Importing a New Asset

Importing a New Asset#

NVIDIA Omniverse relies on the Universal Scene Description (USD) file format to import and export assets. USD is an open source file format developed by Pixar Animation Studios. It is a scene description format optimized for large-scale, complex data sets. While this format is widely used in the film and animation industry, it is less common in the robotics community.

To this end, NVIDIA has developed various importers that allow you to import assets from other file formats into USD. These importers are available through Omniverse Kit workflows, and the URDF/MJCF Python importers can also be used from a standalone pip wheel:

URDF Importer - Import assets from URDF files.
MJCF Importer - Import assets from MJCF files.
Mesh Importer - Import assets from various file formats, including OBJ, FBX, STL, and glTF.

The recommended workflow from NVIDIA is to use the above importers to convert the asset into its USD representation. Once the asset is in USD format, you can use the Omniverse Kit to edit the asset and export it to other file formats. Isaac Sim includes these importers by default. They can also be enabled manually in Omniverse Kit.

Isaac Lab’s URDF and MJCF converter utilities first use the importer APIs from Isaac Sim when the full runtime is installed. In kit-less environments, install the standalone importer wheel as described in Standalone URDF/MJCF importers below. The Kit visualizer and GUI import dialogs still require an Omniverse Kit runtime.

An important note to use assets for large-scale simulation is to ensure that they are in instanceable format. This allows the asset to be efficiently loaded into memory and used multiple times in a scene. Otherwise, the asset will be loaded into memory multiple times, which can cause performance issues. For more details on instanceable assets, please check the Isaac Sim documentation.

Standalone URDF/MJCF importers#

The URDF and MJCF converter scripts can run without Isaac Sim when the standalone isaacsim-asset-isolated wheel is installed in the active environment. The wheel is not published on PyPI, so replace PACKAGE_INDEX_URL with the package index that hosts it:

uv pip install "isaacsim-asset-isolated>=6.0,<6.1" \
  --extra-index-url "PACKAGE_INDEX_URL"

After installing the wheel, run conversion in the kit-less environment. Optionally pass --viz newton (or rerun / viser) to preview the converted asset in a kit-less Isaac Lab visualizer:

python scripts/tools/convert_urdf.py \
  path/to/robot.urdf path/to/output_dir --merge_joints

python scripts/tools/convert_mjcf.py \
  path/to/model.xml path/to/output.usd --merge_mesh

If Isaac Sim is installed in the same environment, Isaac Lab uses the Isaac Sim importer extensions first. The standalone wheel is used only when the full Isaac Sim runtime is not available.

Using URDF Importer#

For using the URDF importer in the GUI, please check the documentation at URDF importer. For using the URDF importer from Python scripts, we include a utility tool called convert_urdf.py. This script creates an instance of UrdfConverterCfg which is then passed to the UrdfConverter class.

Note

The URDF importer was upgraded to version 3.0 in Isaac Sim 6, replacing the previous C++ binding-based API with a Python pipeline (urdf-usd-converter). Assets are now made instanceable by default — make_instanceable is no longer a configuration option. See the Migrating to Isaac Lab 3.0 for a full list of breaking changes.

The URDF importer has various configuration parameters that can be set to control the behavior of the importer. The default values for the importer’s configuration parameters are specified in the UrdfConverterCfg class, and they are listed below. We made a few commonly modified settings to be available as command-line arguments when calling the convert_urdf.py, and they are marked with * in the list. For a comprehensive list of the configuration parameters, please check the documentation at URDF importer.

Articulation and joint structure#

fix_base * - Whether to fix the base of the robot. This depends on whether you have a floating-base or fixed-base robot. The command-line flag is --fix_base where when set, the importer will fix the base of the robot, otherwise it will default to floating-base.
merge_fixed_joints * - Whether to merge the fixed joints. Usually, this should be set to True to reduce the asset complexity. The command-line flag is --merge_joints where when set, the importer will merge the fixed joints, otherwise it will default to not merging the fixed joints.
joint_drive - The configuration for the joint drives on the robot.
- drive_type - The drive type for the joints. This can be either "acceleration" or "force". We recommend using "force" for most cases.
- target_type - The target type for the joints. This can be either "none", "position", or "velocity". We recommend using "position" for most cases. Setting this to "none" will disable the drive and set the joint gains to 0.0.
- gains - The drive stiffness and damping gains for the joint. We support two ways to set the gains:
  - PDGainsCfg - To directly set the stiffness and damping. Both stiffness and damping accept a single float (applied uniformly).
  - NaturalFrequencyGainsCfg - To set the gains using the desired natural frequency response of the system. Deprecated in URDF importer 3.0 — use PDGainsCfg instead.

Geometry, collisions, and materials#

collision_from_visuals - Whether to create collision geometry from visual geometry when no explicit <collision> is defined for a link. Defaults to False.
collision_type - The collision shape simplification to apply. One of "Convex Hull" (default), "Convex Decomposition", "Bounding Sphere", or "Bounding Cube".
self_collision - Whether to activate self-collisions between links of the articulation. Defaults to False.
merge_mesh - Whether to merge meshes where possible to optimize the model. Defaults to False.
link_density - Default density in kg/m^3 for links whose <inertial> properties are missing. 0.0 (default) leaves densities unchanged.

Asset resolution and output#

ros_package_paths - List of ROS package name/path mappings used to resolve package:// URLs in the URDF. Each entry is a dict with keys name and path.
robot_type - Robot type applied by the USD robot schema. Defaults to "Default". Must be one of: "Default", "End Effector", "Manipulator", "Humanoid", "Wheeled", "Holonomic", "Quadruped", "Mobile Manipulators", "Aerial".
run_asset_transformer - Run the asset transformer to convert the flattened USD into a layered USD (interface USD + payloads). Defaults to True.
run_multi_physics_conversion - Also emit MuJoCo-compatible joint attributes alongside PhysX. Defaults to True.
debug_mode - Write intermediate conversion artifacts next to the output USD for inspection. Defaults to False.

Deprecated (no-op in URDF importer 3.0)#

The following options are retained for backwards compatibility but are ignored by the URDF importer 3.0. A warning is logged when they are set.

root_link_name - The link on which the PhysX articulation root was previously placed.
convert_mimic_joints_to_normal_joints - Convert mimic joints to normal joints during conversion.
replace_cylinders_with_capsules - Replace cylinder shapes with capsule shapes during conversion.

For more detailed information on the configuration parameters, please check the documentation for UrdfConverterCfg.

Example Usage#

In this example, we use the pre-processed URDF file of the ANYmal-D robot. To check the pre-process URDF, please check the file the anymal.urdf. The main difference between the pre-processed URDF and the original URDF are:

We removed the <gazebo> tag from the URDF. This tag is not supported by the URDF importer.
We removed the <transmission> tag from the URDF. This tag is not supported by the URDF importer.
We removed various collision bodies from the URDF to reduce the complexity of the asset.
We changed all the joint’s damping and friction parameters to 0.0. This ensures that we can perform effort-control on the joints without PhysX adding additional damping.
The <dont_collapse> URDF tag is no longer supported in URDF importer 3.0. Fixed joint merging is now a Python pre-processing step that merges all fixed joints when merge_fixed_joints is enabled. If you need to preserve a specific fixed joint, disable merge_fixed_joints entirely or restructure the URDF to use a non-fixed joint type (e.g. revolute with zero-range limits).

The following shows the steps to clone the repository and run the converter:

Linux

# clone a repository with URDF files
git clone git@github.com:isaac-orbit/anymal_d_simple_description.git

# go to top of the Isaac Lab repository
cd IsaacLab
# run the converter
python scripts/tools/convert_urdf.py \
  ../anymal_d_simple_description/urdf/anymal.urdf \
  source/isaaclab_assets/data/Robots/ANYbotics/ \
  --merge_joints \
  --joint_stiffness 0.0 \
  --joint_damping 0.0 \
  --joint_target_type none \
  --viz kit

Windows

:: clone a repository with URDF files
git clone git@github.com:isaac-orbit/anymal_d_simple_description.git

:: go to top of the Isaac Lab repository
cd IsaacLab
:: run the converter
python scripts\tools\convert_urdf.py ^
  ..\anymal_d_simple_description\urdf\anymal.urdf ^
  source\isaaclab_assets\data\Robots\ANYbotics\ ^
  --merge_joints ^
  --joint_stiffness 0.0 ^
  --joint_damping 0.0 ^
  --joint_target_type none ^
  --viz kit

Executing the above script will create a USD file inside the source/isaaclab_assets/data/Robots/ANYbotics/anymal/ directory (the subdirectory name is derived automatically from the robot name in the URDF):

anymal.usda - This is the main asset file.

Note

The URDF importer auto-deduplicates the per-robot subdirectory when it already exists. If you re-run the converter against the same usd_dir with a changed configuration (for example, flipping fix_base), the importer writes to a new numbered folder (anymal_1/, anymal_2/, …) rather than overwriting the previous output. usd_path reflects whichever folder the importer actually used. Delete stale subdirectories manually (or wipe usd_dir) if you do not want them to accumulate on disk.

The examples above pass --viz kit to open the converted asset in the Isaac Sim viewport, which requires a full Isaac Sim installation. Name a kitless backend instead – --viz newton, --viz rerun, or --viz viser – to preview the asset without Kit. Omit --viz to exit after the conversion completes.

In Isaac Sim, you can press play on the opened window to see the asset in the scene. The asset should fall under gravity. If it blows up, then it might be that you have self-collisions present in the URDF.

Using MJCF Importer#

Similar to the URDF Importer, the MJCF Importer also has a GUI interface. Please check the documentation at MJCF importer for more details. For using the MJCF importer from Python scripts, we include a utility tool called convert_mjcf.py. This script creates an instance of MjcfConverterCfg which is then passed to the MjcfConverter class.

The default values for the importer’s configuration parameters are specified in the MjcfConverterCfg class. The configuration parameters are listed below. We made a few commonly modified settings to be available as command-line arguments when calling the convert_mjcf.py, and they are marked with * in the list. For a comprehensive list of the configuration parameters, please check the documentation at MJCF importer.

Note

The MJCF importer was rewritten in Isaac Sim 5.0 to use the mujoco-usd-converter library. Settings such as import_sites, import_inertia_tensor, and make_instanceable are no longer needed — the converter now handles these automatically based on the MJCF file content.

Geometry, collisions, and materials#

merge_mesh * - Whether to merge meshes where possible to optimize the model. The command-line flag is --merge_mesh.
collision_from_visuals * - Whether to generate collision geometry from visual geometries. The command-line flag is --collision_from_visuals.
collision_type * - The collision shape simplification to apply. One of "Convex Hull" (default), "Convex Decomposition", "Bounding Sphere", or "Bounding Cube". The command-line flag is --collision_type.
self_collision * - Whether to activate self-collisions between links of the articulation. The command-line flag is --self_collision.

Articulation and physics#

fix_base - Whether to add a fixed joint between the world and the root rigid-body link. Defaults to False.
link_density - Default density in kg/m^3 for links whose <inertial> properties are missing in the MJCF. 0.0 (default) leaves densities unchanged.
import_physics_scene * - Import physics scene properties (gravity, time step, etc.) from the MJCF file. Defaults to False. The command-line flag is --import_physics_scene.

Actuator overrides#

MuJoCo models actuators as an affine transformation tau = gain @ control + bias. The following options override the values parsed from the MJCF on a per-actuator basis. Each defaults to None, which leaves the parsed values unchanged.

override_gain_type - The actuator gain type override (e.g. "fixed").
override_bias_type - The actuator bias type override (e.g. "affine").
override_gain_prm - The actuator gain parameter array override. Example for position control: [kp, 0, 0, 0, 0, 0, 0, 0, 0, 0].
override_bias_prm - The actuator bias parameter array override. Example for position control: [0, -kp, -kd, 0, 0, 0, 0, 0, 0, 0].

Asset resolution and output#

robot_type - Robot type applied by the USD robot schema. Defaults to "Default". Must be one of: "Default", "End Effector", "Manipulator", "Humanoid", "Wheeled", "Holonomic", "Quadruped", "Mobile Manipulators", "Aerial".
run_asset_transformer - Run the asset transformer to convert the flattened USD into a layered USD (interface USD + payloads). Defaults to True.
run_multi_physics_conversion - Convert compatible MuJoCo attributes to PhysX attributes (e.g. actuator gains). Defaults to True.
debug_mode - Write intermediate conversion artifacts next to the output USD for inspection. Defaults to False.

For more detailed information on the configuration parameters, please check the documentation for MjcfConverterCfg.

Example Usage#

In this example, we use the MuJoCo model of the Unitree’s H1 humanoid robot in the mujoco_menagerie.

The following shows the steps to clone the repository and run the converter:

Linux

# clone a repository with MJCF files
git clone git@github.com:google-deepmind/mujoco_menagerie.git

# go to top of the Isaac Lab repository
cd IsaacLab
# run the converter
python scripts/tools/convert_mjcf.py \
  ../mujoco_menagerie/unitree_h1/h1.xml \
  source/isaaclab_assets/data/Robots/Unitree/h1.usd \
  --merge_mesh \
  --viz kit

Windows

:: clone a repository with MJCF files
git clone git@github.com:google-deepmind/mujoco_menagerie.git

:: go to top of the Isaac Lab repository
cd IsaacLab
:: run the converter
python scripts\tools\convert_mjcf.py ^
  ..\mujoco_menagerie\unitree_h1\h1.xml ^
  source\isaaclab_assets\data\Robots\Unitree\h1.usd ^
  --merge_mesh ^
  --viz kit

Executing the above script will create the USD file inside the source/isaaclab_assets/data/Robots/Unitree/ directory:

h1.usd - This is the converted USD asset file.

Note

The MJCF importer auto-deduplicates the per-robot subdirectory when it already exists, matching the URDF importer’s behavior. If you re-run the converter against the same usd_dir with a changed configuration, the importer writes to a new numbered folder (h1_1/, h1_2/, …) rather than overwriting the previous output. usd_path reflects whichever folder the importer actually used. Delete stale subdirectories manually (or wipe usd_dir) if you do not want them to accumulate on disk.

The examples above pass --viz kit to open the converted asset in the Isaac Sim viewport, which requires a full Isaac Sim installation. Name a kitless backend instead – --viz newton, --viz rerun, or --viz viser – to preview the asset without Kit. Omit --viz to exit after the conversion completes.

Ensuring joint drives exist on every joint#

When an articulation uses an Isaac Lab actuator configuration, Newton resolves each configured joint’s target mode before constructing the solver. For an ImplicitActuatorCfg, stiffness-only, damping-only, both-gain, and zero-gain configurations select position, velocity, combined position/velocity, and effort modes respectively. A gain set to None keeps the corresponding imported USD value. An explicit actuator configuration uses effort mode because Isaac Lab computes its effort directly.

Joints without an Isaac Lab actuator configuration retain their imported USD target modes. Consequently, an asset with zero-gain USD drives no longer needs ensure_drives_exist solely to make configured joints actuate in Newton.

The ensure_drives_exist option remains useful when a workflow needs to author placeholder drives independently of an Isaac Lab actuator configuration. When enabled, it assigns a minimal placeholder stiffness (1e-3) to zero-gain drives before simulation startup so that every drive exists in the imported model.

Using Mesh Importer#

Omniverse Kit includes the mesh converter tool that uses the ASSIMP library to import assets from various mesh formats (e.g. OBJ, FBX, STL, glTF, etc.). The asset converter tool is available as an extension to Omniverse Kit. Please check the asset converter documentation for more details. However, unlike Isaac Sim’s URDF and MJCF importers, the asset converter tool does not support creating instanceable assets. This means that the asset will be loaded into memory multiple times if it is used multiple times in a scene.

Thus, we include a utility tool called convert_mesh.py that uses the asset converter tool to import the asset and then converts it into an instanceable asset. Internally, this script creates an instance of MeshConverterCfg which is then passed to the MeshConverter class. Since the mesh file does not contain any physics information, the configuration class accepts different physics properties (such as mass, collision shape, etc.) as input. Please check the documentation for MeshConverterCfg for more details.

Example Usage#

We use an OBJ file of a cube to demonstrate the usage of the mesh converter. The following shows the steps to clone the repository and run the converter:

Linux

# clone a repository with mesh files
git clone git@github.com:NVIDIA-Omniverse/IsaacGymEnvs.git

# go to top of the Isaac Lab repository
cd IsaacLab
# run the converter
python scripts/tools/convert_mesh.py \
  ../IsaacGymEnvs/assets/trifinger/objects/meshes/cube_multicolor.obj \
  source/isaaclab_assets/data/Props/CubeMultiColor/cube_multicolor.usd \
  --make-instanceable \
  --collision-approximation convexDecomposition \
  --mass 1.0

Windows

:: clone a repository with mesh files
git clone git@github.com:NVIDIA-Omniverse/IsaacGymEnvs.git

:: go to top of the Isaac Lab repository
cd IsaacLab
:: run the converter
python scripts\tools\convert_mesh.py ^
  ..\IsaacGymEnvs\assets\trifinger\objects\meshes\cube_multicolor.obj ^
  source\isaaclab_assets\data\Props\CubeMultiColor\cube_multicolor.usd ^
  --make-instanceable ^
  --collision-approximation convexDecomposition ^
  --mass 1.0

You may need to press ‘F’ to zoom in on the asset after import.

Similar to the URDF and MJCF converter, executing the above script will create two USD files inside the source/isaaclab_assets/data/Props/CubeMultiColor/ directory. Additionally, if you press play on the opened window, you should see the asset fall down under the influence of gravity.

If you do not set the --mass flag, then no rigid body properties will be added to the asset. It will be imported as a static asset.
If you also do not set the --collision-approximation flag, then the asset will not have any collider properties as well and will be imported as a visual asset.

Importing a New Asset

Contents

Importing a New Asset#

Standalone URDF/MJCF importers#

Using URDF Importer#

Articulation and joint structure#

Geometry, collisions, and materials#

Asset resolution and output#

Deprecated (no-op in URDF importer 3.0)#

Example Usage#

Using MJCF Importer#

Geometry, collisions, and materials#

Articulation and physics#

Actuator overrides#

Asset resolution and output#

Example Usage#

Ensuring joint drives exist on every joint#

Using Mesh Importer#

Example Usage#