Gearing Up: How ISO 34503 and 34504 Steer Automated Driving Forward

Amir Sahebn
4 min readNov 1, 2023



Automated driving systems (ADS) have the potential to revolutionize transportation and mobility. However, ensuring ADS technology’s safe and reliable operation brings immense technical challenges. Recently, two new ISO standards were published to promote the responsible development and testing of ADS by improving safety, interoperability, and transparency.

ISO 34503 provides a taxonomy and specifications for defining the operational design domain (ODD) of an ADS. Meanwhile, ISO 34504 puts forth a methodology to categorize test scenarios using tags.

Understanding Operational Design Domains (ISO 34503)

ISO 34503 provides a comprehensive taxonomy and specifications for defining an automated driving system’s operational design domain (ODD). The ODD represents the specific conditions under which an ADS is designed to function safely, including restrictions such as geographical location, road types, speed range, weather limitations, and the presence of certain dynamic elements like pedestrians or cyclists. The standard requires ODD definitions to be both machine-readable, allowing automated parsing and analysis and human-readable for interpretation by non-technical stakeholders. It also establishes a system for hierarchically classifying ODD attributes into categories like scenery, environmental conditions, and dynamic elements.

Categorizing Test Scenarios (ISO 34504)

ISO 34504 establishes a methodology for categorizing test scenarios for automated driving systems using tags, which serve as metadata to classify scenarios. The standard puts forth a taxonomy of tags spanning attributes like layout, weather events, dynamic entities, and intended safety validations. These tags allow the creation of a hierarchical system to define categories, subcategories, and scenario variations at different granularities. Scenarios sharing one or more tag classifications are considered part of that scenario category. By providing this standardized tagging scheme, ISO 34504 enables the exchange of labelled test scenarios between regulatory bodies, OEMs, technology vendors, research institutions, and other entities. It also facilitates the assembly of scenario databases, repositories, and libraries that can be utilized for targeted testing and validation of ADS based on operational design domain limitations.

Taxonomy vs. Ontology: Structuring the Semantics of Scenarios

The taxonomy of tags in ISO 34504 provides a hierarchical system to categorize scenarios. An alternative approach is developing an ontology, which defines more complex semantic relationships between concepts. This ontology-focused perspective has been further explored in the works of Bagschik et al. and de Gelder et al. Diving deeper into semantic technologies, the distinctions between ‘taxonomy’ and ‘ontology’ emerge as foundational principles, with their intricacies sometimes being nuanced yet crucial. At its core, a taxonomy resembles a structured tree, neatly offering a hierarchical classification of entities — much like organizing books in a library by genres and sub-genres. In contrast, ontology provides a comprehensive, graph-based blueprint, not only categorizing but also highlighting the intricate relationships and attributes that bind entities. This is analogous to a detailed map that captures not just cities but the intertwining roads, rivers, and landmarks. As we structure the semantics of scenarios, understanding the contrast between these two approaches becomes paramount. While taxonomies present a streamlined, top-down perspective, ontologies ensure every nuance and relationship is meticulously mapped. A notable contrast between these two modes of representation is observed in implementation and scenario generation. While ontology often performs better for scenario generation, taxonomy is more straightforward and easier to implement.

Connecting ODDs, Scenarios, and Safety Cases

While ODDs and scenarios are distinct concepts, ISO 34503 and ISO 34504 are complementary standards. Categorized scenarios can be linked to ODD specifications to support rigorous validation of ADS safety across diverse operating conditions. Structured ODD definitions constrain and guide the creation of relevant test scenarios. Comprehensive libraries of categorized scenario variants enable targeted assessment of an ADS within its operational boundaries. Taken together, the standards provide frameworks and taxonomies to strengthen the safety case for ADS. Codifying ODD limitations and mapping exhaustive scenario sets to those limitations aids system verification, driver training, procurement, regulation, and responsible deployment. Continued development of compatible standards for describing ODDs and scenarios will build confidence in ADS by improving safety, interoperability, and transparency.

Distilling the Infinite Variety of Real-World Driving Scenarios

A key challenge in developing standards for ODDs and scenarios is encapsulating the immense diversity of the real world. Real-world driving data reveals a complex, nuanced, and ever-evolving operational domain. The taxonomic frameworks in ISO 34503 and ISO 34504 aim to provide structures flexible enough to represent this reality. As research institutions and companies accumulate petabytes of logged test miles, these standards must enable the integration of this data into formal ODD and scenario representations. Continued evolution of machine-readable formats can allow automated aggregation, classification, and analysis. But care must be taken to preserve the long tail of edge cases not fully captured in fixed taxonomies. With diligent application and growth, these standards hold promise for formally representing the intricate chaos of the real world.

Key Takeaways and Next Steps

ISO 34503 and ISO 34504 represent initial progress in developing compatible standards to define key elements of the automated driving ecosystem. However, the true success of these standards relies on widespread adoption and continued evolution. To maximize their impact, industry stakeholders, regulators, and technology developers should collaborate to build shared libraries of standardized ODD definitions and categorized scenarios. Using a taxonomy framework may promote greater adoption and collaboration than a more complex ontology approach. Compatible tools and open platforms must be created to share and reuse structured ODD and scenario data seamlessly. As with new standards, broader acceptability and application will steer these toward maturity.