Our products
From an open-source AI drone you can build yourself to a tailored competition that trains a flight model for your own platform.
SOTApilotGeneral UAV navigation
An open-source AI drone navigation stack — obstacle avoidance, objective recognition, and autonomous state-machine control. SOTApilot set a new state of the art on the Swarm benchmark.
SOTApilot
champion 95.34%
Benchmark success
95.34%
The champion score on the open benchmark — one policy, every world, every mission.
Seven worlds, one policy
SOTApilot is scored across seven environments — from dense city to open mountain to the inside of an office. Generalization is the whole game.
City
Dense urban canyons — towers, streets, traffic and streetlights to thread through.
Open
Rolling open terrain with long, exposed sightlines and very few landmarks.
Mountain
Steep ridges, narrow valleys and hard climbs that punish bad altitude control.
Village
Alpine rooftops and sloped, snow-covered streets in a tight settlement.
Warehouse
Indoor racks, conveyors and tight aisle clearances — no GPS, no sky.
Forest
Trees, undergrowth and shifting seasons that clutter every line of sight.
Office
A real office rebuilt as a digital twin — corridors, doorways and furniture at true scale.
InterceptorAir-to-air pursuit
One drone hunts another. The interceptor takes off, flies to a search area and finds a target it has never been given the coordinates of — it sees only what its own camera sees. The task ran until the network beat it, and it is now closed.
Interceptor
winning score 0.9992
Winning score
0.9992
The score that closed the task, out of a possible 1.0000 — reached after 99 models from independent developers had competed for the crown.
Langostino
Langostino is the first AI-autopiloted drone whose development is fully open source — from assembly and wiring to the flight model that acts as its brain.
Follow the open build series below ↓
Simulation League
Partner with Swarm to run a tailored competition inside our Simulation League and train a custom autonomous flight model for your platform.
From pilot to scale
We take autonomous drone systems from initial scope to production deployment, fleet-wide expansion, and continuous improvement.
Pilot · 8–12 weeks
From kickoff to a production-ready v1.
Scale-up · 3–6 months
Expand across missions, sites, and fleets.
Ongoing optimization
Continuous updates, edge-case learning, and model improvement.
