ganatrask/NOVA
Robotics β’ 3B β’ Updated β’ 14
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Expert demonstration dataset for training vision-language-action models on Pollen Robotics' Reachy 2 humanoid robot. Collected in MuJoCo simulation with domain randomization.
This dataset contains 100 episodes of pick-and-place manipulation tasks, designed for fine-tuning NVIDIA GR00T N1.6 or other imitation learning policies.
| Feature | Description |
|---|---|
| Format | LeRobot v2.1 (parquet + H264 video) |
| Episodes | 100 |
| Task Variations | 32 (4 objects Γ 8 colors) |
| Camera Views | 2 (front_cam, workspace_cam) |
| Resolution | 640Γ480 @ 15 FPS |
| Domain Randomization | Position, lighting, camera jitter |
The robot performs pick-and-place tasks with natural language instructions:
NOVA/
βββ meta/
β βββ info.json # Dataset metadata
β βββ stats.json # Normalization statistics
β βββ tasks.jsonl # Task descriptions
β βββ episodes.jsonl # Episode information
βββ data/
β βββ chunk-000/
β βββ episode_000000.parquet
β βββ episode_000001.parquet
β βββ ...
βββ videos/
βββ chunk-000/
βββ observation.images.front_cam/
β βββ episode_000000.mp4
β βββ ...
βββ observation.images.workspace_cam/
βββ episode_000000.mp4
βββ ...
| Field | Dimension | Description |
|---|---|---|
observation.state |
7 | Joint positions (arm) |
Joint Names:
shoulder_pitch (-180Β° to 90Β°)shoulder_roll (-180Β° to 10Β°)elbow_yaw (-90Β° to 90Β°)elbow_pitch (-125Β° to 0Β°)wrist_roll (-100Β° to 100Β°)wrist_pitch (-45Β° to 45Β°)wrist_yaw (-30Β° to 30Β°)| Field | Dimension | Description |
|---|---|---|
action |
8 | Joint positions (7 arm + 1 gripper) |
Gripper: 0 = closed, 1 = open
| Camera | Resolution | FOV | Format |
|---|---|---|---|
front_cam |
640Γ480 | 108Β° | H264 MP4 |
workspace_cam |
640Γ480 | 70Β° | H264 MP4 |
| Field | Description |
|---|---|
annotation.human.task_description |
Natural language task instruction |
| Parameter | Range |
|---|---|
| Object position | Workspace-aware random |
| Lighting intensity | 0.5 - 1.0 |
| Camera jitter | Β±2Β° |
| Object type | Random selection |
| Object color | Random selection |
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset("ganatrask/NOVA")
# Access an episode
episode = dataset[0]
print(episode.keys())
# ['observation.state', 'observation.images.front_cam', 'action', ...]
from datasets import load_dataset
dataset = load_dataset("ganatrask/NOVA")
python -m gr00t.train \
--dataset_repo_id ganatrask/NOVA \
--embodiment_tag reachy2 \
--video_backend decord \
--num_gpus 2 \
--batch_size 64 \
--max_steps 30000
python scripts/data_collector.py \
--episodes 100 \
--output reachy2_dataset \
--arm right \
--randomize-object \
--randomize-color \
--cameras front_cam workspace_cam
| Statistic | Value |
|---|---|
| Total episodes | 100 |
| Avg. episode length | ~150 steps |
| Collection rate | ~2 episodes/min |
| Total size | ~2 GB |
This dataset is released under CC-BY-4.0.
@misc{nova_dataset_2025,
title={NOVA Dataset: Reachy 2 Pick-and-Place Demonstrations},
author={ganatrask},
year={2025},
publisher={HuggingFace},
url={https://huggingface.co/datasets/ganatrask/NOVA}
}