metadata
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
configs:
- config_name: default
data_files:
- split: train
path: data/chunk-000/episode_*.parquet
dataset_info:
features:
- name: timestamp
dtype: float64
- name: frame_index
dtype: int64
- name: episode_index
dtype: int64
- name: index
dtype: int64
- name: task_index
dtype: int64
- name: observation.state
sequence:
dtype: float64
length: 7
- name: observation.velocity
sequence:
dtype: float64
length: 7
- name: observation.effort
sequence:
dtype: float64
length: 7
- name: observation.weight
sequence:
dtype: float64
length: 1
- name: action
sequence:
dtype: float64
length: 7
splits:
- name: train
num_bytes: 16603644
num_examples: 120000
download_size: 16603644
dataset_size: 16603644
FR5 Garlic Manipulation Dataset
This dataset contains demonstrations collected on an FR5 robot teleoperated to manipulate garlic. Each episode pairs synchronized joint trajectories (observation.state, action) with base and wrist camera videos. The layout follows the LeRobot dataset convention with chunked parquet files and per-camera mp4 streams.
Dataset Details
- Total Episodes: 100
- Total Frames: 120,000
- FPS: ~30.01 Hz
- Episode Length: ~1200 frames per episode
- Task: Garlic pick and place manipulation
Data Structure
Each episode contains:
- Joint States (
observation.state): 7D joint positions (6 arm joints + 1 gripper) in radians - Actions (
action): 7D target joint positions (6 arm joints + 1 gripper) in radians - Additional Observations: velocity, effort, and weight measurements
- Videos: Synchronized base and wrist camera videos (MP4 format)
- Metadata: Timestamps, frame indices, episode indices, and task indices
File Structure
demos_garlic_mod/
├── data/
│ └── chunk-000/
│ └── episode_*.parquet
├── videos/
│ └── chunk-000/
│ ├── observation.images.cam_base/
│ │ └── episode_*.mp4
│ ├── observation.images.cam_wrist/
│ │ └── episode_*.mp4
└── meta/
├── info.json
├── episodes.jsonl
├── episodes_stats.jsonl
└── tasks.jsonl
Usage
This dataset is compatible with the LeRobot framework and can be loaded using standard LeRobot data loaders for training imitation learning models.