Deploy ML models to Arm edge devices using Edge Impulse and AWS IoT Greengrass
Introduction
Understand the Edge Impulse and AWS IoT Greengrass deployment architecture
Select and set up your edge device
Set up your Edge Impulse project
Install AWS IoT Greengrass on your Arm edge device
Store your Edge Impulse API key in AWS Secrets Manager
Create the Edge Impulse Greengrass component
Deploy the component to your edge device
Verify inference and view results
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

Create an Edge Impulse account

Create an Edge Impulse account, clone a pre-built project, build a deployment for your Arm device, and generate an API key. Edge Impulse is an ML platform for building, training, optimizing, and deploying models for edge devices.

To create an account:

  1. Navigate to Edge Impulse Studio and select Sign Up:

    Image Alt Text:Edge Impulse Studio login page with the Sign Up button in the upper-right cornerEdge Impulse sign-up page

  2. Fill in the requested information and select Sign Up:

    Image Alt Text:Edge Impulse sign-up form with fields for name, email, and passwordComplete the sign-up form

  3. After a successful sign-up, a confirmation message appears. Select Click here to build your first ML model:

    Image Alt Text:Confirmation message after successful Edge Impulse account creationSuccessful sign-up confirmation

  4. A wizard appears to help you create a default project:

    Image Alt Text:Edge Impulse new project wizard showing initial setup optionsNew project wizard

    Dismiss the wizard by selecting the - button. This reveals your new default project:

    Image Alt Text:Edge Impulse dashboard showing a newly created default projectNew default project

Now that your account is ready, clone an existing project that already has a trained model. You’ll use this model throughout the Learning Path.

Clone the project into your account

To clone a pre-built project into your Edge Impulse account:

  1. Open the Cat and Dog Detector public project .

    Image Alt Text:Edge Impulse public project page for the Cat and Dog Detector modelPublic project page

  2. Select Clone this project. A dialog appears to confirm the clone:

    Image Alt Text:Clone project dialog with default settings and a Clone Project buttonClone project dialog

  3. Leave the default settings and select Clone Project. The cloning process starts:

    Image Alt Text:Progress indicator showing the project clone in progressCloning in progress

  4. The cloning process takes about 12 minutes to complete. When cloning finishes, a completion message appears:

    Image Alt Text:Completion message indicating the project clone finished successfullyClone complete

    Select Dashboard to view the cloned project. The project should look similar to the following:

    Image Alt Text:Edge Impulse dashboard showing the cloned Cat and Dog Detector project with model detailsCloned project dashboard

Build your project deployment

Edge Impulse Studio provides a workflow to build, train, optimize, and deploy ML models. Take a moment to explore the project dashboard:

Image Alt Text:Edge Impulse Studio dashboard showing the project overview with data, impulse, and deployment sectionsProject dashboard overview

Central to Edge Impulse is the concept of an Impulse. An Impulse is a pipeline that defines how sensor data is processed, what model runs on it, and how results are interpreted.

Your cloned project already has an Impulse called Cat and Dog Detector. Select Create Impulse to see the three main parts: the pre-processing block, the model block, and the post-processing block:

Image Alt Text:Create Impulse view showing the three pipeline blocks: pre-processing, model, and post-processingImpulse pipeline structure

Select Object Detection to see details about the model used in the Impulse:

Image Alt Text:Object Detection page showing the model architecture and training resultsObject Detection model details

The Impulse in this project is already created, trained, and optimized. Edge Impulse provides examples and documentation to guide you through creating your own Impulse from scratch:

Image Alt Text:Edge Impulse documentation page showing available guides and tutorialsEdge Impulse documentation

Now, deploy the model to your specific edge device type. Depending on the hardware that you’re using, choose the matching deployment target:

Image Alt Text:Deployment page showing available target device options including Linux AARCH64 and other platformsDeployment target options

Select the appropriate target for your device and select Build. For example, if you’re using a Raspberry Pi 5 or a Graviton-based Amazon EC2 instance, choose Linux (AARCH64) to run the model on the CPU:

Image Alt Text:Build dialog with the Linux AARCH64 target selected and the Build button highlightedBuild deployment

Note

For edge device targets, select the int8 quantization option before selecting Build. The Linux (AARCH64) target is suitable for many Linux-class Arm-based 64-bit devices where the CPU runs the model.

When the deployment build completes, the browser automatically downloads your .eim file to ~/Downloads. From your desktop, copy the .eim file to your edge device. For example, if the file is named pet_detection-linux-aarch64-v137-cat-and-dog-detector.eim, run:

    

        
        
scp pet_detection-linux-aarch64-v137-cat-and-dog-detector.eim <username_on_edge_device>@<edge_device_ip_address>:.

Next, open an SSH session on your edge device:

    

        
        
ssh <username_on_edge_device>@<edge_device_ip_address>
mv *.eim /tmp/currentModel.eim

You’ll see /tmp/currentModel.eim on your edge device and use this file later.

Create your project API key

The Edge Impulse Linux Runner on your device uses an API key to authenticate with your project and download the model. The key connects the Greengrass component to your Edge Impulse project.

To create the key:

  1. Select Dashboard:

Image Alt Text:Edge Impulse project dashboard with the Dashboard link highlighted in the left sidebarProject dashboard

  1. Select Keys:

Image Alt Text:Dashboard view with the Keys tab visible in the project settings areaKeys tab

  1. Select Add new API key.
  2. Enter a name for the key, set the role to admin, and confirm that Set as development key is selected.
  3. Select Create API key:

Image Alt Text:API key creation dialog with fields for name, role set to admin, and the development key checkbox selectedCreate API key

The API key appears on the screen. Copy and save the key immediately — this is the only time that the full key is visible. You’ll store this key later in AWS Secrets Manager.

What you’ve accomplished and what’s next

You’ve created an Edge Impulse account and cloned a pre-built Cat and Dog Detector project. You’ve built a deployment for your Arm device and generated an API key.

Next, you’ll install AWS IoT Greengrass on your edge device.

Back
Next