Industries / Unmanned Systems Unmanned Systems

Perception beyond people.

Air, ground, surface, and subsea platforms produce more video than anyone can watch. MAKRR detects, classifies, and tracks in real time, and speaks MAVLink, so the vehicle acts on what it sees, even in offline and airgapped conditions.

01Field demos
Aerial demoVESSEL DETECTION · DRONE FEED
02MAVLink native

Detections become commands.

MAKRR speaks MAVLink, so a detection on the edge becomes a message the autopilot already understands. The vision layer acts, it does not just observe. And the command set is extendable: define custom MAVLink messages for your platform to loiter on a contact, mark and report it, slew a gimbal, hold station, or trigger a payload.

If a vehicle carries a camera and speaks MAVLink, MAKRR can ride it. The same hardware-agnostic model and message flow run across domains, with no per-platform rewrite.

ON THE VEHICLE · EDGE
Detection
class · track · bearing
MAVLink message
standard + custom
Autopilot
loiter · mark · steer
COMMAND SETEXTENDABLE
  • Aerial drones · UAVFixed-wing and multirotor, onboard or driven from the ground station.
  • Ground rovers · UGVWheeled and tracked platforms for patrol, survey, and inspection.
  • Surface vessels · USVDrone boats and autonomous surface craft holding a track on the water.
  • Underwater · AUV / ROVSubmarines and tethered ROVs, where the link is narrow or intermittent.

Any vision-dependent, MAVLink-enabled platform. One model, one message flow, every domain.

03Deployment modes
M1OnboardDeployment-agnostic edge AI on compact compute on the vehicle, edge AI for drones and rovers alike. Detections happen in the field; only lightweight results come down the link.
M2Ground stationProcess feeds from one or many platforms at the GCS: live overlays, alerts, recording. No internet dependency.
M3Hybrid fleetMix both, managed as device groups from one dashboard, with staged over-the-air model rollouts.
04Designed against
  • The link will drop.Inference runs on the edge device; data syncs when connectivity returns.
  • Targets move.SORT, ByteTrack, and BoT-SORT hold identity across frames at full frame rate.
  • Requirements change.Your operators train new target classes on your own footage, in days.
  • Imagery is sensitive.Everything stays on your hardware, inside your infrastructure. Sovereign by default.
The platforms exist. We build the part that sees.

Operating somewhere demanding?

Tell us the platform and the mission profile. We will tell you exactly what is possible.