Isaac ROS Image Pipeline


This metapackage offers similar functionality as the standard, CPU-based image_pipeline metapackage, but does so by leveraging the Jetson platform’s specialized computer vision hardware. Considerable effort has been made to ensure that replacing image_pipeline with isaac_ros_image_pipeline on a Jetson device is as painless a transition as possible.

System Requirements

This Isaac ROS package is designed and tested to be compatible with ROS2 Foxy on Jetson hardware.


  • AGX Xavier or Xavier NX
  • JetPack 4.6


  • CUDA 10.2/11.2 supported discrete GPU
  • VPI 1.1.11
  • Ubuntu 18.04+

Note: For best performance on Jetson, ensure that power settings are configured appropriately (Power Management for Jetson).


Precompiled ROS2 Foxy packages are not available for JetPack 4.6 (based on Ubuntu 18.04 Bionic). You can either manually compile ROS2 Foxy and required dependent packages from source or use the Isaac ROS development Docker image from Isaac ROS Common. The Docker images support both Jetson and x86_64 platfroms. The x86_64 docker image includes VPI Debian packages for CUDA 11.2.

You must first install the Nvidia Container Toolkit to make use of the Docker container development/runtime environment.

Configure nvidia-container-runtime as the default runtime for Docker by editing /etc/docker/daemon.json to include the following:

    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
    "default-runtime": "nvidia"

and then restarting Docker: sudo systemctl daemon-reload && sudo systemctl restart docker

Run the following script in isaac_ros_common to build the image and launch the container:

$ scripts/ <optional_path>

You can either provide an optional path to mirror in your host ROS workspace with Isaac ROS packages, which will be made available in the container as /workspaces/isaac_ros-dev, or you can setup a new workspace in the container.

Package Dependencies

Note: isaac_ros_common is used for running tests and/or creating a development container. It also contains VPI Debian packages that can be installed natively on a development machine without a container.


  1. Create a ROS2 workspace if one is not already prepared:
    mkdir -p your_ws/src
    Note: The workspace can have any name; the quickstart assumes you name it your_ws.
  2. Clone this metapackage repository to your_ws/src/isaac_ros_image_pipeline. Check that you have Git LFS installed before cloning to pull down all large files.
    sudo apt-get install git-lfs
    cd your_ws/src && git clone
  3. Build and source the workspace:
    cd your_ws && colcon build --symlink-install && source install/setup.bash
  4. (Optional) Run tests to verify complete and correct installation:
    colcon test
  5. Start isaac_ros_image_proc using the prebuilt executable:
    ros2 run isaac_ros_image_proc isaac_ros_image_proc
  6. In a separate terminal, spin up a calibrated camera publisher to /image_raw and /camera_info using any package(for example, v4l2_camera):
    ros2 run v4l2_camera v4l2_camera_node
  7. Observe the rectified image output in grayscale and color on /image_rect and /image_rect_color, respectively:
    ros2 run image_view image_view --ros-args -r image:=image_rect
    ros2 run image_view image_view --ros-args -r image:=image_rect_color

Replacing image_pipeline with isaac_ros_image_pipeline

  1. Add a dependency on isaac_ros_image_pipeline to your_package/package.xml and your_package/CMakeLists.txt. If all desired packages under an existing image_pipeline dependency have Isaac ROS alternatives (see Supported Packages), then the original image_pipeline dependency may be removed entirely.
  2. Change the package and plugin names in any * launch files to use [package name] and isaac_ros::image_proc::[component_name] respectively. For a list of all packages, see Supported Packages. For a list of all ROS2 Components made available, see the per-package detailed documentation below.

Supported Packages

At this time, the packages under the standard image_pipeline have the following support:

Existing Package Isaac ROS Alternative
image_pipeline See isaac_ros_image_pipeline
image_proc See isaac_ros_image_proc
stereo_image_proc See isaac_ros_stereo_image_proc
depth_image_proc On roadmap
camera_calibration Continue using existing package
image_publisher Continue using existing package
image_view Continue using existing package
image_rotate Continue using existing package

See also:

  • isaac_ros_apriltag: Accelerated ROS2 wrapper for Apriltag detection
  • isaac_ros_common: Utilities for robust ROS2 testing, in conjunction with launch_test

Tutorial – Stereo Image Pipeline

  1. Connect a compatible Realsense camera (D435, D455) to your host machine.
  2. Build and source the workspace:
    cd your_ws && colcon build --symlink-install && source install/setup.bash
  3. Spin up the stereo image pipeline and Realsense camera node with the launchfile:
    ros2 launch isaac_ros_stereo_image_proc

ROS2 Package API



The isaac_ros_image_proc package offers functionality for rectifying/undistorting images from a monocular camera setup, resizing the image, and changing the image format. It largely replaces the image_proc package, though the image format conversion facility also functions as a way to replace the CPU-based image format conversion in cv_bridge.

Available Components

Component Topics Subscribed Topics Published Parameters
ImageFormatConverterNode image_raw, camera_info: The input camera stream image: The converted image backends: The VPI backend to use, which is CUDA by default (options: “CPU”, “CUDA”, “VIC”)
encoding_desired: Target encoding to convert to. Note: VIC does not support RGB8 and BGR8 for either input or output encoding.
RectifyNode image, camera_info: The input camera stream image_rect: The rectified image interpolation: The VPI interpolation scheme to use during undistortion, which is Catmull-Rom Spline by default
backends: The VPI backend to use, which is CUDA by default (options: “CUDA”, “VIC”)
ResizeNode image, camera_info: The input camera stream resized/image, resized/camera_info: The resized camera stream use_relative_scale: Whether to scale in a relative fashion, which is true by default
scale_height: The fraction to relatively scale height by
scale_width: The fraction to relatively scale width by
height: The absolute height to resize to
width: The absolute width to resize to
backends: The VPI backend to use, which is CUDA by default(options: “CPU”, “CUDA”, “VIC”)



The isaac_ros_stereo_image_proc package offers functionality for handling image pairs from a binocular/stereo camera setup, calculating the disparity between the two images, and producing a point cloud with depth information. It largely replaces the stereo_image_proc package.

Available Components

Component Topics Subscribed Topics Published Parameters
DisparityNode left/image_rect, left/camera_info: The left camera stream
right/image_rect, right/camera_info: The right camera stream
disparity: The disparity between the two cameras max_disparity: The maximum value for disparity per pixel, which is 64 by default. With TEGRA backend, this value must be 256.
window_size: The window size for SGM, which is 5 by default
backends: The VPI backend to use, which is CUDA by default (options: “CUDA”, “TEGRA”)
PointCloudNode left/image_rect_color: The coloring for the point cloud
left/camera_info: The left camera info
right/camera_info: The right camera info
disparity The disparity between the two cameras
points2: The output point cloud queue_size: The length of the subscription queues, which is rmw_qos_profile_default.depth by default
use_color: Whether or not the output point cloud should have color. The default value is false.
unit_scaling: The amount to scale the xyz points by


RealSense camera issue with 99-realsense-libusb.rules

Some RealSense camera users have experienced issues with libusb rules.


[email protected]:/workspaces/isaac_ros-dev$  ros2 launch realsense2_camera
[INFO] [launch]: All log files can be found below /home/admin/.ros/log/2021-10-11-20-13-00-110633-UBUNTU-piyush-3480
[INFO] [launch]: Default logging verbosity is set to INFO
[INFO] [realsense2_camera_node-1]: process started with pid [3482]
[realsense2_camera_node-1] [INFO] [1633983180.596460523] [RealSenseCameraNode]: RealSense ROS v3.2.2
[realsense2_camera_node-1] [INFO] [1633983180.596526058] [RealSenseCameraNode]: Built with LibRealSense v2.48.0
[realsense2_camera_node-1] [INFO] [1633983180.596543343] [RealSenseCameraNode]: Running with LibRealSense v2.48.0
[realsense2_camera_node-1]  11/10 20:13:00,624 ERROR [139993561417472] (handle-libusb.h:51) failed to open usb interface: 0, error: RS2_USB_STATUS_NO_DEVICE
[realsense2_camera_node-1] [WARN] [1633983180.626359282] [RealSenseCameraNode]: Device 1/1 failed with exception: failed to set power state
[realsense2_camera_node-1] [ERROR] [1633983180.626456541] [RealSenseCameraNode]: The requested device with  is NOT found. Will Try again.
[realsense2_camera_node-1]  11/10 20:13:00,624 ERROR [139993586595584] (sensor.cpp:517) acquire_power failed: failed to set power state
[realsense2_camera_node-1]  11/10 20:13:00,626 WARNING [139993586595584] (rs.cpp:306) null pointer passed for argument "device"


  1. Check if 99-realsense-libusb.rules file exists in /etc/udev/rules.d/
  2. If not, disconnect the camera, copy this file to /etc/udev/rules.d/, then reconnect the camera.


Date Changes
2021-10-20 Migrated to NVIDIA-ISAAC-ROS. Fixed handling of extrinsics in Rectify and Disparity nodes.
2021-08-11 Initial release to NVIDIA-AI-IOT


[1] D. Scharstein, H. Hirschm├╝ller, Y. Kitajima, G. Krathwohl, N. Nesic, X. Wang, and P. Westling. High-resolution stereo datasets with subpixel-accurate ground truth. In German Conference on Pattern Recognition (GCPR 2014), M├╝nster, Germany, September 2014.


View Github