Everything you build on DRONA flies through a compatibility engine, a specs engine, a 3D pipeline, and an AI layer that has to stay honest about what it knows and what it doesn't. This is the research that makes those four layers worth trusting. We publish our methods because engineering tools without transparent methods don't deserve the word "engineering."
DroneConfig type. We report agreement with published reference designs within ±15% on an 18-preset snapshot..vdx is TOML and not JSONThe short answer: humans read it. The long answer involves round-trip invariants, comment preservation, hermetic archivability, and what we learned from FreeCAD's `.FCStd`.
Why specs-core.mjs is a plain ECMAScript module imported by both Node and Vite, and why the small type-erasure cost is worth the single-source-of-truth benefit.
grounded, estimated, conceptualThe three-tier evidence scheme every recommendation passes through, and why we chose to surface the lowest tier to the user rather than pretend certainty.
Our AssistantBlock model treats mid-turn plan, tool-call, checkpoint, and widget events as first-class typed blocks, not as a flat text stream. This makes undo, replay, and partial recovery tractable.
DRONA Labs is a small, rotating authorship group. We publish under the lab name rather than individual names on most papers because our work is collaborative by design and we want the research to travel with the product, not the CV.
Most of our methods are open. If you are doing your own research in drone physics, computer graphics for robotics, or AI-grounded hardware design, we'd love to hear from you — and cite your work where it applies.
Get in touch Follow the lab