Enhancing Scenario-based Testing for Automated Driving Systems:

An Ontology-Based Scenario Modeling Framework

Zhenguo Cui

1,2

, Svetlana Dicheva

, Adam Abdin

, Bernard Yannou

and Jean-Marc Giroux

ALTEN Labs, Sèvres, France

Laboratoire Génie Industriel, CentraleSupélec, Université Paris-Saclay, France

Keywords: Scenario-based Testing (SbT), Scenario Modeling, Ontology Engineering, Automated Driving Systems.

Abstract: Scenario-based Testing (SbT) emerges as a pivotal approach for validating the safe behaviors of Automated

Driving (AD) and Advanced Driver-Assistance Systems (ADAS). Using virtual simulation, SbT allows for

generating and running massive testing cases. This approach gathers typical driving situations and critical

edge cases. Properly modeling representative scenarios is a primary challenge. A scenario model needs to

account for complex components, such as roads, infrastructure, road users, and their behaviors and

interactions. Ontology-based frameworks are proposed to model scenarios in a detailed manner. However,

some limitations exist, such as (i) expressing dynamic behaviors, (ii) the capacity in complex scenario

modeling to achieve more realistic simulation; and (iii) ensuring comprehensive ontology coverage and

plausibility. This paper proposes an ontology framework addressing these shortcomings. A comparative

evaluation is conducted using the developed quantitative metrics to assess the ontology framework against

two other industrial ontologies.

1 INTRODUCTION

Automated Driving (AD) and Advanced Driver-

Assistance Systems (ADAS) present an important

evolution in automotive technologies, aiming to

enhance road safety and reduce human error. The

complexity of real-world driving environments

requires rigorous validation to ensure these

technologies are safe. Scenario-based Testing (SbT)

has the potential to accelerate safety validation. A

scenario describes a specific environment that an

AD/ADAS-equipped vehicle could encounter in the

real world.

Properly modeling representative driving

scenarios is a primary challenge for SbT in both real-

world and simulated environments. Diverse and

complex scenarios can assess the safety of AD/ADAS

in virtual simulation without risks and costs currently

associated with real-world testing. Building a

scenario model requires an in-depth knowledge and

understanding of traffic and environments.

Modeling scenarios using ontologies provides a

suitable framework for validation and testing of

Automated Driving (Armand et al., 2014). Ontology-

based modeling offers a flexible formalism for

managing complex knowledge, which serves as the

foundation for defining, generating, or identifying

scenarios.

However, current ontology-based scenario

modeling frameworks are facing challenges in

covering various elements, expressing dynamism of

driving behaviors, and effectively modeling complex

scenarios.

In the present paper, the challenges of existing

ontologies are presented in Section 2. In the

subsequent Section 3, a new ontology framework for

scenario modeling is proposed to capture the

complexity of real driving environments more

effectively. A preliminary ontology evaluation

method using a comparative approach is conducted in

Section 4, applied on the existing latest industrial

ontology frameworks. Section 5 concludes the

contribution of the presented framework.

2 RELATED WORKS AND

CHALLENGES

Current scenario modeling ontologies are built, from

a structural and organizational perspective, using

hierarchical layered models (Bagschik et al., 2018;

622

Cui, Z., Dicheva, S., Abdin, A., Yannou, B. and Giroux, J.-M.

Enhancing Scenario-based Testing for Automated Driving Systems: An Ontology-Based Scenario Modeling Framework.

DOI: 10.5220/0013438800003941

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

In Proceedings of the 11th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2025), pages 622-630

ISBN: 978-989-758-745-0; ISSN: 2184-495X

Scholtes et al., 2021; Schuldt et al., 2013), block-

based categorical structures (de Gelder et al., 2022;

Erz et al., 2022; Li et al., 2020), or their fusions

(Armand et al., 2014; ASAM, 2022; Chen & Kloul,

2018; Westhofen et al., 2022). Previous research

works have accounted for a wide range of knowledge

within the driving domain, and diverse elements have

been integrated into these ontologies. While existing

models have made notable progress, there are still

limitations leading to the need for new improvements

to overcome these challenges.

2.1 Expressing Dynamics in Scenario

Modeling

The symbolic nature of ontologies limits the

expressiveness of dynamism in driving scenarios. In

previous research, dynamic descriptions are

commonly defined and encapsulated in the layer or

module named Dynamic Objects, Dynamic Entities,

Dynamic Elements, or Traffic Participant (ASAM,

2022; Bagschik et al., 2018; Erz et al., 2022; Scholtes

et al., 2021; Schuldt et al., 2013). The layer

incorporates behavioral descriptions, including (1)

states: position, speed, acceleration; (2) actions:

maneuvers and triggering events; (3) intentions &

interactions: behaviors and activities.

These symbolic concepts lack the ability to

express the spatial-temporal evolution of objects.

Realistic dynamism relies on both temporal and

spatial scales. While typically defined with discrete

scenes, the relationships and dependencies between

scenes are underrepresented in ontologies, failing to

capture the continuous evolution of objects.

On the other hand, the spatial occupation of

objects is often modeled through lane occupancy and

relative positioning, such as when the EGO vehicle is

driving on a lane and another vehicle is in its left rear.

The gap between these high-level symbolic

representations and the fine-grained occupancy grid

(Elfes, 1989) highlights the need for an approach to

bridge discrete symbols and continuous spatial data.

2.2 Complex Scenario Modeling

Traffic participants who do not directly influence the

behavior of the EGO vehicle are often excluded from

actual scenario modeling. While they may not

directly impact the validation of AD/ADAS functions

(Geyer et al., 2014), their presence is necessary to

reflect realistic traffic environments.

In realistic driving environments, multiple traffic

participants are present and interact. A considerable

difficulty is to describe interactive driving behaviors

of multiple road users with a generalized modelling

framework.

The trigger-action mechanism is widely used to

formalize scenarios and is sufficient to model less

complex ones, such as those in the Car-to-Car Rear

(CCR) series in the NCAP protocol concerning

Automated Emergency Braking (AEB) function.

However, this trigger-action practice is not well-

suited for modeling scenarios involving multiple road

participants because of simultaneous interactions or

coordinated behaviors in real-world environments.

Additionally, validating an AD/ADAS function

requires a minimum test duration, such as 5 minutes

for validating the lane-keeping capacity as required

by the regulation (UNECE, 2021).

2.3 Coverage and Plausibility of

Modeled Scenarios

Existing works have not addressed the issue of

ontology coverage. The review (Zipfl et al., 2023)

proposed a categorical coverage measure, using a

checklist-based approach to determine whether an

ontology covers certain categories of elements, such

as lane marking properties of the road. While

informative, this comparative review is insufficient

for comprehensive coverage evaluation.

The problem is further related to the absence of a

baseline ontology for scenario modeling. Under the

Open-World Assumption, it is impossible to claim

that an ontology covers sufficient elements or is

complete when the baseline is missing.

Furthermore, the non-plausibility of scenario

modeling was not discussed in the literature. The

existing approaches bypass the necessary

relationships and constraints that ensure scenarios

remain valid. This can lead to modeling unrealistic

scenarios, such as inappropriate or incompatible

elements.

3 PROPOSED ONTOLOGY

FRAMEWORK

An enhanced ontology-based scenario modeling

framework is proposed and developed for AD/ADAS

validation. The framework articulates refined

ontology concepts in natural language and connects

them through relations and constraints, addressing

dependencies, mutual exclusions, and other factors

relevant to validating AD/ADAS functions.

Furthermore, this framework adapts spatial

definitions to describe traffic participants within

Enhancing Scenario-based Testing for Automated Driving Systems: An Ontology-Based Scenario Modeling Framework

623

scenarios. The framework accounts for the

complexity of real driving environments.

3.1 Refined Ontology Structure and

Modules

In this paper, the proposed framework organizes four

modules in ScenarioDomain to describe dynamic

scenarios: Scenery, Dynamic Elements,

Environmental Conditions, and Goals, as shown in

Fig. 1. Scenery and Environmental Conditions

provide static and environmental descriptions of

scenarios. The Dynamic Elements module refines

Maneuvers, Activities, and Behaviors for dynamic

descriptions. The Goals module outlines validation

objectives related to AD/ADAS requirements. The

ontology elements, including concepts, properties,

and axioms, are developed using Protégé Software

(Noy et al., 2003).

Considering their use in automatic scenario

generation, this framework organizes separately these

descriptive elements with the layered containers.

These containers serve as Components to form

scenarios. A set of defined relationships contributes

to associate elements, encapsulating concepts of

Domain into containers of Components. This

approach ensures the modularity and efficient

composition of elements. For example, a Scenario

occurs in a Zone, which consists of a RoadNetwork,

including one or multiple Roads depending on the

type of RoadNetwork, and so on. The four modules,

organized as Domain, specify these container

concepts. For example, each TrafficParticipant has its

type and a set of Behaviors which involves

Maneuvers, while the interactions of

TrafficParticipants are formed and represented as

Activities.

Figure 1: Concept hierarchy of the proposed ontology-

based framework.

Besides, constraints are integrated in this ontology

framework, to reduce non-plausible scenarios. For

example, a pedestrian crossing road marking cannot

exist on a highway, this mutual exclusion constraint

between elements ensures semantic integrity. And

other constraints improve the logical consistency and

reduce modeling error, such as compatibility

constraints, a speed limit of 110km/h is incompatible

with a roundabout or a crossroad intersection. These

restrictions ensure the validity of scenario modeling.

The actual framework defined more than 200

constraints to improve the modeling quality for

Scenario-based Testing.

The Maneuvers formalize driving actions, such as

Accelerate, TurnRight (heading to right), Stop; and

non-driving ones, like UseTurnIndicator, HonkHorn.

The Activities module includes road user interactions,

from the perspective of each road user.

An Activity is defined as a combination of

maneuvers of multiple road users. A scenario

contains one or multiple activities describing

scenarios, for example, a vehicle CloseUp to another,

then Overtake it.

The Behaviors enhance the dynamic description

of both AD/ADAS-equipped vehicles and other road

users within scenarios. Behaviors are associated with

a series of maneuvers, for instance, the CarFollowing

behavior involves a sequence of maneuvers

Accelerate, Decelerate, ConstantSpeed, and Stop.

This module incorporates the taxonomies from

(NHTSA, 2018), which also include MaintainSpeed,

LaneCentering, ObstacleAvoidance, among others.

For example, MaintainSpeed refers to maintaining a

safe speed set through longitudinal control with

acceptable following distances.

Testing Goals in the Goals module, this module

emphasizes the purpose of modeling traffic

participants, specifically aligning with behaviors of

AD/ADAS functions that should be achieved within

a scenario. They are aligned with the behavioral

competencies of AD/ADAS functions, using the

taxonomies from (AVSC, 2021). A well-defined set

of testing goals helps clearly identify the behaviors

that need to be assessed in scenarios. For example, for

an Adaptive Cruise Control (ACC) function,

RespondingToOtherVehicles is a primary testing

goal. To ensure safety, an ACC-equipped vehicle

must demonstrate behaviors such as MaintainSpeed,

CarFollowing, and ObstacleAvoidance within

scenarios.

3.2 Spatial Segmentation and Layout

A surrounding location layout is proposed to describe

objects' occupation and surrounding locations. The

layout illustrates lanes as a grid-like pattern, where

each cell corresponds to a specific area within the

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

624

road network, defined by the lateral and longitudinal

segmentation.

This layout divides a standard lane into three

parts, as shown as Fig. 2: the Left Adjacent Area (L),

the Central Area (C), and the Right Adjacent Area

(R). The widths of L-C-R are proportional to the total

lane width. For example, if a lane is 3.5 meters wide,

the segments measure 1m-1.5m-1m.

Figure 2: Illustration of the proposed lateral segmentation,

screenshot of a scenario running on the esmini simulator.

When the geometric center of a vehicle remains in

the Central Area of a lane, it is driving in that lane.

This proposed L-C-R lateral segmentation aims to

describe between-lanes behaviors, such as a vehicle

driving in the adjacent area of two neighboring lanes

or a lane change. As shown in Fig. 3 (a), the V1 drives

on the adjacent area between Lane 1 (L1) and L2,

more precisely on the Right Adjacent Area of Lane 1

(L1-R).

The longitudinal length of each layout cell

depends on the distance that the EGO vehicle drives

per N seconds. For instance, when the EGO vehicle is

driving 70 kilometers per hour, fixing N as 1 second,

each cell would be 20 meters long. This segmentation

improves the granularity of behavioral descriptions

This segmentation layout illustrates spatial

surroundings, and cells are numbered following the

same protocol in the standard ISO 34502 (ISO, 2022).

As shown in Fig. 3 (b), a standard three-lane road is

illustrated with respect to the proposed approach. The

central cell represents an AD/ADAS-equipped EGO

vehicle. Surrounding cells are numbered to indicate

their relative positions, facilitating the description of

spatial relations between the subject and other road

users.

This segmentation approach facilitates describing

the spatial occupation of road participants within a

scene. The detailed lane segmentation layout is better

suited for integration with behavioral concepts,

compared to occupancy grids (Elfes, 1989), which are

more aligned with concrete scenario presentations.

Figure 3: (a) Illustration of two vehicles driving within the

grid layout, with their geometric centers marked by red

dots; (b) Numbered spatial layout of a standard three-lane

road w.r.t. the grid numbering protocol of ISO 34502.

3.3 Temporal Sequence of the Spatial

Layouts

The sequence of spatial occupation representations

for traffic participants contributes to shaping a set of

scenarios. Each sequence represents the spatial

interaction of a vehicle from another vehicle’s

perspective and corresponds to an instance of traffic

activity defined within the Activity module of the

proposed ontology. Within a sequence, each spatial

occupation offers a brief snapshot of the relative

positioning and movement dynamics at a specific

moment in the scenario, i.e., the Scene (Ulbrich et al.,

2015). The following example illustrates the utility of

the temporal sequence in conjunction with the

segmentation layout.

As discussed in Section 3.1, ACC functions face

considerable difficulty in scenarios modeled to

evaluate their driving capability to

RespondingToOtherVehicles. The behaviors involve

maneuvers such as Accelerate, Decelerate,

ConstantSpeed, and Stop. A common and challenging

scenario is a cut-in, defined as an activity in the

framework, where another vehicle merges in front of

the ACC-equipped EGO vehicle.

The following Fig. 4 shows the sequence of

layouts to describe this scenario. Initially, V1 drives

parallel in the left lane of the EGO vehicle. V1 then

initiates a lane change, and merges into the lane of the

EGO vehicle. This sequence results from V1

performing the two combinations of Accelerate and

TurnRight, transiting from Fig. 4 (a) to (b), and then

to (c). This is followed by a single combination of

ConstantSpeed and TurnRight, moving from Fig. 4

Following the numbering protocol shown in Fig.

3 (b), the trajectory of V1 during this cut-in activity

follows the sequence of cases 45-21-13-9 in the grid.

Enhancing Scenario-based Testing for Automated Driving Systems: An Ontology-Based Scenario Modeling Framework

625

This sequence is one of possible sequences that

illustrate the cut-in activity. This representation of

behavioral dynamics bridges abstract scenario

descriptions with detailed scenario parameterization,

further aligning with real-world collected trajectory

data.

Figure 4: Sequence of grid layouts illustrating a vehicle

(V1) performs a cut-in activity, where V1 changes lanes and

merges into the same lane as the EGO vehicle (E).

Our ongoing work interests the needs of virtual

validation for existing AD/ADAS functions. We are

generating the occupation sequences to scale down

the scenario space, before exploring the infinite

situation-dependent concrete parameter generation.

The generated sequences are transformed into

filtering conditions, which are then applied to real

world databases for extracting corresponding

trajectories, such as the highD (Krajewski et al.,

2018) dataset. These sequences serve as an

intermediate layer, enabling the integration of

behavioral concepts of the ontology framework with

real-world data. By combining symbolic elements

with time series data, the approach generates the

necessary inputs needed to generate concrete

scenarios in XOSC format (ASAM, 2019).

On the other hand, the AD/ADAS requirements

are linked to behavioral descriptions through the

logical chaining of concepts. In the ACC example, the

Goal-Behavior-Maneuver chain narrows the scenario

space for maneuvers of the EGO vehicle. And

compatible interactions between EGO and non-EGO

vehicles, defined as Activities in the ontology

framework, are combined to complete the dynamic

scenario description.

3.4 Potentials in Complex Scenario

Modeling

The occupation sequences facilitate modelling

dynamics with a more generalized manner. Unlike the

current approach of defining trigger-action pairs in

scenario generation, occupation sequences are more

aligned with the real-world observation about driving

behaviors.

Each sequence represents a driving activity, and

the activity-based representations allow their

concatenation to model complex scenarios. The cut-

in activity, represented by the sequence of cases 45-

21-13-9 in Fig. 4, can be superimposed onto a single

layout representation, as shown in Fig. 5 (a). The

spatial-temporal evolution of V1 is depicted in a

single figure, with four red rectangles marking its

location and three arrows representing its maneuvers

in the scenario.

Figure 5: Illustration of overlaps of occupation layouts: (a)

V1 vehicle’s cut-in activity; (b) V2 vehicle’s close-up

activity and V3 vehicle’s move-away activity; (c) the fusion

of (a) and (b).

Combining activities can generate complex

scenarios. A vehicle V1 performs a cut-in activity due

to a slower leading vehicle V2 in Lane 1. This cut-in

activity is compatible with the V2’s close-up

sequence. Similarly, these activities can be combined

with vehicle V3’s move-away activity, as shown in

Fig. 5 (b). These three activities are logically

compatible, and their fusion allows the creation of a

complex scenario, illustrated in Fig. 5 (c).

Additionally, concatenating sequences also can

produce complex scenarios. For instance, the vehicle

V1 may execute a cut-in and then move away,

increasing its distance from the EGO vehicle. In this

case, the sequences of V1 and V3 could be

concatenated to define a more complex scenario.

4 PRELIMINARY ONTOLOGY

EVALUATION

A comparative approach is conducted to evaluate the

enhanced ontology framework with other existing

ones. Two scientific and industrial ontologies were

compared: the Automotive Urban Traffic Ontology

(A.U.T.O.) (Westhofen et al., 2022) and the ASAM

OpenXOntology (ASAM, 2022). Both ontologies are

accessible from Github and the ASAM website.

The A.U.T.O. ontology, a nested ontology which

implements the 6-layer model (Scholtes et al., 2021),

offers a series of ontology blocks to modularize

distinct domain elements. Inter-module connections

are established through the foundational ontologies,

such as GeoSPARQL for geometry and W3C

standards for temporal aspects.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

626

The ASAM OpenXOntology, within the ASAM

OpenX ecosystem, serves as a semantic foundation

for knowledge representation in the AD/ADAS

domain. It provides a comprehensive structure and

terminology and is compatible with Scenario-based

Testing tools. In contrast to A.U.T.O., this industrial

ontology proposes an upper-level Core block to

interconnect elements in different modules.

4.1 Initial Comparison

Three ontology frameworks, as illustrated in Fig. 6,

share a similar structure of concept organization.

Consequently, this structural review is insufficient for

comparison. It resembles the checklist-based

approach in (Zipfl et al., 2023), which limits the

comparison to the categorical or modular level.

Figure 6: Comparison on the structure of ontologies.

4.2 Quantitative Comparison

A statistical analysis of the proposed ontology was

conducted, along with the two ontologies. Following

an exploratory analysis, the three ontologies are

converted to graph-based representations to analyze

their concept hierarchy and structural organization.

Directed acyclic graphs were built to illustrate and

analyze the overlapping elements among ontologies.

This analysis used an automatic knowledge extraction

tool that we developed with the RDFLib library in

Python.

Four metrics are proposed and applied in these

ontologies: the Connectivity Index, the Property

Utility Ratio, the Redundancy Ratio, and the Branch

Balance.

Connectivity Index (CI) measures the density of

relationships and constraints per concept, calculated

as the total number of relationships and restrictions

divided by the number of concepts. A higher CI

indicates stronger interconnections among concepts.

The Property Utility Ratio (PUR) assesses

expressiveness by dividing the total number of

relationships and constraints by the sum of object and

data properties. A higher PUR implies a more densely

expressed ontology.

The Redundancy Ratio (RR) measures the

proportion of concepts having multiple parent

concepts within the directed acyclic graph. It helps to

identify the degree of overlap and assess unnecessary

structural complexity. The RR is calculated as 1

minus the proportion of unique concepts relative to

the total number of nodes in the ontology.

The Branch Balance (BB) reflects the distribution

of nodes in different branches. It is calculated with the

average entropy of all nodes in the graph, using the

proportion of each node’s subtree size relative to the

total number of nodes. A higher BB score indicates a

balanced structure, which enhances parsing and

searching within the ontology, facilitating its

automatic processing in applications. This metric may

identify imbalances within the ontology, where

branches are insufficiently developed or excessively

complex.

Table 1 presents the statistical analysis of the

ontology elements, the graph-based representations,

and the evaluation of key metrics.

Table 1: Statistics and graph-based analysis.

A.U.T.O.

OpenXOn

tology

Proposed

Ontology

Statistics

Nb. Concepts 284 346 240

Nb. Object

Pro

erties

118 96 37

Nb. Data

Properties

70 2 1

Nb. Individuals 110 348 286

Nb.

Relationships

2682 3030 2283

Nb.

Restrictions

43 24 256

Connectivity

Index

9.59 8.83 10.58

Property

Utilit

Ratio

14.49 31.16 66.82

Graph-based

Nb. Nodes 595 2272 240

Nb. Edges 582 2270 232

Nb. Leaf

Nodes

370 1645 191

Nb. Levels of

Nodes

11 13 7

Redundancy

Ratio

52.3% 84.8% 0%

Branch

Balance

0.86 1.46 1.76

The A.U.T.O. integrates more object and data

properties compared to the others. However, the

interconnections between concepts are limited, as

indicated by the CI and PUR metrics.

A.U.T.O. OpenXOntology Proposed Ontology

L1 Road Network and Traffic

Guidance Objects

L2 Roadside Structures

L3 Temporary Manipulation

of L1 and L2

L4 Dynamic Objects

Traffic Participant and

Behavior

Dynamic Elements

L5 Environment Conditions

L6 Digital Information

XXGoals

Road Topology and Traffic

Infrastructure

Scenery

Environmental Condition Environmental conditions

Enhancing Scenario-based Testing for Automated Driving Systems: An Ontology-Based Scenario Modeling Framework

627

OpenXOntology, while containing an impressive

number of nodes in its directed acyclic graph, suffers

from a high degree of concept overlap, as shown by

the RR metric, which reduces its practicality for

applications.

The proposed ontology framework introduces

fewer concepts and properties, but integrates a

substantial number of relations and restrictions,

particularly reflected in the CI and PUR metrics. Its

acyclic graph representation shows simplified and

well-balanced branches, as highlighted by the RR and

BB metrics, makes it suited for further applications,

such as automated scenario generation.

4.3 Conceptual Consistency

Concerning the absence of a baseline for scenario

modeling ontology, a preliminary approach was

employed to validate the ontology concepts against

international standards. The ISO 34504:2024

standard (ISO, 2024), an internationally recognized

normative document for scenario annotation and

categorization, offers a comprehensive set of

concepts for scenario modeling.

Using ChatGPT

, the terminology and taxonomy

of the ISO 34504 were extracted. This process

generated a taxonomy with 495 distinct concepts,

which was then verified by two human reviewers.

Next, each of these 495 concepts and the concepts

in the three ontologies were transformed into a 3072-

dimensional word vector using ChatGPT’s pre-

trained text-embedding-3-large model. This

transformation enabled us to capture the semantic

nuances and contextual relationships inherent in the

concepts. Cosine similarity between concepts was

calculated. This similarity metric provided a

quantitative measure of the alignment between the

standard and the ontology, offering insights into their

conceptual consistency.

Table 2 presents the aligned concepts between

each ontology and the ISO standard.

Table 2: Aligned concepts between ontologies and ISO

34504 standard (the threshold Cosine similarity > 0.85).

A.U.T.O.

OpenXO

ntology

Proposed

Ontology

Nb. Concepts

aligned with

ISO 34504:2024

10 51 39

https://chatgpt.com/share/66ea2eb1-46e4-800b-9653-

744e1a321ed8

4.4 Discussion

Ontology evaluation poses significant challenges for

real-world applications, particularly in complex

domains such as AD/ADAS validation. In this work,

we have explored and addressed the quantitative

aspects of ontology evaluation, with a focus on

structural and terminological aspects. These

evaluations highlight the importance of building an

ontology that is concise enough for effective scenario

modeling while still being comprehensive enough to

capture a wide range of scenarios.

A key challenge remains in evaluating semantic

relations. The quantity and quality of relationships

and restrictions directly impacts the realism and

plausibility of scenarios. Poorly modeled relations

can lead to oversimplified or non-plausible scenarios,

which reduces the reliability of Scenario-based

Testing for AD/ADAS validation. Our ongoing work

focuses on evaluating the semantic relations between

concepts and their relationships across different

ontologies, using word embedding models. We aim to

capture the nuanced similarities and differences in

how knowledge is represented and interconnected for

scenario modeling.

This evaluation work did not systematically

investigate the well-known modeling issues in

semantic relations, such as those identified by

(Poveda-Villalón et al., 2014). However, during an

initial investigation, a few modeling errors were

identified when comparing ontologies. For example,

in the A.U.T.O. ontology, there is a relation “Cloud

is_part_of exactly 1 Sky,” representing a subset

composition between Cloud and Sky concepts. While

a constraint “Cloud disjoint_with Moon, Sun, Air,

Ground, Air_Particle, Sky, Wind” declares a disjoint

relation between Cloud and Sky. Here, a subset

relation cannot be compatible with a disjoint

constraint.

5 CONCLUSION

This paper identifies critical gaps in existing

ontology-based scenario modeling frameworks that

limit the effectiveness of modeling scenarios for

validating AD/ADAS functions. These challenges

include the issues with ontology coverage and the

plausibility of modeled scenarios, the limitations in

expressing dynamic scenarios, and the capacity to

model complex scenarios that are aligned with the

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

628

operating conditions and requirements of AD/ADAS

validation.

To address these gaps, this paper makes several

contributions. Firstly, the behavioral descriptions of

dynamic objects within scenarios are refined. This

refinement allows a more nuanced representation of

dynamic objects’ actions and interactions. Second,

the ontology framework develops relations and

constraints associated with the Goals, Behaviors,

Maneuvers, and Activities are developed in this

ontology framework. These relationships and

restrictions are particularly useful in determining the

relevance of scenarios for AD/ADAS validation.

Third, lane segmentation and grid layout are

introduced to enhance the modeling capability of real-

world traffic environments. Fourth, activity-based

combinations for scenario modeling have been

introduced. By combining the activities of dynamic

objects, the proposed model allows for a detailed

description of the spatial-temporal changes in a

scenario. Moreover, multiple activities can be

combined in sequence, enabling the concatenation of

scenarios. These contributions enhance the quality of

the ontology framework for valid, detailed, and

complex scenario modeling. This work lays a

foundation for more effective scenario generation for

AD/ADAS validation.

This paper also contributes to quantitative

evaluation of ontologies, offering a systematic

approach to assess the structural and terminological

aspects. This evaluation highlights key strengths and

areas for improvement, supporting the development

of more robust and practical ontology-based scenario

modeling framework for AD/ADAS validation.

ACKNOWLEDGEMENTS

We would like to extend our sincere gratitude to

ALTEN Labs for their invaluable support throughout

this work. This research was also supported by the

ANRT (Association Nationale de la Recherche et de

la Technologie) through a CIFRE (Conventions

Industrielles de Formation par la REcherche)

fellowship.

REFERENCES

Armand, A., Filliat, D., & Ibañez-Guzman, J. (2014).

Ontology-based context awareness for driving

assistance systems. 2014 IEEE Intelligent Vehicles

Symposium Proceedings, 227–233.

https://doi.org/10.1109/IVS.2014.6856509

ASAM. (2019). ASAM OpenSCENARIO.

https://www.asam.net/standards/detail/openscenario/

ASAM. (2022). ASAM OpenXOntology.

https://www.asam.net/standards/asam-openxontology/

AVSC. (2021). AVSC00008202111: AVSC Best Practice

for Evaluation of Behavioral Competencies for

Automated Driving System Dedicated Vehicles (ADS-

DVs) - SAE International. https://www.sae.org/

standards/content/avsc00008202111/

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

based Scene Creation for the Development of

Automated Vehicles. 2018 IEEE Intelligent Vehicles

Symposium (IV), 1813–1820. https://doi.org/10.1109/

IVS.2018.8500632

Chen, W., & Kloul, L. (2018). An ontology-based approach

to generate the advanced driver assistance use cases of

highway traffic. 2, 75–83. Scopus.

https://doi.org/10.5220/0006931700750083

de Gelder, E., Paardekooper, J.-P., Saberi, A. K., Elrofai,

H., Camp, O. O. den, Kraines, S., Ploeg, J., & De

Schutter, B. (2022). Towards an Ontology for Scenario

Definition for the Assessment of Automated Vehicles:

An Object-Oriented Framework. IEEE Transactions on

Intelligent Vehicles, 7(2), 300–314. IEEE Transactions

on Intelligent Vehicles. https://doi.org/10.1109/

TIV.2022.3144803

Elfes, A. (1989). Using occupancy grids for mobile robot

perception and navigation. Computer, 22(6), 46–57.

Computer. https://doi.org/10.1109/2.30720

Erz, J., Schütt, B., Braun, T., Guissouma, H., & Sax, E.

(2022). Towards an Ontology That Reconciles the

Operational Design Domain, Scenario-based Testing,

and Automated Vehicle Architectures. 2022 IEEE

International Systems Conference (SysCon), 1–8.

https://doi.org/10.1109/SysCon53536.2022.9773840

Geyer, S., Baltzer, M., Franz, B., Hakuli, S., Kauer, M.,

Kienle, M., Meier, S., Weißgerber, T., Bengler, K.,

Bruder, R., Flemisch, F., & Winner, H. (2014). Concept

and development of a unified ontology for generating

test and use-case catalogues for assisted and automated

vehicle guidance. IET Intelligent Transport Systems,

8(3), 183–189. https://doi.org/10.1049/iet-

its.2012.0188

ISO. (2022). ISO 34502:2022 Test scenarios for automated

driving systems—Scenario based safety evaluation

framework. https://www.iso.org/standard/78951.html

ISO. (2024). ISO 34504:2024 Road vehicles—Test

scenarios for automated driving systems—Scenario

categorization. https://www.iso.org/standard/

78953.html

Krajewski, R., Bock, J., Kloeker, L., & Eckstein, L. (2018).

The highD Dataset: A Drone Dataset of Naturalistic

Vehicle Trajectories on German Highways for

Validation of Highly Automated Driving Systems.

2018 21st International Conference on Intelligent

Transportation Systems (ITSC)

, 2118–2125.

https://doi.org/10.1109/ITSC.2018.8569552

Li, Y., Tao, J., & Wotawa, F. (2020). Ontology-based test

generation for automated and autonomous driving

Enhancing Scenario-based Testing for Automated Driving Systems: An Ontology-Based Scenario Modeling Framework

629

functions. Information and Software Technology, 117,

106200. https://doi.org/10.1016/j.infsof.2019.106200

NHTSA. (2018). A Framework for Automated Driving

System Testable Cases and Scenarios (p. 180).

https://www.nhtsa.gov/document/framework-

automated-driving-system-testable-cases-and-

scenarios

Noy, N. F., Crubézy, M., Fergerson, R. W., Knublauch, H.,

Tu, S. W., Vendetti, J., & Musen, M. A. (2003).

Protégé-2000: An Open-Source Ontology-

Development and Knowledge-Acquisition

Environment. AMIA Annual Symposium Proceedings,

2003, 953.

Poveda-Villalón, M., Gómez-Pérez, A., & Suárez-

Figueroa, M. C. (2014). OOPS! (OntOlogy Pitfall

Scanner!): An On-line Tool for Ontology Evaluation.

International Journal on Semantic Web and

Information Systems (IJSWIS), 10(2), 7–34.

https://doi.org/10.4018/ijswis.2014040102

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

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

Bollmann, M. H., Körtke, 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, 59131–

59147. IEEE Access. https://doi.org/10.1109/

ACCESS.2021.3072739

Schuldt, F., Saust, F., Lichte, B., Maurer, M., & Scholz, S.

(2013). Effiziente systematische Testgenerierung für

Fahrerassistenzsysteme in virtuellen Umgebungen.

https://doi.org/10.24355/dbbs.084-201307101421-0

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

M. (2015). Defining and Substantiating the Terms

Scene, Situation, and Scenario for Automated Driving.

2015 IEEE 18th International Conference on

Intelligent Transportation Systems, 982–988.

https://doi.org/10.1109/ITSC.2015.164

UNECE. (2021). UN Regulation No. 157—Uniform

provisions concerning the approval of vehicles with

regard to Automated Lane Keeping Systems.

https://unece.org/transport/documents/2021/03/standar

ds/un-regulation-no-157-automated-lane-keeping-

systems-alks

Westhofen, L., Neurohr, C., Butz, M., Scholtes, M., &

Schuldes, M. (2022). Using Ontologies for the

Formalization and Recognition of Criticality for

Automated Driving. IEEE Open Journal of Intelligent

Transportation Systems, 3, 519–538. IEEE Open

Journal of Intelligent Transportation Systems.

https://doi.org/10.1109/OJITS.2022.3187247

Zipfl, M., Koch, N., & Zöllner, J. M. (2023). A

Comprehensive Review on Ontologies for Scenario-

based Testing in the Context of Autonomous Driving.

2023 IEEE Intelligent Vehicles Symposium (IV), 1–7.

https://doi.org/10.1109/IV55152.2023.10186681

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

630