Wind-Resistant UAV Landing Control Based on Drift Angle Control Strategy

Abstract

Addressing lateral-directional control challenges during unmanned aerial vehicle (UAV) landing in complex wind fields, this study proposes a drift angle control strategy that integrates coordinated heading and trajectory regulation. An adaptive radius optimization method for the Dubins approach path is designed using wind speed estimation. By developing a wind-coupled flight dynamics model, we establish a roll angle control loop combining the $L_1$ nonlinear guidance law with Linear Active Disturbance Rejection Control (LADRC). Simulation tests against conventional sideslip approach and crab approach, along with flight tests, confirm that the proposed autonomous landing system achieves smoother attitude transitions during landing while meeting all touchdown performance requirements. This solution provides a theoretically rigorous and practically viable approach for safe UAV landings in challenging wind conditions.

1. Introduction

As the critical phase in UAV flight operations, the landing stage mandates high-precision and robust control systems to ensure mission success and system safety. This phase demands simultaneous multi-objective control under dynamic crosswinds, including trajectory tracking and attitude adjustment, which significantly increase control complexity.

Many researchers conducted research in the topic of aerodynamic disturbances during UAV landing, Lungu M’s team conducted systematic research employing dynamic inversion as the cornerstone methodology. Reference [1] integrated neural networks with pseudo-control hedging to compensate actuator saturation while ensuring robust lateral deviation control, and Reference [2] significantly accelerated transient responses by optimizing roll-yaw coordination efficiency; simultaneously, References [3,4] developed a deep-coupling architecture merging dynamic inversion with backstepping control, incorporating nonlinear disturbance observers and feedforward neural networks to handle wind interference and mass-center perturbations thereby achieving high-precision trajectory tracking during landing.

For carrier-based aircraft operations, refs. [5,6] proposed a nonlinear dynamic inversion (NDI)-based solution utilizing linear reference models to estimate unmeasurable airflow disturbances, injecting compensation states into NDI loops while constructing high-dimensional risk fields for real-time risk quantification to generate dynamic compensatory signals optimizing control commands; this approach nevertheless suffers from substantial computational burdens and lacks robust experimental validation in complex real-world environments beyond idealized simulations. Ref. [7] integrates a Nonlinear Guidance Law (NLGL) for trajectory tracking, employs gain-scheduled adaptive PID controllers for attitude management, and utilizes multivariable H∞ control during flare phase, demonstrating robust crosswind performance. However, practical deployment requires further exploration of fault-tolerant capabilities for complex failure modes and online trajectory replanning mechanisms.

To enhance control performance in complex landing scenarios, subsequent studies have explored more advanced technical pathways. Ref. [8] employed a four-stage cascaded adaptive sliding mode control architecture. This was combined with Kalman filter prediction and tracking differentiator compensation for deck motion. By suppressing air wake disturbances and parameter uncertainties through an adaptive law, the method achieved high-precision landing trajectory tracking for carrier-based UAVs. Ref. [9] realized an anti-windup robust adaptive controller by integrating fixed-time control, fuzzy logic, and integral sliding mode techniques, guaranteeing fixed-time convergence under wind disturbances and parameter uncertainties. These methods harness the strong robustness inherent in sliding mode control as their core, and enhance system adaptability to disturbances/uncertainties by incorporating prediction algorithms and dynamic compensators. Nevertheless, they share common challenges: substantial computational resource demands, high parameter sensitivity, and considerable tuning difficulty. Further developments include [10], which utilized model predictive control (MPC) to dynamically adjust state and control priorities based on real-time landing risk assessment. A BP neural network predicted touchdown risk to optimize lateral attitude and trajectory, achieving coordinated wind disturbance rejection, risk avoidance, and rapid convergence. Ref. [11] designed a hierarchical controller within a fixed-time control (FTC) framework using backstepping. This involved a fixed-time disturbance observer (FTO) for real-time motion disturbance estimation and a fixed-time filter replacing traditional derivative calculations, achieving fixed-time convergence of lateral errors and enhanced robustness. Ref. [12] exploited the error-constraint capability of guaranteed performance control (GPC) and the nonlinear decoupling advantages of backstepping. By decomposing the longitudinal-lateral coupling model, it achieved independent angle-of-attack control and realized lateral-directional dynamic decoupling via thrust compensation. Command filters were introduced to avoid computational complexity in virtual control derivatives, demonstrating excellent performance in carrier landing scenarios. Refs. [13,14] proposed a control scheme based on nonlinear decoupled inner-loop control combined with a comprehensive sideslip-based outer-loop. The approach implemented roll angle and yaw angle command saturation limits that vary with relative height, with control parameters optimized via genetic algorithms. This achieved comprehensive control over sideslip distance, sideslip velocity, roll angle, and yaw angle. However, significant discrepancies between actual wind conditions and simulation assumptions may necessitate adaptive adjustments to wind-related fitness function weights or adoption of online real-time optimization.

Synthesizing the aforementioned advancements and remaining challenges across various technical pathways, refs. [15,16] demonstrate that despite in-depth investigations into lateral-directional control for autonomous landing systems, the field remains dominated by simulation-based studies lacking experimental validation of system reliability.

To address the limitations of existing approaches, which include engineering implementation barriers such as high computational load and poor parameter robustness in real-world scenarios, insufficient environmental adaptability arising from the lack of dynamic adjustment mechanisms under actual wind conditions, and inadequate experimental validation systems marked by over-reliance on simulations and insufficient real flight tests, this paper proposes a novel lateral-directional autonomous landing control system based on a drift angle control strategy. The main contributions are:

• Adaptive Path Planning: A wind-compensated Dubins path planner incorporating an adaptive turning radius derived from coordinated-turn dynamics and wind speed estimation, minimizing initial glide-path deviations in crosswinds.
• Coordinated Control Strategy: A drift angle controller synthesizing the principles of crab and sideslip approaches to minimize last-second corrections while smoothing attitude transitions during landing.
• Validated Simulation Framework: A multi-scale wind field model enabling realistic evaluation of the system’s disturbance rejection capabilities.
• Flight-Tested Validation: Empirical verification through practical flight tests. demonstrating operational applicability under real-world crosswind conditions.

Paper Structure: Section 2 Baseline parameters of the reference UAV are characterized. develops a six-degree-of-freedom (6-DOF) nonlinear dynamics model. followed by the development of a multi-scale wind field model incorporating constant winds, discrete gusts, and turbulence. Section 3 implements Dubins path planning for the approach phase, incorporating an adaptive radius adjustment mechanism to counteract wind disturbances. Trajectory planning is performed for both the glide slope segment and flare phase. Section 4 introduces the $L_1$ nonlinear guidance algorithm and designs a LADRC-based roll attitude control loop. Through rigorous analysis of the limitations inherent in conventional sideslip approach and crab approach, we propose a drift angle control strategy. Section 5 conducts crosswind landing simulations. Performance metrics are evaluated across variable wind speeds. Section 6 validates the Dubins path feasibility and the drift angle control’s disturbance rejection efficacy through flight tests, assessing heading alignment precision, attitude correction effectiveness, and touchdown stability. Finally, Section 7 concludes the research, stating that the proposed integrated solution combining the drift angle control strategy and wind-adaptive trajectory planning enables UAV to achieve smooth attitude transitions and precise trajectory tracking during landing in complex crosswind environments.

2. Model

2.1. UAV Model

This UAV is developed through unmanned modification based on a four-seat manned aircraft platform. Redundant pilot control devices and auxiliary equipment in the cockpit have been removed to free up internal payload space for the installation of the UAV flight control system. While retaining the original aerodynamic characteristics and structural strength of the airframe, it has achieved core unmanned functions such as autonomous takeoff and landing, route tracking, and emergency handling. The airframe is constructed with all composite materials, featuring a twin-engine configuration, a low-wing layout equipped with winglets, and a T-tail design. It adopts a tricycle landing gear with retraction mechanisms, utilizing main wheel braking and nose wheel steering. The flaps are adjustable at three fixed angles: 0°, 20°, and 42° respectively. Figure 1 presents the physical diagram of the UAV, and its basic parameters are listed in

Table 1. Specification of the fixed-wing UAV.

Notations	Definition	Value
$m$	Vehicle mass (kg)	2000
$S_{ref}$	Refernce area (m2)	16.36
$\overline{c}$	Mean aerodynamic chord (m)	1.283
$I_x$	$x-\mathrm{axis}$ moment of inertia (kg·m2)	5093.78
$I_y$	$y-\mathrm{axis}$ moment of inertia (kg·m2)	3414.02
$I_z$	$z-\mathrm{axis}$ moment of inertia (kg·m2)	8198.22
$x_{cg}$	$x$-axis cg position (m)	2.421
$y_{cg}$	$y$-axis cg position (m)	0
$z_{cg}$	$z$-axis cg position (m)	1.427

.

Lateral-longitudinal coupling effects during landing are significant. Applying Newtonian rigid-body kinematics, we establish a 6-DOF motion model:

$$\left\{\begin{array}{l}\dot{x}=f\left(x,t,u\right)\\ y=cx\end{array}\right.$$

(1)

This model integrates three translational and three rotational motions through 12 first-order nonlinear equations [17]:

Translational Dynamics:

$$\left\{\begin{array}{l}\dot{U}=rV-qW-g\sin \theta +{\displaystyle \frac{F_x}{m}}\\ \dot{V}=-rU+pW+g\sin \varphi \cos \theta +{\displaystyle \frac{F_y}{m}}\\ \dot{W}=qU-pV+g\cos \varphi \cos \theta +{\displaystyle \frac{F_z}{m}}\end{array}\right.$$

(2)

where $U$, $V$, $W$ are body-axis velocity components; $\varphi$, $\theta$, $\psi$ denote roll, pitch, and yaw angles; $p$, $q$, $r$ are angular rates; and $F_x$, $F_y$, $F_z$ represent aerodynamic/propulsive force components.

Rotational Dynamics:

$$\left\{\begin{array}{l}\dot{p}={\displaystyle \frac{1}{I_x{I}_z-I_{xz}^2}}\left[I_z\overline{L}+I_{xz}N+\left[\left(I_y-I_z\right)I_z-I_{xz}^2\right]rq+\left(I_x-I_y+I_z\right)I_{xz}pq\right]\\ \dot{q}={\displaystyle \frac{1}{I_y}}\left[\left(I_z-I_x\right)pr-I_{xz}\left(p^2-r^2\right)+M\right]\\ \dot{r}={\displaystyle \frac{1}{I_x{I}_z-I_{xz}^2}}\left[\left[I_x\left(I_x-I_y\right)+I_{xz}^2\right]pq-\left(I_x-I_y+I_z\right)I_{xz}rq+I_{xz}\overline{L}+I_{x}N\right]\end{array}\right.$$

(3)

where $I_x$, $I_y$, $I_z$, $I_{xz}$ are inertia terms; $\overline{L}$, $M$, $N$ denote rolling, pitching, and yawing moments.

Attitude Kinematics:

$$\left\{\begin{array}{l}\dot{\varphi }=p+\tan \theta (q\sin \varphi +r\cos \varphi )\\ \dot{\theta }=q\cos \varphi -r\sin \varphi \\ \dot{\psi }={\displaystyle \frac{1}{\cos \theta }}(q\sin \varphi +r\cos \varphi )\end{array}\right.$$

(4)

Navigation Equations:

$$\left\{\begin{array}{l}{\dot{P}}_N=U\cos \theta \cos \psi +V(\sin \varphi \sin \theta \cos \psi -\cos \varphi \sin \psi )+W(\cos \varphi \sin \theta \cos \psi +\sin \varphi \sin \psi )\\ {\dot{P}}_E=U\cos \theta \sin \psi +V(\sin \varphi \sin \theta \sin \psi +\cos \varphi \cos \psi )+W(\cos \varphi \sin \theta \sin \psi -\sin \varphi \cos \psi )\\ \dot{H}=U\sin \theta -V\sin \varphi \cos \theta -W\cos \varphi \cos \theta \end{array}\right.$$

(5)

where $P_N$, $P_E$, $H$ denote north position, east position, and altitude.

(a)

Flight control systems

The flight control system consists of a flight control computer, sensors, an airborne data transmission link, and an actuator servo system, with a simplified schematic diagram of the system shown in Figure 2.

The flight control computer is designed and developed based on the PowerPC architecture 32-bit MPC5125 processor, boasting an operating speed of up to 400 MHz. It adopts a low-power consumption design with rich interface types, enabling stable operation in environments ranging from −30 °C to 70 °C. Equipped with the VxWorks5.5 real-time operating system, it ensures strict time constraints with short task switching times and minimal interrupt latency. The design of software modules is detailed in the

Table 2. Design of Flight Control Software Modules.

Module	Function	Real—Time Level
Sensor Driver	Data acquisition, filtering, and timestamp alignment	Highest (1 ms)
Control Law Calculation	Control algorithm computation	High (5 ms)
Redundancy Management	Master—slave state synchronization and fault—switching voting	Medium (10 ms)
Communication Protocol Stack	CAN/AFDX protocol encapsulation and parsing	Medium (10 ms)
Fault Diagnosis	BIT self—test, log recording, and alarm reporting	Low (100 ms)

.

The sensor system comprises an inertial navigation system, positioning system, air data computer, and millimeter-wave radar, among other components. The inertial navigation system consists of primary and backup subsystems, providing real-time and precise data such as attitude and heading. The positioning system, which combines the airborne terminal with a ground base station via RTK, delivers centimeter-level real-time positioning. Pitot tubes are installed at the wingtips of both left and right wings as well as the nose, forming a triple-redundancy system. Their data, together with that from the inertial navigation system, is fed into the air data computer to provide accurate airspeed, ground speed, and estimated airspeed data. The millimeter-wave radar provides auxiliary relative altitude data during the UAV’s takeoff and landing phases.

2.2. Wind Field Model

During UAV landing, complex wind conditions significantly compromise attitude stability and trajectory tracking accuracy, posing severe threats to landing safety and precision. To enhance simulation fidelity and validate control algorithm robustness, a high-fidelity wind field model must be deeply coupled with flight dynamics. Based on fluid dynamics theory and empirical wind spectrum data, we establish a wind field model incorporating constant wind, discrete gust and turbulence components.

2.2.1. Constant Wind

Natural wind never remains strictly constant. The operational concept of “constant wind” refers to wind conditions observed over a defined temporal scale during which the mean velocity component exhibits statistical stationarity. Notwithstanding this characterization, instantaneous wind speeds invariably demonstrate stochastic fluctuations about the mean value. In engineering practice, this stochastic process is formally abstracted as a constant wind parameterization.

2.2.2. Discrete Gust

A discrete gust is defined as an airflow disturbance of short duration (generally less than several tens of seconds), characterized by abrupt wind speed variations. Gusts exhibit non-stationarity, randomness, and spatial variability, with wind speeds showing transient discontinuities during atmospheric motion. We adopt the one-minus-cosine (1-cosine) model specified in MIL-F-8785C to describe discrete gusts [18]:

$$V_w=\left\{\begin{array}{ll}0\hfill & h>h_m\hfill \\ {\displaystyle \frac{V_{wm}}{2}}\left(1-\cos {\displaystyle \frac{\pi h}{h_m}}\right)\hfill & 0<h\le h_m\hfill \\ V_{wm}\hfill & h=0\hfill \end{array}\right.$$

(6)

where $h$ indicates relative height, $h_m$ denotes the gust gradient distance, and $V_{wm}$ represents the gust intensity.

2.2.3. Turbulence

Turbulent wind constitutes continuous random fluctuations superimposed on discrete gusts. Like gusts, it varies temporally and spatially, manifesting as random oscillations about a mean value. Stochastic process theory is therefore required for characterization. We implement the von Kármán spectrum for atmospheric turbulence representation. Proposed by von Kármán and extensively validated with empirical data, this model provides superior spectral alignment with real atmospheric turbulence and demonstrates stronger agreement with experimental measurements, effectively capturing inherent turbulence characteristics [19].

Turbulence simulation employs signal processing principles, where white noise input to a shaping filter outputs colored noise. Since conjugate decomposition of the von Kármán temporal spectrum is infeasible, we rationalize the model to obtain the shaping filter transfer function [20]

$$\left\{\begin{array}{l}{G}_u(s)={\sigma }_u\sqrt{{\displaystyle \frac{L_u}{\pi V_a}}}\cdot {\displaystyle \frac{1}{{\left(1+\frac{a{L}_u}{V_a}s\right)}^{5/6}}}\\ G_v(s)={\sigma }_v\sqrt{{\displaystyle \frac{L_v}{\pi V_a}}}\cdot {\displaystyle \frac{1+\sqrt{\frac{8}{3}}\cdot \frac{2a{L}_v}{V_a}s}{{\left(1+\frac{a{L}_v}{V_a}s\right)}^{11/6}}}\\ G_w(s)={\sigma }_w\sqrt{{\displaystyle \frac{L_w}{\pi V_a}}}\cdot {\displaystyle \frac{1+\sqrt{\frac{8}{3}}\cdot \frac{2a{L}_w}{V_a}s}{{\left(1+\frac{a{L}_w}{V_a}s\right)}^{11/6}}}\end{array}\right.$$

(7)

Rational approximation simplifies this to first-order form:

$$\left\{\begin{array}{l}{G}_u(s)={\displaystyle \frac{K_u}{T_{u}s+1}},K_u={\sigma }_u\sqrt{{\displaystyle \frac{L_u}{\pi V_a}}},T_u={\left({\displaystyle \frac{a{L}_u}{V_a}}\right)}^{5/6}\\ G_v(s)={\displaystyle \frac{K_v}{T_{v}s+1}},K_v={\sigma }_v\sqrt{{\displaystyle \frac{L_v}{\pi V_a}}},T_v={\displaystyle \frac{{\left(\frac{2a{L}_v}{V_a}\right)}^{11/6}}{\sqrt{\frac{8}{3}}\frac{2a{L}_v}{V_a}}}\\ G_w(s)={\displaystyle \frac{K_w}{T_{w}s+1}},K_w={\sigma }_w\sqrt{{\displaystyle \frac{L_w}{\pi V_a}}},T_w={\displaystyle \frac{{\left(\frac{2a{L}_w}{V_a}\right)}^{11/6}}{\sqrt{\frac{8}{3}}\frac{2a{L}_w}{V_a}}}\end{array}\right.$$

(8)

where ${\sigma }_u$, ${\sigma }_v$, ${\sigma }_w$ denote turbulence intensities; $L_u$, $L_v$, $L_w$ represent turbulence scales; and $a$ = 1.339. Simulation parameters ${\sigma }_u$ = ${\sigma }_v$ = ${\sigma }_w$ = 2 m/s, $V_a$ = 45 m/s and $L_u$ = $L_v$ = $L_w$ = 100 m, The resulting three-axis turbulence components are shown in Figure 3.

2.2.4. Wind Field Synthesis

The atmospheric environment model is mathematically constructed through the superposition of three components: a baseline constant wind, discrete gusts, and stochastic turbulence. Upon specification of the constant wind magnitude, turbulence intensity is parameterized as 0.1 times this reference value due to their established correlation. Crucially, gust encounters prior to touchdown substantially degrade landing performance. Therefore, a 2 m/s gust component is superimposed when altitude decreases below 100 m. Figure 4 illustrates a synthesized 4 m/s crosswind field.

The wind field is integrated with UAV dynamics via coordinate transformation. In windless conditions, body-axis velocity components are $u$, $v$, $w$. With wind, airspeed components become:

$$\left[\begin{array}{l}{u}_w\\ v_w\\ w_w\end{array}\right]=\left[\begin{array}{l}u\\ v\\ w\end{array}\right]+S\left[\begin{array}{l}{V}_{wx}\\ V_{wy}\\ V_{wz}\end{array}\right]$$

(9)

where $S$ is the Earth-to-body transformation matrix.

3. Trajectory Planning

Landing constitutes a complex and precise operational process requiring coordinated control across multiple phases to ensure safe, smooth touchdown. UAV landing typically progresses through sequential phases: approach phase, glideslope phase, flare phase, and touchdown phase. The longitudinal profile of this process is illustrated in Figure 5.

3.1. Approach Phase

The approach phase aims to position the UAV proximate to the landing runway, aligning its heading with the runway centerline to minimize lateral deviation while maintaining altitude and airspeed within operational limits for transition to the glideslope phase. To satisfy these requirements, Dubins path planning is implemented for approach trajectory generation.

The Dubins path, pioneered by mathematician Lester Dubins, represents the shortest path connecting two points in a 2D plane under curvature constraints and prescribed initial/final tangent vectors. Six optimal path types exist (LSL, RSR, RSL, LSR, RLR, LRL) [21]. This algorithm autonomously generates approach paths based on start/end points and heading directions, with the turn radius $R_a$ as the sole design parameter.

Considering coordinated turns during constant-velocity gliding, the three-degree-of-freedom point-mass aircraft kinematic model yields:

$$\left\{\begin{array}{c}L\sin \varphi ={\displaystyle \frac{m{V}_a^2{\cos }^2\gamma }{R_a}}\\ L\cos \varphi =mg\cos \gamma \\ D=mg\sin \gamma \end{array}\right.$$

(10)

where $L$ is the lift, $D$ the drag, $\varphi$ the roll angle, $m$ the mass, $V_a$ the airspeed, $R_a$ the turning radius and $\gamma$ the glideslope angle. The coordinated turn radius is derived as:

$$R_a={\displaystyle \frac{V_a^2\cos \gamma }{g\tan \varphi }}$$

(11)

During turning maneuvers, continuously varying wind fields and headings complicate disturbance assessment. Strict adherence to the theoretical turn radius may yield poor path-tracking performance or excessive roll angles. Therefore, adaptive radius adjustment is implemented. Wind velocity ${\tilde{V}}_w$ is estimated pre-maneuver using airspeed/groundspeed data, while longitudinal effects are neglected ($\cos \gamma \approx 1$ for shallow glideslopes). The adaptive radius becomes:

$$R_a={\displaystyle \frac{{(V+{\tilde{V}}_w)}^2}{g\tan \varphi }}$$

(12)

The required gliding distance to reach target altitude $H_{cmd}$ from current altitude $H$ is:

$${\mathcal{L}}_{\gamma }={\displaystyle \frac{(H-H_{\mathit{cmd}})}{\tan \gamma }}$$

(13)

Given start point $A(x_i,y_{i,}{\phi }_i)$, endpoint $B(x_g,y_{g,}{\phi }_g)$, and adaptive turn radius $R_a$, coordinate transformation is performed: origin shifts to $A$ followed by rotation $\vartheta$ to align $B$ with the horizontal axis. The schematic diagram of coordinate transformation is shown in Figure 6.

Following coordinate transformation, the start and end points in the new coordinate system are defined as $A(0,0,{\phi }_{ni})$ and $B(D,0,{\phi }_{ng})$, respectively. The transformation parameters are determined through

$$D=\sqrt{\left({(x_i-x_g)}^2+{(y_i-y_g)}^2\right)}$$

(14)

$$\vartheta =\mathrm{atan}\left({\displaystyle \frac{y_g-y_i}{x_g-x_i}}\right)$$

(15)

$${\phi }_{ni}={\phi }_i-\vartheta$$

(16)

$${\phi }_{ng}={\phi }_g-\vartheta$$

(17)

normalizing with turn radius:

$$d=D/R_a$$

(18)

The Dubins Path consists of three basic motions in a plane: left-turn arcs ($L$), right-turn arcs ($R$), and straight segments ($S$), with coordinate transformations of arbitrary points $(x,y,\phi )$ defined as

$$\left\{\begin{array}{l}{L}_v(x,y,\phi )=\left(x+\sin (\phi +v)-\sin \phi ,y-\cos (\phi +v)+\cos \phi ,\phi +v\right)\\ R_v(x,y,\phi )=\left(x-\sin (\phi -v)+\sin \phi ,y+\cos (\phi -v)-\cos \phi ,\phi -v\right)\\ S_v(x,y,\phi )=\left(x+v\cos \phi ,y+v\sin \phi ,\phi \right)\end{array}\right.$$

(19)

For $LSL-type$ paths (other types see Appendix A), start point $A(0,0,{\phi }_{ni})$ left-turn arc of length $t$, Straight segment of length $p$, Left-turn arc of length $q$ reaches $B(d,0,{\phi }_{ng})$. This transformation satisfies the composite operation [22]:

$$L_q(S_p(L_r(0,0,{\phi }_{ni})))=(d,0,{\phi }_{ng})$$

(20)

The resultant coordinate transformation equations are:

$$\left\{\begin{array}{l}-\sin ({\phi }_{ni})+p\cdot \cos ({\phi }_{ni}+t)+\sin ({\phi }_{ni}+t+q)=d\\ \cos ({\phi }_{ni})+p\cdot \sin ({\phi }_{ni}+t)-\cos ({\phi }_{ni}+t+q)=0\\ {\phi }_{ni}+t+q={\phi }_{ng}\end{array}\right.$$

(21)

Solving this system yields the motion parameters:

$$\begin{array}{l}{t}_{LSL}=-{\phi }_{ni}+\arctan {\displaystyle \frac{\cos {\phi }_{ng}-\cos {\phi }_{ni}}{d+\sin {\phi }_{ni}-\sin {\phi }_{ng}}}\left(\mathrm{mod}2\pi \right)\\ p_{LSL}=\sqrt{2+d^2-2\cos ({\phi }_{ni}-{\phi }_{ng})+2d(\sin {\phi }_{ni}-\sin {\phi }_{ng})}\\ q_{LSL}={\phi }_{ng}-\arctan {\displaystyle \frac{\cos {\phi }_{ng}-\cos {\phi }_{ni}}{d+\sin {\phi }_{ni}-\sin {\phi }_{ng}}}\left(\mathrm{mod}2\pi \right)\end{array}$$

(22)

The total path length is given by:

$${\mathcal{L}}_D=t_{LSL}+p_{LSL}+q_{LSL}$$

(23)

The planner computes the lengths of all six potential path types and selects the configuration with the minimum total length. Path planning is then executed according to the determined segment lengths and selected path type. Finally, the approach trajectory is obtained through inverse coordinate transformation.

Upon completion of path planning, the trajectory execution strategy is determined based on altitude differential conditions and the relationship between Dubins path length ${\mathcal{L}}_D$ and glideslope distance ${\mathcal{L}}_{\gamma }$:

(1)

For $H=H_{\mathit{cmd}}$, the UAV traces the Dubins path directly to the terminal point;

(2)

When $H\ne H_{\mathit{cmd}}$ and ${\mathcal{L}}_d>{\mathcal{L}}_{\gamma }$, the aircraft maintains level flight until ${\mathcal{L}}_{remain}={\mathcal{L}}_{\gamma }$, subsequently initiating controlled descent;

(3)

Under $H\ne H_{\mathit{cmd}}$ and ${\mathcal{L}}_d<{\mathcal{L}}_{\gamma }$, continuous descending spirals are executed along the initial arc with dynamic path replanning, transitioning to the updated trajectory when ${\mathcal{L}}_d={\mathcal{L}}_{\gamma }$.

In the approach phase, adaptive Dubins path planning is employed to guide the UAV from its initial position to intercept the glideslope. Under extreme conditions such as sudden gusts, delays or deviations in wind speed estimation may result in heading deviations, altitude fluctuations, or lateral distance offsets at the end of the approach phase. However, the closed-loop control mechanism in the glide slope segment continuously corrects these deviations, enabling convergence before touchdown to ensure landing accuracy. This phased error compensation mechanism effectively reduces reliance on the accuracy of wind speed estimation during the approach phase, significantly enhancing the practical value of the system in complex wind field environments.

This ensures that the UAV reaches the finish line with a specified heading and altitude. Dubins path planning enables flexible approach initiation from arbitrary points/altitudes while minimizing 2D path length, enhancing operational economy and showing significant potential for emergency landings.

3.2. Glide Slope Segment and Flare Phase

During the glide slope phase, the UAV descends at fixed flight path angle γ. Angle selection is critical: excessive γ increases initial sink rate at flare initiation, risking hard landings if descent control is inadequate; insufficient γ extends glide range, demanding larger airspace and reducing economy.

At 15 m relative altitude, the flare phase commences. Here, the aircraft adjusts attitude to reduce vertical velocity, ensuring safe ground contact. Descent rate control maintains sink rate within acceptable limits. The flare trajectory follows an exponential profile that under ideal conditions (zero sink rate at touchdown) satisfies:

$$\stackrel{·}{H}\left(t\right)=-{\displaystyle \frac{1}{\tau }}H\left(t\right)$$

(24)

where $\stackrel{·}{H}\left(t\right)$ represents the sink rate and $\tau$ is the time constant.

Using only this equation would require $t\to \infty$ for touchdown. Practically, we permit sink rate ${\dot{H}}_{cd}$ at main gear contact:

$$\dot{H}\left(t\right)=-{\displaystyle \frac{1}{\tau }}H\left(t\right)+{\dot{H}}_{cd}$$

(25)

This concludes the UAV’s trajectory planning from approach to safe touchdown. Prioritizing descent rate control over position control sacrifices touchdown position accuracy. To ensure landing within runway boundaries with sufficient residual taxiing distance, longitudinal deviations must be constrained through rational touchdown point placement.

4. Controller Design

The overall control system architecture is shown in Figure 7. The aileron loop employs L1 guidance as its outer loop, generating roll angle commands for the inner roll control loop based on yaw angle and cross-track deviation. The rudder loop comprises two components: a drift angle controller and Dutch roll damping.

In the diagram: $\psi$ denotes yaw angle, $PE$ represents lateral path tracking error, ${\varphi }_{cmd}$ indicates roll angle command, $r$ is yaw rate, $\chi$ specifies flight path angle, ${\delta }_{ail}$ gives aileron deflection, ${\delta }_{rud}$ defines rudder deflection.

In practical engineering deployment, the control architecture proposed in this study maintains structural simplicity through modular design, with each core module featuring clear functions and low coupling. The adaptive path planning is implemented based on the geometric analytical method of Dubins paths; the L1 guidance relies on simple trigonometric function calculations and proportional adjustment; the core of the LADRC roll angle control lies in solving the third-order linear extended state observer; and the drift angle coordination logic generates rudder commands only through difference operations. Overall, complex multivariable coupling solving is avoided, resulting in extremely low computational load. The update rate of core control modules can be flexibly adapted to different hardware performances without requiring complex computing power support. Whether for mid-range embedded platforms or low-cost resource-constrained flight controllers, stable operation can be achieved through dynamic task scheduling or lightweight strategies, ensuring a balance between control accuracy and real-time performance, thus demonstrating strong practical engineering application value.

4.1. L1 Guidance

The $L_1$ nonlinear guidance algorithm, proposed by Park [23], selects a virtual target point on the UAV’s predetermined flight path based on relative distance. It calculates the look-ahead distance ($L_1$) to this point and derives the lateral acceleration command ($a_{scmd}$) through a circular transition strategy. This approach features computational simplicity and inherent adaptability through speed-dependent parameter optimization.

Figure 8 illustrates the control logic. A preset target point lies on the desired trajectory at distance $L_1$ from the aircraft, with $\eta$ denoting the angle between the aircraft’s velocity vector and the $L_1$ vector.

Based on the centripetal force equation, the lateral acceleration command for the UAV can be derived as:

$$a_{scmd}=2{\displaystyle \frac{V_a^2}{L_1}}\sin \eta$$

(26)

Depending on the desired path, the lateral acceleration can be expressed in simplified forms. for straight-line trajectories:

$$a_{scmd}=2{\displaystyle \frac{V_a^2}{L_1}}({\displaystyle \frac{d}{L_1}}+{\displaystyle \frac{\dot{d}}{V_a}})$$

(27)

for curved trajectories:

$$a_{scmd}={\displaystyle \frac{2V_a^2{c}^2}{L_1^2}}d+{\displaystyle \frac{2V_{a}c}{L_1}}\dot{d}+{\displaystyle \frac{V_a^2}{R_a}}$$

(28)

$$c=\sqrt{1-{\left({\displaystyle \frac{L_1}{2R_a}}\right)}^2}$$

(29)

where $d$ represents lateral path tracking error. The lateral acceleration $a_{scmd}$ is generated by the aircraft’s lift component during banking maneuvers:

$$a_{scmd}=g\tan \varphi$$

(30)

yielding the roll angle command:

$${\varphi }_{cmd}=\mathrm{atan}({\displaystyle \frac{a_{scmd}}{g}})$$

(31)

This formulation combines variable-gain cross-track error control with lateral velocity regulation. While conceptually similar to traditional PD controllers, the $L_1$ algorithm provides superior performance: smooth convergence under large initial path deviations, inherent robustness to airspeed variations and wind disturbances [24], and adaptive scaling via $V_a$ parameterization that eliminates manual gain scheduling.

4.2. Roll Angle Control

The lateral roll angle inner-loop control of the UAV employs Linear Active Disturbance Rejection Control (LADRC). As an advanced control strategy derived from Active Disturbance Rejection Control (ADRC) theory, LADRC utilizes an Extended State Observer (ESO). Its core principle involves treating unmodeled dynamics and external disturbances collectively as total disturbances, which are estimated by the ESO and actively suppressed through feedback compensation, thus enabling significantly enhanced system robustness and dynamic performance [25]. In scenarios with high model uncertainty and significant disturbances, LADRC demonstrates superior disturbance rejection, simplified parameter tuning, and excellent dynamic response characteristics, serving as an efficient alternative to PID controllers [26]. The block diagram of the roll angle control loop shows in Figure 9.

As detailed in the control diagram, the roll angle controller utilizes two primary inputs: the roll angle error $e_{\varphi }$ and the roll rate $p$. Core control parameters include the roll rate gain $K_p$, roll angle gain $K_{\varphi }$, LESO compensation coefficient $K_{ail}$, and disturbance control input gain $b_{\varphi }$, which collectively enable precise disturbance rejection. The roll angle Linear Extended State Observer (LESO) is formulated as the following state equations:

$$\left\{\begin{array}{l}{e}_{\varphi }=\widehat{\varphi }-\varphi \hfill \\ \begin{array}{l}\dot{\widehat{\varphi }}=\widehat{p}-{\beta }_{1\varphi }{e}_{\varphi }\\ \dot{\widehat{p}}={\widehat{\Delta }}_p-{\beta }_{2\varphi }{e}_{\varphi }+b_{\varphi }{\delta }_{ail}\end{array}\hfill \\ {\dot{\widehat{\Delta }}}_p=-{\beta }_{3\varphi }{e}_{\varphi }\hfill \end{array}\right.$$

(32)

In the above equations, the variables $\widehat{\varphi }$, $\widehat{p}$ and ${\widehat{\Delta }}_p$ represent the estimated values of roll angle $\varphi$, roll angular rate $p$, and the total observed disturbance ${\Delta }_p$ in the roll loop, respectively; while ${\beta }_{1\varphi }$, ${\beta }_{2\varphi }$, ${\beta }_{3\varphi }$ are parameters of the third-order roll-pitch LESO. Given the LESO bandwidth ${\omega }_{\varphi }$ for the roll loop, these parameters satisfy:

$${\beta }_{1\varphi }=3{\omega }_{\varphi },{\beta }_{2\varphi }=3{\omega }_{\varphi }^2,{\beta }_{3\varphi }={\omega }_{\varphi }^3$$

(33)

The disturbance control input gain $b_{\varphi }$ is determined by aerodynamic properties:

$$b_{\varphi }={\displaystyle \frac{1}{2}}{I}_x^{-1}\rho V^2{S}_{ref}c{C}_R^{\delta a}$$

(34)

Integrating these components, the lateral roll channel control law is formulated as:

$${\delta }_{ail}=K_{p}p+K_{\varphi }(\varphi -{\varphi }_{cmd})-K_{ail}•b_{\varphi }^{-1}{\widehat{\Delta }}_p$$

(35)

4.3. Control Strategy

During the landing approach phase, aircraft encountering crosswinds experience lateral deviation from the runway centerline. Subsequent trajectory tracking control forces realignment to the centerline. To maintain stable flight under crosswind conditions, aircraft must implement compensatory techniques—primarily the crab approach or sideslip approach [27]. Building upon these established methods, this paper proposes a Drift Angle Control Strategy. All three strategies employ the aforementioned $L_1$ guidance law and roll angle controller. The schematic illustrating the control strategies is shown in Figure 10.

4.3.1. Sideslip Approach

The sideslip approach achieves runway heading alignment via rudder control. Under crosswind conditions, the resultant sideslip angle generates lateral aerodynamic forces on the UAV. Roll control inputs redirect the lift vector orientation, enabling the horizontal lift component to counteract lateral forces, thereby maintaining stable flight attitude. The advantage of the sideslip approach lies in keeping the aircraft’s nose aligned with the runway direction, thereby reducing the risk of runway excursion after touchdown.

During steady-state flight, the sideslip angle $\beta$ is calculated as:

$$\beta =\arcsin ({\displaystyle \frac{V_w}{V_a}})$$

(36)

where $V_w$ represents Crosswind velocity, $V_a$ enotes true airspeed. The roll angle is expressed as:

$$\varphi =\arctan ({\displaystyle \frac{Y}{L}})$$

(37)

where $Y$ represents lateral force and $L$ denotes lift force. The rudder control law during sideslip approach landing is formulated as:

$${\delta }_{rud}=K_{\psi }\psi +K_{r}r$$

(38)

where ${\delta }_{rud}$ rudder deflection angle, $\psi$ heading error, $r$ yaw rate, $K_{\psi }$, $K_r$ control gains.

4.3.2. Crab Approach

The crab approach leverages directional static stability: during crosswind landings, the aircraft yaws toward the incoming wind direction. This yaw aligns the aircraft nose with the relative airflow, eliminating both sideslip angle and lateral aerodynamic forces. The advantage of the crabbing approach is its ability to automatically correct for crosswinds during flight.

Under steady-state flight conditions, the yaw angle is computed as:

$$\psi =\arcsin ({\displaystyle \frac{V_w}{V_a}})$$

(39)

The sideslip approach requires only integrating yaw rate feedback into the rudder control loop to enhance Dutch roll damping characteristics. Consequently, the rudder control law during sideslip approach landing is expressed as:

$${\delta }_{rud}=K_{r}r$$

(40)

4.3.3. Drift Angle Control Strategy

The crab approach induces significant yaw angles at touchdown, imposing substantial lateral loads on tires during ground roll deceleration, which increases risks of tire damage and runway excursions due to lateral landing inaccuracies. Simultaneously, the sideslip approach generates roll angles that may cause single-wheel touchdown, damaging landing gear structures, while for high-aspect-ratio aircraft, wingtip strikes remain a potential hazard. Under severe crosswinds, both methods risk exceeding aircraft landing performance envelopes, presenting substantial safety hazards.

Therefore, a coordinated method is proposed: using rudder control to regulate the difference between yaw angle and track angle (i.e., drift angle). This synthesized approach enhances crosswind rejection capability while ensuring landing safety. The rudder control law for drift-angle controlled landing is formulated as:

$${\delta }_{rud}=K_p(\psi -\chi )+K_i{\displaystyle {\int }_0^t(\psi -\chi )}dt+K_{r}r$$

(41)

where $\chi$ ground track angle.

4.3.4. Pre-Touchdown Attitude Correction

Prior to touchdown, attitude correction is critical for landing safety. At relative altitudes below 2 m, roll angle commands are subject to magnitude constraints correlated with altitude:

$${\varphi }_{cmd}=\pm \left(2h+1.5\right)$$

(42)

For heading correction, the rudder control law was modified to the sideslip approach formulation with command smoothing filters applied to rudder inputs. This progressively reduces heading deviation, align the aircraft’s longitudinal axis as precisely as possible with the runway centreline.

5. Simulation Analysis

The UAV was initialized at an altitude of 300 m with an airspeed of 54 m/s, heading angle of 0°, and zero roll/sideslip angles. Comparative simulations under a 4 m/s integrated wind field (steady wind, turbulence, and discrete gusts below 100 m altitude) evaluated the proposed drift angle strategy against conventional sideslip and crab approaches. Key state variables during landing are presented in Figure 11.

The statistical results of state quantities of different control strategies in each phase of the landing process are summarized in

Table 3. Statistical Results of State Quantities for Different Control Strategies in Each Phase of the Landing Process.

Strategy	Stage	Roll Angle	Yew Angle	Track Angle	Sideslip Angle	Lateral Deviation
Sideslip	Flare	−5.168	−0.968	0.086	−5.858	2.671
	Correction	−4.999	−1.463	−0.357	−6.602	1.903
	touchdown	−3.516	−1.391	−0.271	−6.659	1.379
Crabbed	Flare	−0.015	−6.895	−0.005	−0.011	−0.037
	Correction	−0.627	−7.106	0.248	0.048	0.839
	touchdown	−0.153	−3.373	0.951	−3.237	1.712
Drift angle	Flare	−3.586	−2.798	0.020	−4.017	1.760
	Correction	−3.642	−3.260	−0.180	−4.520	1.329
	touchdown	−2.949	−2.577	−0.056	−5.209	1.092

.

Simulation results demonstrate that crosswind disturbances induce sideslip, generating lateral forces that produce rolling/yawing moments. Turbulence causes minor state oscillations. During initial descent, the sideslip approach maintains gradual heading convergence with sustained roll/sideslip angles, while the crab approach achieves rapid sideslip convergence but maintains yaw offsets for crosswind compensation.

Below 100 m relative altitude, discrete gusts disrupt system equilibrium, causing fluctuations across all states. During flare, rapid angle-of-attack increases combined with lateral-longitudinal coupling induce persistent oscillations.

To ensure safe touchdown, rudder and aileron-based attitude correction activates below 2 m relative altitude. Prior to correction, the sideslip approach exhibits significant roll angle deviations, while the crab approach demonstrates substantial yaw angle deviations. During correction, both conventional methods undergo abrupt state transitions, presenting safety risks. In contrast, the drift angle strategy initiates correction with smaller initial attitude deviations, requires less control adjustments, and achieves smoother landing transitions.

Pre-touchdown state variables serve as critical performance indicators. Maintaining the established wind field configuration, repeated landing simulations were conducted with varied wind speeds. Statistical distributions of pre-touchdown metrics across wind conditions are presented in Figure 12.

Lateral-directional state variables immediately before touchdown are critically influenced by terminal flare-phase attitude corrections. Although both sideslip and crab approaches exhibit substantial pre-correction deviations, post-correction adjustments nevertheless effectively improve touchdown states. The drift angle control strategy achieves smaller roll angles than sideslip methods and reduced yaw angles compared to crab approaches. Its trajectory tracking outperforms conventional methods under moderate crosswinds while maintaining landing precision in severe crosswind conditions.

In essence, the sideslip approach sacrifices sideslip angle control to achieve runway alignment, whereas the crab approach eliminates sideslip angles at the expense of persistent yaw deviations. Conversely, the drift angle strategy dynamically balances yaw and sideslip parameters.

Comprehensive tests prove the drift angle control strategy enhances UAV landing safety. It integrates trajectory and heading control, achieving smoother attitude changes. Critical touchdown parameters stay within safety limits consistently. Landings remain stable even in crosswind. The strategy also boosts resistance to lateral wind disturbances. This confirms its strong potential for UAV landing systems.

6. Flight Test

To validate that the proposed drift angle control strategy enhances UAV crosswind rejection during landing, this study not only conducts simulation comparisons with conventional crosswind compensation methods but also includes flight test verification. The UAV initiated landing sequencing at 250 m relative altitude with 59 m/s initial airspeed. During the landing phase, the average wind speed is 5 m/s with a maximum wind speed of 7 m/s, and the wind direction is southeast, corresponding to an upwind landing scenario. Figure 13 demonstrates the trajectory tracking performance, with aircraft status during flight-test landing shown in Figure 14.

Flight test recorded zero altitude discrepancy between initial and final approach points, with the aircraft autonomously tracking a predefined Dubins path. Upon landing state activation, flap deflection to 42° increased lift and induced 27.5 m altitude overshoot from the nominal value. Throughout this phase, lateral deviations remained constrained below 20 m while roll angles converged near 15°. Terminal approach measurements indicated 2 m altitude offset, 0.9° yaw angle, and −8.7 m lateral deviation, collectively validating rational adaptive radius planning.

Following approach completion, the aircraft entered the steep glide slope segment where residual roll angles induced peak deviations: 26 m maximum lateral displacement, 11° peak roll angle, and 5.3° maximum yaw angle. All lateral-directional parameters converged upon termination of this segment.

Flight testing confirmed that approach trajectory planning based on the Dubins path algorithm yielded terminal deviations within an acceptable tolerance and satisfied all performance requirements. The implementation of the proposed drift angle control strategy resulted in smoother attitude transitions, improved trajectory tracking, and touchdown parameters bounded within safety limits, contributing to enhanced operational safety margins. These results demonstrate the effectiveness of the proposed strategies in UAV landing control.

7. Conclusions

This study addresses lateral-directional control challenges for fixed-wing UAV landings in complex crosswind environments by proposing an integrated solution combining drift angle control with wind-adaptive trajectory planning. The strategy dynamically coordinates yaw and sideslip angles to resolve inherent conflicts between runway alignment and crosswind compensation in conventional crab and sideslip approaches, enabling smooth attitude transitions and precise trajectory tracking during touchdown.

The wind-adaptive Dubins path planner mitigates initial approach deviations through real-time turning radius optimization based on estimated wind speeds. Simultaneously, the synergistic integration of LADRC-based roll control and $L_1$ guidance significantly enhances system robustness.

Systematic validation via simulations and flight tests demonstrates that this architecture strictly confines touchdown attitude deviations within safety boundaries under challenging crosswind scenarios while maintaining sub-meter positioning accuracy. Compared to existing methods, the drift angle strategy reduces terminal correction efforts and delivers practically viable solutions for high-risk applications such as urban air mobility emergency landings and carrier-based UAV recovery.