Datasetpublic

hiangx/MetaFine_CoRL26

CoRL demos Training data for the MetaFine π0 baseline (T1–T5). Overview Task Variant tree Mixed episodes Mixed frames Cameras State FPS Control / obs T1 grasp_part cap / body / mixed 200 30,670 base + hand, 512×512, FOV ≈ 70° 9-DoF 30 pd_joint_delta_pos / rgb T2 grasp_move_mug left / right / forward / mixed 300 39,395 base + hand, 512×512, FOV ≈ 70° 9-DoF 30 pd_joint_delta_pos / rgb T3 toggle_switch_table red / blue / mixed 200 20,567 base + hand… See the full description on the dataset page: https://huggingface.co/datasets/hiangx/MetaFine_CoRL26.

sourceHugging Faceupdated 1mo agoView on Hugging Face

3likes3kdownloads

Dataset Card

CoRL demos

Training data for the MetaFine π0 baseline (T1–T5).

Overview

Task	Variant tree	Mixed episodes	Mixed frames	Cameras	State	FPS	Control / obs
T1 `grasp_part`	`cap` / `body` / `mixed`	200	30,670	base + hand, 512×512, FOV ≈ 70°	9-DoF	30	`pd_joint_delta_pos` / `rgb`
T2 `grasp_move_mug`	`left` / `right` / `forward` / `mixed`	300	39,395	base + hand, 512×512, FOV ≈ 70°	9-DoF	30	`pd_joint_delta_pos` / `rgb`
T3 `toggle_switch_table`	`red` / `blue` / `mixed`	200	20,567	base + hand, 512×512, FOV ≈ 70°	9-DoF	30	`pd_joint_delta_pos` / `rgb`
T4 `put_blocks_into_boxes`	`red` / `blue` / `green` / `mixed`	300	156,151	base + hand, 512×512, FOV ≈ 57°	9-DoF	30	`pd_joint_delta_pos` / `rgb`
T5 `insert_letter`	`C` / `o` / `R` / `L` / `mixed`	400	115,894	base + hand, 512×512, FOV ≈ 70°	9-DoF	30	`pd_joint_delta_pos` / `rgb`

Per-variant episode counts: T1 cap/body = 100 each; T2 left/right/forward = 100 each; T3 red/blue = 100 each; T4 red/blue/green = 100 each; T5 C/o/R/L = 100 each (mixed merges them).

Checkpoint mapping

LeRobot dataset	Trained checkpoint
`grasp_part/mixed/lerobot`	`outputs/pi0_grasp_mixed_f70/checkpoints/030000/pretrained_model`
`grasp_move_mug/mixed/lerobot`	`outputs/pi0_grasp_move_mug_mixed/checkpoints/030000/pretrained_model`
`toggle_switch_table/mixed/lerobot`	`outputs/pi0_toggle_mixed/checkpoints/030000/pretrained_model`
`put_blocks_into_boxes/mixed/lerobot`	`outputs/pi0_put_blocks_mixed/checkpoints/030000/pretrained_model`
`insert_letter/mixed/lerobot`	`outputs/pi0_insert_letter_mixed/checkpoints/030000/pretrained_model`

Layout

demos/CoRL/
  grasp_part/                 # see grasp_part/README.md
  grasp_move_mug/             # see grasp_move_mug/README.md (T2)
  toggle_switch_table/        # see toggle_switch_table/README.md (T3)
  put_blocks_into_boxes/      # see put_blocks_into_boxes/README.md (T4)
  insert_letter/              # see insert_letter/README.md (T5)
  lerobot/                    # legacy symlinks into the trees above
    grasp_part_cap_n100_f70   → ../grasp_part/cap/lerobot
    grasp_part_body_n100_f70  → ../grasp_part/body/lerobot
    grasp_part_mixed_n200_f70 → ../grasp_part/mixed/lerobot

Each variant folder typically holds:

—*.h5 / *.json — motion-plan source trajectories
—*.rgb.pd_joint_delta_pos.physx_cpu.h5 (+ .json) — RGB replay used for LeRobot conversion
—lerobot/ — HuggingFace LeRobot dataset (meta/, data/, videos/)
—raw/ (T4 / T5) — optional ManiSkill raw episode dump from recording

Eval sensor configs mirror the RGB-replay metadata: see eval/configs/*.yaml.