Isaac Lab Installation Guide (DPPO version)

This guide will walk you through the installation process of Isaac Lab version 4.5.0 with DPPO algorithm using pip installation method.

Prerequisites

• Python 3.10
• CUDA-compatible GPU
• On Windows: GPU driver version 552.86 (for CUDA 12)
• Miniconda (recommended if using Conda)

Installation Steps

1. Virtual Environment Setup

1.1 Create Virtual Environment

conda create -n env_isaaclab python=3.10

1.2 Activate Virtual Environment

Activate when open new terminal

conda activate env_isaaclab

2. Install PyTorch

Install CUDA-enabled PyTorch 2.5.1 (required for Windows, optional for Linux):

pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cu121

3. Update pip

pip install --upgrade pip

4. Install Isaac Sim

Install Isaac Sim packages:

pip install 'isaacsim[all,extscache]==4.5.0' --extra-index-url https://pypi.nvidia.com

5. Install Isaac Lab (DPPO version)

Recommend Create workspace before Install

mkdir workspace_name
cd workspace_name

1. Clone the Isaac Lab repository:

git clone https://github.com/Aitthikit/IsaacLabDPPO.git
cd IsaacLabDPPO

2. Install dependencies (Ubuntu):

sudo apt install cmake build-essential

3. Install Isaac Lab extensions:

cd IsaacLab #go to IsaacLab Directory
./isaaclab.sh --install  # Installs all learning frameworks

Note: You can ignore Isaac Lab template settings file not found error

6. Rsl_rl Install

Clone or copy this project/repository separately from the Isaac Lab installation (i.e. outside the IsaacLab directory):

Note: Install Rsl_rl after finished IsaacLab install.

1. Clone the Rsl_rl repository:

cd ../.. # back to workspace
git clone https://github.com/Aitthikit/rsl_rl.git -b my-fix

2. Install dependencies (Ubuntu):

cd rsl_rl
pip install -e .

Special Notes

For 50 Series GPUs

If you're using 50 series GPUs, use the latest PyTorch nightly build instead:

pip install --upgrade --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128

First Run Information

• The first run will download all dependent extensions (may take ~10 minutes)
• You'll need to accept the NVIDIA Software License Agreement
• Extensions will be cached for subsequent runs

Verification

To verify your installation: Go to IsaacLab directory first and run scripts

1. Run the simulator:

isaacsim

2. Test with a sample script:

./isaaclab.sh -p scripts/tutorials/00_sim/create_empty.py

You should see a simulator window with a black viewport.

3. Verify environment :

./isaaclab.sh -p scripts/tutorials/03_envs/create_quadruped_base.py

You should see a simulator window with Anymal robot. If this appears, your installation was successful!

Training Examples

Once installed, you can try these example training commands:

This might take few minutes to start up.

Train an anymal c to walk with DPPO algorithm:

python scripts/reinforcement_learning/rsl_rl/train.py --task Isaac-Velocity-Rough-DPPO-Anymal-C-Direct-v0 --num_envs 1024

Train an anymal c to distillation Teacher policy:

python scripts/reinforcement_learning/rsl_rl/train.py --task Isaac-Velocity-Distill-DPPO-Anymal-C-Direct-v0 --num_envs 256

Troubleshooting

If you encounter any issues:

• Check the Isaac Sim documentation
• Visit the Isaac Sim Forums

For more information and detailed guides, refer to the official Isaac Lab documentation

For more information and detailed guides about algorithm, refer to the RSL_RL Github