File size: 4,387 Bytes
480daad
 
 
6db0dd0
480daad
6db0dd0
 
480daad
 
 
 
6db0dd0
 
480daad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6db0dd0
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
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
---
license: cc-by-nc-sa-4.0
size_categories:
- 100K<n<1M
task_categories:
- robotics
pretty_name: Vision-Language Accessible Navigation (VLAN) Dataset
---

# Vision-Language Accessible Navigation (VLAN) Dataset

[Paper](https://huggingface.co/papers/2611.18346) | [Code](https://github.com/InSAI-Lab/VLAN)

## Overview

VLAN is a user-conditioned accessible navigation dataset for multimodal navigation research.
This release contains the **evaluation split** used for benchmarking navigation outputs across heterogeneous users:

- `humanoid`
- `blind` / `BVI`
- `users with mobility impairments / wheelchair users`
- `drone`
- `quadrupedRobot`

Each trajectory is paired with:

- egocentric keyframes
- a trajectory description
- user-conditioned accessibility labels
- user-specific target outputs

This dataset is designed to support two related tasks:

1. **Action Prediction Navigation**
   - `drone`
   - `quadrupedRobot`
2. **Language-Guided Navigation**
   - `humanoid`
   - `blind` / `BVI`
   - `users with mobility impairments / wheelchair users`

## Release Contents

This Hugging Face dataset repo is organized into two main folders:

- `vlan/`: compressed full evaluation split archives
- `samples/`: a small browsable subset of scenes and trajectories

### `vlan/`

The `vlan/` folder contains the full evaluation split as **10 compressed archives**:

- `vlan_part01.tar.zst`
- `vlan_part02.tar.zst`
- `vlan_part03.tar.zst`
- `vlan_part04.tar.zst`
- `vlan_part05.tar.zst`
- `vlan_part06.tar.zst`
- `vlan_part07.tar.zst`
- `vlan_part08.tar.zst`
- `vlan_part09.tar.zst`
- `vlan_part10.tar.zst`

These 10 archives together contain:

Each archive can be downloaded and extracted independently.

Example extraction:

```bash
tar --zstd -xf vlan_part01.tar.zst
```

### `samples/`

The `samples/` folder contains a small subset of extracted scenes for quick inspection without downloading the full benchmark.

## Scene and Trajectory Structure

After extraction, the dataset is organized as:

```text
<scene_name>/
  <traj_name>/
    KeyFrames/
      00074.png
      00089.png
      ...
    metadata.json
    trajectory.json
    target.json
```

For example:

```text
0001/
  01/
    KeyFrames/
      00074.png
      00089.png
      ...
    metadata.json
    trajectory.json
    target.json
```

## File Descriptions

### `KeyFrames/`

Contains egocentric RGB keyframes for the trajectory.

### `metadata.json`

Contains:

- `scene_name`
- `traj_name`
- scene taxonomy metadata:
  - `lighting`
  - `scene_type`
  - `path_geometry`
  - `weather`
- embedded keyframe metadata

### `trajectory.json`

Contains the trajectory-level motion and obstacle information, including:

- obstacle geometry
- poses and trajectory states
- user-conditioned accessibility labels

The trajectory-level accessibility labels are stored under:

- `obstacles[0].userAccessibility`

User keys:

- `humanoid`
- `bvi`
- `wheelchair`
- `drone`
- `quadrupedRobot`

### `target.json`

Contains the benchmark target definition for this trajectory, including:

- corrected user-conditioned accessibility labels
- user-specific ground-truth outputs
- obstacle summary
- goal information
- action references for action-prediction users

User keys in `target.json`:

- `humanoid`
- `blind`
- `wheelchair`
- `drone`
- `quadrupedRobot`


## Accessibility Labels

The dataset uses four accessibility labels:

- `Passable`
- `Bypass`
- `HighRisk`
- `Blocked`

These labels are **user-conditioned**. The same scene and obstacle may have different labels for different users.

Examples:

- a low obstacle may be `Passable` for a drone
- the same obstacle may be `Bypass` for a humanoid user
- the same obstacle may be `HighRisk` for a blind user
- the same obstacle may be `Blocked` for a wheelchair user

## Scene Taxonomy

Each scene is annotated with scene-level taxonomy metadata:

- `lighting`
- `scene_type`
- `path_geometry`
- `weather`

These fields are stored in each trajectory's `metadata.json`.

## Intended Use

This dataset is intended for:

- multimodal accessible navigation
- user-conditioned navigation reasoning
- action prediction for embodied navigation agents
- language-guided instruction navigation
- accessibility-aware navigation evaluation

## Notes

- The benchmark targets are stored in `target.json`.
- The included `samples/` folder is for quick inspection and is not a separate split.