Distributed MPC for Cooperative Aerial Load Transportation Deployed over Real Communication Networks

Cooperative transport of cable-suspended loads by multiple UAVs allows heavier payloads and full pose control, but centralized coordination scales poorly and introduces a single point of failure. Distributed model predictive control addresses this, yet existing implementations run on a single computer and neglect the packet losses and limited onboard computation of real wireless networks. This thesis presents a distributed nonlinear model predictive control (MPC) scheme for the Fly-Crane, a cooperative aerial transportation system that carries a load by means of two cables per drone, in which the UAVs agree on the shared load trajectory through a relaxed ADMM that tolerates lost and late messages, without any central coordinator. The work covers the whole chain from formulation to flight: besides a second-order formulation, it delivers a newly developed end-to-end software stack, with event-driven real-time execution, latency compensation, safety interlocks, and containerized deployment on the onboard ARM computers. The framework is progressively validated through simulation, multi-node emulation, hardware-in-the-loop tests, and real flights. In flight, the solvers ran onboard the three quadrotors, with the ground station acting only as mission orchestrator. The swarm completed obstacle avoidance, waypoint tracking, and the traversal of a narrow gap, maintaining accuracy and safety despite synchronization drop rates.