AI @ Edge · Applications in practice
AI at the edge. Built to operate.
Explore how RIoT Secure supported Neuton.ai’s GNSS classification at Stockholm Arlanda Airport and Comau’s robotics-monitoring initiative. Local inference runs on dedicated application hardware, with communication, security and updates managed separately.
AI @ Edge showcases
Real devices. Applied intelligence.
These use cases show AI working close to sensors and local controls. RIoT Secure supplies the independent infrastructure needed to connect, update and operate the device over its lifetime.
Airport ground operations · Stockholm Arlanda
GNSS signal intelligence with Neuton.ai.
Indoor/outdoor detection on an 8-bit MCU, using GNSS signal patterns to support local vehicle speed control.
Explore the use case
Industrial robotics · Sensor-based monitoring
Robotics monitoring with Comau.
TensorFlow Lite inference on a dedicated application MCU, connected through a custom interface board and a separate lifecycle engine.
Explore the use caseAirport ground operations · Stockholm Arlanda
GNSS signal intelligence with Neuton.ai.
At Stockholm Arlanda Airport, RIoT Secure collaborated with Neuton.ai on indoor/outdoor classification for baggage-handling vehicles used by SAS and Swedavia. The AI capability was added to an existing device platform to support local speed-limiting decisions at indoor transitions.
The challenge: recognise the transition.
Vehicles needed to respond when entering indoor baggage-handling areas. In this deployment, coordinate-based geofencing was unreliable around buildings and indoor transitions, while connectivity could not be assumed. The application needed a local indoor/outdoor decision without adding beacon infrastructure.
The input: signal patterns, beyond coordinates.
The approach analysed GNSS GSV (satellites-in-view) data, including satellite visibility, signal-to-noise ratios and signal geometry. Neuton.ai’s classifier used those characteristics to infer whether a vehicle was operating indoors or outdoors, rather than trying to improve its absolute position.
The AI: a C library on an 8-bit MCU.
Neuton.ai supplied a compact C-based classification library, integrated into the existing application on an ATmega2560. The application microcontroller handled data processing, inference and local decisions, including a relay that enabled or disabled vehicle speed limitation.
The classification capability was introduced incrementally alongside existing functions such as driver-badge registration and operational data collection. Inference and the local response did not depend on a cloud round trip.
RIoT Secure’s role: independent device operations.
A separate Arduino MKR GSM 1400 or MKR NB 1500, using a secure element, handled communication, firmware delivery and lifecycle management. This Fusion separation kept the application processor’s resources focused on its local workloads while the RIoT Secure platform handled connectivity and software maintenance independently.
What it demonstrates.
Useful edge AI can run on an 8-bit microcontroller and extend an established device application. The Arlanda implementation used existing GNSS signals for local classification, avoided additional beacon infrastructure and kept lifecycle operations separate from the application’s inference and control logic.
Industrial robotics · Sensor-based monitoring
Robotics monitoring with Comau.
Working with Comau’s in-house team, RIoT Secure supported an initiative to deploy machine-learning-based movement, vibration and anomaly detection near robotic systems in automotive manufacturing environments. The team could develop its AI application while RIoT Secure supplied the independent communication and lifecycle layer.
The challenge: take models into operation.
The team could already develop TensorFlow Lite models using sensor data. Deployment also required secure communication, firmware updates and long-term device management in a production-oriented environment. The architecture needed to support AI iteration without making the data science team responsible for those infrastructure workloads.
The AI: dedicated sensor and inference resources.
An Arduino Nano 33 BLE Sense, based on the Nordic nRF52840, ran sensor acquisition and TensorFlow Lite inference. Its CPU, memory and peripherals remained available to the application, with communication and security processing running on a separate microcontroller.
The developers retained their tools, libraries and native firmware workflow. They could develop and test on a standalone Nano 33 BLE Sense, then carry the same application firmware into the integrated system.
RIoT Secure’s role: bridge the application and lifecycle.
RIoT Secure designed the custom interface board linking the Nano 33 BLE Sense to a separate Arduino MKR WiFi 1010 with a secure element. The MKR module handled secure communication, provisioning, firmware delivery and lifecycle management.
This implemented Fusion’s hardware separation between application execution and infrastructure workloads. The interface board was designed to accommodate different communication modules and network topologies, allowing the AI team and lifecycle team to work independently.
What it demonstrates.
AI teams can retain their model-development workflow and dedicated application resources while using an independent platform to connect and maintain devices. This collaboration brought sensor processing, inference, secure communication and software delivery together around a defined hardware boundary.
The common foundation
Local intelligence. Independent operations.
Both showcases apply the same separation of concerns. The application microcontroller owns sensor processing, inference and device behaviour. A separate microcontroller handles communication, security and lifecycle work.
Fusion provides the hardware boundary, while µTLS and Oasis provide protected communication and lifecycle operations. Teams can evolve native application firmware and AI libraries while retaining a common way to connect, update and manage their devices.
From application to managed device
Bring us your edge AI use case.
Tell us about your sensors, target hardware, inference workload and operating environment. We can help define the boundary between your application and the communication, security and lifecycle infrastructure it needs.