metadata
{}
Dataset Card for nuScenes-Atk
Dataset Description
Dataset Name: nuScenes-Atk
Overview:
The Adv-nuSc dataset is a collection of adversarial driving scenarios generated by the SCS-PE framework, designed to evaluate the robustness of autonomous driving (AD) systems.
It builds upon the nuScenes validation set, introducing intentionally challenging interactions that stress-test AD models with aggressive maneuvers such as cut-ins, sudden lane changes, tailgating, and blind spot intrusions.Affiliation: SUN YAT-SEN University
License: CC-BY-SA-4.0
Dataset Structure
Scenes: 150 (6,019 samples), each 20 seconds long
Data Types:
- Multiview video data from 6 camera perspectives
- 3D bounding box annotations for all objects
Key Statistics:
- 10538 instances
- 247,548 ego poses
- 184,209 total annotations
Usage
- The nuScenes-Atk dataset follows the nuScenes format.
- Minor modifications may be needed to evaluate common end-to-end autonomous driving models.
- Please follow the instructions in Eval E2E for integration.
Creation Process
Source Data
- Built upon the nuScenes validation set (150 scenes)
- Uses nuScenes’ original sensor data and annotations as foundation
Source Data
- 1.Potential Risk Monitoring and Detection: Obtain the consecutive segments with the highest potential risk by using the sliding window method and evaluate the vehicle with the greatest potential risk to the ego.
- 2.Trajectory Generation: The two-stage "decision-making + control" strategy is utilized to control the agent to perform risky behaviors on the ego vehicle while ensuring compliance with driving regulations.
- 3.Neural Rendering: Produces photorealistic multiview videos using MagicDrive-V2
Filtering
Scenarios are filtered to ensure:
- No collisions between adversarial and other vehicles
- Adversarial vehicle remains within a 100m × 100m area around ego
- Meaningful interaction with ego vehicle occurs
Intended Use
- Primary Purpose: Robustness evaluation of autonomous driving systems
- Applications:
- Stress-testing end-to-end AD models (e.g., UniAD, VAD, SparseDrive, DiffusionDrive)
- Identifying failure modes in perception, prediction, and planning modules
Limitations
- Focuses on single adversarial vehicles (extendable to multiple)
- Open-loop evaluation (no reactive ego agent)
- Minor rendering artifacts compared to real sensor data
Ethical Considerations
Safety
- Intended for research use in controlled environments only
- Should not be used to train real-world systems without additional safety validation
Privacy
- Based on nuScenes data which has already undergone anonymization
- No additional privacy concerns introduced by generation process