Observability Analysis and Composite Disturbance Filtering for a Bar Tethered to Dual UAVs Subject to Multi-source Disturbances

Lidan Xu, Dadong Fan, Junhong Wang, Wenshuo Li, Hao Lu, Jianzhong QiaoarXiv 2025 · DOI · arXiv

Abstract

This research addresses the challenge of estimating the full pose of a bar-shaped payload carried by two Unmanned Aerial Vehicles (UAVs) using only the drones' onboard odometry. By modeling the system on geometric manifolds and applying the Observability Rank Criterion, the authors demonstrate that the payload state is observable even under external disturbances. They propose a Disturbance Observer-based Error-State Extended Kalman Filter (DO-ESEKF) that simultaneously estimates system states and disturbances, providing a sensor-lite solution for cooperative aerial transportation.

Introduction

Cooperative aerial transportation—where multiple drones carry a single load—is a promising solution for heavy lifting and complex manipulation. However, precise control requires real-time knowledge of the payload's position and orientation (its "pose").

Currently, most systems rely on external motion capture (limited to indoors) or extra sensors like GPS and IMUs mounted on the payload itself. These additions increase cost, system complexity, and weight. A fundamental question in the field has been: Is the payload pose "observable" (estimable) using only the sensors already on the drones? This is particularly challenging because real-world flight is plagued by "multi-source disturbances," such as aerodynamic ground effects, wind, and unpredictable motor thrust.

Methodology

The researchers tackled this problem through a two-step approach: theoretical proof and algorithmic design.

Observability Analysis

They modeled the two-drone-bar system using the Lagrange D'Alembert variational principle. By linearizing this model around an equilibrium state, they applied the Observability Rank Criterion (ORC) to determine under what conditions the payload pose could be recovered from drone positions and velocities.

Schematic of the two-drone-bar system subject to various disturbances.

DO-ESEKF Design

To put theory into practice, they developed a Disturbance Observer-based Error-State Extended Kalman Filter (DO-ESEKF).

  • Manifold-based: Unlike standard filters that treat states as flat Euclidean vectors, this filter operates directly on the complex geometric "manifolds" (like spheres and rotation groups) that naturally describe drone and cable orientations.
  • Composite Filtering: It treats disturbances as "random-walk" biases, allowing the filter to "learn" and compensate for external forces in real-time.
Nomenclature and physical interpretations of symbols used to define the system's states and disturbances.

Key Insights and Ideas

  • The "Two-Disturbance" Rule: The most significant finding is that the system is fully observable if there are two or fewer types of lumped disturbances. If disturbances hit the drones and the payload's translation and rotation all at once, the system becomes unobservable.
  • Offline Calibration is Key: Because of the rule above, the authors suggest that if you calibrate payload properties (like mass and inertia) offline, you can use the drones' sensors to accurately estimate both the payload's pose and the unpredictable flight disturbances (like wind) acting on the drones.
  • Manifolds Matter: Standard linearized observers often fail during agile maneuvers because they approximate the world as "flat." By using an ESEKF that respects the manifold structure of the configuration space, the authors achieved much higher accuracy during high-dynamic flight.

Experimental Results

The authors validated their work through numerical simulations and real-world experiments using two quadrotors and an indoor lab equipped with an OptiTrack motion capture system (MOCAP) for benchmarking.

  • Simulation Performance: In aggressive trajectory tracking, the DO-ESEKF maintained close alignment with ground truth, while a standard ESEKF (without disturbance estimation) showed significant deviations, particularly in the Z-axis (altitude).
  • Experimental Accuracy: In real-world tests, the filter estimated the payload's attitude with a nonlinear error of less than 0.02, which corresponds to an angular error of less than 2 degrees.
  • Robustness: When a sudden disturbance was "injected" into the simulation, the DO-ESEKF converged to the new true values rapidly, demonstrating its ability to handle transient changes in the environment.

Limitations and Future Work

  • Slow-Varying Assumption: The current filter assumes that disturbances change relatively slowly (a "random walk" model). It might struggle with extremely high-frequency turbulence.
  • Parameter Sensitivity: The method relies on knowing certain system parameters (like cable lengths and payload mass). While the authors suggest offline calibration, a fully autonomous system would benefit from adaptive techniques that estimate these parameters online.
  • Computational Load: The manifold-based calculations are complex. In the experiments, the filter was run on a ground station rather than onboard the drones, though the authors aim to optimize this for future edge deployment.

Summary updated

BibTeX
@misc{xu2025observability,
  author = {Lidan Xu and Dadong Fan and Junhong Wang and Wenshuo Li and Hao Lu and Jianzhong Qiao},
  title = {{Observability Analysis and Composite Disturbance Filtering for a Bar Tethered to Dual UAVs Subject to Multi-source Disturbances}},
  year = {2025},
  eprint = {2512.09377},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.09377},
}
Figure