UXDuck launched an open-source stack for building custom drones

A team building a new drone today has to write most of the ground software itself. The maker's own tools work with the maker's own hardware. UXDuck launched today on Y Combinator with an open-source core that aims to replace that work.

Ben Butterworth is the founder. He was the twelfth engineer at Helsing, the German defence AI company. There he built the software that scaled Helsing's own drones to thousands of units in service. The YC launch page had 21 upvotes at the time of writing.

What the product is

UXDuck sells itself as "AI-native software for teams building and operating custom drones, from planes to submarines". That covers the parts a team needs around the aircraft itself. Simulation during design. Mission planning. Live monitoring of the fleet. A way to connect new hardware and new protocols.

UXDuck Core, the open-source piece, is the software every team can read and change. It is published at github.com/uxduckhq/uxduck-core. A paid version runs on top of it with features for a company that fields many craft.

Butterworth says the stack has logged more than 500 simulated flight hours already. His team has built a working rover and has an aircraft prototype in progress. No customer is named on the launch page, and no price is given.

Why open source

The open-source part is the choice that makes this different from the established drone software stacks. A maker who writes its own closed tools owns the people who buy the craft. UXDuck is offering the opposite: the team that buys a drone also owns the way it is flown. The source code goes with it.

That matters when the same software has to control a submarine, an aircraft and a rover in one fleet. Those machines carry different sensors, different controllers and different safety rules. A closed stack rewrites them from scratch. An open stack lets a small team patch the piece its own craft needs.

Where the AI sits

UXDuck calls itself AI-native, and the launch page points at two places the model does work. It plans missions from a goal statement. It monitors the fleet and flags a craft that is behaving unexpectedly. Butterworth stops short of saying the AI flies the craft on its own. The pitch is that the model handles the work around a human pilot. That is the role with the most time left in it.

A team weighing the stack can read the code before signing anything. That is the only real test of a claim like "AI-native" for software that will carry weight in the air.

Source

UXDuck: The AI-native software for the next billion drones, Y Combinator, by Ben Butterworth.