Building the world's leading lab for autonomous drone AI

Swarm builds the software that flies drones on their own. Our open benchmark finds the best flight models, and our lab puts them on real hardware. We are working towards physical AI: simulations built as twins of real places, where agents learn to fly drones.

simulated worlds
8
drone lab
Andorra
first on-site test
Q1 2027

From simulation to the field

Autonomy cannot be built in simulation alone. A model earns its place there first, then proves it on real hardware.

  1. Simulation

    A simulated drone over a solar park, seen from aboveSimulation

    Models fly thousands of scored missions in simulated worlds. Every score is public.

  2. Lab

    The view from a DJI drone flying at our Andorra test siteReal flight

    The drone Sentinel will fly, a DJI Matrice 4TD with its dock, is in our lab in Andorra. Our team tests there on real hardware.

  3. On site

    A DJI dock and drone set up for a field test in the mountainsReal hardware

    First customer: solar plants in Greece, feasibility confirmed. The first on-site test is planned for Q1 2027.

Eight worlds to fly in

Streets, trees, slopes, shelves and corridors. Each world asks a model something different, so it has to learn to fly, not to memorise a map.

  • City world, rendered in our simulatorCity
  • Forest world, rendered in our simulatorForest
  • Mountain world, rendered in our simulatorMountain
  • Village world, rendered in our simulatorVillage
  • Warehouse world, rendered in our simulatorWarehouse
  • Open terrain world, rendered in our simulatorOpen terrain
  • Office world, rendered in our simulatorOffice
  • Solar park world, rendered in our simulatorSolar park

The same method carries over to security patrols, search and rescue, infrastructure inspection, mapping and flight indoors, where there is no GPS and no sky.

What makes Swarm unique

  • An open competition, not one team

    Developers around the world train flight models and compete on our benchmark. The network pays the ones that fly best.

  • Real-world twins, validated in our lab

    We build simulations as twins of real places, so agents train fast. What they learn is then tested on real hardware in our drone lab.

  • Fitted to your fleet

    We adapt the model to your airframes, sensors and limits, and deliver it to run on board or at the edge.

From concept to autonomy

Bring us a job your drones should do on their own. We take it from a first simulated mission to a model flying on your hardware.

  1. Mission design

    We turn your operation into a simulated mission with clear rules and a score.

  2. Integration

    We fit the flight AI to your drones, sensors, compute and software.

  3. Field validation

    Our lab takes the model through hardware tests and real flights before it goes live.

Real camera · proof of concept

Open source

Real-time fire detection

A drone spots fire from its own camera and marks the evidence, and an agent decides the next move. All of it is open source.

  • Live video from the on-board camera
  • A fire-detection model on every frame
  • An agent that picks the next command
  • A dry-run mode before any flight
divider