ROS 2 (edgeimpulse_ros)

Run Edge Impulse Linux image models in ROS 2 nodes that publish vision messages for robotic perception.

ROS 2 (Robot Operating System) is an open source framework for robot software. It lets the parts of a robot, such as sensing, perception, planning, and motion control, exchange data in real time.

edgeimpulse_ros is an open source ROS 2 package that runs an Edge Impulse Linux .eim model inside a ROS 2 node. The node subscribes to an image topic and publishes the model’s results as vision_msgs messages that the rest of your robot can use.

Edge Impulse vision models running on ROS 2 robots

The node only runs inference. Images go in and vision_msgs messages come out:

camera driver -> edgeimpulse_detector (decode, preprocess, run model, convert) -> vision_msgs topics

It subscribes to a standard sensor_msgs/Image topic, so it fits into an existing pipeline. It supports QoS settings, image_transport, namespaces, and multi-camera setups.

​

Features

• Any camera. Subscribes to any image topic with configurable QoS, and optionally to CompressedImage.
• Common encodings. Decodes bgr8, rgb8, mono8, mono16, bgra8, rgba8, yuyv, uyvy, nv12, and nv21.
• Source timestamps. Copies the input image’s stamp and frame_id to every output, so TF and sensor fusion keep working.
• All image model types. Object detection, image classification, FOMO, and visual anomaly detection (FOMO-AD).
• Original image coordinates. Maps detections from the model’s input size back to the original image, taking the crop, pad, or squash resize into account.
• Standard messages. Publishes vision_msgs/Detection2DArray and Classification, plus latched VisionInfo and LabelInfo.
• Diagnostics. Optionally publishes an annotated debug image and a diagnostic_msgs/DiagnosticArray with FPS, latency, and model health.

​

Prerequisites

• A ROS 2 Jazzy or Rolling workspace. Both ship vision_msgs 4.x. Humble isn’t tested in CI yet.
• The Edge Impulse Linux Python SDK, OpenCV, and NumPy, installed in the same Python interpreter as ROS. The SDK imports pyaudio even for image models, so you also need portaudio19-dev.
• A Linux .eim binary of your model for your device’s architecture.

Install the ROS dependencies and the Edge Impulse Linux SDK for your platform:

Ubuntu

Ubuntu 24.04 pairs with ROS 2 Jazzy. Its system Python is externally managed (PEP 668), so install the SDK into your user site with --break-system-packages:

sudo apt update
sudo apt install ros-$ROS_DISTRO-vision-msgs ros-$ROS_DISTRO-diagnostic-msgs \
                 ros-$ROS_DISTRO-v4l2-camera \
                 python3-opencv python3-numpy portaudio19-dev

pip install --user --break-system-packages edge_impulse_linux pyaudio

Use the x86-64 .eim build, or the aarch64 build on Arm machines.

Qualcomm Linux

On Qualcomm Linux devices, such as the RB3 Gen 2 and other QRB boards, the platform provides ROS 2 and the perception stack. Install the ROS message packages and the Edge Impulse SDK into the platform’s Python interpreter:

sudo apt install ros-$ROS_DISTRO-vision-msgs ros-$ROS_DISTRO-diagnostic-msgs \
                 python3-opencv python3-numpy portaudio19-dev

pip install --user --break-system-packages edge_impulse_linux pyaudio

Use the aarch64 .eim build. See NV12 and Qualcomm QRB cameras to connect the onboard camera.

​

Run the webcam demo

Build the package in your ROS 2 workspace, then launch the demo with a USB camera. The demo needs ros-$ROS_DISTRO-v4l2-camera.

colcon build --packages-select edgeimpulse_ros
source install/setup.bash

ros2 launch edgeimpulse_ros edgeimpulse_with_camera.launch.py \
  model_path:=/absolute/path/to/model.eim \
  publish_debug_image:=true

View the detections and the debug image:

ros2 topic echo /edgeimpulse_detector/detections
ros2 run rqt_image_view rqt_image_view /edgeimpulse_detector/debug_image

​

Use your existing camera node

To use a camera node you already run, launch only the detector and point it at your image topic:

ros2 launch edgeimpulse_ros edgeimpulse_detector.launch.py \
  model_path:=/path/to/model.eim \
  image_topic:=/camera/image_raw \
  image_qos:=sensor_data

The input topic, QoS, resize mode, confidence threshold, and debug overlay are all ROS 2 parameters. The repository includes launch files and a documented config file.

​

NV12 and Qualcomm QRB cameras

The Qualcomm qrb_ros_camera driver publishes NV12 frames. edgeimpulse_ros converts NV12 to BGR inside the node, so you can subscribe to the camera topic without a color conversion node:

ros2 launch edgeimpulse_ros edgeimpulse_detector.launch.py \
  model_path:=/path/to/model.eim \
  image_topic:=/qrb_camera/image \
  image_qos:=sensor_data

​

Visual anomaly detection

FOMO-AD models suit quality control and other tasks that flag unusual scenes. For these models, the node publishes the anomaly grid as boxes on ~/detections, publishes the peak anomaly score on ~/anomaly, and draws the flagged regions on the debug image.

​

Learn more

• Robotics: end-to-end tutorials that use this package.
  • Run Edge Impulse object detection with ROS 2: the edgeimpulse_ros detector reference for the other robotics tutorials.
  • Run Edge Impulse vision on a JetBot with Rubik Pi: perception on a JetBot-class mobile robot.
  • Build a mobile inspection system with Edge Impulse and ROS 2: a rover inspection architecture.
  • Run Edge Impulse pick and place with UNO Q Braccio: robotic arm sorting triggered by detections.
• edgeimpulse-ros on GitHub: setup steps, launch files, examples, and configuration.
• Deploy as a Linux .eim binary.

edgeimpulse_ros is open source under the BSD-3-Clause-Clear license.