ROS + CUDA Docker Images

This repository provides scripts to build Docker images combining ROS (Robot Operating System) and NVIDIA CUDA for GPU acceleration based on Ubuntu images. You can use pre-built images from DockerHub or build your own combinations.

---

Pre-built Images on DockerHub

Some images are available pre-built on DockerHub:

docker pull annazabnus/ros-cuda:{<X.Y>}{-<ros_distro>}

Example:

docker pull annazabnus/ros-cuda:12.6-jazzy

Published Versions

Below is a complete list of pre-built ROS and CUDA versions. Only the listed ROS distributions are supported at the moment.

• noetic: 11.4, 11.8, 12.0, 12.2, 12.4, 12.6, 12.8
• humble: 11.8, 12.0, 12.2, 12.4, 12.6, 12.8
• jazzy: 12.6, 12.8

System Requirements for Using Images

• Docker: Version 20.10+
• NVIDIA GPU Drivers: Compatible with the chosen CUDA version
• NVIDIA Container Toolkit: For GPU support in Docker

---

Building Your Own Images

Run the provided script to build an image for your desired ROS and CUDA versions:

python3 build.py --ros <ros_distro> --cuda <cuda_version>

• <ros_distro>: ROS distribution (e.g., noetic, humble, jazzy). Leave empty or use none for no ROS.
• <cuda_version>: CUDA version in X.Y format (e.g., 12.4). Use none to skip CUDA.

Examples:

• Build ROS Noetic with CUDA 12.4:

  ./build_image.sh --ros noetic --cuda 12.4

• Build ROS Humble while detecting the local CUDA version:

  ./build_image.sh --ros humble

• Build ROS Jazzy without CUDA:

  ./build_image.sh --ros jazzy --cuda none

System Requirements for Building Images

• OS: Linux (Ubuntu 20.04+ recommended)
• Docker: Version 20.10+
• Docker Buildx: Required for caching and multi-stage builds
• Python 3.8+: For helper scripts
• yq: YAML processor for bash scripts

---

Installed Libraries and Tools

Each image includes the following major libraries and tools:

• ROS: ros-<distro>-desktop-full
• CUDA Toolkit: Installed via cuda-toolkit-<version>
• Additional Tools:
  • python3-colcon-common-extensions (for ROS 2)
  • python3-catkin-tools (for ROS 1)
  • build-essential, cmake, and other development utilities

---

Auto-selection of Base Images

For CUDA builds, base images are dynamically pulled from the official NVIDIA DockerHub repository. The selection process:

• Ignores runtime and cudnn images.
• Automatically selects the latest compatible tag based on the requested CUDA version and Ubuntu release.

---

Contributing

Feel free to submit issues or pull requests for new features, supported versions, or bug fixes.

---

License

This project is licensed under the Apache 2.0 License.