Deep Learning in ROS

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 https://github.com/dusty-nv/ros_deep_learning to integrate Nvidia deep learning capabilities with ROS. The code include several ros nodes to deploy networks based on installed jetson inference libraries.

Code Structure

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.

How to Install

Users have to install jetson-inference libraries , ROS and build the ros_deep_learning workspace. The official guide is https://github.com/dusty-nv/ros_deep_learning#installation. For jetson-inference installation, please refer to the previous page.

For ROS melodic, we have to install the prerequisites below:

sudo apt-get install ros-melodic-image-transport ros-melodic-vision-msgs

Then run ‘catkin_make’ for the workspace under the directory: ~/ros_workspace.

How to Run

  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:

cd ~/ros_workspace
source devel/setup.bash
roslaunch ros_deep_learning video_viewer.ros1.launch input:=csi://0 output:=display://0

For input and output settings, refer to https://github.com/dusty-nv/jetson-inference/blob/master/docs/aux-streaming.md for details

  1. Launch the imagenet node for video recognition:

cd ~/ros_workspace
source devel/setup.bash
roslaunch ros_deep_learning imagenet.ros1.launch input:=csi://0 output:=display://0

  1. Launch the detectnode for object detection:

cd ~/ros_workspace
source devel/setup.bash
roslaunch ros_deep_learning detectnet.ros1.launch input:=csi://0 output:=display://0

  1. Launch the segnet for semantic segmentation:

cd ~/ros_workspace
source devel/setup.bash
roslaunch ros_deep_learning segnet.ros1.launch input:=csi://0 output:=display://0

Make sure that you have downloaded necessary networks for jetson-inference. If not, you might consider downloading them manually by following instructions in https://github.com/dusty-nv/jetson-inference/releases.