Build from Source

Create the Conda Environment

conda env create -f environment.yml
conda activate realsim_py

Install the official PyTorch and ensure it is compatible with your GPU driver and CUDA toolkit.

Build RealSim (CMake)

System packages (Ubuntu/Debian)

sudo apt-get update
sudo apt-get install -y \
    build-essential cmake pkg-config \
    libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev \
    libxext-dev libxrender-dev libxfixes-dev \
    libgl1-mesa-dev libglu1-mesa-dev libegl1-mesa-dev \
    libwayland-dev libxkbcommon-dev \
    libglm-dev libeigen3-dev

Please check the CMake and GCC versions, tested on CMake 3.31.4 and GCC 11.4.0.

Third-party libraries

• oneTBB:

cd deps/oneTBB
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=../../onetbb_install -DTBB_TEST=OFF ..
cmake --build .
cmake --install .
cd ../../..

• Boost:

# Remove any existing incomplete boost directory
rm -rf deps/boost
rm -rf deps/boost_install

# Download Boost 1.82.0 (or latest stable version)
cd deps
wget https://archives.boost.io/release/1.82.0/source/boost_1_82_0.tar.gz
tar -xzvf boost_1_82_0.tar.gz
mv boost_1_82_0 boost
cd boost

# Bootstrap the build system
./bootstrap.sh --prefix=../boost_install

# Build and install Boost (this may take a while)
./b2 --prefix=../boost_install --with-system --with-filesystem --with-thread --with-chrono --with-date_time --with-regex variant=release link=static threading=multi install

# Clean up the downloaded archive
cd ..
rm boost_1_82_0.tar.gz
cd ..

The build process may take 10-30 minutes, depending on your system. We are only building the necessary components (system, filesystem, thread, chrono, date_time, regex) to speed up the build process.

• cuBLAS:

cuBLAS is typically already installed if you have previously installed the CUDA toolkit. We have tested on CUDA 12.8.

Add NVIDIA GPG Key and Library:

curl https://developer.download.nvidia.com/hpc-sdk/ubuntu/DEB-GPG-KEY-NVIDIA-HPC-SDK | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-hpcsdk-archive-keyring.gpg
echo 'deb [signed-by=/usr/share/keyrings/nvidia-hpcsdk-archive-keyring.gpg] https://developer.download.nvidia.com/hpc-sdk/ubuntu/amd64 /' | sudo tee /etc/apt/sources.list.d/nvhpc.list

Update Package List and Install NVIDIA HPC SDK:

sudo apt-get update -y
sudo apt-get install -y nvhpc-24-5

• METIS:

Build and install GKlib (METIS dependency) and METIS:

chmod +x ./build_metis.sh
sudo ./build_metis.sh

Verify installation:

ls -la deps/METIS_install/lib/libmetis.a
ls -la deps/METIS/GKlib_build/lib/libGKlib.a

You should see both library files listed.

• assimp:

cd deps/assimp
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=../../assimp_install ..
cmake --build .
cmake --install .
cd ../../..

Compile RealSim

mkdir build && cd build
cmake ..
make -j$(nproc)
cd ..

Install the Python Package

Ensure you are in the repository root:

pip install -e .

Quick Verification

Native binaries:

./build/bin/RealSim
./build/bin/DiffSim

Python bindings:

import realsim_py as rs
sim = rs.RealtimeSim()
sim.load_config("simulation/config/Tests/ParallelEnvTest/scene.json")
sim.set_envs(2)
sim.initialize(suppress_output=True)
for _ in range(100):
    sim.step()