jellyho's picture
Upload merged DROID skill dataset
3c6441b verified
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- DROID
- robotics
- skill-filtered
- merged
- joint-velocity-action
- gripper-position
- stacking
configs:
- config_name: default
data_files: data/*/*.parquet
---
# Merged DROID Skill Dataset: stacking
This dataset merges a language-filtered DROID skill subset with newly collected
environment data for the same skill.
## Repository
- Hugging Face repo: `jellyho/droid_merged_skills_stacking`
- Local merged dataset path: `/scratch/jellyho/droid_merged_skills/stacking`
- Skill: `stacking`
## Sources
```json
[
"jellyho/droid_subsets_stacking",
"jellyho/droid_stack_bar"
]
```
| source_dataset | episodes |
| --- | --- |
| jellyho/droid_subsets_stacking | 300 |
| jellyho/droid_stack_bar | 20 |
## State And Action
The compact LeRobot training fields are standardized as 8D vectors:
```text
observation.state = [joint_position_0..6, gripper_position]
action = [joint_velocity_0..6, gripper_position]
```
For original DROID subset rows, `action` is derived from
`action.joint_velocity` plus `action.gripper_position`. For newly collected
rows, `action` is derived from `observation.joint_velocities[:7]` plus the
collected gripper position from the original `action` vector.
Detailed columns such as `action.joint_position`, `action.joint_velocity`,
`action.gripper_position`, `observation.state.joint_position`, and
`observation.state.gripper_position` remain available in the parquet files.
## Dataset Statistics
- Episodes: `320`
- Frames: `91911`
- Unique tasks: `301`
- `observation.state` shape: `[8]`
- `action` shape: `[8]`
## Top Episode Task Strings
| task | episodes |
| --- | --- |
| ['Put the yellow bar on top of the red cube'] | 20 |
| ['Stack the cups from the top cup to the bottom cup'] | 1 |
| ['Pick up the lid and place it on top of the blue pot.'] | 1 |
| ['Move the white mug from on top of the paper towel to the counter'] | 1 |
| ['Remove the cap from the door coat hanging hook and put it on top of the cabinet'] | 1 |
| ['Stack the items on the tablecloth together.'] | 1 |
| ['Fold the white cloth in half, pick up the white cloth and put it on top of the far right throw pillow'] | 1 |
| ['Place the grey object at the top of the stack'] | 1 |
| ['Stack the cups on the table'] | 1 |
| ['Put the first towel on top of the backrest of the chair'] | 1 |
## Files
```text
meta/info.json
meta/tasks.parquet
meta/episodes/chunk-000/file-000.parquet
data/chunk-000/file-*.parquet
episodes.csv
merge_summary.json
```