1 IsaacLab 2.0 Installation Tutorial:
-
Use Anaconda to manage Python virtual environments
-
Using IsaacSim 4.5: https://docs.isaacsim.omniverse.nvidia.com/4.5.0/installation/download.html
-
Use IsaacLab v2.0.2: https://github.com/isaac-sim/IsaacLab
-
Graphics card driver >= 535
1.1 Unzip IsaacSim
Download the IsaacSim installation package, it is recommended to use 4.5, https://docs.isaacsim.omniverse.nvidia.com/4.5.0/installation/download.html
Unzip isaac-sim-standalone@4.5.0-rc.36+release.19112.f59b3005.gl.linux-x86_64.release.zip
It is recommended to rename the file to isaacsim and place it under ${HOME}
1.2 Clone IsaacLab
git clone -b v2.1.0 https://github.com/isaac-sim/IsaacLab.git
1.3 Create a soft link
# enter the cloned repository
cd IsaacLab
# create a symbolic link
ln -s {path_to_isaac_sim} _isaac_sim
# path_to_isaac_sim is a localisaac_sim path
# recommended: ln -s ~/isaacsim _isaac_sim
1.4 Configure the conda environment
Installing isaaclab in a manually created conda environment may result in errors. It is recommended to use the./isaaclab.sh script to create a new conda environment.
Option 1: The default environment name 'isaaclab' will show if you already have this environment; otherwise, it will help you create this environment
./isaaclab.sh --conda # or "./isaaclab.sh -c"
source activate my_env
Option 2: If you don't want to use this name, you can also create another one
./isaaclab.sh --conda my_env # or "./isaaclab.sh -c my_env"
source activate my_env
1.5 Use apt to install dependencies (only applicable to Linux)
sudo apt install cmake build-essential
1.6 Run the installation command to install all extensions in the source/extensions directory
./isaaclab.sh -i
If the error shown in the figure above occurs during installation, it is most likely due to an unstable network. Simply re-execute the command to complete the installation!
1.7 Installation Verification
Verify whether IsaacSim is running properly according to the test cases on the NVIDIA official website, with the app interface opening normally and mouse operations responding without lag as the criteria
export ISAACSIM_PATH="${HOME}/isaacsim"
export ISAACSIM_PYTHON_EXE="${ISAACSIM_PATH}/python.sh"
# 以上两句话可以写入到.bashrc中
${ISAACSIM_PATH}/isaac-sim.sh
2 Tutorial for Using the Training Project:
2.1 Code Structure Display
pointfoot-Isaaclab-4.5.0/
├── .gitignore
├── LICENCE
├── README.md
├── pyproject.toml
├── rsl_rl/
│ ├── __init__.py
│ ├── pyproject.toml
│ ├── rsl_rl/
│ │ ├── __init__.py
│ │ ├── algorithm/
│ │ │ └── ppo.py
│ │ └── runner/
│ │ └── on_policy_runner.py
│ └── setup.py
├── run_train_sequence.sh
├── scripts/
│ └── rsl_rl/
│ ├── cli_args.py
│ ├── play.py
│ └── train.py
├── start.sh
└── exts/
└── bipedal_locomotion/
├── __init__.py
├── assets/
│ ├── __init__.py
│ ├── config/
│ └── usd/
│ ├── PF_TRON1A/
│ │ └── .asset_hash
│ ├── SF_TRON1A/
│ │ └── .asset_hash
│ ├── WF_TRON1A/
│ │ └── .asset_hash
│ └── .asset_hash
├── docs/
│ └── CHANGELOG.rst
└── tasks/
├── __init__.py
└── locomotion/
├── __init__.py
├── agents/
│ └── __init__.py
├── cfg/
│ ├── __init__.py
│ ├── PF/
│ │ ├── __init__.py
│ │ ├── limx_base_env_cfg.py
│ │ └── terrains_cfg.py
│ ├── SF/
│ │ ├── __init__.py
│ │ ├── limx_base_env_cfg.py
│ │ └── terrains_cfg.py
│ └── WF/
│ ├── __init__.py
│ ├── limx_base_env_cfg.py
│ └── terrains_cfg.py
├── mdp/
│ ├── __init__.py
│ ├── commands/
│ │ ├── __init__.py
│ │ ├── commands_cfg.py
│ │ └── gait_command.py
│ ├── curriculums.py
│ └── rewards.py
└── robots/
├── __init__.py
├── limx_solefoot_env_cfg.py
├── limx_pointfoot_env_cfg.py
└── limx_wheelfoot_env_cfg.py
2.2 Clone the currently open-source IsaacLab training project: https://github.com/limxdynamics/tron1-rl-isaaclab
git clone https://github.com/limxdynamics/tron1-rl-isaaclab.git
2.3 Install the IsaacLab training project open-sourced this time:
python -m pip install -e exts/bipedal_locomotion/
2.4 Install the rl project
python -m pip install -e rsl_rl/
需要卸载IsaacLab安装时，默认安装的rsl_rl依赖
pip uninstall rsl-rl-lib
2.5 Train and Play
Start the training script to begin training:
python scripts/rsl_rl/train.py --task=Isaac-Limx-PF-Blind-Flat-v0 --headless
python scripts/rsl_rl/train.py --task=Isaac-Limx-WF-Blind-Flat-v0 --headless
python scripts/rsl_rl/train.py --task=Isaac-Limx-SF-Blind-Flat-v0 --headless
If the command line prints the following information and continuously updates, it indicates that the training has been running normally
Start the verification script to begin verification:
python scripts/rsl_rl/play.py --task=Isaac-Limx-PF-Blind-Flat-Play-v0 --headless --load_run={} --num_env 10
python scripts/rsl_rl/play.py --task=Isaac-Limx-WF-Blind-Flat-Play-v0 --headless --load_run={} --num_env 10
python scripts/rsl_rl/play.py --task=Isaac-Limx-SF-Blind-Flat-Play-v0 --headless --load_run={} --num_env 10
If you encounter the following errors:
ModuleNotFoundError: No module named 'isaaclab'
You need to perform the following operations in your execution shell window:
source {your_isaaclab_path}/_isaac_sim/setup_conda_env.sh
3 Deploy
3.1 policy file
After play verification, the trained.pt model will be automatically exported as encoder.onnx and policy.onnx, with the location as follows (taking PF as an example)
pointfoot-Isaaclab-4.5.0/
├── logs/
│ ├── pf_tron_1a_flat/
│ │ ├── 2025-07-10_10-29-18（该名称为示例，实际情况根据训练的时间点而定）
│ │ ├── exported
│ │ │ └── encoder.onnx
│ │ │ └── policy.onnx
│ │ │ └── policy.pt
│ │ ├── model_10000.pt
│ ├── sf_tron_1a_flat/
│ ├── wf_tron_1a_flat/
├── rsl_rl
3.2 Download simulation project code
For details, see https://support.limxdynamics.com/en/docs/category/tron-1-sdk-开发指南 Sections 8.1~8.2. Move the trained encoder.onnx and policy.onnx to the corresponding locations according to the instructions in the respective sections, and you will be able to run the trained model in simulation and real-world operation.
4 Known Issue
- **An error occurred during the installation of the environment under the RTX 4070 + Ubuntu 20.04 + NVIDIA Driver 570 + CUDA 12.8 environment, indicating that the Python library is incompatible **
This issue is usually caused by the partial incompatibility between the versions of some Python libraries and the current CUDA and graphics card drivers. It is recommended to reinstall the relevant dependent libraries according to the following versions to resolve the compatibility issue:
pip install numpy==1.26.4
pip install usd-core==23.2
pip install lxml==5.4.0
pip install pydantic==2.11.3
pip install pydantic-core==2.33.1
pip uninstall torchaudio
pip install opencv-python==4.11.0.86
pip uninstall nvidia-srl-usd-to-urdf
pip install setuptools==75.8.0
- **After installing PyTorch in the recommended way, it fails to run properly in the environment of RTX 5070 + Ubuntu 20.04 + NVIDIA Driver 570 + CUDA 12.8 **
This issue may be caused by compatibility issues between RTX 50 series graphics cards and PyTorch 2.5. Although the PyTorch version compatible with CUDA 12.8 was installed according to the conventional method, abnormal operation still occurs under this hardware combination. ** It is recommended to try upgrading to PyTorch 2.7 **, which has demonstrated better compatibility in actual testing and can resolve such issues.
** 3. How to Achieve Compatibility with IsaacSim 5.x Simulation Software **
1. First, it should be noted that the system requires Ubuntu 22.04 or higher;
2. The installation can be done by directly executing the following shell script, where the IsaacSim version and IsaacLab version can be modified within the script;
3. This installation completely abandons the use of Anaconda virtual environments and instead uses a venv environment for installation to avoid copyright concerns.
#!/bin/bash
# Isaac Sim 5.1 + Isaac Lab 2.2.1 完整安装脚本
# 使用 venv + Python 3.11.13 + PyTorch 2.7.0
set -e # 遇到错误立即退出
echo "=========================================="
echo "Isaac Sim 5.1 + Isaac Lab 2.2.1 安装脚本"
echo "=========================================="
# 检查 Python 3.11 是否已安装
if ! command -v python3.11 &> /dev/null; then
echo "错误: Python 3.11 未安装"
echo "请先运行: bash install_python311.sh"
exit 1
fi
echo "✓ Python 3.11 已安装: $(python3.11 --version)"
# 设置环境变量
VENV_DIR="$HOME/.venvs/isaaclab"
ISAACLAB_DIR="$HOME/robot/tool/IsaacLab"
# 1. 创建 venv
echo ""
echo "步骤 1/7: 创建 venv 环境..."
if [ -d "$VENV_DIR" ]; then
echo "venv 已存在，跳过创建"
else
python3.11 -m venv "$VENV_DIR"
echo "✓ venv 创建完成: $VENV_DIR"
fi
# 2. 激活 venv
echo ""
echo "步骤 2/7: 激活 venv 并升级 pip..."
source "$VENV_DIR/bin/activate"
python --version
pip install --upgrade pip
# 3. 安装 PyTorch 2.7.0
echo ""
echo "步骤 3/7: 安装 PyTorch 2.7.0..."
pip install "torch==2.7.0" "torchvision==0.22.0" --index-url https://download.pytorch.org/whl/cu128
python -c "import torch; print(f'✓ PyTorch {torch.__version__} 安装成功')"
# 4. 安装 Isaac Sim 5.1（强制统一版本）
echo ""
echo "步骤 4/7: 安装 Isaac Sim 5.1..."
# 如已安装旧版本 isaacsim，则先卸载，避免多个版本混在一起
if python -m pip show isaacsim &> /dev/null; then
CUR_VER=$(python -m pip show isaacsim | awk '/Version:/ {print $2}')
echo "检测到已安装的 isaacsim 版本: $CUR_VER，先卸载..."
pip uninstall -y isaacsim
fi
echo "开始安装 isaacsim==5.1.0..."
pip install "isaacsim[all,extscache]==5.1.0" --extra-index-url https://pypi.nvidia.com
echo "✓ Isaac Sim 5.1.0 安装成功"
# 5. 检查系统依赖
echo ""
echo "步骤 5/7: 检查系统依赖..."
if ! command -v cmake &> /dev/null || ! command -v g++ &> /dev/null; then
echo "警告: cmake 或 build-essential 未安装"
echo "请运行: sudo apt install -y cmake build-essential"
else
echo "✓ 系统依赖已满足"
fi
# 6. 克隆 Isaac Lab
echo ""
echo "步骤 6/7: 克隆 Isaac Lab 2.2.1..."
if [ -d "$ISAACLAB_DIR" ]; then
echo "Isaac Lab 目录已存在，切换到 v2.2.1..."
cd "$ISAACLAB_DIR"
git fetch
git checkout v2.2.1
else
cd "$HOME"
git clone https://github.com/isaac-sim/IsaacLab.git
cd "$ISAACLAB_DIR"
git checkout v2.2.1
fi
echo "✓ Isaac Lab 2.2.1 已就绪"
# 7. 安装 Isaac Lab
echo ""
echo "步骤 7/7: 安装 Isaac Lab 扩展..."
echo "这可能需要几分钟..."
./isaaclab.sh --install
echo ""
echo "=========================================="
echo "安装完成！"
echo "=========================================="
echo ""
echo "使用方法:"
echo "1. 激活环境: source $VENV_DIR/bin/activate"
echo "2. 进入目录: cd $ISAACLAB_DIR"
echo "3. 运行示例: ./isaaclab.sh -p scripts/tutorials/00_sim/create_empty.py"
echo ""