🛠️Installation

Prerequisites

Below table shows the recommended (tested) requirements of the computer for running this framework.

Component	Recommended (Tested)
CPU	Intel Core i9
GPU	RTX 3080 10GB
OS	Ubuntu 22.04
Python	>=3.8
Nvidia Driver	>=570

LeggedGym-Ex incoporates two simulators into one framework. One can choose either simulator to use, with each simulator requiring a seperate conda environment due to the limitation of python versions. Below is the recommended (tested) environment setting for two simulators:

Component	IsaacGym	Genesis	IsaacSim
Python	3.8	>=3.10	>=3.10
Nvidia Driver	570	570	570
PyTorch	2.4.1+cu121	2.8.0+cu126	2.7.0+cu128

Direct Installation

IsaacGym

# 1. Create a conda environment with python3.8
conda create -n lr_gym python=3.8
conda activate lr_gym
# 2. Install Pytorch
pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu121
# 3. Download IsaacGym Preview4 under /home/username
cd /home/username
wget https://developer.nvidia.com/isaac-gym-preview-4 \
    && tar -xf isaac-gym-preview-4 \
    && rm isaac-gym-preview-4
# Substitute np.float with np.float32 to resolve compatibility
find isaacgym/python -type f -name "*.py" -exec sed -i 's/np\.float/np.float32/g' {} +
# Install isaacgym in this environment
cd isaacgym/python && pip install -e . && cd ../..
# 4. Install LeggedGym-Ex with isaacgym
git clone https://github.com/lupinjia/LeggedGym-Ex.git
cd LeggedGym-Ex && pip install -e ".[isaacgym]"
# 5. Test the installation
python legged_gym/scripts/train.py --task=go2 --num_envs=100

If a window like below appears, the installation is successful.

Genesis

# 1. Create a conda environment with python3.10
conda create -n lr_gen python=3.10
conda activate lr_gen
# 2. Install Pytorch
pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu126
# 3. Install LeggedGym-Ex with genesis
git clone https://github.com/lupinjia/LeggedGym-Ex.git
cd LeggedGym-Ex && pip install -e ".[genesis]"
# 4. Test the installation
export SIMULATOR=genesis
python legged_gym/scripts/train.py --task=go2 --num_envs=100

If a window like below appears, the installation is successful.

IsaacSim

# 1. Create a conda environment with python3.11
conda create -n lr_lab python=3.11
conda activate lr_lab
# 2. Install IsaacSim
pip install --upgrade "isaacsim[all,extscache]==5.1.0" --extra-index-url https://pypi.nvidia.com
# 3. Verify IsaacSim Installation
isaacsim # If a window appears, then it is successful
# 4. Clone and install isaaclab
git clone https://github.com/isaac-sim/IsaacLab.git && cd IsaacLab && git checkout v2.3.2
./isaaclab.sh --install none # do not install rl lib to use our custom rl library
# 5. Verify IsaacLab Installtion
python scripts/tutorials/00_sim/create_empty.py # If a windows with black scene appears, it is successful.
# 6. Install LeggedGym-Ex with IsaacLab
git clone https://github.com/lupinjia/LeggedGym-Ex.git
cd LeggedGym-Ex && pip install -e ".[isaaclab]"
# 7. Test the installation
export SIMULATOR=isaaclab
python legged_gym/scripts/train.py --task=go2 --num_envs=100

If a window like below appears, the installation is successful.

Finally, you need to register a wandb account and set the environment variable:

export WANDB_API_KEY=<your_api_key>

Using Docker

We provide 4 dockerfiles for different scenarios:

Dockerfile	Description	Python	CUDA	PyTorch
Dockerfile	Multiple simulators	3.8/3.10/3.11	12.8	2.4.1/2.8.0
Dockerfile.genesis	Genesis only	3.10	12.6	2.8.0
Dockerfile.isaacgym	IsaacGym only	3.8	12.6	2.4.1
Dockerfile.isaaclab	IsaacSim/IsaacLab only	3.11	12.8	-

If you want lightweight container with only one simulator, then you can choose one of three single-simulator version to use.

If you want to use this framework with full simulator backend support, the multiple-simulators version is more favorable (need at least 70GB of disk).

Building the image

Multiple Simulators

# first download isaacgym and IsaacLab following the instructions above.
# The desired directory tree is like below:
#   ├── IsaacLab/
#   ├── isaacgym/
#   └── LeggedGym-Ex/

cd /path/to/LeggedGym-Ex
docker build -t leggedgym-ex:all -f Dockerfile ..

IsaacGym Only

# first download isaacgym following the instructions above.
# The desired directory tree is like below:
#   ├── isaacgym/
#   └── LeggedGym-Ex/
cd /path/to/LeggedGym-Ex
docker build -t leggedgym-ex:isaacgym -f Dockerfile.isaacgym ..

Genesis Only

docker build -t leggedgym-ex:genesis -f Dockerfile.genesis .

IsaacLab Only

# first download IsaacLab following the instructions above.
# The desired directory tree is like below:
#   ├── IsaacLab/
#   └── LeggedGym-Ex/

docker build -t leggedgym-ex:isaaclab -f Dockerfile.isaaclab ..

Known Issues

IsaacGym window does not show rendered physical world

You may encounter this problem when your nvidia driver version is >= 570. To solve this, use export VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json.

IsaacGym uses Vulkan to render graphics. In a driver with new version, the system may not be able to find correct Vulkan ICD (Installable Client Driver) configuration file.

ImportError: libpython3.8.so.1.0: cannot open shared object file: No such file or directory

Add library path of your current conda environment to LD_LIBRARY_PATH. For example, export LD_LIBRARY_PATH=/home/user_name/miniconda3/envs/lr_gym/lib

ModuleNotFoundError: No module named ‘pkg_resources’

refer to https://github.com/isaac-sim/IsaacLab/issues/4576

Modify source/isaaclab/setup.py:

"flatdict==4.1.0" # change flatdict version from 4.0.1 to 4.1.0

Optional Installation

Unitree Go2 Sim2Sim

Deploying the policy to another simulator can effectively test the robustness of the policy. Also, the code used for sim2sim can oftern be deployed directly to the real robot. To avoid potential collapse on the real robot, it’s better to first test the deployment code in simulation.

Since deployment codes are usually written in C++, a simulator which supports C++ interface is ideal. We offer a sim2sim framework in mujoco based on unitree_sdk2, unitree_mujoco and LibTorch.

For the simulator, you can install unitree_mujoco according to the instructions in README.md.

For the deployment code, you can refer to go2_deploy and go2_deploy_python.

Below is the interface of unitree_mujoco. We have implemented depth image access and publication through dds.

TRON1_PF Sim2Sim

For sim2sim of TRON1_PF, you can install tron1-mujoco-sim and tron1-rl-deploy-python provided by us.