Skip to main content

Installation Guide

DISCOVERSE is a unified, modular open-source 3DGS robot simulation framework supporting the Real2Sim2Real learning workflow. This guide will help you install and configure DISCOVERSE on your system.

System Requirements​

Minimum Requirements​

  • Python 3.8+
  • Operating System: Linux (Ubuntu 18.04+), macOS, or Windows 10+
  • Memory: At least 8GB RAM
  • Python 3.10
  • CUDA 11.8+ (for 3DGS rendering)
  • NVIDIA GPU (8GB+ VRAM recommended)
  • Git LFS (for model file management)

Quick Installation​

1. Clone the Repository​

git clone https://github.com/TATP-233/DISCOVERSE.git
cd DISCOVERSE
Recommendation

It is recommended to download submodules as needed, rather than using the --recursive flag to fetch all at once. This saves time and storage.

2. Create a Virtual Environment​

conda create -n discoverse python=3.10
conda activate discoverse

3. Choose Installation Method​

Select the appropriate installation method based on your use case:

pip install -e .

Includes: MuJoCo, OpenCV, NumPy and other basic dependencies

LiDAR SLAM Research​

pip install -e ".[lidar,visualization]"
  • Features: High-performance LiDAR simulation with Taichi GPU acceleration
  • Dependencies: taichi>=1.6.0
  • Applications: Mobile robot SLAM, LiDAR sensor simulation, point cloud processing

Robotic Arm Imitation Learning​

pip install -e ".[act_full]"
  • Features: Imitation learning, robot skill training, policy optimization
  • Dependencies: torch, einops, h5py, transformers, wandb
  • Algorithms: Other algorithms available with [dp_full] or [rdt_full]

High-Fidelity Visual Simulation​

pip install -e ".[gaussian-rendering]"
  • Features: Realistic 3D scene rendering with real-time lighting
  • Dependencies: torch>=2.0.0, torchvision>=0.14.0, plyfile, PyGlm
  • Applications: High-fidelity visual simulation, 3D scene reconstruction, Real2Sim workflow
pip install -e ".[full]"

4. Download Submodules​

# Automatically detect and download required submodules
python scripts/setup_submodules.py

# Manually specify modules
python scripts/setup_submodules.py --module lidar act

# Download all submodules (suitable for Docker environments)
python scripts/setup_submodules.py --all

5. Verify Installation​

# Basic check
python scripts/check_installation.py

# Detailed information
python scripts/check_installation.py --verbose

Git LFS Setup​

DISCOVERSE model files are managed through Git LFS:

Linux Systems​

curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
sudo apt-get install git-lfs
git lfs install
git lfs pull

macOS Systems​

brew install git-lfs
git lfs install
git lfs pull

Module Feature Comparison​

ModuleInstallation CommandFeature DescriptionUse Cases
Basicpip install -e .Core simulation featuresLearning, basic development
LiDAR.[lidar]High-performance LiDAR simulationSLAM, navigation research
Rendering.[gaussian-rendering]3D Gaussian Splatting renderingVisual simulation, Real2Sim
GUI.[xml-editor]Visual scene editingScene design, model debugging
ACT.[act]Imitation learning algorithmRobot skill learning
Diffusion Policy.[diffusion-policy]Diffusion model policyComplex policy learning
RDT.[rdt]Large model policyGeneral robot skills
Hardware Integration.[hardware]RealSense+ROSReal robot control

Docker Installation​

If you prefer using Docker environment:

# Download pre-built image
# Baidu Cloud: https://pan.baidu.com/s/1mLC3Hz-m78Y6qFhurwb8VQ?pwd=xmp9

# Or build from source
git clone https://github.com/TATP-233/DISCOVERSE.git
cd DISCOVERSE
python scripts/setup_submodules.py --all
docker build -t discoverse:latest .

# Run with GPU support
docker run -it --rm --gpus all \
    -e DISPLAY=$DISPLAY \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    -v $(pwd):/workspace \
    discoverse:latest

High-Fidelity Rendering Setup (Optional)​

If you need 3DGS high-fidelity rendering features:

1. CUDA Installation​

Install CUDA 11.8+ from NVIDIA website, choosing the version compatible with your graphics driver.

2. Build 3DGS Dependencies​

pip install -e ".[gaussian-rendering]"

cd submodules/diff-gaussian-rasterization/

# Apply patches
sed -i 's/(p_view.z <= 0.2f)/(p_view.z <= 0.01f)/' cuda_rasterizer/auxiliary.h
sed -i '361s/D += depths\[collected_id\[j\]\] \* alpha \* T;/if (depths[collected_id[j]] < 50.0f)\n        D += depths[collected_id[j]] * alpha * T;/' cuda_rasterizer/forward.cu

cd ../..
pip install submodules/diff-gaussian-rasterization

3. Download 3DGS Models​

Place model files in the models/3dgs directory.

Troubleshooting​

If you encounter installation issues, please refer to:

  1. Dependency Conflicts: Try creating a new virtual environment
  2. CUDA Issues: Verify GPU driver and CUDA version compatibility
  3. Git LFS Issues: Check network connection and LFS configuration
  4. Permission Issues: Use sudo on Linux/macOS or adjust file permissions

For more detailed troubleshooting information, please refer to the project's troubleshooting documentation.

Next Steps​

After installation, you can: