Drone Dataset for Complex Vehicle-VRU Interactions
A vehicle-VRU interaction dataset with 13,418 trajectories, 87% VRU proportion, and 4,000+ interaction scenarios — built to study how cars, pedestrians, and cyclists actually negotiate shared space.
Drag the slider to compare raw drone video with standardized trajectories on the HD map — vehicle ↔ VRU interactions at an urban intersection (site 22).
VRUD inverts the typical vehicle-centric ratio of autonomous-driving datasets, providing a dense statistical basis for vulnerable road user safety research.
Of 13,418 trajectories, 87% are VRUs — pedestrians, cyclists, and e-bike riders. The dataset is dimensioned for VRU-focused research, not as a side-effect of vehicle recordings.
4,000+ tagged interaction scenarios where a vehicle and at least one VRU influence each other's path or speed — gap acceptance, crossing, conflict, and yielding behaviors.
Top-down drone capture eliminates ego-vehicle occlusion — the bias that distorts ground-level VRU datasets. Clean trajectories at high temporal resolution.
Headline metrics from the VRUD release.
Each release ships with rich kinematic, contextual, and interaction-level fields, ready for direct scenario analysis.
Please cite the VRUD paper in any publication that uses this dataset.
"VRUD: A Drone Dataset for Complex Vehicle-VRU Interactions within Mixed Traffic"
Free for non-commercial academic research. We review every request manually and typically respond within 5 business days.
Application is submitted via our Feishu form. We typically respond within 5 business days.
Companion open datasets from the lab and partner institutions, covering different road types and traffic agents.