Abstract
This paper presents a coordinate-free dynamical model for a team of quadrotors transporting a flexible hose, modeled as a series of discrete links. By employing geometric mechanics on $S^2$ and $SO(3)$, the authors avoid singularities inherent in local coordinate representations. The system is proven to be differentially flat, enabling efficient trajectory generation. A robust tracking controller is developed using variation-based linearization and finite-horizon LQR, which is shown in simulation to significantly outperform standard feed-forward control in dampening hose oscillations.
Introduction
Aerial manipulation is an evolving field, yet most research focuses on rigid payloads or single-cable suspensions. Transporting a flexible hose with multiple quadrotors introduces significant complexity due to the hose's high degrees of freedom and the under-actuated nature of the system. Traditional models using Euler angles often suffer from mathematical singularities (gimbal lock) when the payload swings aggressively. This work addresses these challenges by developing a singularity-free, computationally tractable model for collaborative hose transport.
Methodology
The authors approximate the continuous hose as a chain of $n$ massless links with lumped masses at the joints. To ensure global validity and avoid singularities, the state of the links is represented using unit vectors ($q_i \in S^2$) and quadrotor orientations using rotation matrices ($R_j \in SO(3)$). The equations of motion are derived via the Lagrange d’Alembert Principle.
A key contribution is the proof of differential flatness. The entire system state can be determined from a set of "flat outputs"—specifically the start position ($x_0$), the yaw of the drones, and the tension in specific links—along with their derivatives. This property simplifies trajectory planning into an algebraic problem. For control, the nonlinear dynamics are linearized directly on the manifold using variation-based techniques, followed by the application of a Finite-Horizon Linear Quadratic Regulator (LQR) for path tracking.
Results and Discussion
The proposed framework was validated through numerical simulations in MATLAB. A critical finding was the comparison between the LQR controller and a standard position controller with simple feed-forward (FF) forces. As shown in Figure 5, the LQR controller effectively dampens oscillations and ensures convergence of the link orientations ($\Psi_q$) and quadrotor rotations ($\Psi_R$), whereas the feed-forward approach leads to sustained instability and tracking errors.
The system successfully tracked complex trajectories, such as circular paths, even with initial state errors. However, computational scaling remains a challenge; simulation time increases super-linearly with the number of hose links ($n$). For instance, a 30-link simulation required roughly 600 seconds for a 10-second flight, indicating a need for optimization for onboard deployment.
Conclusion and Future Work
This research provides a rigorous foundation for multi-robot flexible payload transport. Future iterations will address the dynamics of internal fluid flow (water), the mechanical stiffness of the hose, and the development of state estimators to replace the current assumption of full-state knowledge.
Summary updated