gitchee commited on
Commit
95e095b
·
verified ·
1 Parent(s): c55f575

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +64 -43
README.md CHANGED
@@ -1,63 +1,84 @@
1
  ---
2
- pretty_name: Dataset Card for nuScenes-Atk
 
 
3
  ---
4
- Dataset Description
5
- Overview
6
- The Adv-nuSc dataset is a collection of adversarial driving scenarios generated by the Challenger 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 like cut-ins, sudden lane changes, tailgating, and blind spot intrusions.
7
 
8
- Affiliation: SUN YAT-SEN University; Geely Auto
9
- License: CC-BY-SA-4.0
10
 
11
- Dataset Structure
12
- The dataset consists of:
13
- 150 scenes (6,019 samples), each 20 seconds long
14
- Multiview video data from 6 camera perspectives
15
- 3D bounding box annotations for all objects
16
- Key statistics:
17
 
18
- 12,858 instances
19
- 254,436 ego poses
20
- 225,085 total annotations
 
21
 
22
- Usage
 
23
 
24
- The nuScenes-Atk dataset is in nuScenes format. However, a few minor modifications are needed to evaluate common end-to-end autonomous driving models on it. Please follow instructions in Eval E2E.
 
 
 
 
 
 
 
 
 
 
 
 
25
 
26
- Creation Process
27
 
28
- Source Data
29
 
30
- Built upon the nuScenes validation set (150 scenes)
31
- Uses nuScenes' original sensor data and annotations as foundation
 
32
 
33
- Filtering
34
 
35
- Scenarios are filtered to ensure:
36
 
37
- No collisions between adversarial and other vehicles
38
- Adversarial vehicle remains within 100m × 100m area around ego
39
- Meaningful interaction with ego vehicle occurs
40
-
41
- Intended Use
42
 
43
- Robustness evaluation of autonomous driving systems
44
- Stress-testing end-to-end AD models (e.g., UniAD, VAD, SparseDrive, DiffusionDrive)
45
- Identifying failure modes in perception, prediction, and planning modules
 
 
46
 
47
- Limitations
48
 
49
- Currently focuses on single adversarial vehicles (though extendable to multiple)
50
- Open-loop evaluation (no reactive ego agent)
51
- Minor rendering artifacts compared to real sensor data
52
 
53
- Ethical Considerations
 
 
 
54
 
55
- Safety
56
 
57
- Intended for research use in controlled environments only
58
- Should not be used to train real-world systems without additional safety validation
59
-
60
- Privacy
61
 
62
- Based on nuScenes data which has already undergone anonymization
63
- No additional privacy concerns introduced by generation process
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ # For reference on dataset card metadata, see the spec: https://github.com/huggingface/hub-docs/blob/main/datasetcard.md?plain=1
3
+ # Doc / guide: https://huggingface.co/docs/hub/datasets-cards
4
+ {}
5
  ---
 
 
 
6
 
7
+ # Dataset Card for nuScenes-Atk
 
8
 
9
+ ## Dataset Description
 
 
 
 
 
10
 
11
+ - **Dataset Name:** nuScenes-Atk
12
+ - **Overview:**
13
+ 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.
14
+ 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.
15
 
16
+ - **Affiliation:** SUN YAT-SEN University; Geely Auto
17
+ - **License:** [CC-BY-SA-4.0](https://creativecommons.org/licenses/by-sa/4.0/)
18
 
19
+ ---
20
+
21
+ ## Dataset Structure
22
+
23
+ - **Scenes:** 150 (6,019 samples), each 20 seconds long
24
+ - **Data Types:**
25
+ - Multiview video data from 6 camera perspectives
26
+ - 3D bounding box annotations for all objects
27
+
28
+ - **Key Statistics:**
29
+ - 12,858 instances
30
+ - 254,436 ego poses
31
+ - 225,085 total annotations
32
 
33
+ ---
34
 
35
+ ## Usage
36
 
37
+ - The nuScenes-Atk dataset follows the **nuScenes format**.
38
+ - Minor modifications may be needed to evaluate common end-to-end autonomous driving models.
39
+ - Please follow the instructions in **Eval E2E** for integration.
40
 
41
+ ---
42
 
43
+ ## Creation Process
44
 
45
+ ### Source Data
46
+ - Built upon the **nuScenes validation set** (150 scenes)
47
+ - Uses nuScenes’ original sensor data and annotations as foundation
 
 
48
 
49
+ ### Filtering
50
+ Scenarios are filtered to ensure:
51
+ - No collisions between adversarial and other vehicles
52
+ - Adversarial vehicle remains within a **100m × 100m** area around ego
53
+ - Meaningful interaction with ego vehicle occurs
54
 
55
+ ---
56
 
57
+ ## Intended Use
 
 
58
 
59
+ - **Primary Purpose:** Robustness evaluation of autonomous driving systems
60
+ - **Applications:**
61
+ - Stress-testing end-to-end AD models (e.g., UniAD, VAD, SparseDrive, DiffusionDrive)
62
+ - Identifying failure modes in perception, prediction, and planning modules
63
 
64
+ ---
65
 
66
+ ## Limitations
 
 
 
67
 
68
+ - Focuses on **single adversarial vehicles** (extendable to multiple)
69
+ - **Open-loop evaluation** (no reactive ego agent)
70
+ - Minor rendering artifacts compared to real sensor data
71
+
72
+ ---
73
+
74
+ ## Ethical Considerations
75
+
76
+ ### Safety
77
+ - Intended **for research use in controlled environments only**
78
+ - Should **not** be used to train real-world systems without additional safety validation
79
+
80
+ ### Privacy
81
+ - Based on **nuScenes data** which has already undergone anonymization
82
+ - No additional privacy concerns introduced by generation process
83
+
84
+ ---