Rigr AI

Visual autonomy for GNSS-denied airspace

Badb Loop

Give it the objective, not the coordinates.

Aircraft that find their own position without GPS, perceive what matters, and move for a better view — trained in synthetic worlds we build ourselves. A person always sets the objective and closes the loop.

See it navigate Join the waitlist
  • Positioning without GPS — demonstrated
  • Perception in hard conditions
  • Trained in synthetic worlds
  • Built for the edge

Demonstrated — real flight footage

The aircraft recovers its own position from what its camera sees — matched to a map built from earlier flights — flying low, with no satellite fix.

  • Coastland → sea
  • Forested hillrelief & canopy
  • Hill & townbuilt-up
  • Nordiclakeshore
Four environments shown. Many more flown — we've been testing across rural, forested and coastal areas.

Low, and from your own flights.

Most camera-based navigation needs altitude and a satellite map. Ours also works down low, against maps built from your own earlier sorties — where satellite imagery is too coarse and looks from the wrong angle. Because the fix is onboard and visual, the aircraft keeps its bearings with no link back home — so it can work low even when the radio is jammed or silent, and flying low is harder to detect, which helps it survive.

  • Positions itself with no GPS at all — jammed, spoofed, or gone
  • Flies low, using maps built from earlier flights
  • Coast, forest, town — not one kind of terrain

Why we build this: to help keep allied forces and civilians across Britain and Europe safer where GPS is denied. Our systems perceive, navigate, and inform — a person always closes the loop.

Talk to us about navigation

How it's built

One loop, from world to model.

The navigation you just saw was built by this loop — the same worlds we generate to train models. The loop is what you can build on.

01

Synthetic environments

Build worlds you can't safely or affordably film. Compose and re-compose seasons, weather, time of day, and sensor models at will — and reproduce any scene exactly, as many times as you need.

Synthetic frame: an armoured column on a forest road, rendered into a 3D reconstruction of a real site.

Synthetic — rendered scene, real surveyed terrain.

02

Ground-truth datasets

Training-ready datasets with precise labels, balanced for the cases that matter. Segmentation, detection, and depth — exported to the formats your pipeline already speaks.

A near-nadir synthetic frame at 58 m with a ground-truth box on each of the four vehicles in the column.

Synthetic — ground truth from the renderer, at 58 m.

03

Perception AI

Detection and segmentation models tuned for cluttered, low-contrast, real-world scenes — plus the tooling to keep them sharp as conditions drift in the field.

Badb Loop perception models detecting vehicles and personnel in a snow scene.

04

Navigation

Find the aircraft on the map with no GPS, from what the camera sees — against maps built from earlier flights. Demonstrated on real flights.

Badb Loop positioning an aircraft on a map with no GPS.

Available today

Every pixel, labelled. Automatically.

Because the world is synthetic, the labels come from the renderer itself — every vehicle boxed, none missed, no annotation queue. Below: CG armour rendered into a 3D reconstruction of a real forestry road in Finland, from 26 m. Drag the handle to compare the rendered frame with its ground truth.

The same synthetic frame with a ground-truth box on each of the three vehicles, labelled tracked IFV, wheeled APC and main battle tank.

Synthetic — CG armour rendered into a 3D Gaussian-splat reconstruction of a real forestry road we surveyed in Finland, lit from a 360° camera frame captured on the same flight. Boxes are the renderer's own ground truth, not a model's output. Labels name the platform type only.

Hard conditions

Trained for the conditions that break models.

Generate the rare, the dangerous, and the degraded as easily as the everyday — then balance your dataset across the whole long tail.

  • Synthetic autumn woodland with an armoured vehicle detected among the trees. Seasonal woodland
  • Snowbound forest with a truck and personnel segmented in low contrast. Snow & low contrast
  • Dense understory with foliage and trunks segmented for occlusion-heavy scenes. Dense occlusion
  • Broken ground and mud surfaces labelled for traversability. Mud & broken ground
  • Top-down aerial view of a vehicle convoy detected on a forest trail, each vehicle in a labelled bounding box. Aerial convoy detection
  • Overhead view of armoured vehicles segmented among snow patches and dense foliage. Top-down segmentation

Build with Badb Loop

Everything you need to build on it.

Data, models, and the tools to put them to work — packaged for the teams building autonomy and navigation. Take what's ready today; the rest is in preview.

Available

Synthetic flight data

Photoreal rendered flight video with perfect ground-truth telemetry, from 3D reconstructions of real terrain — ready to train and evaluate against.

Explore the data Available

Navigation & detection models

Visual positioning for GNSS-denied navigation and aerial object detection, packaged with edge-ready binaries for the hardware a real vehicle carries.

Browse the models In preview

Developer & integration docs

Reference documentation and integration guides to wire the data and models into your own pipeline. In preview — request access.

Request access Coming soon

Interactive sandbox

A hands-on demo to try the models against sample imagery before you commit. Coming soon.

Request early access