💻 computer science

Dynamic Control Allocation for Dual-Tilt UAV Platforms

This paper proposes a hierarchical control framework with a novel dynamic allocation law for dual-tilting hexarotor UAVs that explicitly models actuator saturation and optimizes propeller tilt angles to achieve robust trajectory tracking.

Original authors: Marcello Sorge, Federico Ciresola, Giulia Michieletto, Angelo Cenedese

Published 2026-04-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marcello Sorge, Federico Ciresola, Giulia Michieletto, Angelo Cenedese

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 a standard drone as a clumsy dancer. It can only push or pull in the direction its body is facing. If it wants to move sideways, it has to tilt its whole body, which is slow and awkward.

Now, imagine a Dual-Tilt UAV (the subject of this paper) as a super-flexible acrobat. This drone has six propellers, but unlike a normal drone, each propeller is mounted on a tiny gimbal. This means every single propeller can swivel left, right, up, and down independently while the drone is flying. It's like having six hands that can push air in any direction, regardless of which way the drone's body is pointing.

This super-flexibility is amazing, but it creates a massive headache for the computer brain controlling it: Too many choices.

The Problem: The "Too Many Options" Paradox

If you want this acrobat-drone to move forward, there are thousands of different ways to tilt its six propellers to achieve that.

Tilt propeller #1 up and #2 down?
Tilt all of them slightly forward?
Spin propeller #3 faster and #4 slower?

The computer needs to pick the best combination from these infinite possibilities. If it picks a bad one, the drone might waste energy, get stuck at the edge of its mechanical limits (like a joint that can't bend further), or crash.

The Solution: A Two-Layer Brain

The authors of this paper designed a "hierarchical" (two-layer) brain to solve this:

The High-Level Coach (The Goal Setter):
This part of the brain doesn't care about the propellers. It just looks at the map and says, "We need to move from Point A to Point B, and we need to do it smoothly." It calculates the total force needed to make that happen. It's like a coach shouting, "Run faster!" without telling the runner exactly how to move their left foot vs. their right foot.
The Allocator (The Tactician):
This is the star of the paper. Once the Coach says, "We need this much force," the Allocator figures out how to get it.
- The Job: It takes the "force command" and breaks it down into specific instructions for all six propellers (how much to tilt, how fast to spin).
- The Trick: It uses the drone's extra flexibility (redundancy) to optimize the flight. It tries to keep the propellers in the "middle of the road" so they don't hit their mechanical limits, and it tries to spin them as efficiently as possible to save battery.

The Creative Analogy: The Orchestra Conductor

Think of the drone's six propellers as an orchestra of six musicians.

The High-Level Controller is the composer who writes the melody (the desired path).
The Allocator is the conductor.

In a normal drone, the conductor just tells everyone to play the same note. But in this Dual-Tilt drone, the conductor has a superpower: they can tell each musician to play a slightly different note or change their volume to create a perfect harmony.

The paper's innovation is that the conductor isn't just making music; they are also managing the musicians' energy.

Avoiding Saturation: Imagine a violinist whose arm can only stretch so far. If the music requires them to stretch too far, they might break. The Allocator ensures the musicians stay in a comfortable range, never stretching to the breaking point (saturation).
Optimization: The Allocator also tries to keep the musicians from getting tired. It finds the arrangement where they use the least amount of energy to play the song.

What Did They Discover?

The researchers ran computer simulations (like a flight simulator) to test their new "Conductor."

It Saves Energy: By constantly adjusting the tilt angles to stay in the "sweet spot," the drone used less battery power than a drone that just randomly picked a solution.
It Handles Limits: They tested what happens when the drone is asked to do something impossible (like move too fast). They found that if the drone hits its mechanical limits, the system gets a little shaky, but it generally holds together. They realized that if the drone is asked to move too fast, the "Conductor" might get overwhelmed, similar to how a human driver might panic if asked to brake instantly at 100 mph.
Asymmetric Control: They tested a scenario where they told the Allocator, "Don't tilt the propellers left/right as much as you tilt them up/down." The system successfully adapted, showing that the drone can be "taught" to prefer certain movements over others without crashing.

The Bottom Line

This paper presents a smarter way to fly super-flexible drones. Instead of just reacting to commands, the drone's brain actively manages its own joints to stay safe, efficient, and ready for the next move. It's the difference between a robot that stumbles through a maze and a gymnast who flows through it effortlessly.

In short: They built a control system that lets a drone with swiveling propellers fly smarter, save battery, and avoid breaking its own joints, all by using a "two-brain" system where one sets the goal and the other finds the most efficient way to get there.

1. Problem Statement

The paper addresses the control challenges associated with dual-tilting hexarotor UAVs. Unlike standard fixed-propeller multirotors, these platforms can independently tilt each propeller along two orthogonal axes ( $\alpha$ and $\beta$ ) during flight. This capability allows for omnidirectional force generation, essential for complex tasks like aerial manipulation and aggressive trajectory tracking.

However, this redundancy introduces significant control complexity:

Nonlinearity: The mapping between actuator inputs (tilt angles and spin rates) and the desired wrench (force and torque) is nonlinear and configuration-dependent.
Redundancy: There are infinite combinations of actuator states that can produce the same desired wrench.
Actuator Constraints: Real-world actuators have physical saturation limits (maximum tilt angles and spin rates).
Dynamic Limitations: Many existing control allocation strategies assume static optimization or simple linear dynamics, neglecting the nonlinear dynamics and saturation effects of the actuators, which can lead to performance degradation or instability.

The core problem is to design a control allocation strategy that tracks a reference trajectory while optimizing the actuator states (to minimize energy or avoid saturation) and explicitly handling actuator saturation dynamics.

2. Methodology

The authors propose a hierarchical control structure consisting of two main layers:

A. High-Level Controller

Function: Generates the desired wrench ( $u_v$ ) required to track a reference trajectory (position and orientation).
Approach: Uses a standard feedback linearization approach based on Euler-Newton dynamics. It computes the required force and torque to drive the tracking errors to zero using proportional-derivative (PD) gains.
Output: A reference wrench vector $u_v^*$ that the lower-level allocator must achieve.

B. Dynamic Control Allocator

Function: Maps the desired wrench $u_v^*$ into specific actuator commands ( $u_a$ ) while optimizing the system state ( $x_a$ ).
Actuator Model: The actuators are modeled as a first-order nonlinear redundant system with explicit saturation limits. The state vector $x_a$ includes tilt angles ( $\alpha, \beta$ ) and spin rates ( $\omega$ ) for all 6 propellers.
Control Law: The allocator uses a gradient-descent-like approach to drive the actuators toward an optimum while ensuring the output wrench converges to the high-level command.
- The control input $u_a$ $u_{a}$ is derived to cancel the system dynamics ( $f(x_a)$ $f (x_{a})$ ) and inject two components:
  1. Tracking Component ( $u_y$ ): Drives the actual wrench to the reference wrench.
  2. Optimization Component ( $u_j$ ): Projects the gradient of a cost function onto the null space of the actuator output Jacobian. This ensures the optimization does not interfere with the primary tracking task.
Saturation Handling: The authors explicitly model saturation using a saturation function $sat(\cdot)$ . They introduce a theoretical insight that when actuators hit saturation, the Jacobian becomes non-invertible, potentially breaking asymptotic convergence. To mitigate this in simulations, a small regularization factor ( $\epsilon$ ) is used in the gradient calculation.
Objective Function ( $J$ ): The optimization minimizes a cost function designed to:
1. Maximize the distance of tilt angles from their saturation limits (keeping them near the center of the operational range).
2. Minimize propeller spin rates (to reduce energy consumption).
3. Allow for asymmetric weighting between the two tilt axes ( $\alpha$ and $\beta$ ) to study their specific roles.

3. Key Contributions

Adaptation of Dynamic Allocation: Successfully adapted a dynamic control allocation architecture (previously used for ground-based ROSPO platforms) to a dual-tilting aerial UAV, explicitly modeling actuator dynamics.
Explicit Saturation Modeling: Unlike many prior works that assume linear dynamics or ignore saturation, this paper explicitly models actuator saturation. It provides theoretical analysis on how saturation affects the invertibility of the allocation matrix and the stability of the control loop.
Asymmetric Optimization: The authors investigate the specific roles of the two tilt angles ( $\alpha$ and $\beta$ ) by defining asymmetric objective functions. This allows the system to prioritize keeping one axis near its center while allowing the other to move more freely, offering flexibility for specific mission constraints.
Theoretical Insights: The paper demonstrates that the hierarchical structure leverages the optimization term ( $u_j$ ) to maintain actuator states within limits at steady-state, though transients may still cause saturation if the trajectory is too aggressive.

4. Simulation Results

Numerical simulations were conducted on a star-shaped hexarotor (2kg mass) tracking a circular trajectory.

Optimization vs. Non-Optimization:
- Tracking: Both optimized ( $\gamma_j = 10$ ) and non-optimized ( $\gamma_j = 0$ ) cases achieved accurate trajectory tracking with negligible difference in position/orientation error.
- Tilt Angles: The optimized case drove the tilt angles to oscillate around the center of their saturation intervals (zero offset), whereas the non-optimized case showed larger offsets.
- Energy: The optimized case significantly reduced propeller spin rates (amplitude reduction), leading to lower energy consumption as reflected in the objective function $J(x_a)$ .
Asymmetric Constraints:
- When the objective function was modified to penalize $\alpha$ more heavily than $\beta$ (and vice versa), the system successfully adjusted the tilt angles accordingly.
- Result: The system kept the penalized angle closer to the center of its range while allowing the other angle to have larger excursions.
- Impact: This asymmetric optimization had a minor impact on propeller spin rates, confirming that the allocator can redistribute the workload between tilt axes without significantly affecting energy efficiency.

5. Significance and Conclusion

This work is significant because it bridges the gap between theoretical dynamic control allocation and practical UAV implementation by addressing actuator saturation and nonlinear dynamics.

Practicality: It proves that dynamic allocation can be used to manage the complex redundancy of dual-tilt UAVs without sacrificing tracking performance.
Efficiency: By actively optimizing actuator states, the system reduces energy consumption and extends the operational envelope by keeping actuators away from saturation limits.
Future Work: The authors note that while steady-state performance is robust, transient saturation can still occur with aggressive trajectories. They suggest future work should integrate the high-level controller and allocator more tightly (similar to anti-windup mechanisms) to dynamically adjust convergence rates and prevent saturation during rapid maneuvers.

In summary, the paper presents a robust, theoretically grounded framework for controlling over-actuated, tilting UAVs, offering a solution that balances precise trajectory tracking with actuator health and energy efficiency.