Evanxw0715/BiCAT-BimanualYAM-Dataset

BiCAT YAM Dataset

Seven teleoperated bimanual YAM robot tasks in LeRobot v3.0 format. Each task folder is a complete dataset with its own meta/, data/, and videos/ directories. Episode indices start at 0 within each task.

Folder	Task instruction	Episodes	Frames	FPS	Size
box/	Put all the metal parts in the box and lift the box.	100	73,252	30	1.24 GB
bottles/	Put all bottles in dustbin.	100	123,558	30	3.87 GB
pen/	Grasp the marker and cap then close.	105	46,924	30	1.24 GB
plate/	Use the appropriate arm to place the bowl on the plate.	100	44,120	30	1.10 GB
handover/	Handover the bottle.	105	63,459	30	1.26 GB
platev1/	Use the appropriate arm to place the bowl on the plate.	110	45,920	30	1.11 GB
blocks/	Put all blocks in the box and move forward the box.	104	55,392	30	0.92 GB

Total dataset size: approximately 10.74 GB, counting files under box/, bottles/, pen/, plate/, handover/, platev1/, and blocks/. GB uses decimal units (1 GB = 1,000,000,000 bytes).

.
├── README.md
├── box/
│   ├── data/
│   ├── meta/
│   └── videos/
├── bottles/
│   ├── data/
│   ├── meta/
│   └── videos/
├── pen/
│   ├── data/
│   ├── meta/
│   └── videos/
├── plate/
│   ├── data/
│   ├── meta/
│   └── videos/
├── handover/
│   ├── data/
│   ├── meta/
│   └── videos/
├── platev1/
│   ├── data/
│   ├── meta/
│   └── videos/
└── blocks/
    ├── data/
    ├── meta/
    └── videos/

All seven datasets use the bi_yam_follower robot. Their features have the same names and shapes:

Feature	Shape	Description
observation.images.left, observation.images.right	480 × 640 × 3	RGB video, 30 FPS
observation.images.top	360 × 640 × 3	RGB video, 30 FPS
observation.state_joint_angles, action_joint_angles	14	Six joints and one gripper per arm
observation.state_eef_absolute, action_eef_absolute, action_eef_delta	16	Position, quaternion, and gripper per arm

The 100 bottles/ episodes are selected demonstrations from three recording sessions (39 + 50 + 11). Combined episode 11 uses original bottlev2 episode 5 in place of original bottle episode 15. The videos were rebuilt from the original recordings; episode frame counts and sampled frames were checked against those recordings.

The 105 pen/ episodes are selected demonstrations from two recording sessions: 47 from penv1_filtered and 58 from penv2_filtered. The sessions were merged into one dataset, with episode indices renumbered from 0 to 104. All episodes use the instruction “Grasp the marker and cap then close.”

The 100 plate/ episodes are selected demonstrations from two recording sessions: 50 from plate_filtered_filtered and 50 from platev1_filtered. They were merged into one dataset, with episode indices renumbered from 0 to 99. All episodes use the instruction “Use the appropriate arm to place the bowl on the plate.” The 105 handover/ episodes are selected demonstrations from four recording sessions: 20 from handover, 13 from handoverv1, 58 from handoverv2, and 14 from handoverv3. The sessions were merged into one dataset, two failed demonstrations were removed, and episode indices were renumbered from 0 to 104. All episodes use the instruction “Handover the bottle.” The 110 platev1/ episodes are selected demonstrations: 50 from platev1_filtered (episodes 0–49) and 60 from plate_right_combined_filtered (episodes 50–109), which itself merges the plate_right and plate_right_v1 sessions with five failed demonstrations removed. Episode indices run from 0 to 109. All episodes use the instruction “Use the appropriate arm to place the bowl on the plate.” The 104 blocks/ episodes are selected demonstrations from two recording sessions: 21 from blocks (episodes 0–20) and 83 from blocksv1 (episodes 21–103). The sessions were merged into one dataset, six failed demonstrations were removed (original blocks episodes 0, 1, 4 and 5, and original blocksv1 episodes 39 and 83), and episode indices were renumbered from 0 to 103. All episodes use the instruction “Put all blocks in the box and move forward the box.”

Load a task locally

The LeRobot loader expects a dataset root containing meta/info.json. Point it at one task folder after downloading the repository:

from pathlib import Path
from huggingface_hub import snapshot_download
from lerobot.datasets.lerobot_dataset import LeRobotDataset

snapshot = Path(snapshot_download(
    repo_id="Evanxw0715/BiCAT-BimanualYAM-Dataset",
    repo_type="dataset",
))
box = LeRobotDataset("local/box", root=snapshot / "box")
bottles = LeRobotDataset("local/bottles", root=snapshot / "bottles")
pen = LeRobotDataset("local/pen", root=snapshot / "pen")
plate = LeRobotDataset("local/plate", root=snapshot / "plate")
handover = LeRobotDataset("local/handover", root=snapshot / "handover")
platev1 = LeRobotDataset("local/platev1", root=snapshot / "platev1")
blocks = LeRobotDataset("local/blocks", root=snapshot / "blocks")

Each task has its own episode numbering and task instruction. The videos are grouped into files; episode video locations and timestamps are recorded in meta/episodes/.