Automated Tuning of Model Predictive Controllers for Cooperative Load Transportation with Quadrotors

This paper proposes Auto-Multilift, a novel framework that automates the tuning of model predictive controllers (MPCs) for multilift systems, where a group of quadrotors cooperatively transport a cable-suspended load. The framework employs deep neural networks to dynamically adjust the MPC hyperparameters online and develops a distributed policy gradient algorithm to efficiently train these neural networks in a closed-loop manner.