🧠Edge-AI Vision — flagship prototype platform
| Compute | STM32N6 with the Neural-ART neural-processing unit (NPU) |
| Model | Compact single-shot detector, 128×128 input (face detection shown as the prototype example) |
| Performance | ~4 – 10 ms inference; ~30 – 65 fps end-to-end (on device) |
| Firmware | TrustZone architecture — separated secure and non-secure domains |
| Imaging | Software auto-gain (AGC) control loop; on-device gamma tone mapping to raise detection confidence |
| Sensors | Monochrome global-shutter and colour image sensors |
The face detector is a prototype example — a reference workload on a platform for
building custom on-device AI-vision solutions: swap the neural model, keep the real-time secure pipeline.
Runs at the edge with no cloud or host PC in the loop.
📡RF Link — sub-1 GHz data link
| Band | 868 / 915 MHz ISM (sub-1 GHz), single fixed channel |
| Transceiver | TI CC1200 with CC1190 PA/LNA range-extension front end |
| Modulation / rate | 2-FSK & 2-GFSK, selectable 1.2 – 50 kbps |
| Security | AES payload encryption (per-link key + IV); CRC-16 on every frame |
| Host interface | USB (CDC virtual COM port) or RS-232 / UART |
| Protocol | Modem-style AT command set — discovery, connect / select, transparent serial bypass mode |
| Networking | Multi-node addressing, heartbeat keep-alive, auto-reconnect, RSSI reporting, configurable TX power |
| Controller | STM32L1 low-power MCU |
🚁Flight Controller
| MCU | STM32F4 (ARM Cortex-M4F) |
| Sensors | 9-axis IMU (gyroscope + accelerometer + magnetometer) and barometric altimeter |
| Firmware | Based on the LibrePilot / OpenPilot autopilot; Madgwick attitude estimation with quaternion stabilization at 100 Hz |
| Actuator I/O | 8 PWM outputs (servo / ESC) |
| Inputs | 6 RC receiver channels; GPS over serial; battery voltage / current monitoring |
| Telemetry | UAVObjects / UAVTalk over an OPLink radio link |
| Storage | On-board SPI flash for logging and configuration |
| Modes | Multirotor & fixed-wing stabilization; camera-gimbal stabilization |
📈Telemetry Module
| MCU | STM32L0 (ARM Cortex-M0+), ultra-low-power |
| Measurements | Pack voltage, bidirectional load current (Hall-effect sensor), temperature |
| Acquisition | 12-bit ADC, DMA, timer-triggered periodic sampling with heavy averaging |
| Power | On-board synchronous step-down regulation + low-dropout MCU rail |
| Output protocol | FrSky Smart Port over UART — emulates current, voltage and temperature sensors |
| Alarms | LiPo low-voltage detection across 2S – 6S packs with LED alarm |
🧩ObjectManager — embedded framework
A universal object-registry and event framework for FreeRTOS systems. The conceptual flow:
flowchart LR
APP["Application tasks"]:::pr --> REG["Register named objects"]:::pr
REG --> OM{{"Object Manager<br/>thread-safe registry"}}:::om
OM --> EV["Event / observer<br/>callbacks"]:::pr
OM --> PS["Persistence<br/>Flash / EEPROM + CRC"]:::pr
OM --> CM["Communication<br/>binding"]:::pr
classDef pr fill:#152039,stroke:#38bdf8,color:#e6edf7
classDef om fill:#1e2a4a,stroke:#818cf8,color:#e6edf7,font-weight:bold
| Type | Object-registry & event framework for FreeRTOS (C99) |
| Data model | Named objects addressed by unique 32-bit ID; per-object mutex — fully thread-safe |
| Events | Observer pattern — multi-subscriber callbacks with change notification |
| Persistence | Save / load to Flash / EEPROM with CRC integrity checking |
| RTOS | CMSIS-RTOS v1 and v2 compatible |
| Portability | Hardware isolated behind a HAL port layer; dual toolchain (IAR + GCC) |
| Footprint | ~1.2 KB RAM / ~8.5 KB Flash (core) |
| Documentation | Fully Doxygen-documented API |