Graph-Informed MPC for Drone Swarms

Collaborative multi-drone payload transport with learned residual modeling

I worked on a research project focused on collaborative multi-drone payload transport using model predictive control. The goal was to improve stability and reduce swing during coordinated transport by combining classical control with learned correction terms.

What I built

  • Designed a model predictive control framework for collaborative drone swarms.
  • Trained a graph neural network to predict nonlinear MPC residuals and cable tension forces.
  • Generalized the learned model across variable swarm sizes and payload weights.
  • Improved robustness of transport trajectories under dynamic load conditions.

Why it matters

This work sits at the intersection of robotics, control, and machine learning. The learned graph-based component helps the controller adapt to changing topology and physical conditions, which is especially important for aerial multi-agent systems.