Controller

Distributed control architecture.

A central brain MCU handles attitude estimation and stabilization; per-axis FOC nodes handle motor control. Communication runs over deterministic SPI, allowing each axis to remain hardware-isolated.

10kHz

Stabilization

40kHz

FOC current loop

21bit

Absolute encoder · 5.1 µs SPI link

Brain STM32H743
FOC nodes 2 × STM32G474
Drivers 2 × DRV8316
Encoders MT6835 absolute
IMU 2 × ICM-42688-P

The controller PCB integrates the brain MCU and both per-axis FOC nodes on a single round board, sized to mount inside the gimbal sphere.

Controller PCB · v1.2

Controller PCB · v1.2

Vision

Onboard detection and tracking.

Detection and tracking happen entirely on the gimbal — no cloud, no ground-station GPU. The video pipeline is split in two: image processing at the sensor head, inference and tracker on the main board where heat can be managed. Both stages communicate over a dedicated video link.

1080p30

sustained

Detection

Multi-target

simultaneous

Tracking

On-device

no cloud, no GPU

Inference

Detection Onboard YOLO-class object detector
Tracking Multi-target, simultaneous
Models Bring your own or use the bundled detector
Image pipeline Full ISP — handles exposure, color and noise on-chip
Sensors RGB today · thermal-ready (B-330)
Architecture Two-stage FPGA · open IP · no vendor lock-in
Integration Ethernet video (RTP) · MAVLink commands

Built on permissively licensed open IP — no encrypted vendor blobs, no single-supplier lock-in. Detection models are swappable: train and deploy your own without replacing the platform.

Vision subsystem · v1.0