T.A.I.L.S. — On Board Node

By Frederick Andrews, Jad Mghabghab, Josué Dazogbo, Maureen Kouassi, Mouad Ben Lahbib, Computer Engineering Students at the University of Ottawa Date: 7 October 2025

Overview

The On Board Node is the aerial intelligence and sensing module of the T.A.I.L.S. (Tactical Aerial Insight and Localization Suite) system. It is mounted on a drone platform and built around a Raspberry Pi 5, equipped with a Hailo-8L AI accelerator, a LoRa-GPS module, and a Raspberry Pi High Quality Camera.

This module performs real-time object detection, GPS-based localization, and LoRa telemetry transmission to the Base Station (Ground Node). The onboard node captures live video streams, detects Points of Interest (POIs) such as humans or vehicles, and transmits both detection results and positional data for mapping in the TAILS mobile app.

Figure 1: Raspberry Pi HQ Camera and Hailo AI Hat mounted on drone

Features

• 🧠 AI Object Detection: Performs onboard inference using a Raspberry Pi and Hailo-8L accelerator.
• 📸 High-Resolution Imaging: Captures real-time video and stills with the Raspberry Pi HQ Camera.
• 📍 GPS Telemetry: Reads coordinates from the LoRa-GPS module and attaches them to detection data.
• 📡 LoRa Transmission: Sends structured telemetry packets to the base station for BLE rebroadcast.
• 🔋 Power-Efficient Operation: Powered by a drone-mounted UPS HAT and LiPo battery system for self sufficiency.

Hardware

• Raspberry Pi 5 (Main processor)
• Hailo-8L AI Accelerator Hat for onboard inference
• Raspberry Pi HQ Camera (12.3 MP sensor, C/CS-mount lens)
• SX1262/L76X LoRa-GPS Module for long-range communication and localization
• 5V UPS HAT for stable power delivery
• Drone Power System (LiPo battery, ESCs, and flight controller)

Software Architecture

The onboard software is written in Python and is organized for modularity and maintainability. It uses computer vision and hardware libraries (for example: OpenCV, Ultralytics/YOLO-style models or Hailo SDK for the Hailo accelerator, and pyserial for serial/GPS/LoRa links). Dependency manifests are kept in src/requirements.txt and src/vision/requirements.txt.

Data flow (high level):

1. Camera Capture — Frames are captured from the HQ Camera.
2. AI Inference — Each frame is processed by the vision pipeline (Hailo-enabled code paths are under src/vision/).
3. Telemetry Construction — GPS coordinates (from src/core/gps) are paired with detections to build a structured telemetry packet.
4. LoRa Transmission — Packets are transmitted by the LoRa driver in src/transmitter/ to the Base Station for further rebroadcast (BLE).

Figure 2: End-to-end test with terminal output captured during integration testing

Repository layout (relevant parts):

OnBoardNode/
├── src/
│   ├── main.py                 # Main entry: initializes camera, model, GPS polling, transmitter
│   ├── core/
│   │   └── gps/                # GPS drivers and manager (L76X driver, gps_manager)
│   │   └── transmitter/        # LoRa transmitter drivers (SX126x)
│   │   └── vision/             # Vision pipelines, Hailo integration, and C++ helpers
│   ├── environment_setup/      # Scripts and venv for environment setup
│   └── utils/                  # Helper utilities (distance, plotting, etc.)
└── README.md

Key scripts and components:

• src/main.py — Project entry point that wires camera, model, GPS, and transmitter.
• src/core/gps/gps_manager.py — GPS polling and parsing logic.
• src/core/gps/L76X.py — Low-level driver for the L76X GPS module (serial parsing, NMEA handling).
• src/transmitter/SX126x.py — LoRa radio control and packet transmit/receive routines.
• src/vision/hailo_apps_infra/detection_pipeline.py — Vision inference pipeline and postprocessing.

This layout reflects the actual repository structure and keeps hardware abstraction, vision inference, and communication layers separated so they can be developed and tested independently.