Skip to main content
This example brings Edge Impulse on-device inference into an immersive Unity XR app for the Meta Quest 3 / 3S (a headset that runs Android). A virtual conveyor belt carries bottles past an inspection station; an Edge Impulse FOMO object-detection model checks each bottle’s cap, and a reject arm shoves the failures off the belt so they smash on the floor. It all runs locally on the headset, with no network connection. Because the Quest runs Android, the same Edge Impulse C++ library you would deploy to a phone is compiled to a native plugin and called from Unity through a small P/Invoke bridge: the Android series NDK approach, wrapped for the Unity engine. The conveyor inspection demo

Example repository

Unity project, native plugin sources and a prebuilt APK.

Public Edge Impulse project

Clone the bottle-cap FOMO model into your own account.

​
What you’ll learn

  • How to run an Edge Impulse FOMO model on-device from Unity through a native plugin (no cloud calls).
  • How the same Android C++ SDK deployment powers a Meta Quest XR experience.
  • How to feed pixels from a Unity camera into the model and act on the result in the scene.

​
How it works

A virtual “inspection camera” in Unity renders the bottle at the inspection point, hands the pixels to the Edge Impulse model, and reacts to the result.
1

Deploy the model as a C++ library

In Edge Impulse Studio, go to Deployment → C++ library and build. A thin wrapper (native/ei_fomo.cpp) exposes a classify entry point.
2

Compile a native plugin per platform

Build the wrapper to libei_fomo.so (Android arm64-v8a, for the Quest) and libei_fomo.dylib (macOS, for the Editor). Prebuilt libraries are committed in the repository.
3

Run inference from Unity

EdgeImpulseFOMO.cs P/Invokes the plugin; the inspection station classifies each bottle’s cap and flags cap_incorrect as a defect.
4

React in the scene

Defective bottles are pushed off the conveyor by a reject arm and shatter on impact, while cap_correct bottles pass through.

​
Requirements

  • Unity 6000.0.32f1
  • A Meta Quest 3 / 3S in developer mode (or an arm64 Android device)
  • Packages (already in the project): URP, XR Interaction Toolkit, OpenXR, XR Hands
  • To rebuild the native library from your own model: the Android NDK r23b

​
Quick start

Download the latest APK from the repository’s Releases page and sideload it onto a Quest 3 / 3S (or an arm64 Android device):
Or open the project in Unity, open Assets/Scenes/FactoryFloorDemo.unity, and press Play to run it in the XR Device Simulator. A single FactoryDemo Bootstrap object builds the whole scene in code, including the conveyor, inspection station, reject arm, and HUD, so it is easy to read and adapt.
The experience is designed to be hardware-agnostic: the same Android build runs on a Quest headset for room-scale XR, and the project can be built for a phone or the browser. This is handy for classroom demos, teaching, and events.

Android series

The NDK + C++ SDK deployment path this example builds on.

ExecuTorch

Deploy PyTorch models to Android with ExecuTorch and Edge Impulse.