Note

This is a Hugging Face dataset. For large datasets, ensure huggingface_hub>=1.1.3 to avoid rate limits. Learn more in the Hugging Face integration docs.

Dataset Card for Grabette Tactile Full Synced (48 episodes)#

Grabette Tactile Full Synced preview

This is a FiftyOne dataset with 48 samples. Each sample is one episode of the task “Pick up the white cube and put it in the correct cup”, recorded with robot type grabette.

Installation#

pip install -U fiftyone

Usage#

Load from the Hub#

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

dataset = load_from_hub("Voxel51/grabette-tactile-full-synced-48ep")

session = fo.launch_app(dataset)

Download a snapshot and load it locally#

import fiftyone as fo
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="Voxel51/grabette-tactile-full-synced-48ep",
    repo_type="dataset",
    local_dir="grabette-tactile-full-synced-48ep",
)

dataset = fo.Dataset.from_dir(
    dataset_dir="grabette-tactile-full-synced-48ep",
    dataset_type=fo.types.LeRobotDataset,
    name="grabette_tactile_full_synced",
)

session = fo.launch_app(dataset)

Dataset Details#

Dataset Description#

48 episodes (19,353 frames) at 50 fps on robot type grabette, all with the task “Pick up the white cube and put it in the correct cup”. Each frame has two 960x720 H.264 camera streams, an 8-D action, an is_lost flag, and two int16 tactile taxel grids (6x6 and 4x8). The source is a LeRobot v3.0 repo.

Shared by: CarolinePascal on Hugging Face
Language(s): en
License: apache-2.0

Dataset Sources#

Repository: https://huggingface.co/datasets/CarolinePascal/grabette-tactile-full-synced

Dataset Structure#

Topology. media_type: multimodal, 48 samples, one per episode, 50 fps, 145-537 frames (2.9-10.7 s). Samples hold a media_reference to the per-frame parquet and 2 video streams.

Field	FiftyOne type	Description
`media_reference`	`LeRobotEpisodeReference`	Pointers into `data/` parquet and the 2 video streams
`episode_index`	`IntField`	Episode index (0-47)
`task`	`StringField`	The single task instruction
`tasks`	`ListField(StringField)`	All task strings of the episode
`length`	`IntField`	Frames in the episode
`duration`	`FloatField`	Episode duration in seconds
`robot_type`	`StringField`	`grabette`
`fps`	`FloatField`	50
Data behind `media_reference` (per-frame, from `meta/info.json`):

Feature	Shape	Description
`observation.images.right_cam0`, `right_cam1`	720x960x3	h264 yuv420p video
`action`	(8,)	`right_x, right_y, right_z, right_ax, right_ay, right_az, right_proximal, right_distal`
`is_lost`	(1,)	`right_is_lost` flag
`observation.tactile.right_sensor_1`	(6, 6) int16	Tactile taxel grid
`observation.tactile.right_sensor_2`	(4, 8) int16	Tactile taxel grid

Label types and why. The FiftyOne LeRobot importer only recognizes video and image features, so the int16 tactile grids are not imported as FiftyOne fields. Multimodal samples have no frame fields, so no tactile values become FiftyOne fields. The arrays stay unchanged in data/chunk-000/file-000.parquet and ship with the exported LeRobot repo.

dataset.info. lerobot entry with format, format_major, episode_count, imported_episode_count, skipped_episodes (empty).

Parsing decisions.

All 48 episodes imported, from a single shard (240 MB). No source file was modified.
Units, taxel layout, and physical meaning of the tactile values are not stated in the source card.

Dataset Creation#

Source Data#

Data Collection and Processing#

The source card states only that it was created using LeRobot.

Annotations#

Annotation process#

A single task string is stored in meta/tasks.parquet.