STATS19-Derived Scenario Set for SafetyPool
STATS19 is the UK’s official road-collision database, published by the Department for Transport (DfT). It records police-reported road traffic collisions and associated casualty, vehicle, road, junction, environmental and severity information. The database contains a large number of annual collision records, including approximately 16,000–20,000 KSI cases, where KSI refers to collisions involving killed or seriously injured casualties.
This scenario set converts selected 2024 STATS19 crash records into simulator-ready scenarios in OpenSCENARIO v1.1 and OpenDRIVE v1.6 formats. The scenarios are generated through a structured toolchain that processes raw STATS19 data, extracts relevant crash context, builds logical scenario blueprints, enriches the road network, and exports validated OpenX scenario files for SafetyPool ingestion.
Because STATS19 contains a very large number of records, converting every case into a scenario would be impractical and would introduce substantial duplication. Many records describe highly similar crash situations. To achieve broad and meaningful coverage, a cluster-based statistical sampling approach is used. STATS19 records are grouped into interpretable crash archetypes using factors such as road environment, road-user type, manoeuvre, junction characteristics, speed limit and collision severity. Each cluster represents a distinct crash pattern and acts as a scenario category.
Example crash archetypes include:
- Pedestrian–vehicle interactions in low-speed urban areas
- Two-wheeler conflicts involving cyclists or motorcyclists
- Junction conflicts, including turning and crossing movements
- Rural high-speed collisions
- Overtaking or lane-changing conflicts
- Vehicle interactions with roadside or static objects
Representative records are selected from each cluster by identifying cases that best match the dominant characteristics of that cluster. For the 2024 dataset, a set of 60 representative scenarios has been extracted and converted for SafetyPool upload.
Each generated scenario preserves the fundamental STATS19 source context, including actor types, road setting, manoeuvre, severity, lighting condition and environment where available. The scenario generation process uses trajectory-based actions to represent the crash-relevant actor movements. The OpenDRIVE road networks are also enriched with contextual objects, such as houses, lampposts and trees, based on urban/rural classification and available environmental cues.
The scenario naming convention follows the format:
actor-type-interaction_road-type_urban-or-rural_day-or-night_unique-scenario-number
For example, a name may indicate the main interacting road users, the road type, whether the case occurred in an urban or rural setting, the lighting condition, and a unique scenario identifier (Example: car-car-single-carriageway-rural-darkness-2024991508972.xosc).
A special note applies to night-time scenarios: some visualisations may appear very dark. This is intentional. These cases follow the source-grounded STATS19 lighting condition, including records marked as darkness or late-night conditions without street lighting. The visual appearance is therefore retained to reflect the original crash context rather than artificially improving visibility.