Research Analyzer
← Back ICRA 2026

Distributed NMPC for Cooperative Aerial Manipulation of Cable-Suspended Loads

Nicola De Carli, Riccardo Belletti, Emanuele Buzzurro, Andrea Testa, Giuseppe Notarstefano, Marco Tognon

PDF

AI summary

Key figure (auto-extracted from paper)
A distributed NMPC framework enables multiple UAVs to cooperatively track the full 6DoF pose of a cable-suspended load without a central coordinator, validated through simulation and real-world experiments.
Distributed MPC Aerial Manipulation Multi-UAV Systems Cable-Suspended Loads ADMM Cooperative Control

Problem

Centralized controllers for cooperative aerial manipulation demand high computational and communication resources, limiting scalability and creating single points of failure. This paper addresses how to achieve scalable, robust, full six-degree-of-freedom pose tracking of cable-suspended loads using a team of UAVs in a fully distributed manner.

Approach

Each UAV solves a local optimal control subproblem and exchanges only the shared load trajectory with neighboring robots via a peer-to-peer network, using a partition-based ADMM algorithm to enforce consensus without a central coordinator.

Key results

  • Optimization complexity remains independent of team size
  • ADMM algorithm tailored to exploit cable-suspended load sparsity
  • Successful real-world validation on the three-robot Fly-Crane platform
  • Significant reduction in communication bandwidth and computational load

Why it matters

Enables scalable, fault-tolerant cooperative aerial manipulation for large-scale applications where centralized computing and high-bandwidth networks are impractical.

Abstract

In this paper, we address the problem of cooperative manipulation of a cable-suspended load by a team of aerial robots. Unlike classical approaches that rely on centralized controllers, we propose a Distributed Nonlinear Model Predictive Control (DNMPC) framework in which the UAVs communicate over a peer-to-peer network a reduced amount of variables. In the proposed method, each robot handles only a small subset of the global optimization problem. The optimal motion computed by the DNMPC loop is then used as a reference for local nonlinear controllers that track the trajectory and compute the robot’s actuation inputs. We validate the proposed scheme both through numerical simulations and real-world experiments on the Fly- Crane system: a rigid platform connected to three robots by pairs of cables.

Index terms

Aerial Systems: Mechanics and Control Multi-Robot Systems Optimization and Optimal Control

Related papers