File size: 2,138 Bytes
a44dd93
 
 
 
 
 
 
6ca540f
a44dd93
 
 
 
6ca540f
 
a44dd93
6ca540f
a44dd93
 
6ca540f
a44dd93
 
6ca540f
a44dd93
 
 
6ca540f
a44dd93
6ca540f
 
 
 
 
a44dd93
6ca540f
a44dd93
6ca540f
 
 
 
a44dd93
6ca540f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a44dd93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: mit
task_categories:
- robotics
- reinforcement-learning
tags:
- metaworld
- short-metaworld
- robotics
- manipulation
- multi-task
- vision-language
- imitation-learning
- r3m
size_categories:
- 10K<n<100K
language:
- en
pretty_name: Short-MetaWorld-VLA (v2+v3)
---

# Short-MetaWorld-VLA (v2 + v3)

## Overview

This dataset contains a short MetaWorld collection used for VLA-style training and evaluation.

Current local structure includes:
- **24 task files** in `r3m_MT10_20` (`12 v2 + 12 v3`)
- **100 trajectories per task**
- **20 or 50 steps per trajectory** (task/version dependent)
- **84,000 total step samples** from PKL action/state streams

## Dataset Structure

short-metaworld-vla/
├── mt50_task_prompts.json
├── short_metaworld_loader.py
├── requirements.txt
├── short-MetaWorld/
│ ├── img_only/
│ │ └── <task>/<trajectory>/<step>.jpg
│ └── r3m-processed/
│ └── r3m_MT10_20/
│ ├── <task>-v2.pkl
│ ├── <task>-v3.pkl
│ └── data.pkl
└── r3m-processed/
└── r3m_MT10_20/

## Data Format

Per step:
- `image`: RGB frame (`.jpg`)
- `state`: **39D** float vector
- `action`: **4D** float vector
- `prompt`: task language instruction (from `mt50_task_prompts.json`)
- `task_name`: task identifier (e.g. `button-press-topdown-v3`)

## Tasks

Includes both `-v2` and `-v3` variants such as:
- basketball
- button-press-topdown
- door-open
- drawer-open / drawer-close
- peg-insert-side
- pick-place
- push
- reach
- sweep
- window-open / window-close
- plus v3-only tasks in this dump (e.g. `handle-pull-v3`, `stick-pull-v3`)


## 🔬 Research Applications

This dataset is designed for:

- **Multi-task Reinforcement Learning**: Train policies across multiple manipulation tasks
- **Imitation Learning**: Learn from demonstration trajectories
- **Vision-Language Robotics**: Connect visual observations with natural language instructions
- **Meta-Learning**: Adapt quickly to new manipulation tasks
- **Robot Policy Training**: End-to-end visuomotor control


## ⚖️ License

MIT License - See LICENSE file for details.