Automated Vehicles developers need to define an Operational Design Domain (ODD) where such vehicles can operate safely. In order to extend the defined ODDs, the developers base their decision after detailed analysis of recorded data from multiple data collection drives. For the acquired data, it is important to know whether it is known trafic situation information (inside the automated vehicle's ODD) or novel information that can be used to expand the ODD. The large amount of data that is generated by a modern vehicle's sensors makes data storage and eficient analysis for expanding ODDs hardly feasible (most of the current approaches record all sensor data and then post-process the data using AI-based methods and finally perform manual checks in order to find the novel data). Hence, there is a need to classify trafic situations as novel at system runtime for an appropriately abstract notion of novelty so that the conceptually same trafic situation, e.g. on two similar days, is not considered novel only because of the diferent date. We propose a new methodology for detection of novel trafic situations at system runtime. The methodology is based on a trafic catalogue that consists of abstract trafic situation descriptions, which are a formalized representation of sets of concrete trafic situations. Continuous, automatic checks for satisfaction of the current trafic situation against the trafic catalogue provides verdicts about the novelty of the current trafic situation. Using an example, we show how domain experts can utilize the detected novelties to create such a trafic catalogue such that the novelties are classified as known in the future. The proposed method doesn't require any pre-training of an AI-based classifier and is human understandable, explainable and traceable.
In the automotive domain, ensuring the safe operation of automated vehicles is a topic of great interest.
One of the ways in which Automated Driving Function (ADF) developers can ensure safe operation of
automated vehicles involves allowing such vehicles to only drive while inside their Operational Design
Domain (ODD). The ODD of an automated vehicle may be defined as the set of operating conditions,
such as environmental factors, trafic conditions, etc., for which the vehicle has been designed and
tested rigorously and has proven capable of safe operation [
ADF developers continuously analyse the ODD of their automated vehicles to pursue their expansion
in order to enable safe operation of such vehicles in additional operating conditions. The usual process
for expanding the ODD involves manual identification of operating conditions by experts that are
not yet covered by the ODD, carrying out training data collection drives aiming to encounter the
desired operating conditions, filtering of the collected data, training the ADFs on the filtered data and
ifnally testing whether the automated vehicle is now capable of safe operation for the new operating
conditions [
This process has certain drawbacks. Data collection is currently carried out by recording all sensor
data during the entire drive. A vehicle equipped with RADAR sensors, LiDAR sensors and cameras
generates around 5 GB of data per second [
Ofline identification of novel operating conditions (i.e. conditions not yet present inside the ODD) from such a large dataset is a cost-, time- and resource-intensive process. This is because the data is available as unorganized and unlabelled collection of sensor data and recordings. This large amount of data must be post-processed, clustered, and then manually checked by experts to determine presence of novel operating conditions.
In this paper, we propose a methodology for automatic identification of novel operating conditions
at system runtime. This allows targeted recording of novel sensor data during data collection. In detail,
our approach proposes the use of runtime monitoring to classify the current trafic situation based on a
trafic catalogue. The proposed trafic catalogue would consist of abstractly specified trafic situations
representing operating conditions that experts have determined to be already present in the ODD.
Concretely, we use the Spatial View (SV) formalism of the Trafic Sequence Charts (TSCs) [
A number of approaches are present in literature dealing with identification of novelties in the
automotive domain. For us, approaches that deal with novelties arising in a trafic situation due to
interaction of trafic participants with each other are of interest. The approaches introduced in [
The research closest to this work is [
In this section, we present our concept for novel trafic situation detection at system runtime. First, we provide definitions for the terminology used in the concept. On the basis of these definitions, we ifrst define a novel trafic situation. Further, we introduce a method for detection of such novel trafic situations at vehicle runtime and recording of novel data. Finally, we present how a trafic catalogue can be created and extended by domain experts on the basis of the recorded novel data. The concept overview is presented in Figure 1.
A trafic situation consists of trafic environment and trafic participants. A trafic environment is defined as the context in which trafic participants operate, e.g. roads, lanes, etc. A trafic participant is an entity that interacts with trafic environment according to a set of behavioural and physical rules and contains attributes, e.g. vehicles, pedestrians, etc. We use an Object Model to model trafic participants and trafic environment. Let = (, , , ), where is a set of classes, is a set of basic types, is a set of typed predicate symbols and is a set of typed function symbols. () is a finite set of typed attributes for each object class. A concrete trafic situation is a spatial arrangement of trafic participants, at a particular point of time, within the trafic environment along with their state. Concrete trafic situations, over the Object Model , are defined as a function over a finite set of object identities ID (pertaining to the objects present in the situation) along with a type-consistent valuation of their attributes. : ID ↛ (() ↛ ), where ID is the set of object identities provided to class instances, is the domain set of attribute types. Hence, ()( : ) ∈ ().
Trafic situation specifications formally describe desired state and spatial properties of concrete trafic
situations. They are defined as a set of well-typed predicate logic formulae over . Examples of
trafic situation specification include Spatial Views [
We define a novelty or a novel trafic situation as a concrete trafic situation which satisfies none of
the trafic situation specifications present in a trafic catalogue. Formally, if ∀ ∈ : N ⊭ , then
N is a novelty. We define novel data as the sensor values contained in the concrete trafic situation
which satisfies none of the formalized situations present in a trafic catalogue. This is the data that
should be recorded and is of interest to domain experts. For detection of novelties at system runtime,
we require a method that can provide us continuous verdicts about whether a concrete trafic situation
satisfies any of the trafic situation specifications present in a trafic catalogue. Such trafic situation
specifications, occurring in the automotive domain, are usually quite complex. The complexity arises
from the need to specify interactions between multiple trafic participants, the trafic environment
and their attributes. In the field of Scenario-based Testing [
Mon(, ) := {︃ ⊤ ⊥ if ∃ ∈ : ⊨ , otherwise (1) When Mon(, ) = ⊤, this implies that the current concrete trafic situation is an already known situation and satisfies at least one of the trafic situation specifications present in the trafic catalogue. When a novelty is encountered i.e. Mon(, ) = ⊥, then the data recording mechanism is triggered to record the corresponding sensor data. The recording of data should be continued till the next Mon(, ) = ⊤ verdict is received.
Once the novel data recordings are available to the domain experts, they are able to analyse and recognize the trafic situations occurring in the recordings. Formalization of a subset of the identified trafic situations into specifications allows them to discuss and decide, by including them in the trafic catalogue, for which set of trafic situations suficient data is present and hence should be filtered out during future data recording.
In this example, we assume that the domain experts want to start from scratch and intend to build their trafic catalogue along the way. Their Data Collection Vehicle (DCV) starts collecting data while driving on a two-lane highway. Since initially, there are no situation specifications present in the trafic catalogue ( = ∅), hence all data is collected as novel data. After the first day of data collection, the domain experts check the collected data and identify certain trafic situations and decide to include them as trafic situation specifications in the trafic catalogue.
For creation of our example trafic catalogue, we use the Spatial View (SV) formalism from the Trafic
Sequence Charts (TSCs) [
= .. > . ∧ .. < . ∧ . = ∅Car where is an abbreviation for the projection of the positional anchor to its second component.
The three trafic situation specifications identified by the domain experts and specified as Spatial Views are added to the trafic catalogue = {1, 2, 3}. The corresponding SVs are presented in d<100[m] (a) 1: Ego vehicle on the left lane of a two-lane highway (b) 2: Ego vehicle on the right lane of a two-lane highway (c) 3: Ego vehicle with other vehicle on left lane at a distance of at least 50 m behind it
We have presented a methodology to detect novel trafic situations based on a trafic catalogue that consists of formally specified known trafic situations. This method is useful for developers of Automated Driving Function (ADF) who want to ensure safe operation of their vehicles by only allowing them to operate within their Operational Design Domain (ODD). The method enables a structured construction of the ODD by recording sensor data only when the operating conditions of the vehicle do not correspond to those present in the trafic catalogue. The lack of AI-based components in our approach makes it explainable, traceable, and understandable to humans. We showed how to apply the concept for creating a trafic catalogue by domain experts on the example of 2-lane highway situations.
Future work is to test the approach rigorously on larger simulation and real world datasets (similar to [24], where runtime monitoring using TSCs is demonstrated for a research vessel), and to implement the monitors on automotive-grade hardware. Further eforts are needed to extend the methodology to detect novel trafic scenarios (i.e., sequences of situations), for tool-support for completeness argumentations for the catalogue, and for adapting the technique for use in other transportation domains such as maritime.
(a) DCV simulation run with empty trafic catalogue (b) DCV simulation run with extended trafic catalogue
Figure 3: Simulation results from the prototypical implementation of our proposed concept
The research leading to these results is funded by the German Federal Ministry of Education and Research under grant agreement No 16MEE044 (EdgeAI-Trust) and by the Chips Joint Undertaking under grant agreement No. 101139892 (EdgeAI-Trust).
During the preparation of this work, the author(s) used LanguageTool in order to: Grammar and spelling check. After using these tool(s)/service(s), the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content. [14] E. Zapridou, E. Bartocci, P. Katsaros, Runtime verification of autonomous driving systems in carla, in: J. Deshmukh, D. Ničković (Eds.), Runtime Verification, Springer International Publishing, Cham, 2020, pp. 172–183. doi:10.1007/978-3-030-60508-7_9. [15] R. Stemmer, I. Saxena, L. Panneke, D. Grundt, A. Austel, E. Möhlmann, B. Westphal, Runtime monitoring of complex scenario-based requirements for autonomous driving functions, Science of Computer Programming 244 (2025) 103301. doi:10.1016/j.scico.2025.103301. [16] A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, V. Koltun, CARLA: An open urban driving simulator, in: Proceedings of the 1st Annual Conference on Robot Learning, 2017, pp. 1–16. [17] D. Grundt, A. Köhne, I. Saxena, R. Stemmer, B. Westphal, E. Möhlmann, Towards runtime monitoring of complex system requirements for autonomous driving functions, in: M. Luckcuck, M. Farrell (Eds.), Proceedings Fourth International Workshop on Formal Methods for Autonomous Systems (FMAS) and Fourth International Workshop on Automated and verifiable Software sYstem DEvelopment (ASYDE), Berlin, Germany, 26th and 27th of September 2022, volume 371 of Electronic Proceedings in Theoretical Computer Science, Open Publishing Association, 2022, pp. 53–61. doi:10. 4204/EPTCS.371.4. [18] J. S. Becker, Safe linear encoding of vehicle dynamics for the instantiation of abstract scenarios, in: A. E. Haxthausen, W. Serwe (Eds.), 29th International Conference on Formal Methods for Industrial Critical Systems, FMICS 2024, volume 14952 of Lecture Notes in Computer Science, Springer, 2024, pp. 3–20. URL: https://elib.dlr.de/205077/. doi:10.1007/978-3-031-68150-9_1. [19] D. Grundt, A. Rakow, P. Borchers, E. Möhlmann, What does ai need to know to drive: Testing relevance of knowledge, Science of Computer Programming 244 (2025) 103297. doi:https:// doi.org/10.1016/j.scico.2025.103297. [20] A. Austel, M. Steidel, B. Westphal, Formal specification of situations in scenario-based testing of maritime assistance systems, in: European Workshop on Maritime Systems Resilience and Security (MARESEC 2024), 2024. URL: https://elib.dlr.de/204947/. [21] M. Wild, J. S. Becker, C. Schneiders, E. Möhlmann, A scenario-based simulation framework for testing of highly automated railway systems, in: VEHITS 2025, Proceedings of the 11th Internationa Conference on Vehicle Technology and Intelligent Transport Systems - VEHITS, SciTePress, 2025, pp. 88–99. URL: https://elib.dlr.de/212746/. doi:10.5220/0013286600003941. [22] J. S. Becker, M. Wild, A. Rakow, E. Möhlmann, How far can we go? capturing trafic rules via trafic sequence charts, 2025. Accepted for publication in Law meets engineering for trustworthy Autonomous Vehicles (LawIng4AV) workshop proceedings, Baden-Baden, Germany, forthcoming. [23] P. Borchers, T. Koopmann, L. Westhofen, J. S. Becker, L. Putze, D. Grundt, T. de Graaf, V. Kalwa, C. Neurohr, Tsc2carla: An abstract scenario-based verification toolchain for automated driving systems, Science of Computer Programming 242 (2025). URL: https://elib.dlr.de/211153/. doi:10. 1016/j.scico.2024.103256. [24] A. Austel, L. Panneke, J. Piotrowski, N. Wetzig, M. Steidel, B. Westphal, Using monitoring of maritime trafic scenarios in the validation of maritime systems, 2025. Accepted for publication in IEEE Symposium on Maritime Informatics and Robotics (MARIS’25) proceedings, Syros, Greece, forthcoming.