Using Isaac Lab with Isaac Sim 5.0 in a Conda Environment

[RandomOakForest]

Show & Tell: Using Isaac Lab with Isaac Sim 5.0 in a Conda Environment

Here’s how to set up Isaac Lab and Isaac Sim 5.0 to work seamlessly with a Conda environment, covering key compatibility, setup, and project creation steps.

1. Isaac Sim 5.0 Requires Python 3.11 and GLIBC 2.35+

Isaac Sim 5.0 is built specifically for Python 3.11. All environments and dependencies must use Python 3.11 for compatibility.
Important: On Linux, Isaac Sim 5.0 also requires GLIBC 2.35+. You can check your GLIBC version with:

ldd --version

This note has been tested in Ubuntu 22.04 using a bash shell. Refer to the official documentation for more about system compatibility and issues on platforms like Ubuntu 20.04.¹²³

2. Setting Up a Conda Environment with Python 3.11

Create and activate your Conda environment:

conda create -n env_isaaclab python=3.11 -y
conda activate env_isaaclab

This environment will be used for all subsequent steps and for running both Isaac Sim and Isaac Lab.

3. Building Isaac Sim 5.0 from Source

To build Isaac Sim 5.0 from source:

# Clone the Isaac Sim repository
git clone https://github.com/isaac-sim/IsaacSim.git 
cd IsaacSim

# Install Git LFS and pull large files
git lfs install
git lfs pull

# Verify that GCC/G++ 11 is being used before building:
gcc --version
g++ --version

# Build
./build.sh

Note: During the first launch or import of Isaac Sim, you may be prompted to accept the NVIDIA Omniverse EULA. If you haven't, run Isaac Sim as follows:

cd _build/linux-x86_64/release
./isaac-sim.sh

You can pre-empt this by setting the following environment variable (here using bash):

export OMNI_KIT_ACCEPT_EULA=YES

If you do not accept, execution will terminate, and if accepted once, you should not be prompted again.¹ You may close Isaac Sim to continue with Step 4.

4. Preparing Isaac Sim for Your Conda Environment

After building Isaac Sim and activating your Conda environment, you need to run the setup script which configures paths and dependencies. From your Isaac Sim directory, where you ran build.sh, the script is found at:

_build/<arch>/release/setup_conda_env.sh

where <arch> is your CPU architecture (e.g., linux-x86_64). For example:

source _build/linux-x86_64/release/setup_conda_env.sh

You must run this every time you start a new terminal session before running Isaac Sim or Isaac Lab.¹ Consider this when setting up your IDE to run tasks and whatnot.

Note: For earlier versions of Isaac Sim and if you are using zsh instead of bash then you need to add this workaround to the Isaac Sim scripts.

5. Using Isaac Lab with Isaac Sim 5.0

To access new features in Isaac Lab leveraging Isaac Sim 5.0, you can now simply use the main branch instead of the feature branch as indicated before the release of Isaac Lab 2.2.0. In this example, if you followed the steps above, cd .. and clone Isaac Lab side by side Isaac Sim, clone and add a symlink to the Isaac Sim build:

git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

# Add symlink
ln -s ../IsaacSim/_build/linux-x86_64/release _isaac_sim

Before proceeding, make sure you have the necessary system dependencies:

sudo apt install cmake build-essential

Then, install Isaac Lab and learning frameworks:

./isaaclab.sh --install

This step will install all supported learning libraries. You can find more information and troubleshooting steps in the Isaac Lab installation docs.

6. Example: Creating a New Project + Isaac Lab Walkthrough

With everything set up:

# Make sure the conda environment is active and setup_conda_env.sh is sourced
conda activate env_isaaclab
source path/to/isaacsim/_build/linux-x86_64/release/setup_conda_env.sh

# From the Isaac Lab directory 
./isaaclab.sh --new

# After following the steps prompted (see the Walkthrough for an example)
cd to/project/name/used/above
python -m pip install -e source/project_name

For further guidance, follow the detailed Walkthrough in the Isaac Lab documentation. This guide takes you through setting up a new extension project, adding robots, customizing environments, and even training a control policy.

Footnotes

1. https://docs.isaacsim.omniverse.nvidia.com/latest/installation/install_python.html ↩ ↩² ↩³

2. https://isaac-sim.github.io/IsaacLab/main/source/setup/installation/pip_installation.html ↩

3. https://github.com/isaac-sim/IsaacLab/issues/480 ↩

[ecstayalive]

@RandomOakForest Following your method, I get an error when I run . /isaaclab.sh --install :

find: ‘${IsaacSimPath}/_build/linux-x86_64/release/source’: No such file or directory

[ecstayalive]

#issue703, this problem has been reported to IsaacSim.

[RandomOakForest]

You have to work on a new copy of Isaac Lab for the branch that supports Isaac Sim 5.0. Specifically, before your install step, do:

git clone -b feature/isaacsim_5_0 https://github.com/isaac-sim/IsaacLab.git

Your issue is different from the one in #703 because there is no "source" directory after building Isaac Sim 5.0 inside the release directory. If you still have this issue, please open a new bug report with steps to reproduce. Thanks.

[ecstayalive]

I'm sure I switched to the feature/isaacsim_5_0 branch correctly by following your steps. However, I still encountered the above error. So far, I've solved the problem using the following method #isaacsim89

[RandomOakForest]

I see, if you are using zsh instead of bash then you need that workaround. I'll add this to the doc above for now, we tagged it as a proposed enhancement for the Isaac Sim team. Thanks.

[ecstayalive]

Thanks.

[KK-Zhou]

Follow the steps above, I can train my own robot using IsaacSim5.0+RTX5090 GPU but with a lower speed than IsaacSim4.5+RTX4090. Mainly slow down on the collection time (IsaacSim4.5+RTX4090:1.25s, IsaacSim5.0+RTX5090: 2.5s). Meanwhile, when I want to play the pretrained model without headless mode, the IsaacSim can't render my robot with the following error:

[RandomOakForest]

If you are not able to go through this Walkthrough without trouble, please file a new issue with your specific steps to reproduce it. Thanks.

[KK-Zhou]

This a multi GPU driver bug, and I have fixed it by running the command:
sudo rm /usr/share/vulkan/icd.d/nvidia_icd.json

[tykim0507]

Hello, is there a way to enable enabling code navigation to python modules in vscode when we use the conda environment?
I ran through the instructions in 6. Example: Creating a New Project + Isaac Lab Walkthrough.
I can run the python codes in the terminal, but the vscode python interpreter cannot navigate to the isaacsim and isaaclab related modules in the codes.

I also ran this command in the vscode, but this doesn't generate the .python.env file.

I think the error message suggests that it doesn't work with conda environments.

I would be helpful I could know if there is a way to automatically setup the code navigation settings with conda!

p.s.) I am using a workaround as manually adding the isaaclab, isaacsim related python paths to PYTHONPATH in the ${workspace}.env file

[RandomOakForest]

This is worth its own discussion post here. Please go ahead and post a new discussion and describe your setup so we can more precisely address your issue.