Local Installation

Isaac Lab installation is available for Windows and Linux. This guide explains the recommended installation methods.

Note

Isaac Lab 3.0 supports kit-less installation. You can install and use Isaac Lab with the Newton physics backend without installing Isaac Sim. Simply clone Isaac Lab and run:

./isaaclab.sh --install   # or ./isaaclab.sh -i

This installs the core Isaac Lab packages and the Newton physics backend. Isaac Sim is not required for this mode. See Kit-less Installation below for which features are available without Isaac Sim.

When you need full simulation features — including PhysX, ROS, URDF/MJCF importers — install Isaac Sim via pip (see the Installation using Isaac Sim Pip Package guide).

Caution

We have dropped support for Isaac Sim versions 5.1.0 and below. We recommend using the latest Isaac Sim 6.0.0 release to benefit from the latest features and improvements.

For more information, please refer to the Isaac Sim release notes.

System Requirements

General Requirements

For detailed requirements, please see the Isaac Sim system requirements. The basic requirements are:

• OS: Ubuntu 22.04 (Linux x64) or Windows 11 (x64)

• RAM: 32 GB or more

• GPU VRAM: 16 GB or more (additional VRAM may be required for rendering workflows)

Isaac Sim is built against a specific Python version, making it essential to use the same Python version when installing Isaac Lab. The required Python version is as follows:

• For Isaac Sim 6.X, the required Python version is 3.12.

Driver Requirements

Drivers other than those recommended on Omniverse Technical Requirements may work but have not been validated against all Omniverse tests.

• Use the latest NVIDIA production branch driver.

• On Linux, version 580.65.06 or later is recommended, especially when upgrading to Ubuntu 22.04.5 with kernel 6.8.0-48-generic or newer.

• On Spark, version 580.95.05 is recommended.

• On Windows, version 580.88 is recommended.

• If you are using a new GPU or encounter driver issues, install the latest production branch driver from the Unix Driver Archive using the .run installer.

DGX Spark: details and limitations

The DGX spark is a standalone machine learning device with aarch64 architecture. As a consequence, some features of Isaac Lab are not currently supported on the DGX spark. The most noteworthy is that the architecture requires CUDA ≥ 13, and thus the cu13 build of PyTorch or newer. Other notable limitations with respect to Isaac Lab include…

1. SkillGen is not supported out of the box. This is because cuRobo builds native CUDA/C++ extensions that requires specific tooling and library versions which are not validated for use with DGX spark.

2. Extended reality teleoperation tools such as OpenXR is not supported. This is due to encoding performance limitations that have not yet been fully investigated.

3. SKRL training with JAX has not been explicitly validated or tested in Isaac Lab on the DGX Spark. JAX provides pre-built CUDA wheels only for Linux on x86_64, so on aarch64 systems (e.g., DGX Spark) it runs on CPU only by default. GPU support requires building JAX from source, which has not been validated in Isaac Lab.

4. Livestream and Hub Workstation Cache are not supported on the DGX spark.

5. Running Cosmos Transfer1 is not currently supported on the DGX Spark.

Note

Build prerequisites on aarch64: Some Python packages (notably imgui-bundle and quadprog) do not ship pre-built wheels for aarch64 and are compiled from source during installation. This requires Python 3.12, OpenGL, and X11 development headers to be installed on the system:

sudo apt install python3.12-dev libgl1-mesa-dev libx11-dev libxcursor-dev libxi-dev libxinerama-dev libxrandr-dev

Without these packages, the build will fail with a CMake error about missing OPENGL_opengl_LIBRARY, OPENGL_glx_LIBRARY, and OPENGL_INCLUDE_DIR.

Troubleshooting

Please refer to the Linux Troubleshooting to resolve installation issues in Linux.

You can use Isaac Sim Compatibility Checker to automatically check if the above requirements are met for running Isaac Sim on your system.

Kit-less Installation

Isaac Lab can be installed and used without Isaac Sim using the kit-less mode. This is the fastest way to get started and is ideal for users who only need the Newton physics backend.

# Clone Isaac Lab
git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

# Install Isaac Lab (Newton backend, no Isaac Sim required)
./isaaclab.sh --install   # or ./isaaclab.sh -i

# Kickoff training with Newton physics and Newton visualizer
./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py \
--task=Isaac-Cartpole-Direct-v0 \
--num_envs=16 --max_iterations=10 \
presets=newton --visualizer newton

Features available in kit-less mode (Newton backend, no Isaac Sim):

• Newton physics simulation (GPU-accelerated, including MuJoCo-Warp solver)

• All manager-based and direct RL environments that support Newton

• RL training with SKRL, RSL-RL, and other frameworks

• Robot assets compatible with Newton

Features that require Isaac Sim:

• PhysX physics backend

• Isaac Sim RTX rendering (not ovrtx)

• Kit visualizer

• Photorealistic rendering workflows

• ROS / ROS2 integration

• URDF and MJCF importers (GUI-based)

• Deformable objects and surface gripper (PhysX-only)

• Teleoperation and imitation learning workflows

To install Isaac Sim, use the pip method described in Installation using Isaac Sim Pip Package.

Selective Install

If you want a minimal environment, ./isaaclab.sh -i accepts comma-separated sub-package names:

Option	What it does
isaacsim	Install Isaac Sim pip package
newton	Install Newton physics + Newton visualizer
physx	Install PhysX physics runtime
ovrtx	Install OVRTX renderer runtime
tasks	Install built-in task environments
assets	Install robot/object configurations
visualizers	Install all visualizer backends
rsl_rl	Install RSL-RL framework
skrl	Install skrl framework
sb3	Install Stable Baselines3 framework
rl_games	Install rl_games framework
robomimic	Install robomimic framework
none	Install only core isaaclab package

Examples:

# Minimal Newton setup
./isaaclab.sh -i newton,tasks,assets

# Newton with OVRTX and RSL-RL only
./isaaclab.sh -i newton,tasks,assets,ovrtx,rsl_rl

# Full Kit install with skrl
./isaaclab.sh -i isaacsim,skrl

OVRTX Rendering

OVRTX provides GPU-accelerated rendering for vision tasks without Kit.

./isaaclab.sh -i ov[ovrtx]

export LD_PRELOAD=$(python -c "import ovrtx, pathlib; print(pathlib.Path(ovrtx.__file__).parent / 'bin/plugins/libcarb.so')")

./isaaclab.sh -p scripts/benchmarks/benchmark_rsl_rl.py \
  --task Isaac-Repose-Cube-Shadow-Vision-Benchmark-Direct-v0 \
  --headless --enable_cameras --num_envs 16 --max_iterations 10 \
  presets=newton,ovrtx_renderer,simple_shading_diffuse_mdl

Running Installation Tests

./isaaclab.sh -p -m pytest source/isaaclab/test/cli/test_cli_utils.py -v

Isaac Sim Installation

For most users, the simplest and fastest way to install Isaac Lab with full Isaac Sim support is by following the Installation using Isaac Sim Pip Package guide. We recommend using uv as the package manager for the fastest and most reliable installation experience.

This method installs Isaac Sim via pip and Isaac Lab from source. If you are new to Isaac Lab, start here.

Choosing an Installation Method

Different workflows require different installation methods. Use this table to decide:

Method	Isaac Sim	Isaac Lab	Best For	Difficulty
Kit-less	❌ not required	💾 source (git)	Newton-only, fastest start	Easiest
Pip (uv) (Recommended)	📦 pip install	💾 source (git)	Most users, full features	Easy
Binary + Source	📥 binary download	💾 source (git)	Users preferring binary install of Isaac Sim	Easy
Full Source Build	💾 source (git)	💾 source (git)	Developers modifying both	Advanced
Pip Only	📦 pip install	📦 pip install	External extensions only (no training/examples)	Special case
Docker	🐳 Docker	💾 source (git)	Docker users	Advanced

Next Steps

Once you’ve reviewed the installation methods, continue with the guide that matches your workflow:

• 🚀 Kit-less Installation (above)

  • Install Isaac Lab without Isaac Sim.

  • Uses the Newton physics backend.

  • Best for getting started immediately or when Isaac Sim is not needed.

• 😃 Installation using Isaac Sim Pip Package (Recommended for full features)

  • Install Isaac Sim via pip (preferably with uv) and Isaac Lab from source.

  • Best for most users who need full simulation capabilities.

• Installation using Isaac Sim Pre-built Binaries

  • Install Isaac Sim from its binary package (website download).

  • Install Isaac Lab from its source code.

  • Choose this if you prefer not to use pip for Isaac Sim (for instance, on Ubuntu 20.04).

• Installation using Isaac Sim Source Code

  • Build Isaac Sim from source.

  • Install Isaac Lab from its source code.

  • Recommended only if you plan to modify Isaac Sim itself.

• Installation using Isaac Lab Pip Packages

  • Install Isaac Sim and Isaac Lab as pip packages.

  • Best for advanced users building external extensions with custom runner scripts.

  • Note: This does not include training or example scripts.

• Container Deployment

  • Install Isaac Sim and Isaac Lab in a Docker container.

  • Best for users who want to use Isaac Lab in a containerized environment.

Asset Caching

Isaac Lab assets are hosted on AWS S3 cloud storage. Loading times can vary depending on your network connection and geographical location, and in some cases, assets may take several minutes to load for each run. To improve performance or support offline workflows, we recommend enabling asset caching.

• Cached assets are stored locally, reducing repeated downloads.

• This is especially useful if you have a slow or intermittent internet connection, or if your deployment environment is offline.

Please follow the steps Asset Caching to enable asset caching and speed up your workflow.