Design to Flight: Autonomous Flight of Novel Drone Design with Robotic Arm Control for Emergency Applications

Abstract

Rapid and precise intervention in disaster and medical-aid scenarios demands aerial platforms that can both survey and physically interact with their environment. This study presents the design, fabrication, modeling, and experimental validation of a one-piece, 3D-printed quadcopter with an integrated six-degree-of-freedom aerial manipulator robotic arm tailored for emergency response. First, we introduce an ‘X’-configured multi-rotor frame printed in PLA+ and optimized via variable infill densities and lattice cutouts to achieve a high strength-to-weight ratio and monolithic structural integrity. The robotic arm, driven by high-torque servos and controlled through an Arduino-Pixhawk interface, enables precise grasping and release of payloads up to 500 g. Next, we derive a comprehensive nonlinear dynamic model and implement an Extended Kalman Filter-based sensor-fusion scheme that merges Inertial Measurement Unit, barometer, magnetometer, and Global Positioning System data to ensure robust state estimation under real-world disturbances. Control algorithms, including PID loops for attitude control and admittance control for compliant arm interaction, were tuned through hardware-in-the-loop simulations. Finally, we conducted a battery of outdoor flight tests across spatially distributed way-points at varying altitudes and times of day, followed by a proof-of-concept medical-kit delivery. The system consistently maintained position accuracy within 0.2 m, achieved stable flight for 15 min under 5 m/s wind gusts, and executed payload pick-and-place with a 98% success rate. Our results demonstrate that integrating a lightweight, monolithic frame with advanced sensor fusion and control enables reliable, mission-capable aerial manipulation. This platform offers a scalable blueprint for next-generation emergency drones, bridging the gap between remote sensing and direct physical intervention.

1. Introduction

Recent advancements in robotics, control systems, and embedded technologies have significantly contributed to the rapid development and deployment of Unmanned Aerial Vehicles (UAVs). These aerial platforms have gained widespread adoption across military, industrial, and commercial sectors, offering solutions in domains such as surveillance, aerial inspection, search and rescue, environmental monitoring, and photography [1,2,3]. This technological evolution has sparked renewed interest in enhancing UAVs’ autonomy and functionality to serve increasingly complex real-world tasks.

1.1. The Core Problem

Despite their versatility, traditional UAVs are primarily limited to observational and navigational tasks. They lack the physical interactivity needed for missions that require manipulation of objects in dynamic or hazardous environments. In emergency scenarios such as natural disasters, urban fires, or rescue operations, there exists a critical need for aerial systems that can not only observe but also interact physically with the environment. Tasks like lifting lightweight debris, delivering emergency supplies with precision, or operating valves and switches remain out of reach for conventional drones.

1.2. Current Technological Framework

UAVs generally fall into two main categories: fixed-wing and rotary-wing. Among rotary-wing types, the quadcopter (or quad-rotor) stands out due to its vertical takeoff and landing (VTOL) capabilities, stationary hover, agility, and ease of control in confined spaces [4,5]. A quadcopter comprises four propellers attached to individual rotors, and flight dynamics are achieved by varying their angular velocities. Vertical movement is controlled by synchronously adjusting the speeds of all four rotors; lateral and longitudinal movement is managed by modulating the speeds of opposing rotors, while yaw motion results from torque differentials generated by counter-rotating propellers.

1.3. Proposed Direction and Solution

To address the limitations of conventional UAVs in situations where human intervention is not possible, recent research has explored the integration of robotic manipulators onto Unmanned Aerial Vehicle (UAV) platforms, creating hybrid systems known as aerial manipulators [6]. These systems combine the maneuverability of drones with the dexterity of robotic arms, allowing physical interaction in mid-air, providing necessary help. This capability has the potential to revolutionize emergency response operations by enabling tasks like object retrieval, supply placement, switch activation, first aid supply and more [7].

1.4. Contributions & Novelty

This work differs from recent aerial manipulation systems (2020–2024) by combining: (i) a monolithic, 3D-printed quadrotor airframe that integrates a lightweight multi-DoF manipulator while preserving a tight weight budget; (ii) a cascaded PX4 control stack augmented with admittance control for compliant interaction; and (iii) outdoor field validation with public logs and a demonstration video. We contrast these choices with recent tilt-rotor/fully actuated platforms, passive capture devices, and learning-centric contact approaches in Section 2.2 and

Table 1. Comparison with recent aerial manipulation systems (2020–2024). NR: not reported.

Work	Platform	Arm DoF	Materials/Build	Control	Validation	Artifacts
[8]	Tilt/fully actuated	1–2	Carbon/composite	Variable-impedance + force	Indoor contact	NR
[9]	Quad (fixed)	1 (passive hook)	Mixed	Geometric + LQ	Indoor transport	Video
[10]	Tilt/fully actuated	3+	Carbon	Admittance (pHRI)	Simulation	NR
[11]	Quad (fixed)	3+ (Delta)	Mixed	Planning + control	Indoor pick/place	NR
[12]	Quad (fixed)	1 (soft gripper)	3D-printed	Learning-based grasping	Indoor/Outdoor	Code
Ours	Quad (fixed)	Multi-DoF	Monolithic 3D-printed	PX4 cascade + admittance	Outdoor transport/contact	Logs + video

. This paper proposes a novel aerial manipulation system featuring an X-frame quadcopter integrated with a robotic emergency arm, and investigates its performance in emergency-relevant scenarios through multiple field flight tests. A demonstration video of the full system is available at https://www.youtube.com/watch?v=QRAUUgJTlR0 (accessed on 17 November 2025); the payload flight sequence with arm operation appears at [00:25–01:15].

1.5. Structure of the Paper

The remainder of the paper is organized as follows. In Section 2, we present a focused literature review of UAVs and aerial manipulation research, highlighting key technical milestones and identifying gaps in existing frame and casing designs. The dynamic model of the quadcopter is explained in Section 3. In Section 4, we describe our system architecture, including the hardware and control strategies, and introduce our own quadcopter frame and robotic emergency arm casing, where the frame is 3D-printed in PLA+ (“PLA+” denotes toughened PLA formulations (trade-name dependent), typically using additive packages that improve toughness and interlayer adhesion; see [13].) to achieve a lightweight yet robust structure. Section 5 provides example flight test case studies and simulations relevant to emergency applications. The remainder of this paper is organized as follows: Section 6 outlines the main findings, emphasizing the novelty of our PLA+ frame, and the following section discusses future directions for deploying such systems in real-world crisis environments.

2. Literature Review

Recent advancements in unmanned aerial vehicle (UAV) research have transitioned from a primary focus on passive sensing to the development of platforms capable of active physical interaction with their surroundings. In this context, aerial manipulators (AMs), multirotor UAVs integrated with robotic arms, facilitate operations such as grasping, tool manipulation, payload transfer, and other tasks of significance to emergency response and infrastructure maintenance [14,15]. This section specifically addresses two aspects from the literature: first, UAVs incorporating robotic manipulators, and second, the associated dynamics and control methodologies, as these areas directly underpin the contribution of this work.

2.1. UAVs with Robotic Arms

Early applications of drones in emergency settings primarily focused on aerial photography, mapping, and reconnaissance. For instance, Ref. [16] demonstrated the use of drones for rapid damage assessment post-earthquakes, while Ref. [17] emphasized UAV deployment for wildfire monitoring. These efforts established UAVs as effective tools for information gathering. However, a growing need for physical intervention, such as delivering aid packages or interacting with infrastructure, led to the evolution toward manipulative aerial systems. The push for physical intervention spurred research into hybrid aerial-manipulation systems.

Early and seminal prototypes demonstrated the feasibility of combining manipulators with UAVs for airborne manipulation tasks such as valve turning, grasping, and constrained surface interaction [18,19]. Design work emphasizes minimizing manipulator mass and inertia while providing sufficient dexterity: lightweight manipulator concepts using carbon-fiber and optimized link geometries report arm masses in the low-kilogram range and prioritize structural compliance to absorb contact impulses [20,21]. These studies illustrate trade-offs between dexterity, payload, and flight endurance that directly inform structural choices for emergency-use platforms.

In emergencies, drones with robotic arms can offer direct physical interventions:

• Search and rescue: Ref. [22] explored systems in collapsed-building scenarios, where UAVs can remove light debris or drop rescue lines.
• Forest fire management: Ref. [23] explored integrating drones with blockchain and swarm robotics to decentralize coordination, secure data sharing, and found that UAVs offer immense benefits in monitoring, detecting, and combating forest fires.

2.2. Coupled Dynamics of Aerial Manipulators

The motion of a manipulator is tightly coupled with the flight dynamics of an underactuated multirotor: manipulator actuation produces time-varying inertial loads and shifts in the center of gravity, which in turn generate disturbance wrenches on the vehicle. Modeling efforts, therefore, treat AMs as coupled multi-body systems with strongly nonlinear, time-varying dynamics; accurate plant models are often required for high-performance controllers and for predicting stability margins during contact-rich tasks [18,21]. Empirical reports quantify how arm motion degrades attitude and position tracking if not actively compensated, motivating the co-design of structure and controller.

Recent Aerial Manipulation Approaches

Recent directions include (a) fully actuated or tilt-rotor platforms for precision contact/wrench control [8]; (b) passive one-DoF capture devices (e.g., hook or perching) to keep mass and power overheads low [9]; and (c) compliance/admittance frameworks and learning-based policies for interaction [10,12]. In contrast, our emphasis is rapid fabrication, low-cost integration of a multi-DoF arm on a monolithic 3D-printed frame, and field validation with reproducible artifacts. We summarize representative works alongside ours in Table 1 and highlight differences in platform actuation, manipulator DoF, materials/build, control approach, validation setting, and artifact availability [8,9,10,11,12].

2.3. Control Strategies for Stability and Interaction

Several control paradigms have been proposed and validated to address the coupled UAV–manipulator problem:

• Model-based dynamic compensation Controllers that compensate manipulator-induced wrenches by adjusting thrust and moments can preserve vehicle stability during manipulation; these approaches typically rely on an accurate dynamic model or online identification [18,19].
• Impedance/admittance control For contact tasks, impedance or admittance controllers at the manipulator end-effector enable compliant interaction and reduce transmitted forces to the vehicle, improving safety during contact with uncertain environments [18].
• Robust/nonlinear/adaptive control. Sliding-mode, adaptive, and passivity-based controllers have been applied to reject disturbances, handle model mismatch, and maintain robustness in outdoor conditions with wind and sensor noise [21,24].
• MPC/NMPC and manipulability-aware control. Model predictive control frameworks (including NMPC) have been explored to enforce actuation and state constraints while optimizing end-effector performance (e.g., manipulability), particularly for constrained or precision tasks [15,25].

These strategies are often combined in hierarchical or cascaded architectures: a high-level planner or MPC provides reference trajectories and manipulability-aware setpoints, while an inner-loop attitude/position controller performs fast stabilization and disturbance rejection.

2.4. Research Gaps and Positioning of This Work

Despite progress, three recurring limitations remain in the literature:

• Payload and endurance constraints: manipulator mass and actuation substantially reduce flight time and available payload capacity, necessitating lightweight structural and actuator choices [20].
• Coupled dynamic instability: many systems still rely on teleoperation or simplified compensation; fully integrated flight manipulation control under real-world disturbances is limited [14,24].
• Limited field validation in emergency scenarios: although simulation and lab tests are common, extensive outdoor validation during realistic emergency tasks is less prevalent [15].

Several recent studies have begun to address these gaps:

• Dynamic compensation for payload and stability: Ref. [26] developed a quadrotor with a 2-DOF arm and implemented real-time dynamic compensation to preserve flight stability under varying loads.
• Admittance control for compliant interaction: Ref. [27] introduced a force-based admittance controller that allows safe, compliant contact with the environment.
• Lightweight structural designs: Ref. [28] proposed carbon-fiber manipulator components to minimize added mass and reduce energy consumption during aerial manipulation.

This paper addresses these gaps by combining a lightweight structural co-design (a PLA+ 3D-printed frame tailored for reduced inertia and rapid prototyping) with a robust flight–manipulation control approach validated in multiple field flight tests. Specifically, our work targets the structural and control trade-offs required for emergency-relevant interactions and provides experimental evidence of performance in outdoor conditions.

Note on Broader UAV Taxonomy

A broader taxonomy of UAV configurations (tricopters, hexacopters, tilt-rotors, ornithopters, etc.) was considered during platform selection, but is outside the core scope of this article and is summarized in Appendix A for context and completeness.

3. Dynamic & Control Model of Aerial Manipulator Robotic Arm Quadcopter

3.1. Dynamic Model

3.1.1. Ensuring Specification Integrity

To formulate the quadrotor’s dynamic equations, we adopt the following simplifying assumptions:

• The vehicle behaves as a rigid body acted upon by a single net lift force and three independent control torques.
• Its design is symmetric about both the x- and y-axes, so the inertia tensor reduces to a diagonal matrix.
• The origin of the body-fixed coordinate frame coincides with the quadrotor’s center of gravity.
• Effects due to aerodynamic drag and rotor gyroscopic moments are considered negligible.

3.1.2. Reference Frames and Kinematic Representation

As illustrated in Figure 1 and detailed in [29], we introduce an inertial frame fixed at a ground reference point and a body-fixed frame attached to the quadrotor’s center of gravity. The vehicle’s Cartesian position in the inertial frame is given by $\zeta ={\left[\begin{array}{c}x,y,z\end{array}\right]}^T$ where x, y, and z denote its coordinates along the respective axes. Its orientation is described by the Euler angles roll ($\varphi$), pitch ($\theta$), and yaw ($\psi$) which we assemble into the attitude vector $\eta =\left[\begin{array}{c}\varphi ,\theta ,\psi \end{array}\right]$. To relate measurements between the body-fixed and inertial frames, we employ the rotation matrix R. This matrix is constructed via a 3–2–1 Euler-angle sequence (yaw–pitch–roll), yielding the coordinate transformation from the body frame to the inertial frame:

$$R(\varphi ,\theta ,\psi )=\left[\begin{array}{ccc}cos\psi cos\theta & cos\psi sin\theta sin\varphi -sin\psi cos\varphi & cos\psi sin\theta cos\varphi +sin\psi sin\varphi \\ sin\psi cos\theta & sin\psi sin\theta sin\varphi +cos\psi cos\varphi & sin\psi sin\theta cos\varphi -cos\psi sin\varphi \\ -sin\theta & cos\theta sin\varphi & cos\theta cos\varphi \end{array}\right]$$

(1)

We adopt a Z–Y–X (yaw–pitch–roll) Euler sequence; R maps body-frame vectors to the inertial frame.

3.1.3. Rotational Kinematics

The correspondence between the inertial-frame Euler-angle rates $\dot{\eta }={[\dot{\varphi },\dot{\theta },\dot{\psi }]}^T$ and the body-frame angular velocity components $\Omega =[p,q,r]$ is expressed by

$$\Omega =W_{\eta }\dot{\eta }$$

(2)

where

$$W_{\eta }=\left[\begin{array}{ccc}1& 0& -sin\theta \\ 0& cos\varphi & sin\varphi cos\theta \\ 0& -sin\varphi & cos\varphi cos\theta \end{array}\right]$$

(3)

By linearizing the nonlinear model about the hover equilibrium and assuming small roll and pitch angles, the matrix $W_{\eta }$ reduces to the $3\times 3$ identity $I_{3\times 3}$. Hence, we obtain

$$\Omega =\dot{\eta }$$

(4)

3.1.4. Translational Dynamics

The quadrotor’s dynamics are then formulated using the Newton–Euler approach. The translational equations of motion, written in the inertial frame, include only the gravitational force G and the body-frame thrust $T_B$.

$$m\ddot{\zeta }=G+R{T}_B$$

(5)

$$\left[\begin{array}{c}\ddot{x}\\ \ddot{y}\\ \ddot{z}\end{array}\right]=-g\left[\begin{array}{c}0\\ 0\\ 1\end{array}\right]+R{\displaystyle \frac{T_B}{m}}\left[\begin{array}{c}0\\ 0\\ 1\end{array}\right]$$

(6)

3.1.5. Rotational Dynamics

In the body-fixed frame, the rotational dynamics are defined by the external torque ${\tau }_B$ applied to the quadrotor. This torque comprises the inertial angular acceleration term $I\dot{\Omega }$ and the centripetal term $\Omega \ast I\Omega$.

$${\tau }_B=I\dot{\Omega }+\Omega \ast I\Omega$$

(7)

3.1.6. Thrust and Torque Generation

A force $f_i$ and a torque $M_i$ are generated perpendicular and around the axis of each motor with angular velocity of $v_i$.

$$F_i=k{\omega }_i^2,M_i=b{\omega }_i^2$$

(8)

where the thrust constant is represented by k, and the drag constant is written by b.

The total thrust ($T_B$) can be determined by the summation of all forces generated by the four rotors.

$$T_B=\sum _{i=1}^4\left|F_i\right|=k\sum _{i=1}^4{\omega }_i^2$$

(9)

Additionally, the body-frame torque ${\tau }_B$ about the x, y, and z axes is given by:

$${\tau }_B=\left[\begin{array}{c}{\tau }_{\varphi }\\ {\tau }_{\theta }\\ {\tau }_{\psi }\end{array}\right]=\left[\begin{array}{c}lk(-{\omega }_2^2+{\omega }_4^2)\\ lk(-{\omega }_1^2+{\omega }_3^2)\\ b(-{\omega }_1^2+{\omega }_2^2-{\omega }_3^2+{\omega }_4^3)\end{array}\right]$$

(10)

Here, l denotes the distance from each rotor to the quadrotor’s center of mass, and ${\tau }_{\varphi }$, ${\tau }_{\theta }$, and ${\tau }_{\psi }$ correspond to the roll, pitch, and yaw torques, respectively.

3.1.7. Control Allocation

The control input vector in terms of the squared rotor speeds is obtained by combining Equations (9) and (10):

$$\left[\begin{array}{c}{u}_1\\ u_2\\ u_3\\ u_4\end{array}\right]=\left[\begin{array}{c}{T}_B\\ {\tau }_{\varphi }\\ {\tau }_{\theta }\\ {\tau }_{\psi }\end{array}\right]=\left[\begin{array}{cccc}0& -\mathsf{\ell }k& 0& \mathsf{\ell }k\\ -\mathsf{\ell }k& 0& \mathsf{\ell }k& 0\\ b& b& b& b\end{array}\right]\left[\begin{array}{c}{\omega }_1^2\\ {\omega }_2^2\\ {\omega }_3^2\\ {\omega }_4^2\end{array}\right]$$

(11)

3.1.8. Complete Nonlinear Equations of Motion

The complete nonlinear dynamics of the quadrotor are then given by:

$$\ddot{x}=\left(cos\varphi sin\theta cos\psi +sin\varphi sin\psi \right){\displaystyle \frac{u_1}{m}}$$

(12)

$$\ddot{y}=\left(cos\varphi sin\theta sin\psi -sin\varphi cos\psi \right){\displaystyle \frac{u_1}{m}}$$

(13)

$$\ddot{z}=-g+\left(cos\varphi cos\theta \right){\displaystyle \frac{u_1}{m}}$$

(14)

$$\ddot{\varphi }=\dot{\theta }\dot{\psi }{\displaystyle \frac{I_y-I_z}{I_x}}+{\displaystyle \frac{u_2}{I_x}}$$

(15)

$$\ddot{\theta }=\dot{\varphi }\dot{\psi }{\displaystyle \frac{I_z-I_x}{I_y}}+{\displaystyle \frac{u_3}{I_y}}$$

(16)

$$\ddot{\psi }=\dot{\theta }\dot{\varphi }{\displaystyle \frac{I_x-I_y}{I_z}}+{\displaystyle \frac{u_4}{I_z}}$$

(17)

In this section, we established a comprehensive six-degree-of-freedom model for the quadrotor by:

• defining inertial and body-fixed reference frames and the corresponding Euler angles and rotation matrix;
• deriving the mapping between body-frame angular rates and Euler-angle rates, including a small-angle linearization for hovering;
• applying Newton–Euler mechanics to obtain both translational and rotational equations of motion;
• characterizing individual rotor thrust and drag torques; and
• formulating the control-allocation matrix that relates squared rotor speeds to the total lift and attitude torques.

The resulting set of nonlinear differential equations (Equations (12)–(17)) captures the essential dynamics needed for simulation, analysis, and controller design across the vehicle’s operating envelope.

3.2. Coupled Dynamics of the Aerial Manipulator

We extend the quadrotor-only model to a floating-base UAV–arm system in which the vehicle and a six-DoF hobby servo arm (MG996R-class actuators, 2 mm aluminum links) are treated as a single articulated rigid body. Let $v_b\in {\mathbb{R}}^6$ denote the vehicle body twist and $q\in {\mathbb{R}}^6$ the arm joint coordinates. Following standard Euler–Lagrange/spatial-vector derivations for aerial manipulators [19,31,32,33,34,35], the equations of motion can be written in block form as

$$\underset{M\left(q\right)}{\underbrace{\left[\begin{array}{cc}{M}_{bb}\left(q\right)& M_{bm}\left(q\right)\\ M_{mb}\left(q\right)& M_{mm}\left(q\right)\end{array}\right]}}\left[\begin{array}{c}{\dot{v}}_b\\ \ddot{q}\end{array}\right]+\underset{C(·)}{\underbrace{\left[\begin{array}{c}{C}_b(v_b,q,\dot{q})\\ C_m(v_b,q,\dot{q})\end{array}\right]}}+\underset{g\left(q\right)}{\underbrace{\left[\begin{array}{c}{g}_b\left(q\right)\\ g_m\left(q\right)\end{array}\right]}}=\left[\begin{array}{c}{W}_{\mathrm{rotors}}+W_{\mathrm{ext}}\\ \tau +J_m^{\top }\left(q\right)W_{\mathrm{ext}}\end{array}\right],$$

(18)

where $M_{bm}$ and $M_{mb}$ (and the corresponding Coriolis terms in C) encode the manipulator–base coupling. This coupling captures center-of-mass shifts, added inertia, and reaction wrenches generated by arm motion and/or end-effector contact. For thrust/torque mapping we use $W_{\mathrm{rotors}}=B{\Omega }^2$ with the standard allocation matrix B for our multirotor geometry.

3.2.1. Arm Model and Parameters

The commercial arm is modeled as a 6R serial chain using a concise DH table (link lengths/offsets measured from the assembled kit). Each joint is a position servo with saturating torque and simple friction:

$${\tau }_i={sat}_{[-{\tau }_i^{max},{\tau }_i^{max}]}\left(K_{p,i}(q_i^{\mathrm{ref}}-q_i)-K_{d,i}{\dot{q}}_i\right)-{\tau }_{f,i}\left({\dot{q}}_i\right),$$

with ${\tau }_i^{max}\approx 1\mathrm{N}\mathrm{m}$ and travel $\approx {180}^{\circ }$ (MG996R class). Link inertias are approximated from 2 mm aluminum plate geometry; the tool/payload is lumped at the end-effector. This pragmatic parameterization is common in prior aerial manipulation work when detailed CAD is unavailable [19,31].

3.2.2. Manipulator-Induced Base Wrench

Partitioning (18) yields a transparent expression for the arm’s disturbance on the base:

$$W_m:=M_{bm}\left(q\right)\ddot{q}+C_{b|m}(q,\dot{q}),$$

(19)

which can also be interpreted via the rate of the arm’s centroidal momentum [33,35]. During aggressive arm motion, $W_m$ produces measurable attitude/position errors if left uncompensated.

3.2.3. Control Compensation

To explicitly address the reviewer’s request, we inject a model-based feedforward term and a residual rejection layer:

$$W_{\mathrm{cmd}}=W_{\mathrm{outer}}-{\widehat{M}}_{bm}\left(q\right){\ddot{q}}_{\mathrm{des}}-{\widehat{C}}_{b|m}(q,\dot{q}),$$

(20)

where $(q_{\mathrm{des}},{\dot{q}}_{\mathrm{des}},{\ddot{q}}_{\mathrm{des}})$ are generated by the arm trajectory planner. This “whole-body” coupling feedforward is consistent with centralized UAV–arm modeling [31,32]. We complement it with a lightweight base-wrench disturbance observer (or centroidal-momentum residual) to reject unmodeled servo nonlinearities, backlash, and contact effects. For readers interested in coordinate-free formulations or flatness-based planning, we refer to [33,35], while COI/SE(3) formulations that treat the UAV–arm as a single body are surveyed and extended in [34].

3.2.4. Identification and Reporting

In experiments, we estimate servo gains and joint friction from base-fixed chirps and regress ${\widehat{M}}_{bm},{\widehat{C}}_{b|m}$ from suspended-frame tests, reporting the magnitude of $W_m$ during representative arm trajectories. The ablation in Section 5.2 compares tracking error and saturation margins with vs. without (20), quantifying the benefit of modeling the coupling terms that are absent from a quadrotor-only model [31].

3.3. Control Model

The quadcopter with the integrated emergency robotic arm employs a PX4 flight controller, which executes a cascaded Proportional–Integral–Derivative (PID) control architecture as shown in Figure 2. This hierarchical control structure is responsible for ensuring stable flight performance across multiple layers of abstraction, ranging from angular rate regulation to precise position control.

The complete control system was validated in outdoor flight tests with a payload and varying altitudes. During experiments, the quadcopter demonstrated stable hover, accurate altitude regulation, and robust mode transitions (e.g., relocation and load-carrying maneuvers). The arm’s integration introduced additional coupling and inertia, yet the cascaded PID controllers effectively compensated, confirming the architecture’s robustness.

3.3.1. Control Architecture

The quadcopter platform employs a cascaded control architecture implemented through the PX4 flight stack. This hierarchical structure consists of nested feedback loops that operate at different bandwidths to ensure both fast attitude stabilization and robust trajectory tracking. The controllers are organized into outer-loop (position and velocity) and inner-loop (attitude, rate, and actuator allocation) layers, as summarized schematically in Figure 3.

The outer loop regulates the vehicle’s trajectory by generating velocity and attitude references based on position errors. These components are jointly illustrated in Figure 3 as Outer-Loop Controllers. The inner loop ensures rapid stabilization of the vehicle’s orientation and angular rates, translating pilot or autonomous commands into actuator inputs. This loop, including attitude, rate, and motor allocation, is summarized in Figure 3 as Inner-Loop Controllers.

During flight testing, this architecture demonstrated effective disturbance rejection and stable tracking of reference commands. Outer-loop position tracking was validated under waypoint navigation tasks, while inner-loop performance was assessed through step-input and disturbance recovery maneuvers. The flight data confirmed the controller’s ability to maintain stability despite the additional payload of the robotic arm.

3.3.2. PID Tunning

PID tuning remains a critical aspect for achieving robustness under external disturbances and the additional dynamic loads introduced by the onboard arm. By iteratively adjusting the proportional (P), integral (I), and derivative (D) gains, the controller compensates for modeling uncertainties, unmodeled coupling effects, and load-induced disturbances. In our implementation, advanced manual PID tuning was performed through flight experiments to balance stability, responsiveness, and energy efficiency.

At the lowest layer, the rate controller manages angular velocities about the roll, pitch, and yaw axes. Desired angular rates are tracked using a PID loop, which translates commands into individual motor thrusts. Building on this, the attitude controller regulates orientation by converting attitude errors into angular rate commands, which are then stabilized by the inner-loop rate controller. Figure 4 shows the cascaded architecture that ensures coordination between layers, resulting in smooth stabilization of the quadcopter.

The velocity controller regulates translational speeds in the x, y, and z directions, converting velocity errors into desired accelerations. These accelerations are subsequently mapped to orientation commands through the attitude controller. Finally, the position controller ensures trajectory tracking by generating velocity setpoints from position errors. This cascaded architecture, as shown in Figure 5, ensures coordination between layers, resulting in optimal trajectory tracking of desired states.

Overall, the flight tests confirm that the cascaded PX4 PID control model provides reliable tracking performance even under emergency-response conditions, such as load carrying and relocation. The architecture’s modularity ensures scalability, allowing adaptation to additional manipulation tasks and more complex mission profiles in future work.

4. Methodology

In this section, a comprehensive overview of the methodology used to design, fabricate, assemble, and operate our custom drone equipped with an integrated robotic arm, tailored for emergency scenarios, is discussed. We start by detailing the design and fabrication process, focusing on the custom 3D-printed frame for the drone and the casing for the emergency arm, along with the fixtures needed for the emergency robotic arm and the integration of electronic components. We then describe the control system used for the robotic arm, which utilizes a nano Arduino and servo motors, and how we implemented remote control functionality to achieve precise in-flight operations.

4.1. Drone Frame: Design, Fabrication, and Setup

The airframe we adopted has an ‘X’ configuration, where two clockwise (CW) and two counterclockwise (CCW) rotors are selected for its well-documented advantages in stability and maneuverability during agile maneuvers and hover [36,37]. By balancing rotor moments across both axes, this layout simplifies the implementation of pitch, roll, and yaw control algorithms on the Pixhawk autopilot. The Computer Aided Design (CAD) model of the complete frame, including the housing for the robotic arm, was parametrically modeled in Siemens NX as shown in Figure 6 to ensure precise alignment of motor mounts, fuselage geometry, and arm interface points. We employed a space-filling Latin hypercube design of waypoint distributions across a cuboidal operational volume as shown in Figure 7.

We selected a toughened PLA (often marketed as PLA+) for printed structural parts because it improves impact resistance and inter-layer adhesion while preserving printability and dimensional accuracy relative to standard PLA, consistent with prior mechanical studies and successful UAV components reported in the literature [38,39,40,41,42]. Unlike traditional top-down machining, additive manufacturing allowed us to fabricate the fuselage and arm enclosure as a single monolithic component, eliminating bolted or bonded joints that can introduce flex and failure points. To further reduce overall mass without sacrificing rigidity, we incorporated strategically placed cutouts and lattice slots in low-stress regions of the arm housing, guided by finite-element stress analysis during the CAD phase.

We conducted a series of print-parameter experiments to calibrate infill density, pattern, and layer height. Critical load paths, such as the motor-mount interfaces and arm base, were printed at 60% hexagonal infill to resist bending moments under maximum lift and payload conditions. Less-stressed volumes used 20% gyroid infill to minimize weight and material consumption. Layer heights of 0.2 mm and 2 perimeters provided a reliable surface finish and dimensional accuracy for snap-fit features. These iterative adjustments were validated through three-point bend tests and thrust-stand measurements, confirming that the printed structure withstands both static and dynamic loads typical of aerial-manipulation tasks.

The culmination of this process is a ready-to-fly, 3D-printed drone whose integrated design maximizes structural integrity, minimizes assembly complexity, and optimizes weight distribution. The seamless transition from CAD to physical prototype reduced lead time and material waste, while the final printed frame, as shown in Figure 7, exhibits the geometric complexity and mechanical robustness required for safe, reliable operation in a variety of flight scenarios.

Material Selection and Terminology

We use a toughened PLA (PLA+) for all printed structural parts. PLA+ offers improved impact resistance and interlayer adhesion compared to standard PLA while retaining low-warp printability and dimensional accuracy that simplify tolerance control for snap-fit and bolted joints. This choice is supported by direct mechanical characterizations of PLA+/Tough PLA [38,39,40] and by successful use of toughened PLA in flight components [41]. We use the term “PLA+” to denote trade-name toughened PLA families rather than a single resin grade; background on additive/toughening strategies for PLA in FDM is reviewed in [13].

4.2. Robotic Arm Integration—Functionality and Application

The manipulator integrated on our platform is a commercial off-the-shelf (COTS) 6-DoF servo arm (Fafeicy; MG996R-class actuators). We designed the mounting adapters, wiring, and end-effector interface, while the arm mechanism itself is not custom. Key parameters used for modeling and integration (mass, overall size, per-joint travel, nominal torque, supply voltage, and structural material) are summarized in

Table 2. Robotic arm specifications used in modeling and integration.

Type	COTS 6-DoF servo arm (Fafeicy)
Actuators	6 × MG996R (analog), 4.8–7.2 V
Nominal stall torque	≈10 kg·cm (≈0.98 N m) per joint
Joint travel	≈180° each joint
Mass	$0.77$ kg (verified post assembly)
Overall size	$31.3\times 9\times 50.5$ cm (supplier spec)
Structure	2 mm aluminum plates with bearings
Used voltage	5–6 V (regulated from power module)
Notes	Mass/limits used in CG and inertia calculations

; an enlarged photo and a dimensioned overlay appear in Figure 8.

The primary role of the arm is to grasp and transport small payloads safely in outdoor scenarios (e.g., delivery of emergency items). The grasping pipeline is:

• Servo actuation and interface. The arm comprises six MG996R-class servos (mechanical travel $\approx {180}^{\circ }$ each; nominal stall torque $\sim 10$ kg·cm). Commands are generated by an Arduino Nano interface and relayed as PWM to the joints; the Nano receives high-level commands from the RC link and sequences joint motions for grasp/open/release.
• Grip adaptation. Joint trajectories are parameterized by object size and estimated mass; position gains and motion profiles are chosen to avoid excessive tip accelerations. The gripper geometry allows quick attachment/detachment for fast pickup and drop-off.

The arm base is mounted at the vehicle’s structural center to minimize configuration-dependent CG shifts and off-diagonal inertia, thereby reducing steady trim demands and cross-axis coupling during manipulation. This design choice is consistent with the coupled UAV–arm model in Section 3.2 and the analysis in Section 4.3, where we quantify the effect of COM offset on hover torque and added inertia.

Once an object is grasped, the Pixhawk autopilot (PX4 cascaded PID baseline) maintains vehicle stability while the operator issues arm commands. Clockwise/counter-clockwise rotor pairs provide balanced thrust/torque, and the arm orientation can be adjusted to avoid obstacles en route. The controller and power-distribution layout follow the wiring shown earlier; mass/limit values from Table 2 are used in the CG/inertia calculations that inform the mounting and tuning choices.

4.3. Effect of Central Arm Placement on Stability and Control Effort

The manipulator is mounted at the vehicle’s structural center to minimize static and configuration–dependent shifts of the center of gravity (CG) and to suppress off–diagonal coupling in the composite inertia. This section formalizes that rationale and quantifies the impact using the identified mass properties of our platform and the off–the–shelf six–DoF servo arm.

4.3.1. Composite CG and Hover Trim

Let $m_u$ and $m_a$ denote the UAV and arm masses, $r_u$ the UAV CG expressed in the body frame, and $r_a\left(q\right)$ the arm’s center of mass (COM) as a function of the joint configuration q. The composite CG is

$$r_{\mathrm{cg}}\left(q\right)={\displaystyle \frac{m_u{r}_u+m_a{r}_a\left(q\right)}{m_u+m_a}}.$$

(21)

In steady hover the thrust vector balances weight, and the required trim torque is

$${\tau }_{\mathrm{trim}}=r_{\mathrm{cg}}\left(q\right)\times \left((m_u+m_a)g\widehat{z}\right),\parallel {\tau }_{\mathrm{trim}}\parallel =(m_u+m_a)g\parallel r_{\mathrm{cg}}^{\perp }\parallel ,$$

(22)

where ${(·)}^{\perp }$ denotes the lateral (in–plane) component relative to the thrust axis. Choosing the body origin at the structural center yields $r_u\approx 0$ and therefore $\parallel r_{\mathrm{cg}}^{\perp }\parallel \approx \frac{m_a}{m_u+m_a}\parallel r_a^{\perp }\parallel$, so the steady trim demand scales directly with the arm’s lateral offset. Central base placement keeps $\parallel r_a^{\perp }\parallel$ small over typical arm motions and reduces the magnitude and variability of ${\tau }_{\mathrm{trim}}$.

4.3.2. Composite Inertia and Cross–Axis Coupling

Let $J_u$ be the UAV inertia about the body origin and $J_{a,\mathrm{loc}}\left(q\right)$ the arm’s local inertia about its COM. The total inertia about the body origin is

$$J_{\mathrm{tot}}\left(q\right)=J_u+J_{a,\mathrm{loc}}\left(q\right)+m_a\left(\parallel r_a{\left(q\right)\parallel }^{2}I-r_a\left(q\right)r_a{\left(q\right)}^{\top }\right).$$

(23)

The third term is the parallel–axis contribution due to the arm COM offset. Central mounting minimizes $\parallel r_a\left(q\right)\parallel$ and consequently the off–diagonal entries ($J_{xy},J_{xz},J_{yz}$) that amplify cross–axis responses when the arm accelerates. In the coupled model of Section 3.2, this choice directly reduces the base–arm coupling terms and the manipulator–induced base wrench required of the attitude/rate loops.

4.3.3. Illustrative Calculation with Measured Masses

Using the mass of the six–DoF arm from our build ($m_a$ = 0.77 kg) and g = 9.81 m/s²,

Table 3. Impact of arm COM offset on hover trim and added inertia for $m_a$ = 0.77 kg.

Offset $\parallel {\mathit{r}}_{\mathit{a}}\parallel$ (m)	$\parallel {\mathit{\tau }}_{\mathbf{trim}}\parallel$ (N m)	$\mathbf{\Delta }{\mathit{J}}_{\mathit{zz}}$ (kg m2)	$\mathbf{\Delta }{\mathit{J}}_{\mathit{yy}}$ (kg m2)
$0.00$	$0.000$	$0.0000$	$0.0000$
$0.12$	$0.906$	$0.0111$	$0.0111$
$0.25$	$1.888$	$0.0481$	$0.0481$

reports the hover trim torque and added inertia for representative lateral COM offsets (assuming the offset lies in the $xy$–plane so $\Delta J_{zz}=m_a{\parallel r_a\parallel }^2$; local link inertias are omitted to isolate the offset effect).

4.3.4. Practical Implications for Control

Central mounting eliminates steady trim when the arm is near the vehicle origin and suppresses off–diagonal inertia, resulting in lower control effort and fewer saturation events during manipulation. While $J_{\mathrm{tot}}$ increases slightly due to the added mass, the reduction in coupling allows the cascaded controller to retain margins with modest gain retuning. A comprehensive posture– and payload–sweep sensitivity study is left for future work; in our outdoor trials this configuration consistently produced quieter transients during aggressive arm motion without sacrificing tracking performance.

4.4. Remote Control Integration

Integration of the remote-control system was crucial to enable seamless operation of both the drone and the robotic arm using a single remote-control unit. This process involved customizing the RC commands and programming the Arduino microcontroller to interpret these signals. Specifically, one of the channels on the RC transmitter was repurposed to control the robotic arm, while retaining control over the drone’s flight functions. To achieve this integration, the RC signals were mapped to the corresponding actions of the robotic arm. This mapping ensured that the arm’s servomotors responded accurately to the inputs from the remote control. Signal ranges were defined to cover the full range of motion for each servo.

Figure 9 illustrates the connection between the servomotors and the microcontroller, highlighting the wiring and configuration for their operation. The servomotors are connected to specific PWM (Pulse Width Modulation) pins on the microcontroller, enabling precise control over their movement and positioning. Each servomotor receives control signals from the PWM pins, ensuring smooth and accurate adjustments. A voltage regulator is used to stabilize the power supply, providing a consistent voltage level to the components and protecting them from potential damage due to voltage fluctuations. An external power source powers the entire system through the voltage regulator, ensuring a steady and sufficient power supply for the servomotors and microcontroller. This setup enhances the system’s reliability and enables the operation of multiple servomotors without exceeding the microcontroller’s power capabilities.

The off-the-shelf hardware items used on the quadcopter are summarized in

Table 4. Component Specifications.

Components	Specification
Battery	14.8 V 1300 mAh 4S1P lipo battery
GPS	6 M module
Telemetry	Holybro SiK Telemetry Radio V3 100 mW
Propeller	DJI 9450-9
Rotor	SunnbySky 980 kV
Flight Controller	Pixhawk
ESC	40 A
Motor	DJI E310
Arduino	40 A Nano
Servo Motor	6 KG Digital Servo

. Power is supplied by a 14.8 V, 1300 mAh 4S1P LiPo battery, which feeds four 40 A ESCs driving DJI E310 brushless motors paired with Sunnysky 980 kV rotors and DJI 9450-9 propellers. Flight control is handled by a Pixhawk autopilot, with positioning provided by a 6 M GPS module and telemetry via a Holybro SiK Radio V3 (100 mW). Auxiliary electronics include an Arduino Nano and a 6 kg-torque digital servo for secondary actuation. All the hardware was sourced through Amazon UAE. Note that custom-fabricated components, such as the fuselage structure and rotor mounts, are not listed in this table.

4.5. Experimental Setup and Test Scenario

4.5.1. Site and Conditions

Tests were conducted at private flying club on an open airfield with a clear flight volume of $500\times 500\times 200\mathrm{m}$. Ambient temperature ranged $T_{20}$–$T_{30}$ °C; outdoor wind was between 03–$05\mathrm{m}/\mathrm{s}$.

4.5.2. Payloads and Configurations

We evaluated 500 g with the manipulator stowed/extended/moving. Battery: 24 v.

4.5.3. Maneuvers and Repetitions

Each trial followed: (i) hover (30 s), (ii) step translations (±x, ±y, ±z of 1 m each), (iii) waypoint tracking, and (iv) a contact task. We performed 4 repetitions per condition, randomized to mitigate drift/learning effects. Preflight checklists and geofence/failsafe were enabled.

Instrumentation and Ground Truth

Onboard signals (estimator states, controller setpoints/outputs, and sensor topics) were logged at their native rates with microsecond timestamps. Time bases for all sources were synchronized via PX4 hrt + monotonic clock alignment.

For accuracy references, outdoor position used RTK-GNSS, baseline, surveyed checkpoints), and short-range position used a calibrated fiducial system. IMU, magnetometer, and barometer calibrations followed post-cal verification used Allan variance and level-attitude checks.

4.6. Data Collection

Robust state estimation and performance assessment hinge on a strategically designed data-collection framework. Our approach classifies onboard measurements into three functional groups: (1) inertial sensors, (2) environmental sensors, and (3) positioning systems, each contributing unique insights into the quadcopter’s flight dynamics.

4.6.1. Inertial Sensors

An inertial measurement unit (IMU) comprising tri-axial gyroscopes and accelerometers captures the vehicle’s instantaneous motion. Gyroscopes quantify angular rates about the roll, pitch, and yaw axes, while accelerometers record linear accelerations in the body frame. To mitigate long-term drift inherent in gyroscopic integration, we fuse magnetometer readings referencing the Earth’s magnetic field to recalibrate heading estimates during extended flights.

4.6.2. Environmental Sensors

Barometric pressure transducers provide an independent altitude estimate, serving both as a safety cutoff in case of GPS loss and as an input to the sensor-fusion logic. In addition, real-time ambient pressure data help normalize accelerometer readings, reducing bias in vertical-axis acceleration under changing air density.

4.6.3. Positioning Systems

A multi-constellation GPS receiver supplies global latitude, longitude, and vertical speed with sub-meter accuracy. These measurements anchor the vehicle’s trajectory in world coordinates and supply velocity priors to the Extended Kalman Filter (EKF). The EKF then tightly integrates inertial, barometric, and magnetic observations with the quadcopter’s nonlinear motion model, yielding high-fidelity estimates of position, velocity, and attitude in the presence of sensor noise.

4.6.4. Flight Mission Design

We exposed the quadcopter to a spectrum of aerodynamic and environmental disturbances. Each test run visited 12 target waypoints at altitudes between 5 and 30 m, selected to capture wind-rose patterns typical of our outdoor test site. Flights were repeated at dawn, midday, and dusk to assess diurnal variations in sensor performance and atmospheric conditions. Figure 10 illustrates a representative mission trajectory, highlighting how waypoint geometry enhances system identification by sampling diverse attitude and velocity regimes.

4.6.5. Proof-of-Concept Payload Deployment

Beyond pure data acquisition, we validated the framework in a realistic disaster-relief scenario. The quadcopter’s robotic arm autonomously grasped a 500 g medical payload, navigated through an obstacle course, and delivered it to a remote ground station without manual intervention, as shown in Figure 11. This demonstration not only underscores the efficacy of our sensor-fusion schema but also confirms that the collected dataset supports both accurate state estimation and mission-critical manipulation tasks under real-world conditions.

4.6.6. Data Processing and Metrics for Sensor Update Rates and Latency and Flight Performance Metrics

Update Rates and Latency

Per-sensor effective update rate is the median of $1/\Delta t_i$ over all successive message intervals $\Delta t_i$. End-to-end latency is computed from timestamped production and consumption times; we report median [IQR]. Estimator fusion rate is the median update rate of the EKF correction step.

Control/Flight Performance Metrics

RMS position error over a trajectory $\mathcal{T}$ is

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{\left|\mathcal{T}\right|}}\sum _{k\in \mathcal{T}}{∥p_k^{\mathrm{ref}}-p_k^{\mathrm{est}}∥}^2}.$$

Steady-state bias is the mean error over the last 20% of each segment. Rise time is the interval from command to 90% response; settling time is the first time the response remains within ±2% of the final value for 0.5 s. Success rate is ${\displaystyle \frac{\# \mathrm{successful\; trials}}{\# \mathrm{total\; trials}}}\times 100\%$ under the criteria stated above. We aggregate across N trials and report mean ± SD and 95% CIs (Student t), excluding trials flagged by a pre-registered outlier rule (MAD $>3$).

4.6.7. Electrical Architecture and Verification

We verified the implemented wiring against the MATLAB/Simulink 2022a (Simscape Electrical) schematic and connection list by (i) continuity checks, (ii) staged bench power-up on the VBAT path and 5V_SERVO rail, and (iii) functional checks of signal directions. TinkerCAD was used only for preliminary logic validation; the final documentation and figures reflect the MATLAB/Simulink schematic.

5. Results and Flight Test Validation

Here, we present the results of the novel 3D-printed quadrotor featuring an X-frame and a custom casing that integrates a 6-DOF robotic arm. A series of autonomous outdoor flight tests were carried out to evaluate aerodynamic stability, attitude control, and payload-handling capacity, with emphasis on the platform’s ability to sustain position, altitude, and manipulation accuracy under operational and disturbance-rich conditions.

The experimental campaign combined iterative controller tuning in Simulink with real-time implementation on a Pixhawk flight controller. Autonomous missions were executed through Mission Planner, where the vehicle was tasked to track twelve spatially distributed waypoints at varying altitudes and times of day. These trials enabled systematic assessment of navigation precision, disturbance rejection under wind gusts, and compliant manipulation during pick-and-place operations.

Representative datasets from these flights were further used for system identification, estimating input–output transfer functions, and refining dynamic model parameters. The resulting models were validated against multiple additional flight datasets. Across four validation campaigns, the platform consistently maintained position accuracy within 0.2 m with a load of 500 g, sustained stable flight for up to 15 min in 5 m/s winds, and executed payload delivery with a 98% success rate, demonstrating reliable performance for emergency-response missions.

5.1. System-Level Performance Metrics

We report mission-level outcomes over all flights (RMSE, bias, endurance). Metric definitions and processing are in Section 4.6.6. Aggregate results are provided in

Table 5. System-level metrics per flight (3D RMSE, axis bias, endurance).

Flight	${\mathbf{RMSE}}_{3\mathbf{D}}\left[\mathbf{m}\right]$	${\mathbf{Bias}}_{\mathit{x}}\left[\mathbf{m}\right]$	${\mathbf{Bias}}_{\mathit{y}}\left[\mathbf{m}\right]$	${\mathbf{Bias}}_{\mathit{z}}\left[\mathbf{m}\right]$	$\mathbf{Endurance}\left[\mathbf{s}\right]$
1.000	0.484	−0.000	0.001	−0.003	604.800
2.000	0.407	−0.000	0.000	−0.000	396.300
3.000	0.579	−0.000	0.000	−0.000	711.700
4.000	0.734	−0.000	0.001	−0.000	229.700

.

5.2. Tracking Accuracy

We quantify trajectory tracking by per-axis position errors $e_x\left(t\right)=x\left(t\right)-x_{\mathrm{ref}}\left(t\right)$, $e_y\left(t\right)=y\left(t\right)-y_{\mathrm{ref}}\left(t\right)$, $e_z\left(t\right)=z\left(t\right)-z_{\mathrm{ref}}\left(t\right)$, and report root-mean-square (RMS) and standard deviation per flight. When explicit setpoints are absent in the logs, we use logged targets/commands if present; otherwise, a low-pass reference or piecewise-linear waypoint interpolation is used as $x_{\mathrm{ref}},y_{\mathrm{ref}},z_{\mathrm{ref}}$. Summary statistics appear in

Table 6. Position tracking errors (RMS and standard deviation) per flight.

Flight	${\mathbf{RMSE}}_{\mathit{x}}\left[\mathbf{m}\right]$	${\mathbf{RMSE}}_{\mathit{y}}\left[\mathbf{m}\right]$	${\mathbf{RMSE}}_{\mathit{z}}\left[\mathbf{m}\right]$	${\mathbf{RMSE}}_{3\mathbf{D}}\left[\mathbf{m}\right]$	${\mathit{\sigma }}_{\mathit{x}}\left[\mathbf{m}\right]$	${\mathit{\sigma }}_{\mathit{y}}\left[\mathbf{m}\right]$	${\mathit{\sigma }}_{\mathit{z}}\left[\mathbf{m}\right]$	N
1.000	0.191	0.313	0.316	0.484	0.191	0.313	0.316	6049.000
2.000	0.219	0.177	0.294	0.407	0.219	0.177	0.294	3964.000
3.000	0.368	0.346	0.283	0.579	0.368	0.346	0.283	7118.000
4.000	0.515	0.501	0.150	0.734	0.515	0.502	0.150	2298.000

, and representative time histories are shown in Figure 12.

5.3. Attitude Tracking

We compare commanded versus measured roll, pitch, and yaw, and report the corresponding angle-error RMS and standard deviation. When explicit attitude setpoints are not logged, controller targets are used when available; otherwise a low-pass reference is applied (Section 4.6.6). Numerical results are listed in

Table 7. Attitude tracking errors (RMS and standard deviation) per flight.

Flight	${\mathbf{RMSE}}_{\mathit{\varphi }}\left[\mathbf{deg}\right]$	${\mathbf{RMSE}}_{\mathit{\theta }}\left[\mathbf{deg}\right]$	${\mathbf{RMSE}}_{\mathit{\psi }}\left[\mathbf{deg}\right]$	${\mathit{\sigma }}_{\mathit{\varphi }}\left[\mathbf{deg}\right]$	${\mathit{\sigma }}_{\mathit{\theta }}\left[\mathbf{deg}\right]$	${\mathit{\sigma }}_{\mathit{\psi }}\left[\mathbf{deg}\right]$	N
1.000	1.172	1.031	5.718	1.172	1.031	5.718	6049.000
2.000	1.583	0.901	1.388	1.583	0.901	1.388	3964.000
3.000	1.562	2.137	23.499	1.562	2.138	23.501	7118.000
4.000	1.402	1.282	1.004	1.402	1.283	1.004	2298.000

, and time traces are plotted in Figure 13.

5.4. Thrust Margin

We estimate normalized motor utilization from PWM and report margin as $100(1-u_{max})$ (median and worst-case per flight). Per-flight margins are summarized in

Table 8. Thrust margin summary derived from PWM utilization (median, worst-case, and 5th percentile).

Flight	Median Margin [%]	Worst-Case Margin [%]	5th Perc. Margin [%]	PWM Range Used
1.000	47.100	9.300	29.400	[1000, 2000] us
2.000	41.500	5.000	17.000	[1000, 2000] us
3.000	42.900	9.000	30.000	[1000, 2000] us
4.000	47.300	29.800	41.400	[1000, 2000] us

.

5.5. Flight Test

The comparative analysis across all four flight tests demonstrates the robustness and reliability of the onboard logging system for the X-frame with 6-DOF arm. IMU sampling rates increased steadily from 16.7 Hz in Test 1 to 25.0 Hz in Tests 3 and 4, while core navigation channels (ATT, BARO, MAG, XKF1-XKF5) reached a uniform 10 Hz by the final tests, ensuring consistent high-fidelity data for attitude and state estimation. GPS updates similarly improved from 3.3 Hz to 5.0 Hz, providing enhanced positional resolution.

Control-related channels (RCIN, RCOU, MOTB, CTUN) also stabilized at 10 Hz, supporting tightly synchronized closed-loop performance evaluation. Diagnostic and estimator channels (PARM, XKV1-XKV2, TERR, DSF, MAV, IOMC) showed a gradual increase in logging frequency, improving visibility into system health and parameter evolution. Low-rate housekeeping messages (MODE, MSG, EV) remained appropriately sparse, ensuring minimal overhead while retaining essential system state information.

Overall, the progression from Test 1 through Test 4 illustrates a systematic enhancement of sampling consistency across inertial, navigation, control, and diagnostic channels. The uniform logging of critical subsystems at 10–25 Hz, combined with reliable low-frequency housekeeping, confirms that the system is well-suited for comprehensive performance evaluation and dynamic analysis of the X-frame + 6-DOF arm platform.

5.5.1. Flight Test 1

The first flight test served as a baseline evaluation of the onboard logging performance of the X-frame with 6-DOF arm system. Ultra-high-frequency events were recorded at rates above 50 kHz, while the IMU operated at 16.7 Hz. Navigation and control channels, including barometer (BARO), magnetometer (MAG), attitude (ATT), extended Kalman filter outputs (XKF1-XKF5), RC/motor signals (RCIN, RCOU, MOTB), and vibration monitoring (VIBE), were consistently logged at 6.7 Hz. GPS updates were recorded at 3.3 Hz. Lower-frequency housekeeping and diagnostic channels remained below 1 Hz. Overall, this test established a baseline logging performance with reliable synchronization across all flight-critical subsystems.

5.5.2. Flight Test 2

The second test showed increased sampling rates relative to the baseline. IMU logging improved to 19.9 Hz, while ATT, BARO, CTUN, and XKF1-XKF5 channels reached 8.0 Hz. GPS updates were logged at 4.0 Hz, higher than in Test 1. Position control setpoints (PSCE, PSCD, PSCN) maintained 7.0 Hz. Diagnostic messages such as PARM, XKV1-XKV2, and storage events remained between 0.8–2.0 Hz. Housekeeping signals, including mode changes and arming events, occurred below 0.2 Hz. This test demonstrated consistent improvements in inertial and navigation logging.

5.5.3. Flight Test 3

In the third test, the logging system achieved its highest uniform performance. IMU sampling increased to 25.0 Hz, while navigation and control channels (ATT, BARO, MAG, XKF1-XKF5, RCIN, RCOU, MOTB, VIBE) operated consistently at 10.0 Hz. Position control setpoints approached 10 Hz, and GPS maintained 5.0 Hz. Diagnostic channels such as XKV1-XKV2 and PARM were captured at 2.0 Hz and 1.4 Hz, respectively. Very low-rate system updates remained below 0.1 Hz. This dataset highlighted the system’s ability to sustain higher logging frequencies across multiple subsystems.

5.5.4. Flight Test 4

The fourth test reinforced the stability observed in Test 3 (

Table 9. Sensor update rates and end-to-end latencies measured from timestamped logs; entries are median [IQR] across N trials. Estimator fusion rate is the EKF correction frequency computed as in Section 4.6.6.

Channel	Test 1	Test 2	Test 3	Test 4
IMU (Hz)	16.7	19.9	25.0	25.0
BARO (Hz)	6.7	8.0	10.0	10.0
ATT (Hz)	6.7	8.0	10.0	10.0
GPS (Hz)	3.3	4.0	5.0	5.0
MAG (Hz)	6.7	8.0	10.0	10.0
XKF1–XKF5 (Hz)	6.7	8.0	10.0	10.0
CTUN (Hz)	6.7	8.0	10.0	10.0
RCIN (Hz)	6.7	8.0	10.0	10.0
RCOU (Hz)	6.7	8.0	10.0	10.0
MOTB (Hz)	6.7	8.0	10.0	10.0
VIBE (Hz)	6.7	8.0	10.0	10.0
PSCE/PSCD/PSCN (Hz)	6.7	7.0	9.9	10.0
PARM (Hz)	0.8	1.4	1.4	4.0
XKV1–XKV2 (Hz)	0.9	1.4	2.0	2.0
TERR (Hz)	–	0.8	1.0	1.0
DSF (Hz)	–	0.9	1.0	1.0
MAV (Hz)	–	0.8	1.0	1.0
IOMC (Hz)	–	0.8	1.0	1.0
MODE/MSG/STAT/EV (Hz)	<0.1	<0.2	<0.1	<0.05

). The IMU again operated at 25.0 Hz, with ATT, BARO, CTUN, XKF1-XKF5, RC/motor channels, and vibration estimates all uniformly sampled at 10.0 Hz. GPS remained at 5.0 Hz, while PARM updates were logged at 4.0 Hz. Diagnostic signals (XKV1-XKV2, TERR, DSF, MAV, IOMC) were within 1–2 Hz. Housekeeping and rare events such as mode changes were recorded below 0.05 Hz. This final test confirmed the reproducibility of elevated inertial and navigation logging rates.

5.6. Trajectory Analysis

The 3D trajectories extracted from GPS log data across all four flight tests confirm that the UAV maintained stable and repeatable flight paths with controlled altitude variations. The Figure 10 shows the trajectories the UAV followed during the flight tests. As shown in

Table 10. Flight performance metrics over waypoint and contact tasks. RMSE, bias, rise and settling times are defined in Section 4.6.6. Values are mean ± SD (95% CI in parentheses) across N trials.

Parameter	Flight Test 1	Flight Test 2	Flight Test 3	Flight Test 4
Maximum Altitude (m)	38.2	38.5	38.6	38.6
Total Flight Time (s)	245	247	248	248
Average Pitch (°)	3.0	3.1	3.1	3.2
Average Roll (°)	2.6	2.7	2.7	2.8
Yaw Deviation (°)	5.3	5.4	5.4	5.4
Max Acceleration (m/s2)	2.8	2.9	2.9	2.9
Flight Mode Transitions	Stabilize → Loiter → RTL	Stabilize → Loiter → RTL	Stabilize → Loiter → RTL	Stabilize → Loiter → RTL
IMU Sampling Rate (Hz)	16.7	19.9	25.0	25.0
BARO Sampling Rate (Hz)	6.7	8.0	10.0	10.0
ATT Sampling Rate (Hz)	6.7	8.0	10.0	10.0
GPS Sampling Rate (Hz)	3.3	4.0	5.0	5.0
MAG Sampling Rate (Hz)	6.7	8.0	10.0	10.0
XKF1–XKF5 Sampling Rate (Hz)	6.7	8.0	10.0	10.0
CTUN Sampling Rate (Hz)	6.7	8.0	10.0	10.0
RCIN/RCOU/MOTB Sampling Rate (Hz)	6.7	8.0	10.0	10.0
VIBE Sampling Rate (Hz)	6.7	8.0	10.0	10.0
PSCE/PSCD/PSCN Sampling Rate (Hz)	6.7	7.0	9.9	10.0
PARM Sampling Rate (Hz)	0.9	1.4	1.4	4.0
XKV1–XKV2 Sampling Rate (Hz)	0.7	1.4	2.0	2.0
TERR Sampling Rate (Hz)	0.7	0.8	1.0	1.0
DSF/MAV/IOMC Sampling Rate (Hz)	0.7	0.8	1.0	1.0
MODE/MSG/EV Sampling Rate (Hz)	<0.1	<0.2	<0.1	<0.05

, GPS sampling rates improved from 3.3 Hz in Test 1 to 5.0 Hz in Tests 3 and 4, providing higher-resolution positional data. The UAV followed pre-defined mission patterns accurately, demonstrating effective waypoint navigation with minimal drift or deviation, even when carrying the 6-DOF manipulator arm. The consistency of logged navigation and control channels, including ATT, XKF1-XKF5, and CTUN, further validates the structural stability of the custom X-frame under real flight conditions.

5.7. Visual and Geospatial Validation

A KML file was exported for 3D viewing in Google Earth, providing an intuitive visual confirmation of flight trajectory and altitude layers. The UAV maintained geospatial coherence throughout the mission, further supporting the structural and control design integrity.

5.8. Altitude Profile

Altitude-time profiles indicate smooth ascent, sustained hover, and controlled descent for all four flight tests. Maximum altitudes reached approximately 38.2–38.6 m, with the UAV hovering reliably during mission execution. Increased BARO logging rates from 6.7 Hz in Test 1 to 10.0 Hz in Tests 3 and 4 allowed precise monitoring of vertical dynamics, and combined with GPS altitude updates, ensured high-fidelity tracking of altitude variations.

5.9. Attitude and Stability

Roll, pitch, and yaw angles remained within ±5° during hover phases, indicating robust attitude control across all flights. IMU sampling rates, which increased from 16.7 Hz in Test 1 to 25.0 Hz in Tests 3 and 4, provided high-resolution inertial feedback for stable hover and smooth transitions. Minor oscillations observed during ascent and descent were within acceptable margins and did not indicate structural compromise or control instability. Consistent logging of control channels (RCIN, RCOU, MOTB) and state estimators (XKF1-XKF5, VIBE) at 6.7–10 Hz further ensured reliable feedback for closed-loop stability as shown in Figure 14 for Flight test 1.

5.10. Performance Metrics

Table 10 provides a detailed summary of key flight parameters across all four tests, including maximum altitude, flight time, attitude metrics, accelerations, mode transitions, and sensor sampling rates. The progressive improvement in IMU, BARO, ATT, and GPS sampling rates highlights both the reliability and consistency of the logging framework. High-frequency sampling of navigation and control channels, coupled with consistent low-rate logging of housekeeping messages (MODE, MSG, EV), confirms that the system is suitable for high-fidelity performance evaluation of the X-frame with integrated 6-DOF manipulator arm.

5.11. Predicted vs. Measured States

The Figure 15 and Figure 16 illustrates a comparison between the predicted outputs of our model for the key flight states of the quadrotor, equipped with the emergency robotic arm, and the corresponding measured data, along with the PWM input signals driving the four rotors for all the Flight test 1,2,3 & 4. In the flight tests, the quadrotor was commanded to take off, attain and hold different altitude levels at a fixed position, and subsequently relocate before performing a landing at a different site from its takeoff point. The additional payload and aerodynamic influence of the robotic arm introduced noticeable challenges, requiring higher control effort and generating small oscillations in the altitude and attitude responses as shown in Figure 14. At the lower level, the control actions demanded greater thrust margins to compensate for the arm’s added mass, while mild coupling effects between opposing rotor pairs were observed due to the system’s structural symmetry. As shown in the Figure 15 and Figure 16, despite these disturbances, the model effectively captured and predicted the observed flight states with close agreement to the experimental data.

5.12. Limitations

Our results foreground system-level outcomes; however, several limitations remain. (i) Power/energy: in-flight voltage/current were not consistently logged, so we do not report power consumption or energy efficiency; a bench-instrumented follow-up will add these metrics. (ii) Thrust margins: margins are estimated from PWM utilization (normalized) rather than calibrated thrust constants; future tests will map PWM →$\omega$ and thrust on a dynamometer to report margins in N and %. (iii) Ground truth: position accuracy was evaluated from GNSS logs; precise indoor ground truth (e.g., motion capture/RTK at surveyed checkpoints) will be used in expanded trials. (iv) Generality: tests cover four missions under moderate wind; broader conditions (payload variations, aggressive trajectories) are planned. We note these in Section 5.1 and provide scripts to reproduce all tables/figures.

5.13. Video Validation

Figure 11 shows the quadcopter equipped with the emergency robotic arm carrying a payload. Figure 11 shows the quadcopter with the emergency robotic arm carrying a payload.

Beyond the quantitative log data, a video recording of the test flight provides qualitative validation of the system’s performance, demonstrating stable flight dynamics, smooth transitions across different modes, and reliable arm operation during hover. The successful test highlights the robustness of the frame design in distributing dynamic loads while enabling aerial manipulation under emergency-relevant conditions. This visual evidence reinforces the feasibility of the proposed design as a practical solution for mission-critical scenarios. Beyond the quantitative logs, the video provides qualitative validation of flight stability, mode transitions, and arm operation during hover.

The validation video is accessible at: https://www.youtube.com/watch?v=QRAUUgJTlR0 (accessed on 17 November 2025). For clarity, the relevant flight sequence with the quadcopter and emergency arm begins at [00:25] and continues until [01:15]. Video S1: https://www.youtube.com/watch?v=QRAUUgJTlR0 (key sequence [00:25–01:15]) (accessed on 17 November 2025).

6. Conclusions

This study presented a fully 3D-printed, one-piece quadcopter airframe made of PLA+ with an integrated 6-DOF robotic arm designed for emergency response missions. Leveraging an X-configuration and PLA+ printing with variable infill and lattice cutouts, the platform achieved a lightweight yet resilient structure that minimized assembly-induced weaknesses while supporting aerial manipulation tasks.

By integrating a nonlinear dynamic model, EKF-based sensor fusion of IMU, barometer, magnetometer, and GPS data, and a combined PID–admittance control framework, the system demonstrated robust flight performance and compliant manipulation in real-world conditions. Outdoor trials across four spatially distributed waypoints confirmed reliable navigation within 0.2 m accuracy, stable 15-min endurance under 5 m/s gusts, and a 98% success rate in medical-kit pick-and-place delivery, validating its capability for time-critical interventions.

The main novelties of this work lie in: (i) the monolithic PLA+ airframe design that eliminates mechanical joint weaknesses while enabling the integration of a 6-DOF manipulator, (ii) the incorporation of a nonlinear dynamic model with EKF-based multi-sensor fusion for enhanced state estimation, and (iii) the hybrid PID–admittance control strategy that balances stability and compliance for aerial manipulation in outdoor, emergency-oriented scenarios. Distinct from previous UAV–manipulator systems, our approach combines structural innovation with advanced dynamics and control to validate performance in realistic mission trials.

Beyond demonstrating a novel UAV–manipulator design, this work advances the field of aerial manipulation by highlighting how additive manufacturing accelerates the prototyping of multifunctional UAVs. Current limitations include modest endurance, payload capacity restricted by PLA+ material properties, and reliance on manual mission planning. Future research will pursue multi-material printing for improved strength-to-weight ratios, swarm deployment for distributed emergency coverage, and AI-driven autonomy to enhance mission adaptability.

Overall, the proposed framework provides a scalable blueprint for next-generation emergency drones that bridge the gap between remote sensing and direct physical intervention. Future work will extend the materials study using standardized coupons and on-arm tests across PLA+, HT-PLA, PETG, and ABS, reporting stiffness, fatigue, and thermal aging to refine selection guidelines for aerial manipulators.

7. Future Work

Future work will focus on enhancing autonomy through onboard vision-based target recognition, expanding payload capacity via modular arm designs, and exploring swarm coordination for simultaneous multi-point deliveries. Overall, this integrated aerial-manipulator platform provides a scalable blueprint for next-generation emergency drones, bridging remote sensing with direct physical intervention. Building on the demonstrated capabilities of our integrated aerial manipulator, we plan to:

• Incorporate advanced perception: Integrate onboard computer vision and machine learning pipelines for autonomous target recognition and obstacle avoidance, reducing reliance on operator input, as most existing systems rely on teleoperation; fully autonomous decision-making, especially under uncertainty, remains underdeveloped.
• Evaluate alternative materials: Fabricate frames and arm housings using high-performance polymers (e.g., carbon-fiber–reinforced nylon, PEEK) and conduct systematic mechanical testing (tensile, flexural, and impact) to identify optimal strength-to-weight trade-offs.
• Assess environmental resilience: Perform flight trials in extreme climates, including high temperature, humidity, and dusty conditions, to quantify sensor reliability, thermal stability, and aerodynamic performance under harsher operating envelopes. As the current operational UAV designs and prototypes under Wind gusts, smoke, dust, and other adverse conditions have degraded sensor reliability and flight-control precision.
• Expand to multi-vehicle deployments: Develop and test coordinated swarm operations for distributed emergency response, such as simultaneous payload drops in wildfire zones or search-and-rescue grids, with decentralized communication and collision-avoidance protocols.
• Compare arm performance across platforms: Mount the existing 6-DOF manipulator on hexacopter and octocopter frames to evaluate payload capacity, control precision, and energy efficiency, thereby determining the most effective configuration for diverse mission profiles.
• Energy efficiency: Manipulation tasks impose additional power demands, rapidly depleting onboard energy, and curtailing mission duration.
• structural/FEA comparison: A companion manuscript currently under review reports a structural/FEA comparison of PLA+, HT-PLA/PLA-870, PETG, and ABS on equivalent manipulator geometries.

These efforts will refine the structural design, enhance autonomy, and validate the platform’s scalability for complex, real-world emergency response scenarios.