3.0 Results

Adaptive algorithms for drone flight control are computational methods that adjust a drone’s control system in real time based on changing environmental conditions and operational parameters. These algorithms allow drones to dynamically respond to factors such as wind, obstacles or varying payloads by modifying flight behaviours to maintain stability and efficiency. Adaptive algorithms for drone flight control under communication constraints and incomplete information are crucial to ensure the safety and efficiency of autonomous flights. Adaptive drone flight control algorithms represent a key element in ensuring efficient and safe drone operations under conditions of limited communication and insufficient information. Such situations require the development of intelligent systems capable of adapting to variable environmental conditions and making decisions based on limited data. Some key principles and technologies that can be utilised in adaptive drone control algorithms in such scenarios (Fig. 1).

Figure 1. Adaptive drone flight control algorithms that can successfully operate in situations of limited communication and insufficient information. Source: compiled by the author.

The use of machine learning techniques such as neural networks or reinforcement learning algorithms allows drones to adapt to changing conditions and make decisions based on experience. These algorithms can learn from existing flight data as well as in real time using sensor information and external data [Reference Kondratenko, Klymenko and Sidenko7]. The development of algorithms that allow drones to make decisions independently based on localised information reduces dependence on a centralised controller and allows the drones to continue performing tasks even if communication with the base station is lost. The use of probabilistic methods, such as Kalman filters or particle filters, allows the drone to estimate the state of the environment and adjust the drone’s actions based on this estimate. This allows drones to adapt to variable conditions and maintain flight stability even with limited information.

Developing reinforcement learning or teacher learning algorithms allows drones to improve their control skills based on experience and sharing information with each other. This also allows drones to adapt to new situations and make effective decisions even with limited communication with the base station or operator. The implementation of redundancy and autonomous decision-making systems allows drones to remain operational and continue to perform missions even in the event of communication failures or incomplete information [Reference Hysa8]. This ensures mission continuity and enhances drone reliability in a communications-constrained environment. In general, the successful operation of drones in situations of limited communication and insufficient information requires a set of adaptive control algorithms capable of adapting to variable conditions and making effective decisions based on limited data. This enables drones to perform their tasks efficiently and ensures their reliability and safety in different usage scenarios.

The development of an autonomous control system with route planning in the MATLAB/Simulink software package began with the development of a system model, which is a sequence of individual blocks combined to provide optimal control. This approach allowed the creation of an autonomous control system capable of adapting to different situations and ensuring optimal fulfilment of tasks (Fig. 2).

Figure 2. General control system that is immune to external interference.

An important aspect of such a control system is the ability to adapt to variable conditions and maintain stable operation even in the presence of unfavourable external influences. This is achieved by applying various techniques and methods such as data filtering, noise compensation, prediction of changes and correction of control signals. Simulation modelling under different conditions such as weather changes, obstacles and other factors have been used to test the system performance and efficiency in different scenarios. Simulations of different weather conditions such as strong wind, rain or fog allowed the evaluation of how the control system reacted to such changes. The results showed how efficiently and reliably the drone adapted to the changing atmospheric conditions and continued to fulfil its tasks. In the case of high winds, the control system automatically adjusted the flight path and controlled the drone’s motors to compensate for wind-induced drift. For rainy weather, the system took into account moisture and water that could get on the drone’s cameras or sensors and took steps to ensure that the electronics were clean and working properly. In the case of fog, the system relied on other sensors and location data such as Global Positioning System (GPS) to provide a safe and accurate navigation solution.

When compared to conventional drone control techniques, the simulation results showed that the adaptive control algorithms performed noticeably better. Important performance indicators, including trajectory tracking precision, flight stability and mission completion rate, demonstrated significant improvements in the evaluated environmental conditions with limited communication and insufficient information. In the presence of high winds and obstructions, for instance, the adaptive algorithms were able to maintain an average trajectory variation of less than 2m, but in similar conditions, the classic control methods showed deviations of more than 5m. Furthermore, the adaptive algorithms’ mission completion percentage was above 90%, whereas the older methods’ rate was just 75%.

During the simulation, obstacles such as buildings, trees or other objects that could be encountered in the drone’s path were introduced. The control system had to correctly detect these obstacles, choose safe avoidance routes, and continue with the mission. Simulation results demonstrated the effectiveness of the route planning and obstacle avoidance algorithms. When obstacles were detected, the control system automatically calculated optimal avoidance routes, considering their size, shape and other characteristics. This allowed the drone to avoid collisions and maintain a safe distance from obstacles. The simulations also considered other factors such as communication limitations, unexpected changes in the environment or the behaviour of other drones in the general flight area. This approach allowed assessing the level of adaptability of the control system and its ability to make appropriate decisions in real time to ensure safe and efficient flight. Setting up the simulations in this way allowed assessing how the control system responded to different scenarios and conditions that could occur during flight. For example, if communication with the operator or other drones was lost, the system had to be able to decide on further actions based on available information and pre-defined parameters.

During the simulations, the research team encountered several key challenges that had to be addressed. Accurately simulating the dynamic environmental conditions, such as changing wind speeds, precipitation, fog and the existence of impediments like buildings, was one of the biggest problems. In order to solve this, the researchers integrated sensor data and created intricate models of atmospheric impacts to give the drone’s control system real-time environmental information. Making sure the control algorithms could adjust to changing circumstances fast enough to prevent instability was another difficulty. In order to strike a balance between responsiveness and stability, the team experimented with various feedback systems and control law tweaking. They also had to optimise the algorithms to perform well on the on-board processors, taking into consideration the drone’s limited computational resources. The researchers created adaptive control techniques that could successfully manage the dynamic and resource-constrained environment simulated in MATLAB/Simulink through an iterative process of simulation, analysis and refining.

The results of the simulations allowed the performance and effectiveness of the control system to be assessed under different conditions and helped identify potential problems or improvements that could be made to the system. This helped to ensure a high level of reliability and performance of the drone systems in real-world environments. Prediction plays a key role in modern drone control, especially in a variable and unpredictable environment [Reference Radzki, Nielsen, Golińska-Dawson, Bocewicz and Banaszak9]. The use of machine learning and data analysis techniques opens the door for drones to adapt to various changes, which significantly improves their efficiency and safety. Prediction allows drones to react quickly to changes in the environment. Machine learning techniques can analyse large amounts of data, such as data from drone sensors and external weather or traffic data, and use it to make predictions about potential changes [Reference Falko, Gogota, Yermolenko and Kadenko10]. For example, drones can predict weather changes to avoid flying in unfavourable conditions, or predict traffic flows to choose the best route. Prediction helps drones adapt to limited information. In environments where complete information about the environment is not always available, data analysis techniques allow drones to fill in information gaps and make assumptions based on available data. For example, drones can predict object trajectories or weather changes based on analysing historical data.

Decentralised algorithms allow drones to make decisions more flexibly and quickly. Instead of relying on a centralised controller, each drone can analyse local information, such as sensor data and the results of its actions, and make decisions based on this information. This allows drones to quickly adapt to changing conditions and respond effectively to unforeseen situations. Decentralised algorithms reduce the risk of the entire system failing due to the failure of a central node. Since each drone makes decisions independently of the others, the failure or loss of communication of one drone does not disrupt the other drones. This increases the reliability and resilience of the system as a whole. Decentralised algorithms allow drones to better adapt to communication constraints. In situations where there is limited or no communication between drones or with the central controller, decentralised algorithms allow drones to continue to perform their tasks based on available local information.

Feedback and trajectory correction play an important role in providing reliable and safe drone control [Reference Yang, Zhou and You11]. This process involves using feedback from sensors on board the drone to continuously monitor its position, motion and environment. Based on this information, the drone then adjusts its flight path in real time to ensure reliable and accurate task execution even under unfavourable conditions. The use of feedback allows the drone to receive continuous updates on its current position and environment. This includes data on altitude, speed, tilt angles and other parameters collected from inertial sensors, GPS and other sensors. By receiving this data, the drone can estimate its position relative to the target trajectory and make necessary adjustments. Feedback allows the drone to respond to changes in the environment and unforeseen situations. For example, if the drone encounters strong winds or a sudden change in weather conditions, feedback will help it adapt to the new conditions and maintain stable flight. This is especially important in the event of a loss of communication with the base station or operator, when the drone must be able to make decisions on its own to ensure a safe return or mission completion. Feedback and trajectory correction ensure the drone’s reliability and stability in a variety of operating conditions. Even in the face of temporary communication interference or environmental changes, the drone retains the ability to accurately control and execute missions, enhancing its effectiveness and safety in a wide range of usage scenarios.

The use of probabilistic methods such as Kalman filters or particle filters provide powerful tools for estimating the state of the environment and correcting drone actions based on this estimation [Reference Zitar, Mohsen, Seghrouchni, Barbaresco and Al-Dmour12]. Probabilistic methods allow estimating the state of the environment, taking into account different sources of uncertainty. Drones equipped with sensors collect data about their position, surrounding objects and flight conditions. However, this data often contains noise and measurement errors. Probabilistic methods can effectively filter out such noise and provide accurate estimates of the environment even in the presence of inaccurate data. Probabilistic methods allow drones to make decisions based on statistical models of the environment. Kalman filters, for example, can predict the future state of an object based on its current position and velocity, while accounting for random changes and noise in the data. This allows drones to make informed decisions and respond to changing conditions with a high degree of accuracy and reliability. Probabilistic methods provide the ability to adapt to different scenarios and flight conditions. Particle filters, for example, use a Monte Carlo (a statistical technique that uses repeated random sampling to obtain numerical results and solve problems that might be difficult to solve analytically) method to efficiently update system state estimates based on new data. This allows drones to adapt to unexpected changes and dynamically changing conditions in real time. The use of probabilistic methods contributes to the safety and reliability of drones. By incorporating uncertainty and probability distributions into environmental estimates, drones can avoid collisions with obstacles, navigate in limited visibility conditions, and effectively ensure safe flight for themselves, surrounding objects and people.

Distributed algorithms allow drones to improve their control skills based on their own experience and sharing information with other drones [Reference Venturini, Mason, Pase, Chiariotti, Testolin, Zanella and Zorzi13]. Distributed learning algorithms facilitate continuous improvement of drone control skills. Reinforcement learning algorithms, such as Q-learning or deep learning methods, allow drones to adapt to different scenarios and environments by systematically learning the results of their actions and adjusting their behavioural strategy based on their experience. This allows drones to effectively adapt to changing environments and improve their performance characteristics over time. Distributed learning algorithms facilitate the sharing of knowledge and experience between drones. By sharing their experiences and learning outcomes, drones can learn from each other and adapt their behaviour based on the collective experience. This collective learning approach allows drones to reach optimal control strategies faster and more efficiently, and improves the overall performance level of the system. Distributed learning algorithms contribute to the flexibility and adaptability of drones in different application scenarios. Learning with a teacher allows drones to quickly learn new knowledge and skills based on feedback from human operators or other data sources. This allows drones to efficiently adapt to new tasks and conditions, which is especially important in dynamic and unpredictable environments. Distributed learning algorithms contribute to the autonomy and self-reliance of drones. Trained drones can make decisions and perform tasks without constant human intervention, increasing their efficiency and real-time performance. This is especially important in situations where there is limited or no communication with the operator.

Implementing redundancy systems allows drones to remain operational even in the event of failures in core systems or loss of communication with the operator. Automatic redundancy systems, such as power or control redundancy, can provide the drone with sufficient time to take action to prevent emergencies or return to a safety base. Autonomous decision-making systems allow drones to autonomously analyse the situation and take the necessary actions to ensure safe flight [Reference Quinones-Grueiro, Biswas, Ahmed, Darrah and Kulkarni14]. For example, autonomous control systems can switch to alternative routes, avoiding dangerous areas or obstacles, or even automatically return to base if malfunctions are detected. The introduction of redundancy and autonomous systems allows drones to maintain mission continuity even in the face of limited connectivity or incomplete information. Drones can continue their missions by relying on localised data and decision-making algorithms, increasing their efficiency and autonomy in real time.

Redundancy and autonomy contribute to increased trust and acceptance of drones as reliable tools in various applications. By enabling autonomous decision-making and safety in case of contingencies, drones are becoming more attractive for commercial use as well as security, medical, search and rescue and other applications. Performance evaluation methods for complex systems are powerful tools to evaluate the performance, safety and functionality of autonomous systems including drones and other unmanned devices [Reference Alam and Oluoch15, Reference Yermolenko, Klekots and Gogota16]. Performance evaluation methods provide a clear definition of the criteria to be considered in the design and use of such systems. This includes both technical parameters and parameters related to the safety of both the device itself and the environment. Such analysis helps engineers and developers to focus on key aspects, ensuring optimal performance and minimising risks. Evaluation methods provide the ability to analyse test results and verify the performance of control algorithms. By comparing actual data against predefined criteria, engineers can evaluate the performance and reliability of the system under different conditions. This identifies possible bottlenecks and problems that may arise during operation. Evaluation methods serve as a tool to identify and analyse weaknesses in the system and suggest ways to improve them. Conducting systematic analyses enables the identification of vulnerabilities and weaknesses and the development and implementation of appropriate remedial measures. This contributes to improving the reliability, safety and efficiency of autonomous systems as a whole. Performance evaluation methods for complex systems represent an integral part of the engineering process, ensuring quality control and continuous improvement of technologies, including the development and use of autonomous systems such as drones.

Unmanned aerial vehicle (UAV) flight control plays an important role in the modern aviation industry by providing the basic principles and methods required for effective flight control [Reference Zuo, Liu, Han and Song17]. This aspect of aviation technology has a significant impact on the development and utilisation of drones, enhancing their applications in various fields. UAV flight control provides basic principles that can be adapted for use with drones. These principles include an understanding of the basic principles of aviation, aerodynamics and flight mechanics, which is fundamental to the development of effective drone control algorithms. Meanwhile, basic control principles and techniques developed for UAVs can be successfully adapted to drone control, given their specific characteristics and capabilities. UAV flight control involves various techniques such as autopiloting, navigation and flight control, which are key components for the development of adaptive drone control algorithms. Autopiloting, for example, allows certain aspects of flight control to be automated, which improves the accuracy and stability of drone flight. Navigation techniques enable the location and direction of drones to be determined, which has a significant impact on their efficiency and safety. Flight control provides the ability to regulate drone motion and behaviour in real time, which is an important aspect in the development of adaptive control algorithms [Reference Babak, Babak, Eremenko, Kuts, Myslovych, Scherbak and Zaporozhets18, Reference Kharlamov, Krivtsun, Korzhyk, Ryabovolyk and Demyanov19]. UAV flight control represents a fundamental element for the development and utilisation of drones. The basic principles and methods, as well as the autopiloting, navigation and flight control techniques provided by this aspect of aviation technology, are the basis for the development of adaptive drone control algorithms that are able to function effectively in a variety of environments and ensure the safety and reliability of autonomous flight.

Significant advantages in terms of environmental effect and energy efficiency were also shown by the application of adaptive control techniques. The adaptive system minimised needless energy expenditure and optimised power usage by enabling the drones to dynamically modify their flight settings in response to changing conditions. For instance, the algorithms could recognise opportunities to glide or coast instead of sustaining constant effort, or they could recognise tailwinds and modify thrust accordingly. Comparing this to conventional control methods resulted in longer flight periods and less battery consumption. Furthermore, by avoiding unnecessary manoeuvres and navigating around obstructions, the adaptive algorithms helped reduce noise, pollution and other environmental disruptions brought on by the drone operations.

Satellite communications and navigation play a key role in the development and control of drones, providing information on location, direction and other parameters required for the effective operation of these systems [Reference Nguyen, Rohacs and Rohacs20, Reference Biliuk, Shareyko, Savchenko, Havrylov, Mardziavko and Fomenko21]. This aspect of technology provides a wide range of capabilities and limitations that need to be considered when developing control algorithms. Satellite communications and navigation provide information on various navigation technologies including GPS, GLONASS and other systems. These technologies provide highly accurate and reliable real-time positioning of drones, which is a key aspect for successfully controlling their flight. Satellite communications and navigation allow the capabilities and limitations of these technologies to be considered when developing drone control algorithms. For example, some areas may have limited access to satellite signal due to geographical features or atmospheric conditions, which can affect the accuracy and reliability of drone navigation. Satellite communications and navigation may be useful for implementing positioning and control algorithms in environments with limited satellite signal availability [Reference Azarov, Kolesnyk and Krupelnitskyi22]. For example, additional sensors or correction algorithms may be included in the drone control system to compensate for loss of satellite communication and provide reliable positioning and control of the drone. Satellite communication and navigation play an important role in the development of drone control algorithms by providing essential location and directional information. By considering the capabilities and limitations of these technologies, researchers and developers can create more efficient and robust control systems that can successfully operate in different environments and ensure the safety and stability of autonomous drone flight.

The use of these materials combined with state-of-the-art methods and technologies will enable the development of more robust and adaptive drone flight control systems that can function effectively in challenging environments, including communication limitations and incomplete information. Adaptive control systems are an important class of systems capable of adapting to changing environmental conditions and task requirements without the need for external intervention [Reference Javaid, Saeed, Qadir, Fahim, He, Song and Bilal23, Reference Han, Shi, Zhang, Korzhyk and Le24]. They play a key role in various fields such as industrial automation, transport, robotics, artificial intelligence and others. The classification of adaptive control systems is based on various criteria such as the ability to change parameters, structure and control strategies in response to external or internal changes (Fig. 3).

Figure 3. Classification of adaptive control systems. Source: compiled by the author.

Systems can be divided into static and dynamic adaptive systems. In static systems, adaptation occurs by changing the control parameters, while in dynamic systems the control structure itself is changed to adapt to changing conditions. Systems can be divided into self-adapting and self-learning systems. In self-tuning systems, control parameters are automatically adjusted based on feedback from the control object. Self-learning systems go further, they are able to change their structure and control strategies based on accumulated experience or training data. In addition, adaptive systems can be categorised according to their principle of operation. This includes feedback systems and feedforward systems. Feedback systems use information about the control error to correct actions, whereas direct-link systems operate without feedback and rely on pre-defined rules or models. The classification of adaptive control systems is an important tool for understanding and systematising different approaches to adaptation in control. It allows the selection of the most appropriate methods depending on the specific task and operating conditions.