Deep Learning in ROS
Last updated
Was this helpful?
Last updated
Was this helpful?
Note: This tutorial is applicable for Kerloud UAV products equipped with Jetson Nano only.
To develop robotics applications, we can follow the official repository in to integrate Nvidia deep learning capabilities with ROS. The code include serveral ros nodes to deploy networks based on installed jetson inference libraries.
Main directories are listed below for the ros deep learning repo:
(1) launch/: launch files to deploy ROS nodes for deep learning tasks.
(2) src/: source codes for ROS nodes:
node_detectnet: ROS node to deploy the detectnet network for object localization.
node_imagenet: ROS node to deploy the imagenet network for visual recognition.
node_segment: ROS node to deploy the segnet network for semantic segmentation.
node_video_source: ROS node to handle the video input and publish image messages.
node_video_output: ROS node to create video stream with overlayed images.
image_converter.cpp: class to convert images to various ros messages.
Users have to install jetson-inference libraries , ROS and build the ros_deep_learning workspace. The official guide is . For jetson-inference installation, please refer to the previous page.
For ROS melodic, we have to install the prerequisites below:
Then run 'catkin_make' for the workspace under the directory: ~/ros_workspace.
(1) Before proceeding, if you're using ROS Melodic make sure that roscore is running first.
(2) Launch the video viewer to check whether the video stream is OK:
(3) Launch the imagenet node for video recognition:
(4) Launch the detectnode for object detection:
(5) Launch the segnet for semantic segmentation:
For input and output settings, refer to for details
Make sure that you have downloaded necessary networks for jetson-inference. If not, you might consider downloading them manually by following instructions in .