Autonomy software for defence and demanding civil missions. We build for UAVs operating where GPS and communication links cannot be relied on: onboard decisions, mission logic designed for verification, and clear interfaces to your aircraft.
Three foundations connect perception to mission execution. Each module has a defined role, clear interfaces and behaviour that can be examined and tested.
Perception, state estimation, navigation and mission execution are separate modules with defined interfaces. Individual components can be adapted or replaced without rebuilding the whole system.
We design around the aircraft’s flight controller, sensors and onboard computer, keeping the hardware setup lean. Integration happens at the interfaces, allowing the autonomy software to evolve across platforms.
Perception and sensor fusion provide a live estimate of where the aircraft is and what surrounds it, including where GPS is unavailable. Onboard mapping turns sensor observations into a geometric picture of free space and obstacles.
Planning and decision-making run on board, turning that understanding into navigation and mission actions. Shared mission execution was demonstrated with two aircraft in competition; larger fleets are a development goal.
Behaviour trees express mission logic explicitly and remain amenable to formal verification. Clear module boundaries let us test individual components and assess how they interact as a complete system.
[object Object]
When the world differs from the plan, autonomy needs more than another preset rule. We are developing an AI layer that proposes new behaviour, with a classical verification step between the proposal and execution. Scene understanding and agentic reasoning build on our flight-tested navigation and mission software.
A map shows where an obstacle is. A scene graph adds what it is and how it relates to the mission: a doorway, a room, a route between them. We are integrating objects, places and memory so decisions can use context beyond the latest sensor reading.
A blocked entrance should prompt another approach. The agentic layer we are developing uses scene context, mission objectives and domain knowledge to propose what to try next, using the aircraft’s available capabilities.
Our architecture separates proposing a behaviour from authorising it. The verification step is designed to check proposed changes against mission rules and operating limits before the behaviour-tree executive runs them, with a record of the proposal, checks and outcome.
Our focus is defence: reconnaissance, counter-drone awareness and support for responders in contested environments. These are the applications we are developing towards, with the same software foundations supporting civil protection and infrastructure security.
Build a picture of unknown buildings and terrain without GPS. Combine onboard mapping and scene understanding to give operators useful information, with less dependence on continuous piloting or a ground connection.
Bring airborne sensing and object tracking together to support early warning around protected sites and critical infrastructure. Give operators a clearer picture of nearby drone activity and how it changes.
Search unfamiliar spaces, identify people and hazards, and share locations with responders. Coordinated aircraft can extend coverage for defence support and civil rescue missions where access and connectivity are limited.
The brain is not tied to one airframe or one mission. If your problem involves aircraft, autonomy and environments where GPS cannot be trusted, we would like to hear about it.
Start with your aircraft and one mission. Together, we can define the integration scope, agree success criteria and plan a flight demonstration.
Three researchers from robotics, navigation and software security, with an advisory board from both universities.
Leipzig was a warning. We are building towards earlier detection, clearer awareness and more time to act around critical infrastructure.