Bavovna
UKR
YOLO
TensorFlow.NET
Terrestrial Navigation
Object recognition
In-house solution, not ArduPilot

Tactical Navigation & Intelligence in Denied Environments

Bavovna Defence Systems engineers autonomous flight-control, neural vision, and Terrestrial Navigation terrain-matching systems that operate completely independently of GNSS, satellite communication, or external signals.

Explore Product Specifications

Next-Generation Autonomy for Contested Space

Bavovna delivers a completely closed-loop, jam-resistant, autonomous control suite for unmanned systems. We unify a deterministic hardware runtime, real-time visual intelligence, and absolute terrain contour referencing into a single high-performance ecosystem.

System

An ecosystem of our own

Built from scratch for a focused range of tasks — a complete ecosystem for autonomous flight control.

  • Flexible sensor integration
  • Fast data processing
  • Multi-platform

Navigation in GNSS-denied environments

Full operation on terrain and visual landmarks without GNSS.

Terrestrial Navigation
OptimisationHardware

Built to run on constrained resources

ExampleNVIDIA JetsonRaspberry Pi + AI ModuleOrange PiESP32

Why us

ArduPilot
  • Hard to integrate with different sensor types
  • Limited data processing and execution speed
  • Requires dedicated flight controllers
Bavovna Flight Control
  • Flexible integration with different sensor types
  • Fast data processing
  • A complete solution for autonomous flight control

Tactical Product Suite

Bavovna Runtime

Deterministic flight control loop

A proprietary, high-reliability execution engine built from the ground up for real-time UAV control on physical airframes. A tick-driven, high-integrity component model designed for safety-critical execution.
  • Integrated sensor abstraction: Unified interface linking all physical sensors, estimators, and actuators seamlessly
  • Deterministic hot-path: High-speed, lock-free ticking using zero-allocation scheduling and hard real-time safety for absolute timing accuracy
  • Complete platform independence: Runs multirotor, fixed-wing, and hybrid platforms identically on budget companion computers or SBCs
  • Independent codebase: 100% owned IP and free of licensing restrictions, optimized for direct hardware compilation
Bavovna.Visio

Real-time video stream processing

Used for landmark recognition and terrain-contour deviation tracking, target detection and lock-on.

Hardware

The solution is built to run on NVIDIA Jetson and Raspberry Pi with an AI module, with all processing done directly on the airframe.

The neural part runs on TensorFlow.NET and YOLO models; the classical part on OpenCV (for ESP32 or Raspberry Pi builds without an AI module).

Neural detection

Real-time object detection and classification models trained both on public datasets (for example, labelled crossroads) and on specialised closed datasets from Brave1.

Capabilities

  • Neural and classical detection: AKAZE, BRISK, ORB, SIFT, Shi-Tomasi, Harris.
  • Attention regions: contrast-based saliency maps and cascaded contour filtering by area, distance and clustering.
  • Real-time map mosaicking into an orthophoto of the terrain.
  • Multi-object tracking.
  • Media transforms and HUD overlays on top of the video stream.
Bavovna.Ganymede

Unjammable terrain-referenced navigation (TRN)

A GPS-independent, unspoofable absolute positioning system that fixes the aircraft's coordinates by matching physical ground elevation scans against on-board maps. It requires no external signals and has zero drift over long distances.
  • Real-time contour-matching: Continuous terrain-clearance profile matching that maintains navigation hypothesis and tracks position with high accuracy
  • Optimized lookup database: Offline preprocessed signature layers allowing lightning-fast positioning fixes on any flight path
  • Continuous fused navigation: High-fidelity fusion filter integrating multiple positioning sensors to achieve robust tactical accuracy
  • Autonomous cold start: Acquires location from an unknown launch coordinate inside a 1 km-diameter uncertainty box and reports time-to-first-trusted-fix
  • Built-in integrity checking: Scales confidence by terrain uniqueness and rejects sensor anomalies via smart validation checks

Contacts

Get in touch

Tell us about your platform, operating conditions and mission profile, and we will propose a Bavovna module configuration for it.

Email
contact@bavovna.solutions
Location
Ukraine
Send us a message