⚡ electrical engineering

Observer-Based Estimation and Hydrostatic Inertia Modeling for Cooperative Transport of Variable-Inertia Loads with Quadrotors

This paper presents a control framework for cooperative quadrotor transport of variable-inertia fluid payloads that combines a geometric tracking controller with an observer-based estimator, utilizing a pre-computed hydrostatic inertia surrogate indexed by fill level and attitude to approximate complex internal fluid dynamics without requiring real-time identification of the full inertia tensor.

Original authors: Jacob Goodman, Leonardo Colombo, Juan Giribet

Published 2026-03-02

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jacob Goodman, Leonardo Colombo, Juan Giribet

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are part of a team of four drones trying to carry a heavy, wobbly bucket of water from one place to another.

Here is the problem: As they fly, the water inside the bucket sloshes around. Sometimes the bucket is full, sometimes half-empty, and the water is constantly shifting. This changes two critical things:

How heavy it is (Mass).
How hard it is to spin (Inertia).

If the drones don't know these numbers are changing, they will crash. They might try to lift a "heavy" bucket that has suddenly become light, or they might try to turn a "stiff" bucket that has suddenly become floppy.

This paper presents a clever solution to this problem using a mix of math, physics, and a little bit of magic (estimation).

The Three Main Tricks

1. The "Smart Guessing" Game (Mass Estimation)

Usually, if you want to know how much a bucket weighs, you put it on a scale. But these drones don't have scales, and the water is leaking or being poured out while they fly.

Instead, the researchers taught the drones to play a game of "guess the weight" in real-time.

The Analogy: Imagine you are pushing a shopping cart. If you push hard and it moves slowly, you guess it's heavy. If you push with the same force and it zooms away, you guess it's light.
How it works: The drones constantly measure how fast they are accelerating and how hard they are pushing. They compare this to their "expected" behavior. If the math doesn't add up, they instantly update their internal guess of the weight. It's like a self-correcting GPS that says, "Wait, I'm moving faster than I should be for a 10kg load. I must be carrying 5kg now."

2. The "Frozen Water" Cheat Code (Inertia Modeling)

Knowing the weight is one thing, but knowing how the water sloshes to change the "spin-ability" (inertia) is incredibly hard. The water is a fluid; it moves in complex, chaotic waves. Calculating the exact physics of every ripple in real-time would require a supercomputer.

The researchers found a shortcut. They realized that if the drones fly smoothly (without sudden jerks or sharp turns), the water inside doesn't have time to slosh wildly. It acts like it's frozen in place, just tilted by gravity.

The Analogy: Think of a glass of water in a car. If the car drives smoothly, the water stays flat. If you slam on the brakes, the water spills. The researchers decided to drive the "car" (the drones) so smoothly that the water stays flat.
The Cheat Code: Because the water stays flat, they didn't need to calculate the messy waves. Instead, they pre-calculated a giant lookup table (like a menu at a restaurant) before the flight even started.
- Scenario: "If the bucket is 50% full and tilted 10 degrees to the left, the 'spin-ability' is X."
- During flight: The drone just checks its current tilt and estimated weight, looks up the answer in the table, and instantly knows how to spin. No complex math needed while flying.

3. The "Dither" Dance (Making the Drones Wiggle)

There is a catch. For the "Smart Guessing" game (Step 1) to work, the drones need to move enough to get good data. If they just hover perfectly still, they can't tell if the weight changed.

The Analogy: If you are trying to guess the weight of a suitcase by lifting it, you need to actually lift it, not just hold it still.
The Solution: The researchers programmed the drones to add tiny, invisible "wiggles" (called dithers) to their flight path. It's like a dancer doing a tiny, subtle shimmy. To the human eye, the flight looks smooth, but to the math, these wiggles provide just enough data to keep the weight estimate accurate.

The Result

The paper shows that by combining these three ideas:

Guessing the weight based on movement.
Using a pre-made menu for how the weight spins (because they fly smoothly).
Adding tiny wiggles to keep the guess accurate.

...a team of drones can successfully carry a leaking, sloshing bucket of water through the air, adjusting their flight plan on the fly without crashing.

Why This Matters

This isn't just about buckets of water. This technology could be used for:

Search and Rescue: Dropping supplies that might be leaking or changing weight.
Construction: Moving liquid concrete or fuel between sites.
Space Exploration: Transporting fuel tanks where the fuel is constantly sloshing in zero gravity.

In short, the paper teaches robots how to be flexible thinkers when carrying messy, changing loads, turning a chaotic fluid problem into a manageable, smooth flight.

1. Problem Statement

The paper addresses the challenge of cooperative aerial transportation using a team of quadrotor UAVs carrying a rigid payload suspended by elastic cables. The core difficulty lies in handling time-varying mass and inertia, specifically scenarios involving fluid-carrying payloads (e.g., water tanks, chemical transport) where:

Mass ( $m_L(t)$ ): Changes over time due to leakage, discharge, or evaporation.
Inertia ( $J_L(t)$ ): Changes dynamically due to the internal redistribution of fluid (sloshing) and the changing center of mass.

Key Limitations of Existing Work:

Most geometric control models assume constant mass and inertia.
Real-time identification of the full inertia tensor is impractical for fluid loads because internal fluid motion is governed by high-dimensional partial differential equations (PDEs) and is not directly measurable with standard sensors.
Standard controllers degrade in performance when these parameters vary, leading to instability or poor tracking.

2. Methodology

The authors propose a unified framework combining geometric control theory, online parameter estimation, and offline hydrostatic modeling.

A. Geometric Modeling

Configuration Space: The system is modeled on the manifold $Q = (SO(3) \times \mathbb{R}^3) \times (S^2)^N \times (SO(3))^N$ , representing the load, $N$ cables, and $N$ quadrotors.
Reduced Dynamics: Using variational calculus on manifolds, the authors derive reduced Euler-Lagrange equations that account for time-varying mass ( $\dot{m}_L$ ) and inertia ( $\dot{J}_L$ ).
Control Architecture: A geometric tracking controller is designed. The control inputs are decomposed into components parallel and perpendicular to the cable vectors to decouple load dynamics from cable dynamics. The controller utilizes feedback linearization terms based on estimated states.

B. Online Mass Estimation (Observer Design)

Since direct mass sensing is often unavailable or unreliable, an adaptive observer is developed:

Parametric Model: The mass is modeled as a function of parameters $\theta$ (e.g., constant mass, Torricelli leak, or viscous leak).
Gradient Descent: An update law for $\hat{\theta}(t)$ is derived by minimizing a quadratic cost function based on the residual between commanded forces and the measured kinematic response ( $\dot{v}_L$ ).
Stability Analysis: The error dynamics are analyzed using a Lyapunov-like approach. The authors establish that the estimation error is exponentially input-to-state stable (eISS) provided a persistent excitation (PE) condition is met. This implies the trajectory must have sufficient acceleration/velocity variation to "excite" the mass dynamics.

C. Hydrostatic Inertia Surrogate

Instead of solving complex fluid dynamics online, the authors propose a computationally efficient surrogate:

Assumption: Under conditions of low jerk and moderate acceleration, the fluid reaches hydrostatic equilibrium. The free surface is approximated as a plane orthogonal to the body-frame gravity vector ( $g_L = -R_L^T e_3$ ).
Offline Pre-computation: Using the known tank geometry and fluid density, the inertia tensor is pre-calculated for various fill levels and orientations.
Online Lookup: During flight, the system estimates the current mass (fill level) and measures the load attitude. It queries a lookup table to retrieve the corresponding inertia tensor $J_L(t)$ .
Validity Conditions: The approach is valid when translational jerk ( $\dddot{x}_L$ ) and angular acceleration ( $\dot{\Omega}_L$ ) are limited, preventing significant sloshing that would violate the hydrostatic assumption.

D. Trajectory Planning Trade-off

The paper highlights a critical trade-off:

Optimal Control: Minimum-effort trajectories (e.g., cubic splines) minimize excitation, which is bad for mass estimation.
Estimation Requirement: Mass estimation requires "persistent excitation" (non-trivial acceleration/velocity variations).
Solution: The authors propose using smooth nominal references (tension splines) augmented with small, bandwidth-limited dither signals. This ensures the system remains smooth (limiting sloshing) while providing enough dynamic variation for the mass observer to converge.

3. Key Contributions

Geometric Model with Time-Varying Parameters: Derivation of structure-preserving reduced equations for multi-quadrotor systems with elastic tethers and time-varying mass/inertia.
Online Mass Observer: A geometric observer that estimates time-varying mass using standard state estimates and control inputs, with rigorous stability guarantees under PE conditions.
Hydrostatic Inertia Surrogate: A novel method to handle variable inertia without real-time PDE solving, utilizing a pre-computed lookup table indexed by estimated fill level and attitude.
Path-Planning Guidelines: Identification of the conflict between control efficiency and parameter identifiability, proposing a "dither" strategy to balance low-jerk motion with necessary excitation.

4. Simulation Results

Numerical simulations involving four quadrotors transporting a variable-mass load demonstrated:

Mass Estimation: The observer successfully tracked both constant mass (converging from large initial errors) and exponentially decaying mass (leakage scenario).
Inertia Tracking: The hydrostatic surrogate accurately updated the inertia tensor in real-time, matching the true diagonal components of the inertia matrix as the mass decreased.
Robustness: The closed-loop system maintained stability and tracking performance despite:
- Initial parameter mismatches.
- Sensor noise (position, velocity, attitude).
- External wind disturbances.
Trajectory Behavior: The system successfully stabilized the load to the desired equilibrium, with transient errors resolving as the mass estimate converged.

5. Significance and Future Work

Significance: This work bridges the gap between theoretical geometric control and practical fluid-carrying logistics. It provides a viable solution for missions where payloads change weight and shape dynamically (e.g., firefighting, chemical transport, agricultural spraying) without requiring heavy, complex onboard sensors.
Future Directions:
- Experimental validation with multi-UAV hardware.
- Improved onboard state estimation for load acceleration.
- Extending the model beyond the hydrostatic regime (e.g., using data-driven corrections for significant sloshing).
- Refining trajectory planning to balance actuator limits, delays, and unmodeled aerodynamics.

In summary, the paper presents a robust, mathematically rigorous framework for cooperative aerial transport of fluid payloads, replacing complex real-time fluid dynamics with a smart, geometry-based lookup strategy while ensuring parameter convergence through careful trajectory design.