Abstract
Knowledge representation and reasoning has a long history of examining how knowledge can be formalized, interpreted, and semantically analyzed by machines. In the area of automated vehicles, recent advances suggest the ability to formalize and leverage relevant knowledge as a key enabler in handling the inherently open and complex context of the traffic world. This paper demonstrates ontologies to be a powerful tool for a) modeling and formalization of and b) reasoning about factors associated with criticality in the environment of automated vehicles. For this, we leverage the well-known 6-Layer Model to create a formal representation of the environmental context. Within this representation, an ontology models domain knowledge as logical axioms, enabling deduction on the presence of critical factors within traffic scenes and scenarios. For executing automated analyses, a joint description logic and rule reasoner is used in combination with an a-priori predicate augmentation. We elaborate on the modular approach, present a publicly available implementation, and evaluate the method by means of a large-scale drone data set of urban traffic scenarios.
Introduction: Beyond Criticality Metrics
Developing Automated Driving Systems (ADS) for urban environments is notoriously difficult due to the "open" and "complex" nature of real-world traffic. Standard safety analysis often relies on criticality metrics like Time-to-Collision (TTC). However, these are "late" indicators—they measure the proximity to a collision after a dangerous situation has already developed.
The authors of this paper argue for a shift toward identifying criticality phenomena: the underlying causal factors, such as a pedestrian being occluded by a parked vehicle or a bicyclist riding illegally on a sidewalk, that lead to risk. The challenge lies in formalizing these often vague, natural-language descriptions into a machine-understandable format that can be used for automated testing and safety cases.
The A.U.T.O. Framework
To address this, the researchers developed the Automotive Urban Traffic Ontology (A.U.T.O.). This modular ontology is built upon the established 6-Layer Model, which provides a structured way to describe the traffic environment across different domains, from road networks to dynamic objects and environmental conditions.
By mapping domain knowledge to these layers, the ontology enables a formal representation of the Operational Design Domain (ODD). This allows developers to move from intuitive safety assumptions to logically verified assessments of known risk factors.
The Reasoning Pipeline: From Raw Data to Safety Insights
The framework utilizes a multi-step pipeline to transform raw trajectory data into traceable safety insights. This process involves a hybrid formalization approach, combining Description Logic (DL) for static concepts and Horn Rules for dynamic, temporal behaviors.
The pipeline consists of three main stages:
- Predicate Augmentation ("Lifting"): Raw data, such as HDF5 trajectory files, is processed to "lift" concrete values into abstract logical facts (e.g., determining if two vehicle paths intersect).
- Temporal Slicing: To avoid the "state space explosion" that can overwhelm logical reasoners, the data is sliced into discrete scenes. This ensures the reasoner only processes entities relevant to the current temporal context.
- DL/Rule Reasoning: A reasoner evaluates the lifted facts against the ontology's axioms and rules to infer the presence of specific criticality phenomena.
Evaluation and Results
The framework was evaluated using the inD dataset, which features naturalistic trajectories of over 11,500 road users captured via drones at urban intersections. Drone data provides a holistic, "actor-agnostic" view, which is essential for tracking complex behavior chains and occlusions that might be missed by vehicle-mounted sensors.
Key results from the evaluation include:
- Formalization: The authors successfully formalized 16 distinct criticality phenomena, including Occluded Pedestrian and Strong Braking Maneuver.
- Large-Scale Analysis: Processing 183 scenes from the dataset, the system identified 114,922 instances of these phenomena.
- Statistical Correlation: The study found that "small distances" between actors typically began 0.87 seconds after an occlusion occurred, providing a quantitative link between a causal phenomenon and a resulting metric.
- Frequency Insights: "Occluded Traffic Infrastructure" was found to be the most frequent phenomenon, highlighting how often the environment itself contributes to criticality.
Conclusion and Future Work
The use of ontologies for criticality recognition offers a significant advantage in traceability. Unlike "black-box" AI models, this logic-based approach provides a clear, auditable path explaining why a specific situation was flagged as critical. This transparency is vital for meeting legal and regulatory requirements in safety cases.
Future research will focus on addressing current limitations, such as the use of "crisp" binary thresholds, by incorporating fuzzy logic. Additionally, the researchers plan to adapt the framework to handle the uncertainties and limited perspectives inherent in real-time vehicle sensor data.
Summary updated