Onboard AI for drones and UGVs
We are building an onboard AI layer for drones and unmanned ground vehicles (UGVs). Our vision goes beyond waypoint navigation: robots that understand your intent, reason about what they see and adapt their next step, with an operator in control.
Aerial and ground autonomy. One intelligence layer.
Approach
Some of the most important work happens in places that are remote, hazardous or costly to reach. We see a future where robots help people understand these environments and act safely within them.
We are developing an intelligence layer built on small language models (SLMs) and vision-language models (VLMs) running on the vehicle's edge computer. It connects what you ask with what the robot sees, so it can judge whether it has enough evidence and decide what to look at next.
Understand the request. Reason from evidence. Adapt.
Turn a natural-language request into a defined task, grounded in what the robot's sensors can see.
Weigh the evidence, quantify uncertainty and choose the most useful next observation.
Propose a new viewpoint or a bounded action. The flight controller keeps low-level control and the operator can override at any time.
Compare observations, confirm whether the task is complete and report what remains unresolved.
Platforms
The same onboard reasoning runs on aerial and ground vehicles: it links language, perception and bounded action, while existing autopilots and safety systems stay in charge of control.
AIRWe are bringing onboard language and vision models to uncrewed aircraft. Give a drone an objective, let it interpret what its camera sees and have it reposition for another view when the first does not answer the question. The flight plan becomes part of the task, not the definition of autonomy.
GROUNDThe same intelligence on a ground vehicle: interpret a request, inspect an asset at close range, compare viewpoints and propose the next useful action. Onboard reasoning turns close-up observations into a clear answer, and flags when more evidence or operator input is needed.
Use cases
The question changes; the principle stays the same: understand the objective, gather the right evidence and work out what comes next. These are the applications shaping our roadmap.
01“Take a closer look at that.”
An unclear observation should trigger a better view. We envision drones that examine a finding, recognise missing detail and capture the evidence an inspector needs to sign it off.
Discuss an inspection task
02“What has changed since the last survey?”
Move from collecting footage to understanding it. Our aim is to compare surveys over time, investigate meaningful changes and deliver a concise report on what needs attention.
Discuss a monitoring task
03“Check it before we send someone in.”
Combine aerial reach with close-range ground views to inspect confined spaces and hazardous areas. The goal is to resolve uncertainty remotely and flag where human judgement is still required.
Discuss a hazardous-area task
04“Can the robot work out the next step?”
Explore how language, vision and physical action work together. We are building towards robots that interpret an objective, choose a useful observation and verify their own results within defined limits.
Discuss a research projectContact
Choose a use case, tell us where the work happens and describe the outcome you need. We will reply with where Kytona could fit and a practical next step.
info@kytona.com