πŸ›°οΈ UAV Airspace Digital Twin

Up here the world is empty. The only question is what else is moving near your flight path.

A BVLOS courier flies a planned route by itself. Beside it runs a Doppler-radar detect-and-avoid twin β€” a safety-cage that tracks every intruder, forecasts closest-point-of-approach, and re-plans by air-law right-of-way when a conflict is coming. It is fully client-side: fly it yourself in the browser, no backend, no install.

πŸ“‘ Doppler, not LiDAR

Radial velocity is measured per return, directly β€” robust to rain, fog and dust, and exactly the "detect-and-avoid radar" a SORA authorization is built around.

βš–οΈ Right-of-way autopilot

On a forecast conflict the ego gives way or stands on by the rules of the air β€” turning, climbing or diving under a crossing track to maximize the miss distance.

🧊 Adaptive voxel twin

The radar rebuilds the static world as an adaptive octree β€” flat ground as big cells, masts and rotor discs sharp β€” while a plausibility veto keeps clutter out.

Research rig Β· synthetic scenario Β· real DGM terrain + OSM obstacles Β· not a certified detect-and-avoid system

β–Ά

Fly it β€” live co-simulation

The ego UAV flies itself on a looping route; intruders (a crossing drone, a descending bird, a helicopter transit) fly their own tracks. Watch the course change grey β†’ red (fresh) β†’ blue (settled) as the twin re-plans. Drag to orbit, scroll to zoom.

β†— Open in a new tab Best on a desktop GPU. The scene runs entirely in your browser.
?

How the twin reads the sky

A sparse-airspace safety-cage: the static world comes from a terrain and obstacle database, the radar finds the movers, and physics decides what is real.

🎯 Persistent tracks that coast

Intruders are tracked in 3D (constant velocity, Doppler-fused). On signal loss a track dead-reckons and its uncertainty cone grows, until a fresh return re-acquires and snaps it back.

🚫 Newtonian plausibility veto

Every return is cleared against inertia and geometry before it may update a track. A teleporting clutter point or a physics-violating jump is an error, not a detection.

🧭 CPA conflict + air-law routing

Each track is forecast with its cone and run through closest-point-of-approach against the ego. Give-way turns and passes behind; stand-on holds β€” so two aircraft never oscillate.

πŸŒ€ Rotor micro-Doppler, discriminated

A spinning turbine is mapped as obstacle geometry yet told apart from a free flyer by its micro-Doppler signature, so a rotor disc never becomes a phantom intruder track.

Talk through applying this to your operation

BVLOS detect-and-avoid, airborne-radar tracking, or a safety-cage digital twin for your own route network β€” if that is where your release argument stalls, let's compare notes. One message to me, a short reply, nothing more. No sales database.

Method, data sources, and honest limits

The world you fly over is real β€” DGM terrain plus OSM turbines and power lines, baked into a small geodata file and inlined into the page. The scenario (the ego's route and the intruder tracks) is synthetic and deterministic: no public airborne-Doppler dataset exists, so the moves are hand-authored while the physics β€” tracking, the plausibility veto, CPA conflict, the right-of-way logic β€” is real. Doppler is blind to purely tangential motion, so the mover gate is a strong positive signal only; position tracking runs for everything.

  • Sensing β€” FMCW / Doppler radar: radial velocity per return, ego-motion compensation, a mover gate.
  • Tracking β€” 3D constant-velocity Kalman filter, position fused with the Doppler scalar, with a growing uncertainty cone.
  • Safety logic β€” closest-point-of-approach conflict detection and avoidance that re-plans by the rules of the air.
  • SORA / BVLOS β€” "detect-and-avoid" is the specific mitigation a BVLOS operating authorization is built around.

This is a research rig for the sensing and navigation ideas β€” not an airworthiness artifact, a conformity assessment, or a certified detect-and-avoid system. Confirm any operational scope with a qualified assessor.