easyminnn/opentouch-lerobot

OpenTouch (LeRobot v2.0)

2,958 egocentric human-hand demonstrations across 26 real-world scenes with a 16×16 pressure grid, converted to LeRobot v2.0.

Part of a set of tactile manipulation datasets converted to a single common LeRobot v2.0 layout so they can be mixed in one training run.

At a glance

Episodes	2,958
Frames	324,072 (30 fps)
Tasks	26 — one per source recording (eat_mcdonalds, fablab_ml_p1, grocery_plant, …)
Video	ego_view (observation.images.main) — 640×480, Aria egocentric
Tactile	array — observation.tactile_grid, shape [16, 16]
Action	57-D human hand pose (HumanHandPoseLeRobotDataset layout)
Tactile type	array, needs a taxel encoder

Notes

• Pressure is normalised and sign-corrected. The raw source grid is an ADC reading where 3072 means no contact and lower means more pressure. This conversion stores (per-cell 99th-percentile baseline − raw) / baseline, clipped to [0, 1], so higher is now more pressure — the opposite of the raw files. Do not re-invert.
• The 57-D action vector is partially masked. info.json carries action_valid_slots: wrist orientation and camera pose are not measured by this source, so those slots are zero-filled and should be masked out of the loss rather than treated as targets. Only the marked slots (wrist position and finger landmarks) are real.
• Frame rate verified from per-frame timestamps: median 30.0 Hz across all 26 recordings.

Format

GR00T-flavoured LeRobot v2.0: per-episode Parquet under data/chunk-{:03d}/ and per-episode MP4 under videos/chunk-{:03d}/{video_key}/, with meta/info.json, meta/episodes.jsonl, meta/tasks.jsonl and meta/modality.json.

from datasets import load_dataset          # metadata / parquet only
# or read directly:
import pyarrow.parquet as pq, json
info = json.load(open("meta/info.json"))
tbl  = pq.read_table("data/chunk-000/episode_000000.parquet")

meta/modality.json names the tactile stream and the state/action slot layout, so the dataset can be loaded by an Isaac-GR00T style loader without extra configuration.

Provenance & license

This is a format conversion of rayxsong/opentouch (https://huggingface.co/datasets/rayxsong/opentouch), released under CC-BY-4.0. All sensor data originates from the authors of that dataset; please cite their work. No frames were dropped or re-timed beyond what is listed under Conversion above.