How it works
We find where your robot’s vision fails, find out why — and fix it with the right data.
We render the conditions your model hasn’t seen (dust, night, rain, mud, fog) on real frames with physics rather than filters, and measure what breaks. Then we trace the failure to what’s missing in your training data and fill that gap — with real data where it exists, simulation where it doesn’t — and prove the gain on your own real frames.
Your frames
A few hundred labelled camera frames from your platform — or 3–5 site photos for a first look. Your model as PyTorch or ONNX if you want a failure map.
Depth per pixel
A monocular depth model estimates how far every pixel is. Haze and dust depend on distance, so this is what makes them physically right.
Physics-based degradation
Dust, night, rain on the lens, mud on the lens and fog, each at three severities. Your labels carry over unchanged — the ground is still where it was.
Output A · Failure map
How much, where, and on which classes your model breaks
- Your model scored on your frames under every condition and severity
- Checked with independent (held-out) generators too, so the result isn’t an artefact of our own simulator
- Per-class and drivable-surface IoU, a failure gallery and a ranked fix list
Output B · The fix
The right data, a retrained model, and proof
- Real data for the gaps: commercially licensed datasets that cover what your model never saw, mapped to your classes
- Simulation where nothing can be captured: physics-based hard cases rendered on your own scenes, labels kept pixel-exact
- Your model retrained, measured against a control run on your real held-out frames — the gain that survives the field
What it did on public data
SegFormer-B0 trained on clear RELLIS-3D frames, then tested on degradations it never saw:
How an engagement runs
- Brief (day 0). You send photos or frames and what goes wrong in the field. We sign an NDA if you need one.
- First look and quote (≤ 2 business days). A few of your own frames under the conditions you care about, and a fixed price for the audit.
- Audit (≈ 10 business days). Your model under every condition, why it fails, and a fix plan with a fixed quote.
- Fix. The right data for the gaps, your model retrained, and a before/after on your real validation frames.
what we won’t claim
- Simulated weather alone rarely fixes real-world failures. In our own tests the right real data did far more — so we only sell simulation where it measurably helps.
- Synthetic data doesn’t replace real adverse-weather testing. It finds weaknesses early and cheaply; your real validation frames confirm the fix.
- Gains measured on our own simulator are larger than on independent ones — so we always report both, and lead with the smaller number.
- Your data is used only for your project and deleted on request. See privacy.