Cooperative drone transport — ARMADA project (LS2N)
Multiple drones carrying a shared payload through a deformable net — a problem where classic model-based control breaks down.
Context
The ARMADA project at LS2N explores cooperative aerial transport: several drones carrying a single shared payload suspended in a deformable net. The elastic coupling between the vehicles makes the system hard to model, and a classic model-based controller struggles to keep the formation stable. I took over an existing research controller that was unstable in flight and reworked it from the ground up, working in close collaboration with the LS2N researchers and one robotics engineer, who carried out the deployment and tests on the real platforms.
What was delivered
I migrated the controller from Python to C++ and replaced the rigid-model approach with a model-free design: per-drone PD control combined with a Moving Horizon Estimation observer that estimates the net disturbances in real time. The work was validated first in Gazebo simulation, then in real indoor flights in the LS2N arena.
- Up to 4 drones carrying a 1 kg shared payload.
- Full pick-up-and-deposit cycles while holding formation within ±15 cm.
- Real indoor flights in the LS2N arena, backed by Gazebo simulation.
A research paper on this work is currently being written; photos and flight video will follow its publication.
Tech stack
C++ control stack running on ROS2, with Gazebo for simulation. This engagement is part of long-term work on low-level UAV control — a C++ and ROS2 stack, Gazebo simulation across multiple vehicle types, and fleet-control applications built on top.