What happens when you cross a Qualcomm-powered Rubik Pi dev kit, some Arduino Nicla boards, and Edge Impulse’s machine learning tools? You get one of the coolest vision-and-sensor fusion demos we’ve seen at The Things Conference 2025.
The Edge Impulse Rubik Pi demo turns a racetrack into an AI playground. Overhead, a camera runs Edge Impulse’s FOMO (Faster Objects, More Objects) model to track multiple cars in real time. On each car, an Arduino Nicla Sense ME board with an onboard IMU detects crashes via Bluetooth Low Energy (BLE). Both models run on a single device.
At the heart of this setup is the Rubik Pi, built around the Qualcomm QCS6490 processor. This chip combines a Cortex-A class application processor, an NPU (Qualcomm’s QNN accelerator), and a GPU, all in a Raspberry Pi-compatible form factor. The dev board hosts not just the FOMO model, but also a full web server handling race logic, lap timing, and BLE sensor fusion.
This isn’t just a flashy show-floor demo. It’s a proof-of-concept for real edge AI deployments. The Edge Impulse platform makes it simple to train robust models—even when your dataset includes unpredictable objects like the developer’s cat wandering onto the track. Whether you’re working with vision sensors, IMUs, or BLE, this flow is designed for engineers who want to get working fast.
The training process starts in the Edge Impulse Studio—just upload images or time-series data, label your classes (like “driving”, “crashed”, “idle”), and deploy to your device. The models are optimised to run on constrained hardware and can be quantised to fit low-power targets.
While this race used Arduino and Qualcomm hardware, the workflow is platform-agnostic. Edge Impulse supports a wide range of MCUs and system-on-modules (SoMs) across industrial, consumer, and IoT applications.
From crash detection in micro-mobility to vision-based tracking in robotics, the demo shows how edge intelligence is ready to leave the lab and hit the road.
Comments are closed.
Comments
No comments yet