datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hand_tracking_pv_carton_dual_view
hand_tracking_pv_carton_dual_view
The raw PressureVision carton recordings. 54 episodes, 13,369 frames, 10 Hz,
front and side RGB, six-joint state and action. LeRobot format v3.0.
This is the source the rest of the carton line derives from, which is why it
is here: hand_tracking_pv_carton_phase_b (30 reviewed episodes) and its
_train24 / _train24_onset_qreplace splits are all built from these
recordings by
build_phase_b_training_dataset.py.
A derived set can be rebuilt from this… See the full description on the dataset page: https://huggingface.co/datasets/stevenzenith/hand_tracking_pv_carton_dual_view.hand_tracking_pick_placeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 10,
"features": {
"action": {
"dtype": "float32",
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_roll.pos",
"gripper.pos"
],
"shape": [
6… See the full description on the dataset page: https://huggingface.co/datasets/stevenzenith/hand_tracking_pick_place.hand_tracking_pv_carton_middle_standard
hand_tracking_pv_carton_middle_standard
13 episodes, 2,902 frames, 10 Hz, front and side RGB, six-joint state and
action. LeRobot format v3.0.
Recorded with the middle-pose convention, and used together with
hand_tracking_pv_carton_phase_b
to fine-tune
act_carton_middle_labels_5k
from a 50k ACT base. The exact episode selection and the hand-made recovery
labels live in
training/train_middle_labeled_act.py:
episodes 0-7, 10 and 11 of this dataset, plus 1, 9, 14, 20 and 26 of… See the full description on the dataset page: https://huggingface.co/datasets/stevenzenith/hand_tracking_pv_carton_middle_standard.g1_handtrackinghand_tracking_pv_carton_phase_bThis dataset was created using LeRobot.
Dataset Description
Private Phase B training dataset for adaptive 250 g paper-carton grasping with an SO-101 arm. It contains 30 reviewed formal episodes at 10 Hz with front and side RGB video, six-joint observations, and six-joint actions. Audit-only PressureVision/teacher fields and constant grip-context dimensions are intentionally excluded. See source_episode_map.csv and phase_b_training_manifest.json for provenance and coverage… See the full description on the dataset page: https://huggingface.co/datasets/stevenzenith/hand_tracking_pv_carton_phase_b.