⚡ electrical engineering

Global Sensitive-Based Input Shaping for UAV-Payload Precision Motion Control

This paper proposes a global sensitivity-based input shaping method enhanced by Shapley values for 3D UAV-payload systems, demonstrating through numerical simulations that it significantly improves robustness against uncertainties in payload mass and rope length compared to non-robust, standard robust, and minimax designs.

Original authors: Karan Baker, Sanjay Maharjan, Tariq Hlayel, Oladapo Ogunbodede, Dutch Dunphy, Adrian Stein

Published 2026-07-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Karan Baker, Sanjay Maharjan, Tariq Hlayel, Oladapo Ogunbodede, Dutch Dunphy, Adrian Stein

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 trying to carry a bucket of water on a stick while running a race. If you move too jerkily, the water sloshes out. If you move too slowly, you lose the race. This is the daily struggle of a specific branch of robotics: getting flying machines to move heavy things without making them swing wildly. Scientists call these flying machines "Unmanned Aerial Vehicles" or UAVs (drones), and the things they carry are "payloads."

The tricky part is that the payload isn't glued to the drone; it hangs on a rope. This creates a pendulum effect. When the drone turns, the load swings. When the load swings, it pulls the drone off course. It's a chaotic dance where the dancer and the partner are constantly tripping over each other. To fix this, engineers use a trick called "Input Shaping." Think of it like a conductor telling a musician exactly when to hit a note so that the sound cancels out any unwanted echo. By timing the drone's movements perfectly, the shaking stops. However, this trick usually only works if the engineer knows the exact weight of the bucket and the exact length of the rope. In the real world, those numbers change, or nobody knows them until the drone is already in the air. This paper asks: Can we teach the drone to handle the swinging even when we don't know the exact details of the load?

The Paper's Mission
This paper presents a new way to design those "conducting" instructions for drones carrying hanging loads. The researchers, working with a 3D drone system, wanted to create a control method that stays calm even when the payload's mass (weight) or the rope's length is uncertain. They didn't just guess; they used a mathematical tool called "Global Sensitivity Analysis" (GSA) to figure out exactly how much the weight and the rope length mess up the drone's stability.

To make this even smarter, they borrowed a concept from game theory called the "Shapley value." Imagine a group of friends trying to split a pizza fairly. The Shapley value is a formula that calculates exactly how much each person contributed to the group's success. In this paper, the "friends" are the uncertain factors (the weight and the rope length), and the "pizza" is the total amount of swinging. The researchers used this math to ask: "How much of the swinging is actually caused by the rope length versus the weight?" By knowing the answer, they could design a controller that specifically ignores the things that cause the most trouble.

What They Did and Found
The team built a computer simulation of a drone carrying a payload. They tested four different ways to control the drone's movement:

Non-robust: A basic method that assumes the weight and rope are exactly what they expect.
Robust: A method that tries to handle a range of possible weights and lengths.
Minimax: A method that tries to minimize the worst-case scenario.
The New GSA/Shapley Method: The approach proposed in this paper.

They ran thousands of simulations where the weight of the payload varied between 0.2 kg and 1 kg, and the rope length varied between 0.2 m and 1.8 m.

The results showed that the standard methods (non-robust, robust, and minimax) did a decent job, but they still left the payload swinging a bit. For example, with the standard "robust" method, the swinging angle (how far the load swung side-to-side) had an average range of about 8.98 degrees.

However, the new Global Sensitivity-based Input Shaper performed significantly better. In their simulations, this new method reduced the average swinging range to just 7.83 degrees. Even more impressively, it made the results much more consistent. The "standard deviation" (a measure of how much the results jumped around) dropped from over 1.2 degrees in the other methods to just 0.89 degrees. This means the drone was not only steadier on average, but it was also much more predictable.

The Shapley Twist
The researchers then took it a step further. They used the Shapley value to force the controller to be extra careful about the rope length, which they found was the biggest culprit for swinging. They added a rule to the math: "Make sure the rope length doesn't have more than 85% of the influence on the swinging."

When they did this, the results for the rope length became incredibly stable. The average swinging range stayed low (around 7.97 degrees), but the consistency improved even more, with a standard deviation of only 0.78 degrees. The trade-off? The swinging related to the other direction (the second angle) got slightly worse, which is expected when you focus so hard on fixing one specific problem. But overall, the team showed that by using these advanced math tools, they could make the drone's control system "desensitized" to the unknowns.

What This Means
The authors are careful to note that these results come from computer simulations, not real-world flights yet. They explicitly state that while the math looks promising, they need to test this on actual hardware to prove it works in the wind and turbulence of the real world. They plan to do this next, using a drone with a 0.6 kg payload and a 1-meter rope.

In short, this paper suggests that by using game theory to understand why a drone swings, we can teach it to fly steadier even when we don't know exactly what it's carrying or how long the rope is. It's a step toward drones that can deliver packages, build structures, or film movies without the camera or the package swinging like a pendulum.

Technical Summary: Global Sensitive-Based Input Shaping for UAV-Payload Precision Motion Control

Problem Statement
Unmanned Aerial Vehicles (UAVs) transporting suspended payloads offer significant advantages in logistics and infrastructure tasks but face fundamental control challenges due to their underactuated and nonlinear dynamics. The suspended load introduces additional degrees of freedom, creating complex coupled dynamics between the vehicle and the payload. Small trajectory deviations can excite oscillatory payload motion, which feeds back into the vehicle dynamics, often resulting in residual swing that compromises tracking accuracy and safety. These challenges are exacerbated by parameter uncertainties, specifically variations in payload mass and rope length, which are often unknown prior to flight or change during operation. While existing control strategies (e.g., passivity-based, adaptive, sliding mode) offer robustness, they are typically tuned for nominal operating points and may degrade under broad or unstructured uncertainties.

Methodology
The paper proposes a Global Sensitivity Analysis (GSA)-based input shaping framework designed to suppress payload swing while accounting for uncertainties in payload mass ( $m_l$ ) and cable length ( $L$ ). The methodology involves three primary components:

System Modeling: The UAV-payload system is modeled as a point mass UAV with a suspended point mass payload connected by a rigid cable. The dynamics are derived using the Lagrangian formulation, accounting for translational motion, payload swing angles ( $\alpha, \beta$ ), and rotational energy. Uncertainties are modeled as uniform distributions: $m_l \sim U[0.2, 1]$ kg and $L \sim U[0.2, 1.8]$ m.
Input Shaper Design: The study compares several feedforward input shaping techniques:
- Non-robust: A standard shaper designed for nominal parameters.
- Robust: A shaper placing two sets of zeros at the nominal pole location to widen the notch.
- Minimax: A shaper designed to minimize residual energy over a defined frequency range.
- GSA-based: An optimization-based shaper where the objective is to minimize the expected range of swinging angles ( $R_\alpha, R_\beta$ ) over the maneuver time, subject to constraints on magnitude and switch times.
Shapley Value Integration: To systematically account for uncertainties, the authors incorporate the Shapley value concept from cooperative game theory. In this context, the uncertain parameters ( $m_l$ and $L$ ) are treated as "players" in a game. The Shapley value quantifies the marginal contribution of each parameter to the system's output variance (swinging angle range). A specific constraint is introduced to limit the Shapley value of the cable length's influence on the swing angle range ( $Sh^R_\alpha_L \leq 0.85$ ), effectively desensitizing the controller to variations in rope length.

Key Contributions

GSA-Based Input Shaping: The work demonstrates the feasibility of applying GSA to design input shapers for 3D UAV-payload systems, moving beyond nominal or simple robust designs.
Shapley Value Constraints: The paper introduces a novel method of using Shapley values as optimization constraints within the input shaper design. This allows for the systematic attribution of uncertainty contributions and the targeted reduction of sensitivity to specific parameters (e.g., rope length).
Comparative Analysis: The study provides a comprehensive numerical comparison between non-robust, robust, minimax, standard GSA, and Shapley-constrained GSA input shapers.

Results
Numerical simulations were conducted for a point-to-point maneuver with a final time of $T_f = 3.7898$ s. Key findings include:

Sensitivity Analysis: Without input shaping, the swinging angle range showed a strong dependency on cable length ( $L$ ) compared to payload mass ( $m_l$ ).
Performance Improvement: The standard GSA-based input shaper significantly outperformed the non-robust, robust, and minimax designs. It achieved a lower mean ( $\mu$ ) and standard deviation ( $\sigma$ ) for the swinging angle ranges. For instance, the standard deviation for $\alpha$ was reduced to $0.8959^\circ$ (compared to $1.2303^\circ$ for minimax).
Shapley Constraint Effectiveness: Enforcing the Shapley value constraint ( $Sh^R_\alpha_L \leq 0.85$ ) successfully reduced the sensitivity of the swing angle range to cable length uncertainty. While this resulted in a slight increase in the mean swing angle for $\alpha$ and a trade-off in the performance of $\beta$ (increased variance), it significantly reduced the standard deviation of $\alpha$ to $0.7807^\circ$ .
Trade-offs: The results highlight a trade-off in multi-objective optimization; desensitizing the system to one parameter (cable length) for one objective ( $\alpha$ ) can impact the performance of other objectives ( $\beta$ ).

Significance and Claims
The authors assert that the proposed GSA-based input shaper offers a promising framework for uncertainty-aware control in aerial payload transport. The primary significance lies in the ability to desensitize UAV-payload motion controllers to system uncertainties without increasing maneuver time, providing a robust alternative to conventional methods.

The paper modestly notes that while the Shapley-based approach provides critical insights into handling parameter variability, the reliance on Monte Carlo simulations makes the offline design computationally intensive. Furthermore, numerical deviations in Shapley value summation (due to the nonlinear nature of the system and finite sample sizes) were observed. The authors conclude that future work should focus on experimental validation using a physical drone and exploring surrogate models, such as polynomial chaos expansion, to reduce computational costs. The ultimate goal is to validate these strategies for applications requiring high-accuracy delivery, stable cinematography, and precision surveying.

More like this