Datasets:

WorldEngineAI
/

WEB-Dataset

Name: WEB-Dataset
Creator: WorldEngineAI
License: https://choosealicense.com/licenses/cc-by-nc-4.0/

Modalities:

Video

Size:

100K - 1M

Libraries:

Datasets

License:

Dataset card Data Studio Files Files and versions

xet

Community

Request access to the dataset

Please complete the form below. We will review your request as soon as possible.

By requesting access, you agree to use this dataset only for permitted research purposes and not to redistribute it.

WorldEngine Bimanual Dataset for Post-training

A large-scale, language-annotated real-robot bimanual manipulation dataset for post-training robotics foundation models. It spans 90 everyday manipulation tasks collected with a bimanual YAM follower arm teleoperated by a GELLO leader, recording joint state, action, and three synchronized camera streams at 60 Hz.

Shared lineage, different story. This dataset shares its hardware, teleoperation setup, and recording pipeline with the RSS 2026 Post-training Challenge dataset (3 benchmark tasks with expert / failure / human-in-the-loop subsets). Where that release targets a focused HIL challenge, this release is a broad, diverse pre/post-training corpus: 90 tasks of expert demonstrations enriched with hierarchical natural-language annotations — one task-level prompt plus per-frame subtask labels.

Scale

Statistic	Value
Tasks	90
Episodes	50,266
Frames	199,332,169 (about 199M)
Duration	about 923 hours @ 60 Hz
Cameras	3 (overhead + dual wrist)

Dataset Structure

The dataset follows the LeRobot v3.0 format. Each of the 90 tasks is a top-level folder:

<task>/
  meta/
    info.json        # codebase_version v3.0, robot_type "yam", fps 60, features
    tasks.parquet    # the natural-language task prompt (the instruction)
    episodes/...     # per-episode metadata
    stats.json
  data/chunk-000/file-*.parquet    # per-episode frames (state, action, subtask, ...)
  videos/<camera>/chunk-000/file-*.mp4   # AV1-encoded RGB video

These are all expert teleoperated demonstrations (no failure / rollout subsets).

Key Features

14-dimensional joint state & action (bimanual: 7 per arm, gripper included).
Three RGB camera streams — observation.images.cam_high (overhead), observation.images.cam_left_wrist, observation.images.cam_right_wrist — stored as AV1 MP4 at 60 Hz.
Hierarchical language annotations:
- Task prompt — a natural-language instruction per task (in meta/tasks.parquet), e.g. "Plug cable connectors into an extension board.".
- Per-frame subtask — a subtask column in the data parquet giving the fine-grained step each frame belongs to, e.g. for cable-assembling: "Insert the USB cable into its socket." → "Insert the Double-cable into its socket." → …
robot_type: yam.

Subtask annotation notes

Subtask labels are derived from per-episode video-second annotations and mapped onto the dataset's frames. Because the LeRobot frames are a (nearest-neighbor) synchronized subset of the original recording, segment boundaries are linearly scaled to each episode's actual frame count (boundary error is a few frames, < ~0.5 s).
subtask = "" (empty string) marks frames with no applicable label: lead-in before the first segment, gaps, "outlier behavior" segments, and episodes that are not yet annotated.
45,933 / 50,266 episodes (~91%) carry subtask labels; the remainder have an empty subtask (annotation pending). See the task list below.

Usage

Each task is a subfolder of this single repository (not a separate repo), so load a task by downloading its subfolder and pointing root at it:

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

REPO = "WorldEngineAI/Bimanual-Dataset-for-Posttraining"
task = "cable-assembling"

# download only this task's subfolder
local = snapshot_download(REPO, repo_type="dataset", allow_patterns=f"{task}/*")

ds = LeRobotDataset(REPO, root=f"{local}/{task}")
print(ds.num_episodes, ds.num_frames)

To fetch only metadata/parquet and skip the large videos, add ignore_patterns="*.mp4" to snapshot_download.

The data also reads with standard tooling (PyArrow / pandas for parquet, an AV1-capable decoder such as torchcodec / PyAV for the videos).

Tasks & Prompts

Task	Prompt
`accessory-return`	Hang the accessory on the jewelry stand.
`arrange-shoes-in-pairs`	Arrange all shoes into matching pairs.
`baby-clothes-sorting`	Sort baby clothes into organized piles.
`badminton-clean-up`	Put the badminton shuttlecock into the ball bucket.
`bag-packing-with-zipper-closure`	Pack items into a bag and close the zipper.
`bagging-groceries`	Bag groceries into shopping bags.
`balance-scale`	Balance objects on a scale by putting weights.
`bowl-pouring`	Pour contents from one bowl into another.
`bracelet-return`	Return the bracelets to the jewelry stand.
`breakroom-essentials-organization`	Organize breakroom supplies and essentials in the cup and condiment station.
`brick-stacking`	Stack bricks into pairs.
`bussing-table`	Clear and bus dishes and items from a dining table.
`cable-assembling`	Plug cable connectors into an extension board.
`clean-up-board-game`	Remove chess pieces from chess board.
`clothes-rack-return`	Hang clothes on the rack.
`coffee-pods-organization`	Organize coffee pods into their storage carousel.
`coffee-pods-retrieval`	Retrieve all coffee pods from the storage carousel.
`color-sorting-items`	Sort items into plates with the same color.
`connect-pipe-segments`	Connect pipe segments together into a complete unit.
`cooking-utensils-organization`	Organize cooking utensils neatly into the container.
`decorate-a-cake_toy_version`	Decorate a toy cake with toppings.
`desk-rack-organization`	Organize items on a desk rack.
`dice-roll-block`	Roll dice and sort blocks with the corresponding color and amount.
`dinnerware-organization`	Sort dinnerware plates and bowls by color.
`dry-asparagus`	Dry fresh asparagus with the paper towel and put them on the plate.
`fill-piggy-bank`	Insert coins into a piggy bank through its opening.
`grocery-bag-folding`	Fold grocery bags and stack them into compact bundles.
`hang-baseball-hats`	Hang baseball hats on a rack.
`hanoi-tower-ring-placement`	Solve the Tower of Hanoi puzzle by moving rings.
`hardware-sorting`	Sort bolts and screws by types labelled on the plate.
`insert-mouse-battery`	Insert a battery into a mouse.
`jenga-game-preparation`	Set up a Jenga tower.
`jewlery-rings-return`	Return jewelry rings to their storage plate.
`lift-a-pot`	Lift a pot and put it on the shelf.
`load-dishware-rack`	Load dishes into a dishwasher rack.
`magnetic-structure`	Build magnetic components to match the sample structure.
`mechanical-lock-puzzle`	Open the mechanical tool box and remove the middle layer.
`mining-balls`	Sort balls from a pile into plates labelled with color.
`multiple-dice-ordering`	Roll and arrange dice by their numeric values in an ascending order.
`office-desk-rack-setup`	Set up and organize an office desk rack.
`open-lock-box`	Open a locked box with a key.
`open-sealed-box-and-take-something-out`	Open a sealed box and remove contents.
`organize-office-supplies`	Organize office supplies into their designated areas.
`pack-bento-box`	Pack food items into a bento box compartments.
`pack-lunch-boxes`	Pack lunch containers with food items.
`pack-sports-bag`	Pack a sports bag with equipment and gear.
`pack-suitcase-for-trip`	Pack items into suitcase.
`paddle-bouncing-a-ping-pong-ball`	Bounce a ping pong ball on a paddle.
`paper-towel-setup`	Set up a paper towel holder.
`place-books-into-shelves`	Place books vertically onto shelves.
`pour-instant-coffee`	Pour instant coffee in a cup.
`prepare-new-shoes`	Unpack new shoes for wearing.
`prepare-sandwich`	Prepare a sandwich by assembling ingredients.
`push-buttons-in-sequence`	Push buttons into a sequence.
`restaurant-takeout`	Prepare takeout order and package it.
`return-necklaces`	Return necklaces to the jewelry rack.
`return-new-shoes`	Return shoes to the shoe box.
`roll-dice-write-on-blackboard`	Roll dice and check the result on a blackboard.
`scoop-with-containers`	Scoop items into containers using a utensil.
`seal-water-bottle-cap`	Seal a water bottle by closing its cap.
`separate-sports-equipment`	Sort sports equipment to their plates.
`set-up-board-game`	Set up all components of a chess board.
`set-up-checkerboard`	Set up a checkerboard with game pieces.
`shape-sorter`	Sort shapes into their matching openings.
`shopping-basket-loading`	Load items into a shopping basket.
`simple-puzzle`	Solve a simple jigsaw puzzle.
`sock-matching-and-folding`	Match socks into pairs and fold them.
`sort-fruits-and-vegetables`	Sort fruits and vegetables into categories by type.
`sort-kitchen-utensils`	Organize kitchen utensils by type into the container.
`sort-lego-by-color`	Organize LEGO pieces by color.
`sort-toy-vehicles`	Sort toy vehicles by type.
`stack-3d-prints`	Stack 3D printed objects vertically.
`stack-nesting-cups`	Stack cups into nested formations.
`tableware-stacking`	Stack tableware plates and bowls.
`tennis-ball-cleanup`	Clean up and collect tennis balls into their tube.
`tennis-preparation`	Arrange tennis balls onto a tray.
`tool-collection-organization`	Organize tools in a toolbox.
`towel-folding-and-stacking`	Fold towels and stack them neatly.
`tower-of-hanoi-game`	Play the Tower of Hanoi puzzle game.
`toy-robot`	Assemble a toy robot.
`transfer-items-with-a-tray`	Transfer items using a tray.
`transfer-storage-items`	Transfer items between storage containers.
`trash-sorting`	Sort trash into recycling and trash bins.
`unload-dish-rack`	Unload dishes from a dishwasher rack.
`untie-shoelaces`	Untie shoelaces on shoes.
`usb-cable-disassembling`	Unplug USB cables from the extension board.
`usb-plugging`	Plug USB cables into their ports.
`vacuum-dirty-surface`	Vacuum a dirty surface to clean it.
`whiteboard-supplies-organization`	Organize whiteboard markers and erasers.
`wrap-a-toy-for-storage`	Wrap a toy in a piece of cloth for storage.

License

Released under CC BY-NC 4.0.

Citation

TODO: add BibTeX once the paper is available.

Downloads last month: 1,690

Total file size:

2.36 TB