Installation Guide

Installation all-in-one Script

Clone the repository and run the installation script scripts/install.sh to set up the environment and install dependencies all at once. The script will create a conda environment named UniVTAC and install Isaac Sim, Isaac Lab, TacEx, cuRobo, and other necessary packages.

git clone https://github.com/univtac/UniVTAC.git
cd UniVTAC
bash scripts/install.sh

Manual Installation Instructions

Requirements

• System: Linux with NVIDIA GPU
• Python 3.10
• NVIDIA Isaac Sim 4.5 + Isaac Lab 2.1.1
• NVIDIA cuRobo
• TacEx: Must be built from the local third_party/TacEx source (contains project-specific modifications)

Installation & Setup

Step 1: Clone the Repository

git clone https://github.com/univtac/UniVTAC.git
cd UniVTAC

Step 2: Create a Conda Environment

conda create -n UniVTAC python=3.10 -y
conda activate UniVTAC

Step 3: Install cuRobo

cuRobo is used for GPU-accelerated collision-aware motion planning. Follow the official cuRobo Installation Guide.

Step 4: Install TacEx (Modified Source)

Important: Do not install TacEx from the public repository. UniVTAC requires a modified version of TacEx that is bundled in third_party/TacEx. Some internal APIs have been adapted for UniVTAC's tactile sensor pipeline.

cd third_party/TacEx

If you have a working Isaac Lab environment, you can directly install TacEx. Otherwise, you need to install Isaac Sim 4.5 and Isaac Lab 2.1.1. Below is a quick summary, but here is the full installation guide.

# install cuda-enabled pytorch
pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cu118
pip install --upgrade pip
# install isaac sim packages
pip install 'isaacsim[all,extscache]==4.5.0' --extra-index-url https://pypi.nvidia.com

# install dependencies via apt (Ubuntu)
sudo apt install cmake build-essential
git clone https://github.com/isaac-sim/IsaacLab
cd IsaacLab
# use Isaac Lab version 2.1.1
git checkout v2.1.1
# activate the Isaac Sim python env
conda activate UniVTAC
# install isaaclab extensions (with --editable flag)
./isaaclab.sh --install # or "./isaaclab.sh -i"

conda activate UniVTAC
python scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Ant-v0 --headless

Installing TacEx [Core]

1. Activate the Isaac Env

conda activate UniVTAC

2. Install the core packages of TacEx

# Script will pip install core TacEx packages with --editable flag)
./tacex.sh -i

You can install the extensions one by one via e.g. python -m pip install -e source/tacex_uipc

3. Verify that TacEx works by running an example:

python ./scripts/demos/tactile_sim_approaches/check_taxim_sim.py --debug_vis

And here is an RL example:

python ./scripts/reinforcement_learning/skrl/train.py --task TacEx-Ball-Rolling-Tactile-RGB-v0 --num_envs 512 --enable_cameras

You can view the sensor output in the IsaacLab Tab: Scene Debug Visualization > Observations > sensor_output

Installing TacEx [UIPC]

The tacex_uipc package is responsible for the UIPC simulation in TacEx.

1. Install the libuipc dependencies:

• If not installed yet, install Vcpkg

mkdir ~/Toolchain
cd ~/Toolchain
git clone https://github.com/microsoft/vcpkg.git
cd vcpkg
./bootstrap-vcpkg.sh -disableMetrics

• Set the System Environment Variable CMAKE_TOOLCHAIN_FILE to let CMake detect Vcpkg. If you installed it like above, you can do this:

# Write in ~/.bashrc
export CMAKE_TOOLCHAIN_FILE="$HOME/Toolchain/vcpkg/scripts/buildsystems/vcpkg.cmake"

• We also need CMake 3.26, GCC 11.4 and Cuda 12.4 to build libuipc. Install this into the Isaac Sim python env:

# Inside the root dir of TacEx repo
conda activate UniVTAC
conda env update -n UniVTAC --file ./source/tacex_uipc/libuipc/conda/env.yaml

If Cuda 12.4 does not work for, try updating your Nvidia drivers or try to use an older Cuda version by adjusting the env.yaml file (e.g. Cuda 12.2).

2. Install tacex_uipc

# This also builds `libuipc` and pip installs the python bindings.
conda activate UniVTAC
pip install -e source/tacex_uipc -v

You can also install all TacEx packages with ./tacex.sh -i all.

3. Verify that the tacex_uipc works by running a data collection example:

bash collect_data.sh grasp_classify demo 0