Counter-UAS autonomy · UAS Groups 1–3

Counter the drone threat — at the edge.

OAK Defense Drone AI is a counter-drone autonomy stack: a 15-app suite, a real drone-type classifier, an onboard agent that drives your EW payload, a ground-station C2 console with mutual-TLS, and a reactive sense → classify → decide engine. CPU-only — it runs on a companion computer, no GPU.

Request a demo See the architecture
DECISION SUPPORT // the brains — the operator owns the licensed RF payload & authority
15
integrated counter-UAS apps
3
UAS groups covered (1–3)
0
GPU required · pure-numpy edge
92%
drone-type classifier accuracy*

*Honest k-fold cross-validation on synthetic data with a 0% bird false-alarm rate; real-data validation labelled pending. See the validation report.

The kill chain

Detect → track → identify → mitigate

The Counter-UAS console scores the full chain against the small/medium drone threat — layered radar, RF, EO/IR and acoustic sensing, then a scored mitigation.

01

Detect

Radar by RCS, passive RF link, EO/IR signature and acoustic — layered.

02

Track

Motion-tracked EO and fused multi-INT tracks build the picture.

03

Identify

Drone-type classifier — 8 classes incl. a bird confuser — from RF, kinematic and micro-Doppler features.

04

Mitigate

RF control-link jam and GNSS denial scored as effect probabilities — gated, never auto-fired.

Architecture

One agent. Your whole payload.

A ground station tasks a headless onboard agent over a tactical data link; the agent drives multiple receive and effector devices through a pluggable adapter + protocol layer. The command/telemetry link is mutually authenticated with TLS.

Ground-station C2 ↔Link-16 / SATCOM · token · mTLS↔ Onboard agent →ESM / ELINT / COMINTRF jammerGNSS jammer

Receivers are receive-only. Effectors are control / set-point only, behind an ARM + dry-run + role + audit interlock — no waveform synthesis, no transmission in the software.

Know the drone

Tell a quadcopter from a bird

  • Eight-class drone-type classifier from RF-link, kinematic and micro-Doppler / acoustic features
  • A dedicated "bird (confuser)" class — 0% bird false-alarm in cross-validation
  • Train it yourself: the AI Model Trainer learns on synthetic or real public RF datasets
  • Explainable AI advisor and a learning cognitive-EW agent on top
AI drone-type identification — classifies drones vs a bird confuser
Onboard agent

The brains, on the platform

  • Headless, GUI-free, pure-numpy — runs on a Linux / ARM companion computer
  • Wire real devices by config, not code: TCP-JSON, serial, SCPI and UDP transports
  • Pluggable wire-protocol codecs — drop in a vendor PDW format as a plugin
  • An on-platform pre-flight panel brings the payload up and checks it out — audited
On-platform pre-flight panel — receivers and gated effectors
Ground-station C2

Decide on the loop, not in it

  • Operator console: per-effector tasking cards discovered live from telemetry
  • The agent recommends — the operator accepts, rejects or holds, with an engagement log
  • Transmit is always a separate, gated action — the software never auto-transmits
  • Mutual-TLS, token auth and auto-reconnect on the command / telemetry bus
Ground-station C2 console — linked to drone-01 over SATCOM
Reactive engine

Sense an event — recommend a response

  • An ordered rules-of-engagement table turns a live track into a recommended response
  • Swarm saturation, INS-autonomous escalation, GNSS / RF defeat — first-match-wins, with a rule trace
  • ADVISE or AUTO (dry-run) — live effector dispatch stays gated
  • Cognitive-EW model reaches ~95% of optimal vs ~58% for a fixed doctrine
Reactive tactical response — ROE decision and rule trace
What's inside

Fifteen apps — one installable product

Sense & identify 6

  • Counter-UAS Console
  • Radar — Echo Generator & Receiver
  • EO/IR Sensor
  • Video Tracker (EO)
  • Multi-INT Fusion
  • Drone-Type ID

Effect & defeat 2

  • Comms Jam + GNSS denial
  • NAVWAR / PNT resilience

Effect probabilities only — no waveform or transmit.

AI core 4

  • AI Model Trainer
  • AI EW Advisor
  • Autonomous Cognitive-EW Agent
  • Adversarial-ML / Counter-AI

Deploy & command 3

  • Ground-Station C2 Console
  • Reactive Response Engine
  • On-platform Pre-flight Panel

Edge footprint

  • CPU-only — no GPU / CUDA / torch
  • ~95–115 MB RAM per app
  • Single ARM core · Jetson / Pi class
  • Windows x64 or run from source on Linux / ARM

Governance

  • Role-based sign-in (RBAC)
  • Classification banners
  • Append-only audit log
  • Effector transmit role-gated & audited
Honest by design

The brains of the fight — not a weapon

Drone AI senses, decides and recommends. It controls the operator's own licensed RF payload behind an ARM + dry-run + role + audit interlock — it contains no exploit code, no waveform synthesis and no transmission. Pilot-ready today; operational fielding is gated on real-hardware test & evaluation, accreditation and transmit / export authority.

Receive-only sensing Gated effectorsMutual-TLS C2 RBAC + auditDry-run pilot plan CPU-only edge

Put it on your platform

From a bench dry-run to a companion-computer integration on your airframe and payload — let's scope a pilot.

Request a demo
DECISION SUPPORT // CONTROLLER, NOT A WEAPON — receive-only sensing · gated effectors · honest by design