Installation Guide

Follow these steps to set up your environment for GenManip.

0. Prerequisites

1. Ubuntu 22.04
2. RTX Series GPU with CUDA 12.1 (Recommended)
3. Latest MiniConda

   mkdir -p ~/miniconda3
   wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
   bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
   rm ~/miniconda3/miniconda.sh

About GUI

Note that if you are using a server without a graphical user interface (GUI), please keep this in mind. Although GenManip can run all its functionalities entirely without a GUI, there are still certain cases where whether the GUI is enabled matters. We will clearly indicate these situations in the relevant sections throughout the documentation, so please pay attention to those notes.

1. Install Dependencies

GenManip requires Conda for environment management and specific Python packages, including PyTorch and Isaac Sim.

1. Create and activate a Conda environment. We recommend Python 3.10.

   conda create -n genmanip python=3.10
   conda activate genmanip

2. Clone the GenManip repository.

   git clone https://github.com/InternRobotics/GenManip.git
   cd GenManip

3. Install Required Python packages.

   bash scripts/install.sh # modify the script if you are not using CUDA 12.1
   
   # Install genmanip_client (your model environment should also install this)
   # It contains only common deps and is unlikely to cause conflicts; it helps you communicate with the GenManip Server.
   cd standalone_tools/packages/genmanip_client
   pip install -e .
   cd ../../..

4. (Optional) Install rerun.io support

   rerun.io provides a clean way to visualize data. In Evaluation, we support saving intermediate rerun files, and will dump them into the result path when rerun-sdk is installed. Since it may require manual dependency conflict resolution and takes ~100MB, we list it as optional for users who want synchronized visualization of state/action and frames.

   pip install rerun-sdk==0.28.4
   pip install numpy==1.26.4 # rerun-sdk installs numpy>2; pin back to 1.26.4

2. Download and Prepare Assets

You need to download the scene assets for the benchmarks.

1. Download Scene Assets. The assets contain the 3D models and textures used in the benchmark scenes.

   huggingface-cli login
   
   # Download assets you need
   python standalone_tools/download_assets.py --dataset basic # assets for installation verification

3. Verify Installation (Render Test)

To ensure your installation and environment are set up correctly, run the demos/example_case/test_demoscene.py script. This script will attempt to load a basic scene and render an image.

python demos/example_case/test_demoscene.py

If the script executes without errors, it will save a rendered image at tmp/test.png. Successfully generating this image indicates that Isaac Sim is likely installed and configured correctly.

You are now ready to configure your environment and run the benchmarks.

4. (Optional) Download More Assets

You can download more assets by running the following command:

python standalone_tools/download_assets.py --dataset all-robot # download all robot assets and corresponding cuRobo configurations