Data Assets

Data Assets#

EmbodiChain provides a comprehensive set of pre-built data assets hosted on HuggingFace, covering robots, end-effectors, objects, scenes, materials, and more. Assets are automatically downloaded on first use, but you can also pre-download them using the built-in CLI tool.

Data Root Directory#

By default, assets are stored in ~/.cache/embodichain_data. You can override this by setting the EMBODICHAIN_DATA_ROOT environment variable:

export EMBODICHAIN_DATA_ROOT=/mnt/shared/embodichain_data

Similarly, the dataset recording root (used by LeRobotRecorder) defaults to ~/.cache/embodichain_datasets and can be overridden with:

export EMBODICHAIN_DATASET_ROOT=/mnt/shared/embodichain_datasets

Download CLI#

The embodichain.data module provides a command-line interface for managing assets.

List Available Assets#

# List all assets across every category
python -m embodichain.data list

# List assets in a specific category
python -m embodichain.data list --category robot

The output shows each asset name and whether it has already been downloaded (✓):

[robot] (18 assets)
  [✓] ABB
  [ ] ARX5
  [ ] Agile
  [✓] Aubo
  ...

Data root: /home/user/.cache/embodichain_data

Download Assets#

# Download a single asset by name
python -m embodichain.data download --name CobotMagicArm

# Download all assets in a category
python -m embodichain.data download --category robot

# Download everything
python -m embodichain.data download --all

Downloaded files are saved to <data_root>/download/ and automatically extracted to <data_root>/extract/. Non-zip assets (e.g. .glb files) are copied into the extract directory.

Asset Categories#

Category

Description

Examples

robot

Robot URDF models

CobotMagicArm, Franka, UniversalRobots, UnitreeH1

eef

End-effector / gripper models

DH_PGC_140_50, Robotiq2F85, InspireHand

obj

Manipulable objects and furniture

ShopTableSimple, CoffeeCup, TableWare

scene

Full scene environments

SceneData, EmptyRoom

materials

Rendering materials, IBL, and backgrounds

SimResources, CocoBackground

w1

DexForce W1 humanoid robot and components

DexforceW1V021, DexforceW1ChassisV021

demo

Demo scene data

ScoopIceNewEnv, MultiW1Data

Using Assets in Code#

Use get_data_path to resolve asset paths in your configuration. It accepts a relative path in the format <AssetClassName>/<subpath>:

from embodichain.data import get_data_path

# Resolves to the URDF file, downloading if necessary
urdf_path = get_data_path("CobotMagicArm/CobotMagicWithGripperV100.urdf")

get_data_path resolves paths in the following order:

  1. Absolute path — returned as-is.

  2. Local data root — if the file exists under EMBODICHAIN_DATA_ROOT, it is returned immediately without triggering a download.

  3. Data-class download — falls back to the registered asset class, which downloads and extracts the asset from HuggingFace.

You can also instantiate asset classes directly:

from embodichain.data.assets import CobotMagicArm

dataset = CobotMagicArm()
print(dataset.extract_dir)  # Path to extracted files

Adding Custom Assets#

To add a new asset:

  1. Create a class in the appropriate file under embodichain/data/assets/ (e.g., robot_assets.py for a robot):

class MyRobot(EmbodiChainDataset):
    def __init__(self, data_root: str = None):
        data_descriptor = o3d.data.DataDescriptor(
            os.path.join(EMBODICHAIN_DOWNLOAD_PREFIX, robot_assets, "MyRobot.zip"),
            "<md5_checksum>",
        )
        prefix = type(self).__name__
        path = EMBODICHAIN_DEFAULT_DATA_ROOT if data_root is None else data_root
        super().__init__(prefix, data_descriptor, path)

  1. The class is automatically discovered by the download CLI and get_data_path — no additional registration is needed.

Contents