4.1.1 Wind disturbances

Wind disturbances present a significant challenge for UAV operations in agricultural, rural and forested environments, where wind conditions are often variable and unpredictable. For UAVs that conduct precision spraying, mapping or payload transport, stability under wind disturbances is crucial to maintaining accuracy and safety. Unpredictable gusts and varying wind patterns in open agricultural fields or in uneven forest terrain introduce lateral forces and rotational disturbances that can impact UAV flight paths and compromise mission objectives if not properly managed [Reference Geronel, Botez and Bueno87].

In rural and agricultural settings, wind disturbances disrupt the uniform application of pesticides and fertilisers, creating drift effects that hinder even distribution. Studies emphasise that optimising UAV altitude and dynamically adjusting position can mitigate wind-induced drift, ensuring more consistent pesticide application while minimising off-target effects [Reference Hou88, Reference Ibrahim, Sciancalepore and Di Pietro89]. Furthermore, integrating real-time wind data from sensors like anemometers and GPS-enabled drift sensors into UAV control systems enables micro-adjustments to counteract wind forces, enhancing stability and accuracy during operation.

Model Reference Adaptive Control (MRAC) and Sliding Mode Control (SMC) have been effectively adapted for UAVs to maintain stability under variable wind conditions [Reference Wang, Chen and He90]. These methods allow UAVs to dynamically adjust flight parameters, reducing drift and maintaining accurate trajectory tracking even in challenging wind conditions. For more turbulent disturbances, Model Predictive Control (MPC)-based methods offer an advanced approach by predicting and compensating for wind influences on UAV trajectories, although they require higher computational power, limiting real-time applications in some cases [Reference Shi, Liu, Mao, Shen and Li69].

In recent years, machine learning-based control systems have shown promise in managing wind disturbances by learning from historical data to predict and pre-emptively adjust for common wind patterns. For example, reinforcement learning (RL) algorithms have shown improved control performance in high-disturbance environments, enabling UAVs to better handle lateral wind forces and navigate efficiently in forested and agricultural terrains [Reference Phadke, Medrano, Chu, Sekharan and Starek91]. Furthermore, route optimisation algorithms designed specifically for agricultural applications can minimise UAV exposure to crosswinds by planning paths that avoid high-wind areas, thus conserving energy and ensuring efficient spray coverage across fields [Reference Azid, Ali, Kumar, Cirrincione and Fagiolini21].

In conclusion, managing wind disturbances is vital for UAVs operating in agricultural and rural environments. Using adaptive control techniques, integrating advanced sensing technologies and using machine learning, UAVs can achieve stable and precise operations despite wind variability, ultimately enhancing their reliability and performance in precision agriculture.

4.1.2 Mission planning

Multi-rotor UAVs are widely recognised for their manoeuverability and portability, making them highly effective for navigating complex environments. These UAVs can optimise flight paths by minimising turning angles and achieving maximum roll and pitch angles for efficient mission execution. However, determining the optimal shortest path for multi-rotor UAVs is a challenging task, as it requires balancing feasibility, safety, energy consumption and computational efficiency. These complexities arise due to the inherent nonlinear dynamics of UAVs and the constraints imposed by real-world applications. A taxonomy diagram for the path planning algorithms can be seen in Fig. 4.

Figure 4. Taxonomy of path-planning techniques [Reference Basiri92].

A mission planning framework tailored for precision agriculture is presented in Ref. (Reference Zhai, Martínez Ortega, Lucas Martínez and Rodríguez-Molina93). This approach leverages a multi-agent system with an improved federal architecture, incorporating captain, sub-captain and normal agents to enhance coordination and task execution. The model ensures seamless UAV operation by considering boundary conditions that allow UAVs to return to farming areas when necessary. To address mission planning as a multiobjective optimisation problem, the MP-PSOGA algorithm is employed, which integrates particle swarm optimisation and genetic algorithms. The algorithm achieves rapid convergence and effectively determines Pareto optimal solutions. A coalition mechanism further improves UAV cooperation by facilitating dynamic re-planning and mission auction processes, enhancing system robustness and adaptability.

Coverage path planning (CPP) is another critical aspect of mission planning, particularly for applications such as precision agriculture and search-and-rescue operations. In Ref. (Reference Apostolidis, Kapoutsis and Kapoutsis94), an improved version of the grid-based spanning tree coverage (STC) algorithm is proposed, utilising simulated annealing for optimisation. The discrete area partitioning (DARP) algorithm is then applied to allocate grid sections to UAVs based on their operational capabilities. Implemented within a robust software platform, the proposed method was validated through field experiments, demonstrating a significant improvement in mission duration while preserving UAV operational characteristics.

The choice of a UAV platform significantly impacts mission efficiency. A comparative study [Reference Ristorto, D’Incalci, Gallo, Mazzetto and Guglieri95] evaluates three UAV configurations (Q4P Rotor, AGRI-2000, and Q4L-Rotor) for crop monitoring, considering factors such as flight endurance, sensor type and coverage efficiency. The results indicate that for small-scale fields (5 ha), the Q4P Rotor is optimal, whereas the AGRI-2000 is more suitable for larger areas (10–50 ha). Notably, fixed-wing UAVs are deemed unsuitable for low-altitude, high-resolution missions, making multirotor platforms preferable for irregular terrains.

In Ref. (Reference Mukhamediev96), a path optimisation methodology is introduced utilising a modified genetic algorithm, mhCPPmp. This method improves UAV mission planning by incorporating mobile platforms for in-flight replenishment, leading to a 10%-12% improvement in efficiency. The algorithm operates offline, computing the waypoints that UAVs follow during missions. Although effective in reducing flyby costs, limitations include the exclusion of fuelling considerations and real-time path adjustments.

Multi-UAV mission planning is further examined in Ref. (Reference Ramirez-Atencia, Bello-Orgaz and R-Moreno97), where the challenge of coordinating multiple UAVs at different ground control stations (GCS) is addressed. The problem is formulated as a temporal constraint satisfaction problem (CSP) and solved using a hybrid MOGA-CSP approach, which combines a multi-objective genetic algorithm (MOGA) with CSP constraints. This method successfully minimises mission make-up, UAV count and fuel consumption while ensuring feasible task allocations. Experimental results validate the efficiency of the approach, though increased computational resources are required for more complex mission scenarios.

In Ref. (Reference Pradeep, Park and Wei98), a multiphase optimal control framework is developed for agricultural UAV missions, focussing on maximising the coverage area while minimising energy consumption. Using pseudo-spectral numerical methods, the study determines optimal flight speeds for the DJI Phantom 4.0 UAV, highlighting trade-offs between maximum coverage and minimal energy expenditure. Findings suggest that multi-rotor UAVs are best suited for detailed precision agriculture tasks rather than large-scale coverage due to increased power demands at higher speeds.

In Ref. (Reference Skobelev, Budaev, Gusev and Voschuk99), a distributed multi-agent planning prototype is introduced for UAV coordination. The system facilitates real-time adjustments in UAV missions, allowing dynamic task reallocation in response to operational disruptions. The prototype, tested on 3DR IRIS UAVs with Pixhawk PX4 flight controllers, demonstrated robust adaptability for large-scale agricultural monitoring.

Finally, Ref. (Reference Ramirez-Atencia, Rodriguez-Fernandez and Camacho100) presents a decision support system (DSS) designed for multi-criteria multi-phase planning problems (MCMPP). This system ranks mission plans using fuzzy multi-criterion decision-making (MCDM) algorithms, with Fuzzy VIKOR performing best among tested methods. The DSS streamlines operator workload by filtering redundant solutions, enabling efficient decision-making. However, the final selection of the mission remains dependent on the judgement of the operator, highlighting the need for improved interface designs to improve usability.

4.1.3 GPS-Denied navigation and obstacle avoidance

Agricultural UAVs frequently operate in environments where Global Positioning System (GPS) signals are weak or unavailable, such as dense orchards, greenhouses and remote farmlands. These settings present dynamic obstacles, including tree branches, irrigation systems and farm infrastructure, complicating autonomous navigation. This subsection examines recent advancements in UAV navigation under GPS-denied conditions, emphasising adaptive control strategies, obstacle avoidance techniques and alternative localisation methods.

In the absence of GPS, UAVs rely on adaptive control algorithms that compensate for communication constraints and incomplete environmental data. Traditional control methods struggle with limited connectivity to the base station or fluctuating sensor inputs, increasing navigation errors. Recent approaches incorporate predictive modelling, decentralised decision making, feedback mechanisms, probabilistic reasoning and reinforcement learning to enhance autonomy, improve operational safety, reliability and efficiency in agricultural applications [Reference Li101].

Stereo-vision systems play a crucial role in real-time obstacle detection. Stefas et al. [Reference Stefas, Bayram and Isler102] developed a stereovision framework that allows UAVs to autonomously navigate orchards while avoiding tree branches. Similarly, Ref. (Reference Vanegas, Gaston, Roberts and Gonzalez103) proposed a UAV exploration system for unknown GPS-denied environments using a partially observable Markov decision process (POMDP) solver. Their framework, integrated with the TAPIR software and the ABT-POMDP algorithm, dynamically updates an occupancy map via depth sensors, optimising navigation safety and information acquisition.

To address localisation challenges, Ref. (Reference Xu, Li, Kang, Meng and Niu104) introduced a multimodal positioning strategy combining scale-invariant feature transform (SIFT) with iterative closest point (ICP) algorithms. Their approach demonstrated superior accuracy in obstacle avoidance compared to traditional ICP methods while mitigating cumulative drift. However, gravitational acceleration remains a persistent source of error in inertial navigation systems (INS), highlighting the need for further research into vertical drift correction.

Beyond navigation, UAVs are increasingly deployed for precision agricultural tasks such as fruit harvesting. Kumar and Behera [Reference Kumar and Behera105] presented Drone-Bee, an autonomous aerial manipulation system for GPS-denied environments. This platform employs embedded computing for real-time object detection, positioning, and control, achieving high-precision grasping in cluttered agricultural settings.

With the rise of global navigation satellite system (GNSS) jamming, robust interference detection is critical. Pleninger et al. [Reference Pleninger, Topkova and Steiner106] developed a detection framework that uses automatic dependent surveillance-broadcast (ADS-B) data, achieving 98.40% accuracy by cross-referencing the Navigation Accuracy Category for Position (NACp) indicators with Horizontal Dilution of Precision (HDOP) values. Future work may focus on accelerating detection speeds and adapting the system for dual-frequency GNSS receivers.

Environmental landmarks have also been explored for UAV localisation. Costley and Christensen [Reference Costley and Christensen107] demonstrated that the integration of vineyard trellises with inertial measurement unit (IMU) data significantly reduced positional uncertainty, offering a viable alternative to GPS-based navigation.

Efficient path planning is essential for UAVs in complex terrains. Thoma et al. [Reference Thoma, Thomessen, Gardi, Fisher and Braun108] evaluated cost functions for local path planning, finding that minimising worst-case scenarios at each step reduced the probabilities of failure. Future advancements may incorporate machine learning techniques, such as neural networks or bioinspired algorithms, to further refine collision avoidance strategies.

Zhao et al. [Reference Zhao, Xu, Wang, Du and Shen109] proposed a vision-based navigation system that integrates differential GPS with deep learning to avoid dynamic obstacles. Their method employs feature extraction and inverse kinematics to optimise flight paths in real time.

For indoor or low-altitude UAV operations, GPS-denied navigation requires alternative solutions. Youn et al. [Reference Youn, Ko, Choi, Choi, Baek and Myung110] developed an error state Kalman filter (ESKF) integrated with an INS dynamic model and RTAB-Map, achieving high-precision localisation with minimal computational overhead. Meanwhile, studies on low-altitude navigation [Reference Dlouhy, Lev and Kroulik111] demonstrated the reliability of artificial landmarks but highlighted weather-induced limitations.

Convolutional Neural Networks (CNNs) have shown promise in aerial visual localisation. A recent review [Reference Al-Jarrah, Shatnawi, Shurman, Ramadan and Muhaidat112] highlighted hybrid models that combine CNNs with recurrent neural networks (RNNs), significantly improving accuracy in settings denied GPS. However, challenges persist in data set scarcity, computational efficiency and real-time deployment.

Despite progress, key challenges remain:

• • Environmental Interference: Adverse weather conditions (e.g., fog, rain) degrade sensor performance, necessitating multi-sensor fusion.

• • Computational Constraints: Neural networks and fuzzy logic systems require extensive datasets and processing power, limiting the applicability in real time.

• • Simulation-to-Reality Gap: Many algorithms are trained in simulated environments, which requires more realistic virtual datasets.

• • Sensor Synchronisation: Precise alignment of LiDAR, IMU and vision sensors is critical for reliable data fusion [Reference Ahmed, Qiu, Ahmad, Kong and Xin113].

Future research should prioritise robust algorithms, efficient computation, and advanced sensor integration to enhance UAV reliability in GPS-denied agricultural operations.

4.2.1 Liquid tank modelling

Fluid sloshing in tanks during UAV spraying operations presents significant modelling challenges. The movement of liquid within the tank can cause shifts in the centre of mass of the UAV, resulting in instability and erratic behaviour. For small-scale UAVs, even minor changes in mass distribution can substantially impact flight stability, necessitating accurate modelling of sloshing dynamics [Reference Uzun, Bilgic, Çopur and Çoban114].

Traditional modelling approaches, such as single and multi-pendulum models, approximate the behaviour of the liquid by representing sloshing as pendular motion [Reference Kurode, Spurgeon, Bandyopadhyay and Gandhi115–Reference Dong, Qi, Guo and Li117]. These models offer simplicity and computational efficiency, making them suitable for real-time applications where stability is critical [Reference Guerrero-Sánchez, Mercado-Ravell, Lozano and García-Beltrán118].

More detailed approaches employ finite element methods (FEM) to simulate fluid behaviour within a tank, providing a finer resolution of liquid movement and mass shifts [Reference Nichkwade, Harish and Ananthkrishnan119]. FEM-based models allow for high accuracy in capturing sloshing dynamics but can be computationally intensive, limiting their use in real-time UAV operations.

Smoothed particle hydrodynamics (SPH) is another approach that treats the liquid as a collection of particles interacting according to fluid dynamics principles. SPH is particularly useful for simulating free-surface flows and handling complex tank geometries, as demonstrated in applications for aerospace and marine systems [Reference Panferov, Nebylov and Brodsky120].

Additionally, models based on multi-mass spring systems provide a compromise between simplicity and accuracy by representing the liquid as a series of interconnected masses and springs [Reference Feddema, Dohrmann, Parker, Robinett, Romero and Schmitt121, Reference Reyhanoglu and Rubio Hervas122]. This method captures essential sloshing behaviour while maintaining lower computational requirements, making it applicable to UAV applications where computational resources may be constrained.

Recent studies have introduced advanced modelling techniques to address the complexities of sloshing dynamics. Yano and Terashima [Reference Yano and Terashima123] developed a trajectory control design method to suppress residual vibration in liquid container transfer systems without directly measuring vibration. Their approach involves shaping the frequency characteristics of the controller using simple structures like notch and low-pass filters, and determining parameters through optimisation with constraints in both time and frequency domains. This method was applied to a liquid container transfer system, demonstrating effective sloshing suppression during high-speed transfers along three-dimensional paths.

Zang et al. [Reference Zang, Huang and Liang124] proposed two methods to reduce an infinite number of sloshing modes in a moving liquid container. The first method utilises command smoothing to eliminate slosh by targeting the first-mode frequency, while the second combines input shaping and command smoothing. The input shaper reduces slosh for the first mode, and the smoother command suppresses slosh for the third and higher modes. Both methods effectively eliminate transient and residual slosh, with the combined controller achieving a shorter rise time. Simulations and experimental results validated the effectiveness of these methods.

Each modelling technique offers unique benefits depending on the specific application and payload requirements of the UAV. Pendulum-based models and multi-mass spring systems are often preferred for their simplicity in real-time control scenarios, while SPH and FEM are suited for detailed off-line analyses where computational power allows for finer resolution. Advanced methods, such as those proposed by Yano and Terashima [Reference Yano and Terashima123] and Zang et al. [Reference Zang, Huang and Liang124], provide robust frameworks for modelling and suppressing sloshing dynamics, enhancing UAV stability and precision in agricultural spraying applications.

4.2.2 Center of gravity (COG)

An unstable COG presents a significant challenge for UAVs in agricultural applications, especially when carrying liquid payloads for spraying or transporting slung payloads. In these scenarios, the COG is dynamic, often changing as liquid levels decrease or the payload moves, leading to instability and complicating flight control. The assumption that the COG coincides with the UAV’s geometric centre and remains constant is often unrealistic, as practical applications frequently involve continuous changes in payload distribution. These shifts in COG can result in erratic flight behaviour, reduced manoeuvrability and an increased risk of errors in the application of pesticides or the transport of the payload [Reference Johnson and Singhose125].

To address the complexities of COG variability, advanced control strategies that dynamically account for changes in the mass distribution are essential. Lee et al. [Reference Andrew J. and Roger M.127] developed a model for a quad rotor that experiences COG variations due to fluid movement in the payload tank, highlighting the need for robust proportional derivative control (PD) to maintain stability in the presence of liquid-induced COG shifts [Reference Lee, Giri and Son126]. Their research emphasises the importance of control methodologies that respond dynamically to COG changes, particularly in agricultural UAVs, where stability is critical for precision.

Furthermore, estimation techniques such as Kalman-Bucy filters, as explored by Stanley and Goodall [Reference Andrew J. and Roger M.127], provide a framework for predicting and compensating for COG shifts in real time. This approach is highly relevant in applications where COG can shift unpredictably, such as in UAVs that carry liquid payloads with complex motion dynamics [Reference Andrew J. and Roger M.127]. Kalman filtering offers a reliable method for improving control stability and accuracy, which is essential for precision spraying or stable transport of agricultural materials.

Additional approaches, such as the adaptive fault-tolerant control method proposed by Tan et al. [Reference Tan, Shen and Yu128], further enhance UAV stability in response to COGs varying over time. Tan’s research introduces a control methodology that adapts to COG changes and potential faults, providing a robust framework for UAVs tasked with dynamic payloads or long-duration missions where the mass distribution may vary significantly over time [Reference Tan, Shen and Yu128]. Similarly, Patel et al. [Reference Sreelalitha, Jayalakshmi, Lakshmi and Patel129] explored adaptive control methods under variable mass configurations, applicable to UAVs with changing COG due to liquid payload depletion or shifting slung loads, enhancing the responsiveness and flight stability of the UAV [Reference Sreelalitha, Jayalakshmi, Lakshmi and Patel129].

Iqbal et al. [Reference Iqbal, Javed and Shahzad130] studied COG shifts in airships, which, while different from UAVs, share the challenge of significant COG variation due to heavy payloads. Their findings highlight the relevance of adaptive weight distribution strategies, which could be applied to UAVs that carry large or mobile agricultural payloads that alter the balance and control requirements of the vehicle during flight [Reference Iqbal, Javed and Shahzad130].

In practical applications, COG shifts affect the precision of UAV operations, particularly in tasks such as the application of targeted pesticides, where a consistent and steady flight path is essential for uniform coverage. Tumari et al. [Reference Tumari131] designed a cascade structure with a proportional integral derivative (PID) controller to adapt to COG variations, effectively stabilising UAVs under shifting payload conditions. Their study illustrates the potential of structured control approaches to mitigate the instability associated with COG changes,have become an integral part of precision agricultur which is critical for agricultural spraying missions where precision and stability directly impact application efficiency and environmental safety [Reference Tumari131].

In conclusion, addressing COG variability in UAVs for agricultural spraying and slung payload transport is vital to maintaining operational precision and stability. By integrating advanced estimation techniques, such as Kalman filtering, and adaptive control frameworks, UAVs can dynamically adjust to COG changes, ensuring accurate and reliable performance in complex agricultural environments. These developments not only enhance the reliability of UAVs in precision agriculture, but also expand their potential applications in scenarios where payload dynamics play a central role.