Technology

Detect. Decide. Neutralize. Unattended.

A decoupled edge architecture — a Brain and a Body — that runs the entire kill chain with zero human in the loop and zero cloud.

01 · Live Demo

The system in the field.

Recorded engagement footage of the LT-CUAS platform tracking and engaging aerial targets.

LT-CUAS · live-fire demonstration 1080p · edge replay
02 · The System

Walk a threat through the chain.

Drag the slider to move a hostile track toward the perimeter. Each stage executes on-device — no cloud, no RF link, no operator.

01DetectDual optical/thermal
02ClassifyEdge-AI vision
03FuseEKF tracking
04TrackPan-tilt turret
05NeutralizeKinetic effector
Kill-chain stage
Detect-to-track
Turret latency
Safe mode on loss
The Brain

Edge-AI Fusion & Tracking

  • NVIDIA Jetson Orin AGX / Orin NX ruggedized compute
  • ROS 2 middleware, TensorRT Int8/FP16 inference
  • Dual CMOS/LWIR thermal ingestion, PTP time-sync
  • 3D Extended Kalman Filter fusion, ≤15ms per-frame inference
  • Auto-switches to thermal when optical contrast drops
The Body

Real-Time Actuation

  • STM32H7 (Cortex-M7) MCU on FreeRTOS, ISO-C / Rust
  • Field-Oriented Control on high-torque BLDC gimbal
  • 100 Hz target command bus (CAN-FD / isolated Ethernet)
  • 100ms watchdog → instant "Safe Hold" cutout
  • Multi-stage Ready → Armed → Fire handshake
03 · Subsystems

Engineered for the kill chain.

Switch the protocol to inspect each subsystem. Every layer is modular — if optical fails, thermal keeps tracking.

Optical + thermal fusion & EKF

Dual optical + LWIR thermal streams are fused through a 3D Extended Kalman Filter. PTP hardware timestamping prevents drift during coordinate extrapolation, and predictive intercept math generates live turret vectors. No radar = no RF emissions, no export controls.

  • Inference latency≤ 15 ms
  • Time syncPTP / IEEE 1588
  • Sensor mixCMOS + LWIR (no radar)

Edge-AI object classification

Quantized Int8/FP16 models on the Jetson classify Micro-UAV, Fixed-Wing, Bird and Aircraft with zero cloud reliance. The system auto-blends to thermal when optical contrast collapses — night, fog, or dust storms.

  • Classes4 (incl. bird vs drone)
  • Optical1080p @ 60 fps
  • Thermal640×512 @ 30 fps

Real-time kinematic control

A 100 Hz command bus drives FOC BLDC pan-tilt motors with absolute magnetic-encoder feedback. A 100 ms watchdog cuts power to "Safe Hold" on Brain loss, and geofencing prevents the mechanism from striking its mount.

  • Command rate100 Hz
  • Watchdog100 ms safe-hold
  • Encoder14-bit absolute

Power & low signature

LiFePO₄ / solid-state cells deliver a 24–48h watch cycle. Silent hot-standby at 10W, waking to 50W on detection. 3kW portable units are silent, low-thermal, hot-swappable and daisy-chainable for estate or yacht defense.

  • Watch cycle24–48 hrs
  • Hot standby10 W · silent
  • Power hub3 kW portable · MIL-SPEC
ltcuas@edge:~ — live engagement log
$ ltcuas track --mode autonomous
› optical detect @ 2.4 km · micro-UAV ✓
› EKF fusion locked · contrast low → thermal blend
✓ class: quadcopter · hostile · track locked
$ ltcuas engage --handshake
› Ready → Armed → Fire (CRC-16 verified)
✓ neutralized @ 11ms turret latency · 0 link
$
LayerStackWhy it matters
ControlC++20 / Rust · ROS 2 · rt-kernelDeterministic, memory-safe real-time loops
PerceptionPython · TensorRT · OpenCV · ONNXSub-15ms edge inference, no cloud
Fusion3D EKF · PTP/IEEE 1588Drift-free coordinate extrapolation
ActuationFreeRTOS · STM32H7 · FOC BLDCZero-jitter kinematic control
LinkCAN-FD · CycloneDDS / FastDDSTuned QoS: best-effort sensors, reliable fire

Want the full engineering deep-dive?

Request a technical briefing →