Isaac Lab: Install, Versions, First Training Run

Isaac Lab is an open-source, GPU-accelerated framework for robot learning (reinforcement learning, imitation learning and post-training of robot policies). Researchers searching for it usually need to decide which release matches their hardware and Isaac Sim version, install it from the official source, and run a first training task whose version they can later report. This page covers those three jobs, using the project’s own documentation as read on 2026-11-07.

What Isaac Lab is used for in research

The project’s GitHub repository describes Isaac Lab as a framework for robot learning built for GPU-accelerated simulation. It supports multiple simulation backends: Newton, OVRTX, OVPhysX and NVIDIA Isaac Sim. It evolved from the earlier Orbit framework. The reference publication is Mittal et al., “Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning,” arXiv:2511.04831 (2025), the citation given in the repository.

Isaac Sim vs Isaac Lab

Isaac Sim is NVIDIA’s simulator; Isaac Lab is the learning framework that runs on top of it or, in current releases, on other backends. In the 3.0.0 documentation, Isaac Sim is an optional extra (isaacsim) rather than a hard requirement: a Newton-based path installs without Isaac Sim. If you only need a physics backend for reinforcement learning, you can start without Isaac Sim; if you need Isaac Sim’s Kit tooling, you install the extra.

Version compatibility with Isaac Sim

The repository publishes this compatibility mapping (checked 2026-11-07). The install page states that Isaac Sim 5.1 and older are unsupported on the current documentation.

Isaac Lab branch or tag	Isaac Sim version
main / 3.0.0	6.2
v3.0.0-EA	6.1
v3.0.0-beta2	6.0
v2.3.X	4.5 / 5.0 / 5.1
v2.2.X	4.5 / 5.0
v2.1 / v2.0	4.5

Licences for institutional use

Isaac Lab is released under BSD-3-Clause. The isaaclab_mimic extension is Apache 2.0 but requires cuRobo, which is proprietary. Workflows that use Isaac Sim include proprietary components, and the Docker base container depends on the Isaac Sim image, which is covered by the NVIDIA Software License Agreement. If your institution restricts software licences, review these terms before choosing the Isaac Sim route over the Newton route.

Install or access (official source only)

The official installation page is the Isaac Lab installation documentation and the source is github.com/isaac-sim/IsaacLab. The older /setup/installation/index.html address only redirects to the page above.

Requirements (3.0.0 documentation)

Python 3.12, managed by uv.
Operating systems: Ubuntu 22.04 or newer on x86_64, Linux aarch64 (including DGX Spark), or Windows 11 x86_64.
GPU: a CUDA-capable NVIDIA GPU. The default CUDA 13 stack needs Turing or newer (compute capability 7.5+).
Drivers listed: 595.58.03 (Linux x86_64), 580.159 (Linux aarch64), 595.97 (Windows).
For the Isaac Sim extra: GLIBC 2.35 or newer; on x86_64 the page lists at least 32 GB RAM and 16 GB VRAM.
Isaac Sim 6.2.0 is the version used by the optional isaacsim extra; Newton is installed by default (with Warp 1.17).

Option A: clone the repository

curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/isaac-sim/IsaacLab.git --branch main
cd IsaacLab
uv sync --extra isaacsim

Option B: install the wheel

uv venv --python 3.12 env_isaaclab
source env_isaaclab/bin/activate
uv pip install isaaclab==3.0.0 --index https://pypi.nvidia.com --torch-backend cu130

Option C: Docker

Start and enter the container with ./docker/container.py start and ./docker/container.py enter. The base container depends on the Isaac Sim image. A kitless profile exists that contains no Isaac Sim or Kit.

Step-by-step: run a first training task

These commands come from the current installation page. Older train.py scripts and Isaac-Cartpole-v0 task names do not appear on it.

From the cloned IsaacLab directory, confirm the install by opening an empty simulation: uv run --extra isaacsim python scripts/tutorials/00_sim/create_empty.py --viz kit
Train the cartpole task on the Newton backend (no Isaac Sim needed): uv run isaaclab train --rl_library rsl_rl --task Isaac-Cartpole-Direct physics=newton_mjwarp
Or train it on Isaac Sim’s PhysX backend: uv run --extra isaacsim isaaclab train --rl_library rsl_rl --task Isaac-Cartpole-Direct physics=isaacsim_physx
Play back the result: uv run isaaclab play --rl_library rsl_rl --task Isaac-Cartpole-Direct --viz newton_gl

Output and how to read it

The train command runs the reinforcement-learning library you named with --rl_library (here rsl_rl) on the named task, and play loads the trained policy and renders it with the visualiser you pass in --viz. For a reproducible record, report: the Isaac Lab release or commit, the matching Isaac Sim version (or that the Newton backend was used), the task name, the physics= setting, the RL library, the GPU and driver version, and cite arXiv:2511.04831. Because the physics= setting changes the simulator underneath the same task, results from newton_mjwarp and isaacsim_physx are different experimental conditions and should not be pooled without saying so. See reproducible AI experiment and reproducibility infrastructure for workflows, containers and code sharing.

Troubleshooting

Isaac Sim version mismatch. Match your Isaac Lab branch to the table above; Isaac Sim 5.1 and older are unsupported on the 3.0.0 documentation, and v2.3.X is the line for Isaac Sim 4.5 to 5.1.
GPU not supported. The default CUDA 13 stack needs compute capability 7.5 or higher (Turing or newer).
Wrong Python. The page specifies Python 3.12; create the environment with uv venv --python 3.12.
Isaac Sim not wanted. Use the Newton path (physics=newton_mjwarp) or the kitless Docker profile.
Copied commands fail. Tutorials that use train.py or Isaac-Cartpole-v0 predate the current command form.

Free alternatives

These are other open-source simulators used in robotics and reinforcement-learning research, with details taken from their own repositories (read 2026-11-07). They are different tools, not drop-in replacements for Isaac Lab’s task suite.

Tool	What its repository says	Licence	Python install
MuJoCo (Google DeepMind)	General-purpose physics engine that aims to facilitate research and development in robotics	Apache 2.0 (code)	`pip install mujoco` (Python 3.11+)
PyBullet (Bullet Physics SDK)	Python bindings for the Bullet SDK, recommended in its README for robotics, reinforcement learning and VR	zlib	`pip3 install pybullet --upgrade --user`

FAQ

What is the difference between Isaac Sim and Isaac Lab?

Isaac Sim is the simulator; Isaac Lab is the robot-learning framework that can use Isaac Sim as one of its backends. In version 3.0.0, Isaac Sim 6.2 is an optional extra.

Which Isaac Sim version does Isaac Lab 3.0.0 need?

Isaac Sim 6.2, according to the repository’s compatibility mapping.

Is Isaac Lab open source?

Yes, under BSD-3-Clause, though some Isaac Sim workflows and the cuRobo dependency of isaaclab_mimic are proprietary.

How do I cite Isaac Lab?

Mittal et al., “Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning,” arXiv:2511.04831 (2025). Add the release version you ran. For background see what is robotics and what is machine learning.

Sources

Isaac Lab installation documentation, read 2026-11-07.
isaac-sim/IsaacLab repository, read 2026-11-07.
google-deepmind/mujoco, read 2026-11-07.
bulletphysics/bullet3, read 2026-11-07.

Ask about Isaac Lab: Install, Isaac Sim Versions and a First Training Run

Ask your first 2 questions free below. Subscribers get 150 a day for $29 a month.

An AI assistant specialized in research administration. It cites the sources behind every answer, labels web answers and says when it can't answer.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.