Applied research
Navigation that doesn't depend on a satellite.
A small research line, separate from our product work: helping autonomous systems know where they are when satellite positioning is unreliable, degraded or unavailable.
Where this actually stands
This is early-stage research, not a product you can buy. There is a working prototype and a set of open questions. We are not a defence contractor, we hold no contracts we can point to, and we would rather say that plainly than dress a research project up as a capability. If that is the stage you want to talk at, we would like to hear from you.
The problem
Most autonomy assumes it can always ask where it is.
Almost every small autonomous platform in service today gets its position from a satellite constellation. It is cheap, accurate and available — right up until it isn’t. Signal is weak by design, easily overwhelmed, and reflects badly in terrain with steep relief. When the fix goes, a system built around it doesn’t degrade; it stops.
The alternative is for the platform to work out its own position from what it can see. That idea is not new — it is how a pilot navigates with a chart — but doing it continuously, on embedded hardware, over terrain that changes with season and weather, is still genuinely unsolved at the size and power budget that matters.
That is the space we are working in. It is a long research problem and we are early in it.
What we’re working on
Four open threads.
Terrain-relative positioning
Estimating position by matching what the camera sees against known terrain features, rather than trusting a satellite fix. The interesting problems are the boring ones: seasonal change, cloud shadow, and how gracefully the estimate degrades when the ground stops looking like the map.
Inference at the edge
Everything has to run on the airframe. That is a hard constraint — a model that needs a datacentre is not a candidate — so most of the work is in making perception small and predictable enough to run on embedded compute within a fixed power budget.
Degraded-signal behaviour
What a system should do when its primary reference disappears mid-flight is a design question, not just an algorithmic one. We are interested in handover: detecting that the satellite fix has become untrustworthy, and switching reference frames without a discontinuity.
High-altitude conditions
Thin air, extreme cold and low-contrast snow-covered terrain break assumptions that hold at sea level. Indian geography makes this a natural thing for an Indian team to work on, and a hard one to study anywhere else.
Collaboration
We’re looking for the right conversations.
Academic groups working on visual odometry and terrain matching, teams with access to flight-test environments, and anyone who has spent real time on embedded perception. If that is you, write to us and say what you’re working on.