💻 computer science

Acceleration-Based Control of Fixed-Wing UAVs for Guidance Applications

This paper presents an acceleration-level outer-loop control framework that converts commanded tangential and normal accelerations into executable body-rate and normalized thrust commands for fixed-wing UAVs, enabling the practical deployment of acceleration-based guidance laws like proportional navigation through an energy-based thrust formulation and real-flight validated mapping.

Original authors: Jixiang Wang, Siyuan Yang, Ziyi Wu, Siqi Wei, Ashay Wakode, Agata Barcis, Hung Nguyen, Shaoming He

Published 2026-03-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jixiang Wang, Siyuan Yang, Ziyi Wu, Siqi Wei, Ashay Wakode, Agata Barcis, Hung Nguyen, Shaoming He

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 the captain of a very fast, very efficient airplane (a fixed-wing drone). You have a brilliant navigator sitting in the cockpit who knows exactly where you need to go. This navigator is a master of physics and says, "Captain, in the next second, I need you to accelerate up at 5 meters per second squared and forward at 2 meters per second squared."

This is the dream scenario. But here's the problem: You don't have a "push-up" or "push-forward" button.

Your plane doesn't work like a helicopter or a rocket that can just blast in any direction. It works like a real airplane:

To go up, you have to tilt your wings (roll) and pull back on the stick (pitch) to redirect the air pushing you up.
To go forward, you have to twist the throttle (thrust).
To go backward, you can't just reverse the engine; you have to slow down or dive.

If you try to tell a standard airplane autopilot (like the ones used in hobby drones or commercial systems) "accelerate up at 5," it will get confused. It only understands "roll left at this speed" or "throttle to 70%."

This paper is the translator. It builds a "smart interpreter" between the Navigator (who speaks "Acceleration") and the Pilot (who speaks "Roll, Pitch, and Throttle").

Here is how they solved the puzzle, broken down into three simple parts:

1. The "Banking" Trick (Normal Acceleration)

The Problem: The navigator says, "Turn sharply to the left!" (This is a sideways acceleration).
The Solution: The paper explains that to turn a fixed-wing plane, you don't just turn the nose; you bank (tilt) the wings.

The Analogy: Think of a cyclist leaning into a turn. If you want to turn left, you lean left. The paper creates a formula that says: "If the navigator wants a specific turn force, here is exactly how much you need to lean (roll) and how fast you need to pull back (pitch)."
The Magic: It does this without needing a complex physics textbook. It uses simple geometry: "To get this much lift sideways, tilt the wings this much."

2. The "Energy" Guess (Tangential Acceleration)

The Problem: The navigator says, "Speed up!" or "Slow down!" (This is forward/backward acceleration).
The Solution: Usually, engineers need a perfect model of the engine, the propeller, and the wind to know how much throttle gives you how much speed. But engines are messy, and wind changes.

The Analogy: Instead of trying to calculate the engine's math perfectly, the authors treat the plane like a battery. They look at the plane's "Total Energy" (how high it is + how fast it's going).
The Magic: They flew the plane around, tried different throttle settings, and watched what happened. They said, "Okay, when we set the throttle to 50% at this speed, the plane gains energy at this rate." They built a simple lookup chart based on real flight data, not theory. This means they don't need to know the exact engine specs; they just need to learn from the plane's own behavior.

3. The "Traffic Cop" (Handling Conflicts)

The Problem: Sometimes the navigator asks for two impossible things at once. For example: "Dive steeply to slow down and turn sharply to the left."

The Dilemma: If you dive to slow down, you might not have enough power to turn sharply. If you turn sharply, you might lose too much speed. The plane has limits (it can't stall, it can't fly upside down forever).
The Solution: The paper adds a "Traffic Cop" logic.
- Mode A (The Hunter): If the mission is to catch a target, the plane prioritizes turning (Normal Acceleration). It will turn as hard as possible, even if it means the speed gets a bit messy.
- Mode B (The Glider): If the mission is to manage fuel or land safely, the plane prioritizes speed (Tangential Acceleration). It will adjust its nose up or down to control speed, even if the turn isn't perfect.
- This ensures the plane never tries to do the impossible and crash.

The Result: A Real-World Test

The team tested this on a real drone that can take off vertically like a helicopter but fly like a plane.

They told the drone to follow a "Proportional Navigation" path (a classic missile-style guidance law used to intercept targets).
The Outcome: The drone successfully intercepted a virtual target with a miss distance of less than half a meter. It did this by listening to the "Acceleration" commands and translating them instantly into "Wing Tilts" and "Throttle Twists."

Why This Matters

Before this paper, if you wanted to use advanced guidance algorithms (like those used for missiles or high-speed interception) on a standard drone, you had to rewrite the entire flight controller software. That's hard and risky.

This paper provides a "plug-and-play" adapter. You can now take any high-level guidance system that speaks "Acceleration" and plug it into any standard drone autopilot (like PX4 or ArduPilot) without changing the core software. It bridges the gap between the "Brain" (Guidance) and the "Muscles" (Flight Control), making fixed-wing drones much smarter and more agile.

1. Problem Statement

Fixed-wing Unmanned Aerial Vehicles (UAVs) are widely used for long-range missions, but their guidance and control remain challenging due to underactuation, strong nonlinear dynamics, and flight-envelope constraints.

The Core Mismatch: Many high-level guidance algorithms (e.g., Proportional Navigation, collision avoidance, motion planning) output acceleration commands ( $a_c$ ). While multirotor UAVs can track these directly, fixed-wing UAVs cannot. They must generate acceleration indirectly by modulating attitude (to redirect lift) and thrust.
The Engineering Bottleneck: Mainstream autopilots (e.g., PX4, APM) provide high-bandwidth inner-loops for body angular rates ( $p, q, r$ ) and normalized thrust ( $T_c \in [0,1]$ ), but they lack a native interface for acceleration tracking.
The Challenge: How to translate high-level acceleration commands into executable body-rate and thrust commands without modifying the existing inner-loop controller, while accounting for aerodynamic coupling, saturation limits, and the lack of precise propulsion models.

2. Methodology

The authors propose an acceleration-level outer-loop control framework that decomposes the desired acceleration into normal (perpendicular to velocity) and tangential (parallel to velocity) components.

A. Normal Acceleration Tracking (Attitude Control)

Objective: Map desired normal acceleration ( $a_n^c$ ) to roll ( $p_c$ ) and pitch ( $q_c$ ) rate commands.
Mechanism:
- Under small angle-of-attack ( $\alpha$ ) and sideslip ( $\beta$ ) assumptions, the lift vector is redirected to generate normal acceleration.
- The required lift vector is calculated as $a_L = a_n^c + g$ .
- A bank-to-turn (BTT) logic determines the commanded roll angle ( $\phi_c$ ) based on the angle between the required lift vector and the vertical axis.
- A proportional law converts the roll angle error into a roll rate command ( $p_c$ ).
- The pitch rate ( $q_c$ ) is derived directly from the vertical component of the normal acceleration and airspeed ( $V$ ), approximating $q_c \approx -a_{n,z}^c / V$ .

B. Tangential Acceleration Tracking (Thrust Control)

Objective: Map desired tangential acceleration ( $a_t^c$ ) to a normalized thrust command ( $T_c$ ) without relying on complex propulsion models.
Mechanism:
- Utilizes a Total Energy Control System (TECS) perspective. The authors define an "energy acceleration" ( $a_{TE}$ ) that accounts for the coupling between speed and altitude changes: $a_{TE} = \dot{V} - g(V_z/V)$ .
- Data-Driven Identification: Instead of modeling thrust and drag analytically, the system identifies an empirical linear relationship between $a_{TE}$ , airspeed squared ( $V_a^2$ ), and $T_c$ directly from flight data:
  $a_{TE} = k_V V_a^2 + b_V T_c$
- This allows the system to compute the required $T_c$ online to achieve a specific energy acceleration, avoiding the need for thrust-bench calibration.

C. Priority Handling and Saturation

Problem: During aggressive maneuvers or steep climbs/descents, the normal and tangential channels couple, and actuator limits (max thrust, stall limits) may make simultaneous tracking impossible.
Solution: The framework implements two priority strategies:
1. Normal-Acceleration Priority: Prioritizes tracking $a_n^c$ (critical for interception). Thrust is saturated, and the aircraft accepts speed deviations.
2. Tangential-Acceleration Priority: Prioritizes speed/energy management. The pitch angle is adjusted (within bounds) to satisfy $a_t^c$ even if it requires sacrificing the normal acceleration target.

3. Key Contributions

Acceleration-to-Command Mapping: A novel outer-loop that converts desired normal/tangential accelerations into executable body-rate and normalized thrust commands compatible with standard autopilots (PX4/APM).
Model-Light Thrust Realization: An energy-based method that identifies the thrust-acceleration relationship directly from flight data, eliminating the need for explicit propulsion modeling or pre-flight calibration.
Engineering Validation: Successful real-world flight experiments on a VTOL fixed-wing platform demonstrating accurate acceleration tracking and the practical implementation of Proportional Navigation (PN).

4. Experimental Results

The framework was validated on a SkyFury VTOL fixed-wing UAV equipped with a PX4 autopilot.

Thrust Identification: Flight data confirmed a linear relationship between energy acceleration and thrust command across various airspeeds, validating the empirical model.
Acceleration Tracking:
- The UAV successfully tracked step commands in normal acceleration with stable transient response.
- In coupled scenarios, the system correctly prioritized normal acceleration when thrust saturated, resulting in expected speed deviations consistent with the control logic.
Proportional Navigation (PN) Demonstration:
- The UAV intercepted a virtual target located 600m horizontally and 30m vertically away.
- Result: The mission achieved a miss distance of only 0.58 meters, proving that high-level acceleration-based guidance can be effectively deployed on fixed-wing UAVs using this framework.

5. Significance

Bridging the Gap: This work solves a critical integration problem, allowing advanced acceleration-based guidance algorithms (common in robotics and defense) to be deployed on commercial fixed-wing autopilots without rewriting the low-level flight controller.
Practicality: By using data-driven identification for thrust and standard autopilot interfaces, the solution is robust, adaptable to different airframes, and does not require deep knowledge of the specific propulsion system.
Mission Capability: It enables fixed-wing UAVs to perform agile, guidance-driven maneuvers (like interception and collision avoidance) that were previously difficult to implement due to the mismatch between guidance outputs and control inputs.

Future Work: The authors plan to relax the small-angle assumptions, explicitly model wind disturbances, and develop formal stability guarantees for the priority-switching logic.