Close
0%
0%

Open-Source 5-Drone PX4 Swarm, Onboard ESP32-P4&S3

Capture-the-flag swarm — dual-ESP32 drones (P4 onboard vision, S3 comms), PX4 flight, ROS2 ground station. Open-source.

Similar projects worth following
We are team Mach Mind, a two-person team — Mindaugas (firmware, software, and electronics) and Hauke (mechanical design and integration). In our spare time we built an autonomous drone swarm for the Swarm Drone Challenge 2026, organised by MBDA and brigkAIR — an indoor capture-the-flag contest flown entirely by autonomous drones, with no pilots in the loop.

What makes it unusual: each drone sees for itself, with computer vision running directly on a bare ESP32-P4 microcontroller — no Raspberry Pi, Jetson, or other Linux companion computer on board. A second ESP32-S3 handles MAVLink and micro-ROS comms, PX4 for flight and a ROS 2 ground station coordinates the swarm.

We were one of six finalists, and at the finals on 11 June at ILA Berlin we finished 4th.

Everything is public: firmware (v1.0 and v2.0), the ROS 2 ground station and swarm algorithms, and the full hardware set — drawings, PCB, and printable STLs. The codebase is mostly C++ and C,

Build details are provided on GitHub together with source code, drawings, photos.


  • 5 × ESP32-P4 Swarm Stack Vision Module
  • 5 × ESP32-S3 Swarm Stack Communication Module
  • 5 × MicoAir NxtPX4v2 PX4 Flight Controller
  • 5 × Cinebot30 Drone Frame
  • 5 × MicoAir MTF-01 Optical Flow & LiDAR Sensor

View all 12 components

  • Project Log

    Mach Mind06/18/2026 at 22:11 0 comments

    Timeline

    • Dec 2025 — Participation confirmed
    • Jan 2026 — Swarm concept
    • Feb 2026 — First prototype built
    • Mar 2026 — First test flight
    • Apr 2026 — Qualifying (2 drones, proof of concept)
    • May 2026 — First test flight with all five drones (Video)
    • Jun 2026 — First swarming behavior (Video)
    • 11 Jun 2026 — Finals (ILA Berlin)

    What's next

    Finishing the ArUco-EKF localization so the drones hold an absolute position reference (which also unlocks real coordinate tasking from the LLM layer) and getting ready for next challenge.

View project log

Enjoy this project?

Share

Discussions

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates