CudaRobotics Docs

Getting started

Install CudaRobotics

Use the published v0.3.0 GitHub assets, build the current source from a checkout, or run the Nav2 CUDA MPPI demo from GHCR.

Release assets

Install a wheel

Download the matching Linux x86_64 wheel from the published v0.3.0 release.

Source build

Develop from checkout

Use an editable install when changing Python bindings or CUDA core code.

Container

Run the demo image

Use GHCR for the packaged Nav2 plugin smoke and benchmark path.

Requirements

  • Linux x86_64 for the published wheel assets.
  • CUDA Toolkit 12.x for source builds.
  • A compatible NVIDIA driver at runtime.
  • CMake 3.18 or newer and a C++17 compiler for local builds.
  • ROS 2 and Nav2 build dependencies when building the controller plugin.

Published Release Assets

The quality-gated v0.3.0 release is the latest published release.

python -m pip install ./cudarobotics-0.3.0-*.whl
python -c "import cudarobotics as cr; print(cr.__version__)"

Download assets from GitHub Releases. PyPI publishing is not currently part of the supported distribution path.

Current Source Checkout

git clone https://github.com/rsasaki0109/CudaRobotics.git
cd CudaRobotics
python -m pip install -e python/
python -c "import cudarobotics as cr; print(cr.__version__)"
python examples/python/mppi_quickstart.py
python examples/python/registration_quickstart.py

Maintainers should run `./scripts/sync_python_core.sh` before building release artifacts so the Python package contains the current CUDA core. See the v0.3.0 release checklist for the retained release evidence and validation commands.

Docker Demo

The published image loads the Nav2 plugin and runs the controller benchmark. The development image also contains the complete short CudaNav loop.

docker run --rm --gpus all ghcr.io/rsasaki0109/cuda-mppi-controller-demo:v0.3.0

# current source checkout: end-to-end CudaNav development smoke
docker build --pull --no-cache -f docker/Dockerfile -t cudarobotics .
docker run --rm --gpus all -v "$PWD/out:/out" cudarobotics cudanav

The `cudanav` mode runs GPU KISS-ICP, voxel mapping, typed ESDF, the Nav2 CUDA MPPI controller, and the command-driven simulator. It writes a machine-checked JSON summary and is a short integration smoke, not the retained 10-minute v1.0 result.

See the Docker README for benchmark modes and output capture. The machine-readable v1.0 support matrix tracks when this development path becomes a release-supported path.

Colab

No local setup needed: open the Colab quickstart.