File size: 4,433 Bytes
2f6555d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8d4a935
 
 
2f6555d
8d4a935
 
2f6555d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8d4a935
2f6555d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f1de1b1
71b03a2
 
2f6555d
 
 
 
 
 
d085ab3
8d4a935
 
 
 
 
 
 
 
 
 
 
2f6555d
 
8d4a935
 
 
2f6555d
 
8d4a935
 
 
 
 
2f6555d
 
 
da0b103
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
---

pretty_name: "Robusto-1 Extended"
tags:
  - video
  - urban-driving
  - peru
  - traffic
  - autonomous-driving
license: apache-2.0
task_categories:
  - video-classification
---



# Robusto-1 Extended Dataset

## Description

The **Robusto-1 Extended** dataset contains videos recorded from 3 different cities in Peru: mainly Lima, Cusco and Cajamarca; 
along with detailed information about vehicle actions, external factors, and driving conditions. 
This dataset is useful for tasks such as vehicle behavior analysis, traffic condition detection, and driving scenario simulation.

Each video record includes data on the ego-vehicle's behavior, external conditions (traffic signs, weather, etc.), and more. 
The videos were recorded under various traffic conditions, with each one labeled with its corresponding metadata.

### Features:
- **ID**: Unique identifier for the video.
- **Video_Name**: Filename of the video.

- **Place**: Location where the video was recorded.

- **Ego_Vehicle**: Detailed information about the ego-vehicle’s actions during the recording.
- **External_Factors**: External conditions such as traffic signs, weather, etc.

- **Video**: Path to the video file in `.mp4` format.



## Metadata Example



Each video in the dataset is associated with metadata containing relevant information about the driving conditions at the time of recording. Below is an example of one metadata record:



```json

{

    "ID": "Robusto 1_1",

    "Place": "Lima",

    "Ego_Vehicle": {

        "Vehicle Actions": "Advancing Intermittently",

        "Driving Action Reasoning": "Following Vehicle Ahead",

        "Vehicle Motion Behavior": "Stop-and-Go"

    },

    "External_Factors": {

        "Traffic Signs": "",

        "Traffic Lights": "Red",

        "Weather Conditions": "Clear",

        "Road Surface Conditions": "",

        "Road Structures": "",

        "Static objects": "Buildings",

        "Other Vehicle Behaviors": "Cutting Off",

        "Pedestrian Behavior": "",

        "Unexpected Obstacles": "",

        "Emergency Situations": "",

        "Lighting Conditions": "Daylight",

        "Traffic Conditions": "Heavy Traffic",

        "Driving Environment": "Urban"

    }

}

```



## Usage



This dataset can be used for vehicle behavior analysis, improving autonomous driving algorithms, or training/testing computer vision models that require data on traffic and vehicle behavior in various conditions.



### Load the Dataset



To load this dataset from Hugging Face, you can use the following code:



```python

from datasets import load_dataset



dataset = load_dataset("path/to/robusto-1_extended")

```



## File Structure



The structure of the dataset files is as follows:



```

robusto-1_extended/

├── README.md

├── metadata.jsonl

└── videos/

    ├── Robusto 1_1.mp4

    ├── Robusto 1_2.mp4

    └── ...

```





- `README.md`: Contains the dataset description and YAML metadata at the top.  

- `metadata.jsonl`: File storing video metadata in JSON Lines format.  

- `train/`: Directory containing all `.mp4` video files.





## Short Q&A:



**Q**: What is the main difference between the Robusto-1 Extended and the Robusto-1 dataset?





**A**: They are both composed of small 5 second clips of 10 frames per seconds of out-of-distribution driving scenes in Peru. The main difference is that the Robusto-1

dataset is composed of 7 video clips with both Human and Machine responses and video meta-data, while the Robusto-1 Extended is only 

composed of 200 videos with their meta-data to facilitate further research in VQA research so that different groups can create their own questions.



## Citation



If you use this dataset in your research, please cite our ArXiv pre-print that was recently accepted at the **CVPR 2025 Workshop on
Distillation of Foundation Models for Autonomous Driving**.





```

@article{cusipuma2025robusto,

  title={Robusto-1 Dataset: Comparing Humans and VLMs on real out-of-distribution Autonomous Driving VQA from Peru},

  author={Cusipuma, Dunant and Ortega, David and Flores-Benites, Victor and Deza, Arturo},

  journal={arXiv preprint arXiv:2503.07587},

  year={2025}

}

```

```

This `README.md` provides an overview, usage instructions, and metadata example, along with the necessary information to cite the dataset in academic work.