Skip to main content

​
Robotics with Edge Impulse and ROS 2

Robotics hardware used throughout the Edge Impulse robotics tutorial series
This series shows how to build practical robotics systems with ROS 2 and Edge Impulse. The tutorials start with environment setup and model deployment, then progress into full robot workflows for mobile robots, camera-based inspection, and robotic-arm pick-and-place tasks. Across the series, the common pattern is to train a model in Edge Impulse, run inference locally on the robot or host computer, and pass the result into ROS 2 so robot-specific nodes can decide what to do next. Depending on the workflow, that local runtime can be either a Linux .eim deployment or a Python/Linux runner command.

​
What the series covers

  • Setting up Linux and ROS 2 for robotics development.
  • Exporting Edge Impulse models for local Linux inference.
  • Publishing object-detection results as ROS 2 messages.
  • Connecting perception output to robot-specific behavior nodes.
  • Testing robot motion and perception separately before integrating them.
  • Moving from simulation and bench tests to physical robot workflows.

​
Foundations and setup

Run Edge Impulse Object Detection with ROS 2

Run a Linux .eim object-detection model on a ROS 2 image topic and publish detection messages.

​
Robot tutorials

Run Edge Impulse Vision on a JetBot with Rubik Pi

Combine JetBot-class motor control on Qualcomm Rubik Pi with Edge Impulse object detection.

Build a Mobile Inspection System with Edge Impulse and ROS 2

Design a ROS 2-native inspection rover with detection, anomaly scoring, gimbal control, logging, and safety layers.

Run Edge Impulse Pick and Place with UNO Q Braccio

Use Edge Impulse object detection, ROS 2, and an Arduino UNO Q to sort colored blocks with a Braccio arm.
Simulation, topic classes, and what’s nextThese tutorials also work with simulated robotic arms — you can develop and test the full perception-to-motion pipeline in Gazebo before touching hardware, then point the same nodes at a physical arm by swapping the camera and joint-command topics.Across the series you work with a consistent set of ROS 2 message classes: perception input as sensor_msgs/Image, detection and classification output as vision_msgs (Detection2DArray, Classification), model health as diagnostic_msgs/DiagnosticArray, and actuation as std_msgs, sensor_msgs/JointState, and trajectory_msgs. Learning this mapping once carries across every robot in the series.We plan to cover imitation learning for robotic arms in an upcoming tutorial, extending this detection-driven workflow toward learned, demonstration-based manipulation.
Start with the shared Edge Impulse ROS 2 detector tutorial. It explains the .eim deployment, image-topic subscription, output topics, and runtime parameters used by the JetBot workflow and by generic inspection systems. The UNO Q Braccio workflow uses a repo-native Edge Impulse runner pipeline instead of the shared edgeimpulse_ros .eim wrapper. Read the shared detector page for background, but follow the Braccio tutorial for the exact arm launch commands and topics.

​
Prerequisites

Before starting the tutorials, have the following ready:
  • A supported Linux development environment.
  • Basic familiarity with ROS 2 nodes, topics, launch files, and workspaces.
  • An Edge Impulse account and an object-detection or anomaly-detection project.
  • Access to the target robot or a simulation environment for the tutorial you follow.
  • Required sensors and actuators, such as a USB camera, motor controller, gimbal, or gripper.

​
Safety notes

Robotics examples can move physical hardware. Test perception, networking, and motion separately before connecting them into a live workflow. Keep physical access to emergency power, use conservative speeds, and avoid unattended operation until the behavior has been validated in the final environment.

​
Resources