Adv-nuSc is a collection of adversarial driving scenarios generated by the Challenger framework to evaluate the robustness of autonomous driving systems. It builds upon the nuScenes validation set, introducing challenging interactions like cut-ins and sudden lane changes. The dataset was created by Pixtella and last updated on May 21, 2025.
Use Cases
- Stress-testing autonomous driving models based on adversarial maneuvers like cut-ins.
- Evaluating model robustness based on scenarios involving sudden lane changes.
- Benchmarking system performance based on interactions like tailgating and blind spot intrusions.
- Training or validating safety-critical driving systems based on intentionally challenging scenarios.
Strengths
- Dataset is designed specifically for evaluating robustness of autonomous driving systems.
- Scenarios include concrete adversarial maneuvers such as cut-ins, sudden lane changes, tailgating, and blind spot intrusions.
- Dataset builds upon the established nuScenes validation set.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Pixtella via Hugging Face.
- Collection Method
- Generated by the Challenger framework, building upon the nuScenes validation set.
- Freshness
- Last updated 2025-05-21 17:17:31; freshness should be verified.