File size: 3,137 Bytes
9e0c675
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aede4b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9e0c675
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- DROID
- robotics
- skill-filtered
- picking
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# DROID Skill Subset: picking

This is a skill-filtered subset of the local DROID 1.0.1 LeRobot-format dataset.
It was generated with `droid_filter.py` from `/scratch/jellyho/prsl_skills`.

## Repository

- Hugging Face repo: `jellyho/droid_subsets_picking`
- Source dataset path used locally: `/scratch/jellyho/droid_1.0.1`
- Output subset path used locally: `/scratch/jellyho/droid_subsets/picking`

## Filtering

```json
{
  "skills": [
    "picking"
  ],
  "skill_keywords": {
    "picking": [
      "pick",
      "pick up",
      "grab",
      "grasp",
      "take",
      "lift",
      "remove"
    ]
  },
  "queries": [],
  "exclude": [],
  "regex": false,
  "match_all": false,
  "min_length": null,
  "max_length": null,
  "max_episodes": 300,
  "sample_mode": "balanced",
  "seed": 0
}
```

## Dataset Statistics

- Episodes: `300`
- Frames: `88076`
- Unique tasks: `300`
- Sampling mode: `balanced`
- Max episodes: `300`
- Seed: `0`
- Mean episode length: `293.58666666666664`
- Min/max episode length: `85` / `1450`
- Episode success rate: `1.0`

## Preserved Action Fields

Filtering only selects episodes/rows. It does **not** drop action-space columns.
The following action columns are preserved from the original DROID parquet files:

```text
action.gripper_position
action.gripper_velocity
action.cartesian_position
action.cartesian_velocity
action.joint_position
action.joint_velocity
action.original
action
```

`action` is the compact 8D policy action used by LeRobot-style loaders. The
additional `action.*` columns remain available in the parquet files for
inspection or custom loaders.

## Top Tasks

| task_text | episodes |
| --- | --- |
| Take the utensils out of the weaved basket | 1 |
| Remove the purple marker from the box and put it on the counter | 1 |
| Pick up the white remote and move it to the right | 1 |
| Remove the plastic spoon from the small mug and put it on the counter | 1 |
| Pick the fork and put it in the drawer | 1 |
| Remove the glass lid from the black pot | 1 |
| Take the lid off the pot | 1 |
| Pick up the brown and yellow medicine bottle from the orange plastic and put it on the table | 1 |
| Take the spray bottle in the box and put it on the table | 1 |
| Take the strawberry off of the tray | 1 |

## Files

This repository keeps the LeRobot-style layout:

```text
meta/info.json
meta/tasks.parquet
meta/episodes/chunk-000/file-000.parquet
data/chunk-000/file-*.parquet
episodes.csv
episodes.jsonl
summary.json
stats.json
task_counts.csv
length_histogram.csv
source_episode_files.csv
```

The subset keeps compact `episode_index`, `task_index`, and `index` values for
loader compatibility. `episodes.csv` keeps `original_episode_index` for tracing
rows back to the original DROID dataset when available.

## Notes

Video files are not included unless explicitly materialized with video symlinks
or copies. These subsets are intended for lightweight skill-conditioned
fine-tuning experiments.