenario-based testing of ATO-Systems was

presented at (Greiner-Fuchs et al., 2022). Based on

this, our paper outlines a method for generating

appropriate scenario sets. Figure 1 shows the

proceeding of the Scenario Generation. The following

subsections present the current state of our research

on the knowledge-based scenario generation process.

First, based on the ATO-System, the associated

Operational Design Domain (ODD) is discussed.

This is followed by the process of creating a

knowledge-based dataset of the necessary scenario

elements. The Scenario Description is then

introduced as the basis for Scenario Combination.

During this part, the combinatorics and actual

generation are discussed, from which the scenario set

is finally created. Lastly, the derivation of concrete

test scenarios is described.

Figure 1: Proceeding of knowledge-based Scenario

Generation for ATO-Systems.

The structure should ensure a high degree of

automation and be used for different kind of ATO-

Systems. The challenge is to fully integrate and consi-

der all necessary knowledge sources while ensuring

consistency throughout the scenario generation

process. In our research, the focus is on the evaluation

of automated on-sight train operation that work in

combination with sensor units. In main line operation,

the use of visual track monitoring is limited due to the

long braking distances. Nevertheless, there are still

situations that require sensor-based environment

detection in driverless operation. These include, for

example, shunting movements for train preparation,

monitoring train entry and exit at platforms, or critical

situations where safe operation cannot be guaranteed

by train control. In addition to main line operations,

other important applications include operating in

shunting yards, industrial and port facilities, and the

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

396

operation of trams or other slow-moving rail vehicles.

For the application testing of the method, we focus on

automated shunting operations (ASO).

3.1 Operational Design Domain

The first step of the generation process, located in the

upper part of Figure 1, is to define the boundaries and

scope of the HADS. For this purpose, an ODD of the

ATO-System is formulated. Within the ODD, the area

of application of the HADS is determined, thus

defining the conditions under which the system can

be used. For automotive applications in this purpose,

the PAS 1883 standard (British Standards Institution

(BSI), 2020) defines the fields of scenery,

environmental conditions and dynamic elements. The

standard provides a taxonomy as a basis for setting up

the System-ODD.

For the formulation of a railway-specific ODD-

Systematic the PAS 1883 standard is used as input

and system definitions from current research on ATO

applications are analysed. Initial approaches to the

definition of an ODD in rail transport can be found,

e.g., in (Tonk et al., 2021), where a proposal for the

definition of an ODD for securing remote driving

trains based on the PAS 1883 standard was

introduced. Another approach by (Meng et al., 2021)

presents an ODD for high-speed ATO-Systems in the

context of an analysis of derived scenarios.

In the test development of the ASO-System, we

created a first draft of an ODD, which is shown in

Table 1. The ODD is based on the PAS 1883

standard, with specific adaptations and extensions for

the field of ASO. In relation to the functional

requirements of the system, implicitly the

specifications, as well as the operational area as

shunting yard in Germany, the individual aspects of

this ODD are determined. The three categories

scenery, environmental conditions and dynamic

elements are defined to the extent necessary to

represent the scope and operating conditions of the

ASO-System. Amongst the surroundings, conditions

and occurring elements under which the system is to

function

faultlessly, information is provided on data

Table 1: ODD for Automated Shunting Operations.

Knowledge-Based Approach to Generate Scenarios for Testing Highly Automated On-Sight Train Operations

397

communication, details on the missions and

movements of the ego-vehicle and other dynamic

elements. Besides the functional description of the

ASO-System, it is also useful to derive its specific

UC. These are complementary to the definition of the

System-ODD and also serve to derive the test

scenarios in an efficient manner. (Hofmeier et al.,

2022) describes how UCs for the example of ASO can

be derived.

Further research will define a railway-specific

ODD-Systematic with the assistance of the ODD for

ASO. This taxonomy must be able to map ODDs

from all areas of rail transport and represent the limits

and scope of a specific ATO-System.

3.2 Railway-Specific Dataset

The System-ODD in combination with railway-

specific knowledge sources are used to create a

dataset that forms the foundation for the Scenario

Description. The dataset must include all entities

required for testing in various scenarios. This

encompasses a comprehensive list of necessary static

and dynamic objects associated with the defined

ODD, as well as descriptions of the track topologies

to be considered. The dataset thus serves to specify

the aspects defined in the System-ODD and provides

the foundation for Scenario Generation. The structure

and process for filling the dataset can be used across

different ODDs, but a separate or customized dataset

must be created for each specific ODD.

To achieve comprehensive ODD-specific test

case coverage, it is essential to aim for a high degree

of completeness when setting up the dataset. This task

requires the use of multiple data sources to populate

the dataset. For our example of ASO at a German

shunting yard, this analysis will consider operational

and situational conditions through formalities and

guidelines such as the train service instruction (DB

Netz AG, 2021), signal regulation (DB Netz AG,

2020), and railway construction and operating

regulations (Eisenbahn Bau- und Betriebsordnung:

EBO, 1967), as well as existing operational datasets

and expert knowledge. Additionally, exceptional

cases and potential errors in regular operation will

also be taken into account. This will be achieved

through the use of statistics on accidents and

operational analyses. Examples of well-known

sources include the investigation reports of the

German Federal Bureau of Railway Accident

Investigation and its open dataset on hazardous

incidents in railway operations that have been finally

investigated (BEU, 2024). Own experiences from

system development and recorded data during our

research are also considered. Furthermore, small

existing public and self-collected measured datasets

are used as a supplement. However, the current

amount of data is insufficient for a complete

measurement-based derivation of scenarios.

Therefore, the data serves only as an additional source

of knowledge.

3.3 Scenario Description

To set up scenarios, a logical process is used based on

the collected objects in the dataset. Following the 6-

Layer Model (Scholtes et al., 2021), a railway-

specific Scenario Description was published by

(Greiner-Fuchs et al., 2023). In this description, a

model for rail transport is developed and extended

specifically for the example of ASO to the 7-Layer

Shunting Model. The model provides the

fundamental framework for an organized scenario

layout. The entities from the railway-specific dataset

are assigned to seven distinct layers. The sorting

process is performed simultaneously with the dataset

creation, resulting in a sorted database that can be

directly applied to the method.

Starting with Layer 1 “Railway system and

signals”, the scenario's scenery is first defined. Layer

2 “Stationary objects” defines all non-moving objects

that occur in the environment of the situation. Layer

3 “Temporary changes of Layers 1, 2 (& 4)” contains

temporary deviations from the previous layers, e.g.

due to a tree falling onto the track or also derailed

railway vehicles. Last mentioned are usually assigned

to Layer 4 “Dynamic objects”, which contains all

movable elements. Weather, light and soil conditions

can be found in Layer 5 “Environmental conditions”.

Layer 6 “Digital information” deals with digital data

such as localisation signals, digital maps or the status

of railway traffic lights. In addition, Layer 7

“Shunting order” has been added for the ASO-System

as a higher-level information and status layer. The

shunting order describe the work tasks and the

associated movement of the locomotive, including

data on the driving route, travel speed in various track

sections and track release notification. In addition,

indirectly safely occurring objects, such as wagons or

shunting staff, can be derived for scenario definition.

When using the method, it is possible to create a

wide variety of scenarios based on the sorted

database. Thereby, a scenario can always consist of a

defined number of objects per layer. The use of this

ordered approach is shown in the following Scenario

Combination.

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

398

3.4 Scenario Combination

The complete scenario set is created by following the

structuring and sorting of the scenario description

above. Starting from the lowest level of abstraction,

the scenarios are implemented in semantic form

according to (Bagschik et al., 2017). To achieve this,

a program flow combinatorics is developed, which is

shown in Figure 2 as a simplified flowchart. In the

following, the individual steps of combinatorics are

listed and the chosen order of the object combination

is explained.

Figure 2: Flowchart - Combination Process.

The first step is the Scenario Initialization, where

basic information of the scenario is generated. The

objects in the database are also pre-filtered directly,

referenced to the specific UC. Unnecessary entities

are removed from the combinatorics to increase the

relevance of generated scenarios. Based on the UC,

additional information through Layer 7 is determined.

Safely occurring objects are added to the

combinatorics and defined limits, such as speed

specifications, are set. The starting conditions of the

ego vehicle are also specified. The inclusion of these

additional information enables a more targeted and

efficient scenario generation. The track topology and

scenery are created using the objects specified in

Layer 1. During the combination process, the objects

are linked in advance and saved as a “Map”. These

are generated to cover the spatial operation area of the

HADS. In addition, the restriction of certain objects

from appearing in the defined area limit the object

database. By combining the conditions of the UC,

Layer 7 definitions and maps, an initial set of

scenarios is created. This set is used as input for the

subsequent combination process.

During the part of adding Layer 5 objects, each

initial scenario is combined with all possible and

sensible environmental conditions, following specific

rules to ensure logical conditions. For instance, the

occurrence of snow is excluded in combination with

high temperatures. Each final scenario requires

precisely one environmental condition to guarantee

unambiguous definitions. This leads to the first final

set of semantic scenarios. The objects from the

remaining layers are added in ascending order, using

the most recent semantic scenario set as input. The

newly created scenarios are saved to the scenario set

and used as additional initial input for the next layer.

Rules for adding the objects are also established with

the help of the knowledge sources.

For Layers 2 – 4, the first step is to determine the

potential placement of the new object within the

scenario. This is achieved by developing a logic of

referencing. Starting from the initial position of the

ego vehicle, objects are positioned laterally and

longitudinally to the movement path. Simultaneously,

a check is carried out to identify specific object

combinations or irrelevant constellations. It is

important to avoid conflicts by ensuring that no

position in the scenario is filled twice. Certain objects

may have limitations or specific rules depending on

their intended use. For instance, there may be a

predetermined maximum number of an object

allowed in a given scenario, or restrictions on its

possible positions. In the case of Layer 3 objects, it is

important to use a reasonable combination to

determine the maximum number of temporary

changes that can occur simultaneously. When dealing

with Layer 4 objects, it is also important to consider

their movement. Dynamic objects may have a defined

trajectory and move within the scene. It is essential to

avoid unwanted collisions between objects and

ensure that the movement vectors and trajectory paths

are properly set up. Furthermore, the mobility of the

objects enables additional positioning in relation to

the object's orientation and pose. In the final

combination section, Layer 6 objects are included.

These objects are intangible and do not appear in the

scene, but describe states and changes in state of

existing objects or provide additional information for

the automated system.

Our research currently focuses on elaborating

specific combinatorics for ASO. Using the described

procedure and the predefined object database, a first

approach to completeness can be established.

However, it is necessary to set up the combination

rules in a well-founded and comprehensible manner

and to justify any limitations. The implementation of

combinatorics is dependent on the defined ODD and

knowledge base. These affect the limitations of the

database during the combination steps, as well as the

specific rules for adding objects. The basic structure

Knowledge-Based Approach to Generate Scenarios for Testing Highly Automated On-Sight Train Operations

399

of the combination process can be applied to various

ATO-Systems, but object-related specifications and

rules must be adapted or supplemented accordingly.

In the end the output is a final semantic scenario set

that serves as input for further Test Scenario

Derivation.

3.5 Test Scenario Derivation

In order to execute test cases, both in the field and in

a virtual environment, it is necessary to concretize the

abstract semantic scenarios. According to (Menzel et

al., 2018), therefore first logical and then concrete

scenarios are generated by parametrization. Based on

this reliable process, the derivation of test scenarios is

described in the final step of the Scenario Generation.

To implement parameterization, each semantic

variable must be assigned a defined parameter space

that outlines different states of the description. For

example, this enables more detailed specification of

object properties such as size, speed, and positioning.

By defining an increment within the parameter

spaces, specific values can be assigned to each

variable, allowing for the generation of concrete

scenarios for each logical scenario. Finding the

appropriate step size is crucial as it greatly impacts

the number of concrete scenarios. It is important to

strike a balance between generating useful diverse

scenarios and avoiding situations that are only

slightly diffuse. When establishing a meaningful and

logical definition and differentiation in the parameter

space, it is important to consider the influence of

knowledge sources.

At present, we have analysed the derivation of

concrete test scenarios in theoretical form. We are

developing a demonstrative combinatorics based on

our example of ASO. Based on a first proven

semantic scenario set, concrete scenarios will be

derived and tested in a railway-specific virtual

environment (Schäfer et al., 2023).

4 CONCLUSION

The testing of HADS in rail transport is still in its

infancy. New and customised methods need to be

developed and proven in practice. This paper presents

a general process and structure of knowledge-based

railway scenario generation. A proceeding for

defining and deriving scenarios in a systematic, step-

by-step manner has been introduced. However,

further research and work are required to elaborate on

the individual sub-steps, in order to verify the

methodological approach using the example of ASO.

The level of detail in the ODD and the dataset derived

from knowledge should be regularly reviewed. The

combinatoric requires a formal definition of how

individual objects are linked in a given scenario.

Specific rules are established based on the knowledge

sources to manage this data linkage. For this purpose,

a semantic description language is used to define the

dependence of the objects. The implementation of the

semantic language for the combination needs to be

examined more closely. It is crucial to determine

whether an ontology or another semantic conversion

would be more efficient. Moreover, the combination

process must consider object placement and object

variation in more detail. It is important to clarify how

much variability in object positioning and appearance

is required to cause a significant difference in the

scenario. This directly affects the number of scenarios

generated and the performance of the combinatorics.

Once the process has been fully developed, it is

essential to demonstrate its applicability for scenario-

based testing. For the ASO example, it is necessary to

create a comprehensive scenario database, test it, and

evaluate it in a virtual test environment. The results

will be used to verify the meaningfulness and

representativeness of the generated scenarios, as well

as their direct transferability to the field test.

Currently, the method offers a theoretical approach to

generate specific scenarios in the railway sector.

However, it is crucial to conduct thorough virtual

testing and practical evaluation to confirm its

effectiveness.

ACKNOWLEDGEMENTS

This work was partly accomplished within the project

VAL, FKZ 5320000013, EBA Az. 8fd/003-

1255#008-VAL, funded by the German Federal

Ministry of Digital and Transport.

REFERENCES

ASAM e. V. (2024). ASAM OpenSCENARIO® DSL.

https://www.asam.net/

Bagschik, G., Menzel, T., & Maurer, M. (2018). Ontology

based Scene Creation for the Development of

Automated Vehicles. https://arxiv.org/pdf/1704.0100

6.pdf

Bagschik, G., Menzel, T., Reschka, A., & Maurer, M.

(2017). Szenarien für Entwicklung, Absicherung und

Test von automatisierten Fahrzeugen. https://www.pe

gasusprojekt.de/de/lectures-publications

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

400

BEU. (2024). Bundesstelle für Eisenbahnfallunter-suchung.

https://www.eisenbahn-unfalluntersuchung.de/EUB/DE

/Publikationen/publikationen_node.html

British Standards Institution (BSI). (2020). Operational

design domain (ODD) taxonomy for an automated

driving system (ADS). Specification. BSI Standards

Limited. https://www.bsigroup.com/en-GB/

DB Netz AG. (2020). Richtlinien 301 – Signalbuch: Ril301.

DB Kommunikationstechnik GmbH. https://www.dbin

frago.com/web

DB Netz AG. (2021). Richtlinie 408—Züge fahren und

Rangieren: Ril408. DB Kommunikationstechnik

GmbH. https://www.dbinfrago.com/web

Eisenbahn Bau- und Betriebsordnung: EBO, Bundes-

gesetzblatt II (1967). https://www.gesetze-im-

internet.de/ebo/EBO.pdf

Fremont, D. J., Yue, X., Dreossi, T., Ghosh, S.,

Sangiovanni-Vincentelli, A., & Seshia, S. A. (2018).

Scenic: Language-Based Scene Generation. Electrical

Engineering and Computer Sciences University of

California at Berkeley. https://www2.eecs.berkeley.edu/

Greiner-Fuchs, L., Schäfer, S., Hofmeier, T., & Cichon, M.

(2023). 7-Layer Shunting Model: Generische

Szenariobeschreibung automatisierter Rangierfunk-

tionen. International Rail Symposium Aachen, 4. Band.

https://doi.org/10.18154/RWTH-2024-00257

Greiner-Fuchs, L., Schäfer, S., Hofmeier, T., & Cichon, M.

(2022). Database-supported methodical approach for

the development of a toolchain for the evaluation of

ATO functions using a scenario-based test

methodology. Proceedings of the Fifth International

Conference on Railway Technology: Research,

Development and Maintenance. https://doi.org/10.42

03/ccc.1.13.4

Guo, J., Kurup, U., & Shah, M. (2018). Is it Safe to Drive?

An Overview of Factors, Challenges, and Datasets for

Driveability Assessment in Autonomous Driving.

http://arxiv.org/abs/2311.10937

Hao, K., Liu, L., Cui, W., Zhang, J., Yan, S., Pan, Y., &

Yang, Z. (2023). Bridging Data-Driven and

Knowledge-Driven Approaches for Safety-Critical

Scenario Generation in Automated Vehicle Validation.

http://arxiv.org/abs/2311.10937

Hofmeier, T., Greiner-Fuchs, L., Schäfer, S., & Cichon, M.

(2022). Task analysis of a shunting locomotive to derive

use-cases for scenario based tests of ATO Functions.

9th Auto Test Conference, Stuttgart. https://www.re

searchgate.net/

IEC. (2006). IEC 62290-1:2006 Railway applications –

urban guided transport management and

command/control systems – part 1: System principles

and fundamental concepts (IEC 62290–1).

Meng, Z., Tang, T., Wei, G., & Yuan, L. (2021). Analysis

of ATO System Operation Scenarios Based on

UPPAAL and the Operational Design Domain.

Electronics, 10(4). https://doi.org/10.3390/electronics1

0040503

Menzel, T., Bagschik, G., Isensee, L., Schomburg, A., &

Maurer, M. (2018). Detaillierung einer stich-

wortbasierten Szenariobeschreibung für die Durch-

führung in der Simulation am Beispiel von Szenarien

auf deutschen Autobahnen. Workshop Fahrer-

assistenzsysteme und automatisiertes Fahren, 12.

https://www.uni-das.de/

Pegasus. (2020). Projekt zur Etablierung von generell

akzeptierten Gütekriterien, Werkzeugen und Methoden

sowie Szenarien und Situationen zur Freigabe

hochautomatisierter Fahrfunktionen. https://www.pega

susprojekt.de

Powell, J. P., Fraszczyk, A., Cheong, C. N., & Yeung, H.

K. (2016). Potential Benefits and Obstacles of

Implementing Driverless Train Operation on the Tyne

and Wear Metro: A Simulation Exercise. Urban Rail

Transit, 2(3–4). https://doi.org/10.1007/s40864-016-

0046-9

Riedmaier, S., Ponn, T., Ludwig, D., Schick, B., &

Diermeyer, F. (2020). Survey on Scenario-Based Safety

Assessment of Automated Vehicles. IEEE Access, 8.

https://doi.org/10.1109/ACCESS.2020.2993730

SAE International. (2021). SAE J3016. 2021 SAE

International/ISO. https://www.sae.org/content/12-03-

01-0003/

Schäfer, S., Greiner-Fuchs, L., Hofmeier, T., Koch, P., &

Cichon, M. (2023, April). Virtual Validation Method of

Automated On-Sight Driving Systems for Shunting

Operations. BOOK OF ABSTRACTS. 10th International

Conference on Railway Operations Modelling and

Analysis (ICROMA), Belgrade, Serbia. https://doi.org/

10.37528/FTTE/9788673954677/RailBelgrade.2023.ZE

Scholtes, M., Westhofen, L., Turner, L. R., Lotto, K.,

Schuldes, M., Weber, H., Wagener, N., Neurohr, C.,

Bollmann, M. H., Kortke, F., Hiller, J., Hoss, M., Bock,

J., & Eckstein, L. (2021). 6-Layer Model for a

Structured Description and Categorization of Urban

Traffic and Environment. IEEE Access, 9.

https://doi.org/10.1109/ACCESS.2021.3072739

Schuldt, F. (2017). Ein Beitrag für den methodischen Test

von automatisieren Fahrfunktionen mit Hilfe von

virtuellen Umgebungen. http://publikationsserver.tu-

braunschweig.de/

Tonk, A., Boussif, A., Beugin, J., & Collart-Dutilleul, S.

(2021). Towards a Specified Operational Design

Domain for a Safe Remote Driving of Trains.

Proceedings of the 31st European Safety and Reliability

Conference (ESREL 2021). https://doi.org/10.3850/97

8-981-18-2016-8 504-cd

Ulbrich, S., Menzel, T., Reschka, A., Schuldt, F., & Maurer,

M. (2015, September). Defining and Substantiating the

Terms Scene, Situation, and Scenario for Automated

Driving. 2015 IEEE 18th International Conference on

Intelligent Transportation Systems. https://doi.org/

10.1109/ITSC.2015.164

VVM. (2022, September 28). Verification Validation

Methods. VVM Projekt. https://www.vvm-projekt.de/

Zintel, M., Pluchet, J., Hensler, A., Bamberger, V., Baron,

R., & Watson, S. (2023). Rail 2040. Arthur D. Little.

https://www.adlittle.com/sa-en/insights/report/rail-

2040

Knowledge-Based Approach to Generate Scenarios for Testing Highly Automated On-Sight Train Operations

401