Control Algorithms, Kalman Estimation and Near Actual Simulation for UAVs: State of Art Perspective

Abstract

The pervasive use of unmanned aerial vehicles for both commercial and military operations has undergone rapid development in the recent past. When designing unmanned aerial vehicles, it is highly desirable for them to be able to complete their missions with minimal human intervention. Reaching full autonomy requires a reliable and efficient control algorithm that can handle all flight conditions. Due to the confidential nature of UAV design and development, there is a lack of comprehensive literature on the subject. When it comes to the practical application of the ideas presented in the literature, the situation is even bleaker. This research not only examines the flight phases in which controllers and estimators are used for UAVs but also provides an in-depth analysis of the most recent and state-of-the-art control and estimate techniques for UAVs. Research opportunities and challenges specific to UAVs were also examined in this study in an effort to raise the bar for UAV design as a whole and smooth the way for researchers to go from simulation-based research to practical applications. This review paper establishes a foundation that not only investigates the inherent flight dynamics, control architecture, and Kalman estimators utilized in the development of UAVs but also points out the shortcomings that currently exist in research. A number of design considerations for realistic applications and potential studies are presented in the conclusion.

1. Introduction

Unmanned Aerial Vehicles (UAVs), also known as drones, have become one of the most attractive branches of the aviation industry in recent years due to their versatility and capabilities for a variety of applications [1,2,3,4,5]. Typically, the military employs UAVs for tasks like surveillance, reconnaissance, combat/strike missions, intelligence gathering, etc. [6,7]. But as new technologies have emerged, their potential for use in the public sector has also grown. This is especially true in applications like disaster management, wireless sensor networks [8,9], surveillance and monitoring [10,11], search and rescue operations [12], Internet of Things (IoT) [13], remote sensing [14] and commodities transportation [15] etc. As the mission complexity of unmanned aerial vehicles (UAVs) increases, so does the difficulty of their development, particularly with the goal of achieving fully autonomous flight (i.e., with minimal to zero human interaction). Due to the versatility of UAVs, the mission profile will change depending on the mission’s objective. Takeoff and landing are the two critical flight phases of a fully autonomous UAV flight. Furthermore, numerous applications necessitate that UAVs operate autonomously in unpredictable and changing environments, where the UAV must rely on their available sensors to perceive their surroundings and efficiently complete their mission [16]. The three fundamental subsystems necessary for UAV autonomy are guidance, navigation, and control (GNC).

An unmanned aerial vehicle’s (UAV) “pilot” is its guidance system, which plans and makes decisions to carry out missions and achieve objectives. Usually, a reference trajectory for the UAV is constructed using information from both the mission planner and the present state of the vehicle. The current position and speed of the UAV are also taken into account alongside the intended location and environmental factors like wind speed that could affect the UAV’s flight. UAVs utilize a variety of different types of guidance algorithms depending on their mission objectives. These algorithms developed a reference trajectory that improves the UAV’s performance in accordance with user-specified goals, such as reducing travel time to the target and reducing fuel consumption. Common UAV guidance algorithms include waypoint guidance, proportional navigation, and pure pursuit.

Information regarding UAV’s state and its surrounding is provided by the navigation system [17]. The UAV’s location, speed, and orientation are continuously monitored by the navigation system using data collected by onboard sensors. The UAV’s guidance and control systems rely on accurate state estimates provided by the navigation system. In order to estimate the UAV’s current states, the navigation system frequently consists of up of a suite of sensors and algorithms. Common sensors used in UAV navigation systems include GPS, IMU, vision sensor, and pitot tube. These sensor readings are then used by State estimators within the navigation algorithms to ascertain the current states of the UAV. State estimators estimate the UAV’s state using data from sensors in conjunction with the UAV’s mathematical model and this estimate is more accurate than the original sensor data. In order to execute the necessary control actions, rapid and accurate state estimations are required due to the fast dynamics of the UAV. However, some system states are not observable and some measurements are unreliable because of sensor errors [18]. Efficient and precise state estimation is crucial for enabling autonomous UAV flights. Even though it’s crucial for the UAV’s control performance and safety, state estimate has been mainly ignored until now [19].

Unmanned Aerial Vehicles (UAVs) guidance, navigation, and control (GNC) rely heavily on the control algorithm. Actuator inputs required to produce moments and forces in accordance with guidance system commands are handled by the control system, which is comprised of control laws that take into account the current state of the UAV as determined by the navigation system (sensors) and the UAV dynamics. Maintaining stability and controllability of the UAV in the presence of exogenous inputs and wind disturbances is in the hands of the control algorithm. An unmanned aerial vehicle’s trajectory is maintained via a control command calculated from the UAV’s current state and the pre-planned trajectory. The ultimate objective of UAVs is to perform the required missions with minimal human assistance. Fully autonomous control of UAVs is more challenging because it does not require human support during autonomous operation. As a result, problems with flight safety and accidents are more likely to occur in unmanned aircraft compared to planes flown by humans [20]. Creating and implementing sophisticated and reliable control algorithms is crucial for preventing these failures and enhancing autonomy. Numerous control algorithms, from the more basic PID [21] controller to the more complex Neural network and fuzzy logic controllers [22], have been developed and implemented for the autonomous flight of UAVs.

UAVs quickly expanding fleet, and their increasing utility presents a significant challenge to engineers in terms of creating efficient and robust GNC algorithms. However, the development of dependable and robust systems is made possible by technological advances in the aerospace industry [9,23,24,25] and ground vehicles [26,27,28,29,30,31,32,33,34,35,36]. Figure 1 depicts the typical UAV GNC architecture. To identify and address the potential faults and errors in the designed algorithm prior to its practical implementation, model-in-loop, software-in-loop, processor-in-loop, and hardware-in-loop, along with different visualization software, are generally used. With the use of simulation and visualization tools, developers can evaluate the GNC algorithm’s functionality and performance in a variety of scenarios, ensuring its robustness and dependability before committing to costly and time-consuming flight testing. Additionally, these realistic simulation techniques can be utilized to assess the GNC algorithm’s robustness to complex and dynamic situations, assuring that it can deal with unforeseen changes and disruptions [37]. There is a wide variety of unmanned aerial vehicles (UAVs), each with its own set of features and capabilities to meet the desired mission requirement. Fixed-wing, rotary-wing, and Vertical Takeoff and Landing (VTOL) UAVs are the three primary types of unmanned aerial vehicles (UAVs). Like conventional airplanes, fixed-wing UAVs rely on a pair of wings to provide lift, they can carry more payload as compared to their counterparts however, they require a runway for take-off and landing. Alternatively, rotary-wing UAVs mimic the flight characteristics of helicopters by using a rotor to create lift and steer the UAV. Vertical takeoff and landing unmanned aerial vehicles (VTOL UAVs) are a hybrid of fixed-wing and rotary-wing aircraft, combining their respective strengths to provide both efficient forward flight and vertical takeoff and landing [38]. In this work, however, we shall examine only fixed-wing UAVs.

1.1. Objective and Contents

The objective of this paper is to present an analysis of flight control algorithms utilized for the autonomous flight of fixed-wing UAVs, with a focus on landing and takeoff. Also discussed is the estimator used to compensate for inaccurate sensor data and sensor failure. It is clear from the articles consulted and the literary analysis that most UAV studies conducted by researchers have concentrated on one aspect of flight, such as takeoff or landing and the work also lacked the implementation of a state estimator. In addition, the design architecture remained simulation-based without any near-actual simulation work, such as Processor-in-the-loop or Hardware-in-the-loop simulations, which are the basis for the practical implementation of the onboard flight controller. This is an essential component for successfully implementing the control architecture to physical hardware. Due to the aforementioned highlighted observations, it is necessary to present a consolidated systematic review that summarizes and discusses all such essential components for the actual development of UAVs. Here, we provide a comprehensive overview of the simulation methods and flight simulators that have been employed by academics in their efforts to verify the efficacy of flight controllers for fixed-wing UAVs. This paper will assist both the general public and industries in moving from simulation-based work to practical implementation.

1.2. Paper Organization

In Section 2, we provide a review of the published survey papers on UAVs. Section 3.1 and Section 4 respectively cover the control algorithms and state estimators utilized. Section 5 provides information and review on realistic simulation techniques and visualization software used for fixed-wing UAVs. In Section 6, we discuss the existing challenges as well as future work directions. In the final section, Section 7 we sum up this manuscript.

2. Relevant Studies

Flight controller design, motion planning algorithms, traffic surveillance, communication networks, vision-based navigation, and Kalman filtering are the most studied topics when it comes to UAVs. Dadkhah et al. [40] presents an overview of motion planning techniques for unmanned aerial vehicles, focusing on their application for autonomous guidance. The authors describe the main sources of uncertainty and practical techniques to cater to these uncertainties. This review focuses mostly on algorithms that can be utilized for UAV guidance. The author also discusses the main challenges in autonomous UAV guidance. A survey on UAV’s application for traffic management and the ongoing research on UAVs by the different universities is presented in [41]. The author highlighted that in the field of traffic management, unmanned air vehicles outperform other approaches due to their maneuverability and wireless network communication. Moreover, the barriers to UAV development are also discussed. The author also discusses various types of vision sensors and their types of processing.

Ollero et al. [42] presents a survey on various UAV platforms. The survey also presents control architectures, issues faced during the implementation of control algorithms, and computer vision techniques used in UAVs. Perception methods for UAVs were the primary focus of the study. The author also provided a brief overview of recent developments in multi-robot systems. Chen et al. [43] present a survey on the concept of autonomous control and the Autonomous Control Level (ACL) metrics to assess the autonomy of unmanned aerial vehicles. The architecture for autonomous control for UAVs is also discussed. Emami et al. [44] in their paper, conducted a comprehensive review of the design and development of intelligent flight control systems for UAVs, with a special emphasis on neural network-based controllers. The mathematics of neural network-based controllers is laid out in detail, and both the challenges and issues of these systems are also explored. In addition, a clear design guideline for an intelligent control system is also presented. Some of the review papers, along with their areas of research, are shown in

Table 1. Consolidated studies: UAV.

Reference	Research Focus
[45,46,47,48,49,50,51,52]	Control algorithms
[53,54,55,56]	Motion planning techniques algorithms
[57,58,59,60,61]	Applications of UAV
[62,63]	Collision avoidance strategies
[64,65]	Navigation techniques
[66,67]	Guidance and Control algorithms
[68]	Kalman Filtering Techniques
[69]	6G UAV communication
[70]	Open-source Hardware and Software Flight Control Platforms

.

3. UAV Dynamic Modeling and Control Architecture

The mathematical models of UAVs serve as the basis for a number of different GNC algorithms. These models eradicate the need for costly and time-consuming physical testing in order to predict and assess the UAV’s performance in a variety of scenarios. Unmanned aerial vehicles can be represented with either three degrees of freedom (3-DOF) or much more complex and comprehensive six degrees of freedom (6-DOF) models. A 3-degrees-of-freedom model represents the UAV as a point mass that can move in three dimensions (x, y, and z), which is computationally inexpensive but significantly inaccurate. Due to its low computational cost, the 3-DOF model has been widely used by researchers in the design of UAV control systems. Models with three degrees of freedom typically fall into one of two categories: Sach’s equations of motion [71] and Zhao’s equations of motion [72].

However, in the 6DoF model, UAV is considered to be a rigid body having six degrees of freedom: three translations (along the x, y, and z axis) and three rotations (i.e., roll, pitch, and yaw). A 6DoF model is more accurate than a 3DoF model and can simulate a wider range of maneuvers and flight situations since it takes into consideration aerodynamic forces and moments as well as the UAV’s mass and moment of inertia. The 6DoF equation of motions for a UAV under the assumption of a flat earth model, as found in [73,74]. Where U,V and W are the translational velocities in the body axis. The body axis angular velocities are denoted by P, Q and R. Euler angles $\varphi$, $\theta$ and $\psi$ define the UAV’s attitude. $P_E$, $P_N$ and h are the inertial position of the UAV, and m is the mass of the UAV. The moment of inertia about the x, y, and z axes are denoted by $J_X,J_Y$ and $J_Z$, respectively; J represents the moment of the inertia matrix. $J_{XZ}$ represent the product of inertia and g represents the gravitational acceleration.

These 12 coupled first-order differential Eqs accurately describe the motion of a UAV. A fixed-wing UAV has four actuators (throttle, ailerons, elevator, and rudder) to control these 12 states. The throttle (${\delta }_t$) controls the forward acceleration of the UAV, the ailerons (${\delta }_a$) control the bank angle/roll angle, the elevator (${\delta }_e$) controls the pitch angle, and the rudder (${\delta }_r$) controls the yaw of the UAV. Similarly, in the wind axis system, the states are UAV speed ($V_T$), angle of attack ($\alpha$), and slide slip angle ($\beta$). The fixed-wing UAV is an underactuated system since it only has four control inputs and six degrees of freedom.

3.1. Flight Control Algorithms

Unmanned aerial vehicles (UAVs) rely heavily on flight control algorithms to enable precise and autonomous control throughout their flights. Several research examining various flight phases of UAVs using various control strategies have been conducted. As shown in Figure 2, control strategies can be categorized as model-based or model-free. Model-based flight control algorithms are further subdivided into linear and non-linear controllers. Common linear flight controllers include PID, LQR, $H_2$/$H_{\infty }$, and Gain Scheduling. While Non-linear controllers include Feedback Linearization, Model Predictive Control, Backstepping, Sliding Mode Control, and Adaptive Control [49]. Because of their inherent limitations, Linear and Nonlinear flight controllers are being replaced by Learning based/Model free controllers [75], which are more complicated and intelligent. Fuzzy logic and Neural networks are types of Model-free controllers [76]. Both model-based and model-free control strategies have been employed in diverse ways to achieve desired goals [77].

Both linear and non-linear controllers have been utilized in diverse ways by researchers to achieve the desired results. In this regard, Carnes [21] has developed a low-cost flight controller based on a PID controller for the autonomous takeoff and landing of a fixed-wing UAV. Moreover, Hardware-in-Loop simulations are used to validate the efficacy of the designed controller. Similarly, Chen et al. [78] suggested a PID-based longitudinal landing control system for UAVs. The simulations were conducted using MATLAB/Simulink, and the outcomes were good. However, the complete dynamics of UAVs and the impact of wind disturbance were not investigated. Pokswat et al. [79] developed a Gain-Scheduled PID controller for a fixed-wing UAV. To determine the controller gains, an automatic tuning algorithm is used. The gain-scheduled control algorithm for the fixed-wing UAV was tested in a wind tunnel to demonstrate its effective implementation. Jetley et al. [80] designed an LQR controller for the autonomous landing of fixed-wing UAVs. The designed controller gave satisfactory performance up to 25% headwind and 10% crosswind. Similarly, Santoso et al. [81] proposed a Linear Quadratic Gaussian (LQG) controller to optimally control the longitudinal flight channel of a fixed-wing UAV. Simulation results showed that the LQG controller provides satisfactory altitude holding, takeoff, and landing performance. Manjarrez et al. [82] designed the autonomous takeoff and landing control architecture based on feedback linearization and PI controller. In the presence of winds, the performance of the developed controller is satisfactory. Similarly, Lesprier et al. [83] designed an auto-landing $H_{\infty }$ based controller for UAVs.

In regards to non-linear control techniques, Qayyum et al. [84] suggested a model predictive controller (MPC) for the autonomous landing of UAV based on the Laguerre function. A comparative analysis between MPC and PID was also carried out. The results of the simulation demonstrated that the MPC controller outperformed the PID controller. However, wind disturbances and gusts were not catered. Lungu [85] Lungu proposed an auto-landing architecture for a tailless and blended-wing unmanned aerial vehicle subject to external disturbances and sensor measurement errors. The predetermined landing trajectory is tracked using a backstepping-based attitude angle controller, a dynamic inversion-based speed controller, and an adaptive disturbance observer for the estimation of wind disturbances. Similarly, Zhu et al. [86] designed an active disturbance rejection controller (ADRC) for a small fixed-wing UAV’s entire flight regime and compared the simulated results to those of a PID controller. The simulation results demonstrated that ADRC offers superior anti-interference capabilities in comparison to PID. The authors also performed the flight experiments and illustrated that the ADRC controller showed satisfactory performance on stability and tracking accuracy during landing in the presence of external disturbances. However, the ADRC controller was only designed for the longitudinal UAV channel.

The world is now shifting towards model-free/learning-based controllers. In contrast to traditional controllers, model-free controllers don’t need any plant model. Instead, they adapt their behavior based on what they have learned through experience. In this regard, Nho et al. [87] presented a Fuzzy Logic based controller for the autonomous Landing of UAVs. The designed controller was implemented on both the linear longitudinal and 6-DOF non-linear aircraft models. The designed controller was sufficiently robust to provide excellent performance in terms of stability and low steady-state error. Similarly, Kurnaz et al. [22] proposed a Fuzzy Logic based controller for the autonomous cruise and landing of small, fixed-wing UAVs. Autonomous navigation on the pre-defined path was also achieved by fuzzy logic-based systems. However, the effects of wind disturbances were not catered for in this work. Details on the controllers utilized in autonomous UAVs are provided in

Table 2. Control algorithms employed on fixed-wing UAV.

Flight Phase	Control Technique	Reference
Landing	PID	[78,88,89,90,91,92,93,94,95,96]
	Fuzzy Logic	[97]
	Sliding Mode Control	[93,98,99]
	Model Predictive Control	[84,90,100]
	LQR	[80,94,101,102,103]
	Backstepping	[104,105,106]
	Feedback Linearization	[101,105,106,107]
	Adaptive Controller	[108,109]
	ADRC	[86,88]
Take-off	LQR	[103]
	ADRC	[86]
	LQG	[81]
	PID	[95,96]
	Adaptive Controller	[108,109]
Cruise	LQR	[110]
	Feedback Linearization	[111]

, along with an indication of the stages of flight in which these controllers are utilized while

Table 3. Advantages and disadvantages of control algorithms.

Control Algorithms	Advantages	Disadvantages
PID	Easy to implement and tune	Sensitive to noise and disturbances
LQR	Provides optimal control solutions, engineering friendly, guarantees stability margins	Full state feedback is required, requires an accurate model of the system
$H_{\infty }$	Handle uncertainties and disturbances	Require accurate model of the system, computationally expensive, tuning is very time-consuming
Gain Scheduling	Handle non-linear systems effectively, coverage of a wide range of operating conditions and flight envelops is possible	Stability issues if the transition between gains is not smooth, slow and laborious design process
Adaptive Control	Can handle uncertain and time-varying systems, can handle disturbances and unmodeled system dynamics	Computationally expensive, good knowledge of system dynamics is needed, tuning of parameters requires expertise
Backstepping	Excellent tracking performance and disturbance rejection capabilities, can handle under-actuated systems effectively	Computationally expensive, requires an accurate mathematical model of the system
Model Predictive Control	Can handle systems with constraints on inputs and states, can handle multivariable control problems with multiple objectives	Performance depends heavily on the accuracy of the prediction model.
Feedback Linearization	Handle non-linear systems effectively, good tracking performance and disturbance rejection capability	Require accurate mathematical model of the system, computationally expensive
Sliding Mode Control	Can handle uncertainties and disturbances effectively, good tracking performance, disturbance rejection capability, does not require an accurate model of the plant	Chattering effect

presents the advantages and disadvantages of the control algorithm.

4. Exploration of State Estimation Techniques

UAVs rely heavily on state estimators, referred to as filters, to reliably estimate the UAV’s internal states from sensor data. The real-time estimations of position, velocity, orientation, and other critical variables provided by these estimators play a key part in UAV navigation, control, and guidance. The main goal of incorporating a state estimator into the GNC algorithm is to increase system dependability by mitigating noise in sensor readings and accommodating sensor failures. Additionally, state estimators are also utilized for sensor integration. A mathematical model of the plant is used by the state estimator to predict the state based on previous states and the current control inputs and then adjusts its prediction using available measurements to generate a more precise estimate of the plant states. The quality of the sensor measurement data and the accuracy of the mathematical model of the plant determine how accurate the state estimation will be. The Kalman filter (KF) is frequently employed by researchers for accurate state estimates in systems when sensor data are noisy or information about the state cannot be measured directly [68]. R.E. Kalman introduced the Kalman filter concept in his paper [112] on linear filtering. This recursive algorithm predicts the state of the plant and then uses those predictions together with data from the sensors to produce an accurate estimate. It has since been the focus of numerous academic and practical applications. The Kalman filter (KF) has multiple variants, including the extended Kalman Filter (EKF), the Unscented Kalman Filter (UKF), and the Cubature Kalman Filter (CKF). These filters are employed in numerous applications, including control systems, robotics, computer vision, and navigation [113]. In this regard, authors of [21,81,103,110,114] have employed the Kalman filter for state estimation purposes. Yang et al. [115] present an EKF to estimate the complete state of the UAV. The author’s estimation of attitude, velocity, position, airspeed, and horizontal wind speed yields encouraging results. Lie et al. [116] calculated Airspeed, Angle of Attack, and Side Slip Angle using the Extended Kalman filter, with a very low RMSE between the real and estimated data. Similarly, authors of [111,117,118,119,120,121] used EKF for state estimation. Xiaoqian et al. [122] present the design of a Cubature Kalman Filter (CKF) and Unscented Kalman Filter (UKF) based attitude estimation algorithm for a fixed-wing unmanned aerial vehicle. The simulation outcomes showed that CKF was better than UKF in accuracy, robustness, non-linear performance, and estimation of attitude information. Similarly, authors of [123] utilize CKF in UAV’s GNC algorithm.

Similarly, Marina et al. [124] proposed an AHRS based on a UKF using the Fast Optimal Attitude Matrix (FOAM) algorithm as an observer model. UKF gave suitable results for the attitude estimation of a lightweight fixed-wing UAV. In [125] author proposed an AHRS based on the UKF using the three-axis attitude determination (TRAID) algorithm as the observation model. The author also compared the performance of UKF with EKF. The results indicated that the microcontroller’s real-time performance was satisfactory with low computational complexity. Authors of [125] also used UKF in their work. The various state estimators utilized in GNC of fixed-wing UAVs are summarised in

Table 4. FlightGear and X-plane comparison.

Features	FlightGear	X-Plane
Product Price and availability	Open source software (Free)	Paid software (USD 59.99)
Operating System	Linux, Windows, and MAC	Linux, Windows, and MAC
Aircraft Catalogue	Vast selection of user-contributed aircraft models are available for free download	Compared to FlightGear, they have lesser airplanes available by default, a large collection of payware aircraft from independent developers are available
Flight dynamics model	Use JBSim model	Based on blade element theory
Co-simulation	Easy integration with MATLAB/Simulink, batch file generation is required	Easy integration with MATLAB/Simulink
Scenery Quality	High-quality graphics, use of OpenGL rendering engine, less detailed and realistic view than X-Plane, highly customizable	High-quality graphics, use of HDR and PBR rendering for realistic weather and detailed texture, excellent depiction of world and aircraft
Customization	Highly customizable	Less customizable

. The study of the relevant literature review reveals that many GNC algorithms were developed and proposed without utilizing state estimators [22,78,79,80,82,83,84,86,89,90,91,92,93,94,95,96,97,98,99,100,101,102,105,108,109,126], Reviewing the pertinent literature, we find that many GNC algorithms were created and suggested without using state estimators, on the false premise that sensors are always present and free of noise, despite the fact that state estimators are of paramount significance in UAV’s GNC algorithm.

5. Simulation and User Adaptation Techniques

Tuning the gain of a flight controller is a laborious but essential component of its design and development. The fine-tuning of controller gains requires extensive testing. Tuning the controller’s parameters during actual flight is the ideal method, but it is also quite time-consuming and costly. Model-in-loop simulations (MILS), software-in-loop simulations (SILS), processor-in-loop simulations (PILS), and hardware-in-loop simulations (HILS) are near-actual simulation techniques that can help speed up the process and reduce the number of actual flight trails [37]. Furthermore, several visualization software are employed to simulate flight conditions as close to real. The functionality and performance of an algorithm can be tested by simulating a wide range of scenarios, including varying weather and terrain variations.

5.1. Realistic Simulation Techniques

Model-in-Loop Simulation (MILS) is the first step in the design and development of any flight controller. In MILS, a mathematical model of the actual plant which is to be controlled is built in a simulation environment like MATLAB/Simulink. Next, the controller model is constructed in the same simulation environment as the plant model, and its performance (whether or not the controller is effective enough to control the plant as planned) is evaluated. The main goal of MIL simulations is to identify potential issues in the system, optimize system performance, and assess different design options before committing to hardware implementation. MIL simulation offers a low-risk testing environment, enabling engineers and researchers to test and refine system models without placing physical systems at risk. This technique can save time and money by allowing iterative design and testing cycles before the system is built. Nevertheless, MIL simulation does have some limitations and challenges, such as the simulation’s accuracy is dependent on the accuracy of the model, which can be challenging to develop for complex systems and mathematical models not fully represent the actual physical model. A lot of research is based only on Model-in-loop simulation (MILS). MILS was used by the authors of [80,94] to validate the LQR controller’s performance in autonomous UAV landing. Similarly, authors of [22,78,81,82,83,84,89,91,92,93,98,99,100,101,102,104,105,106,107,109,110,111] performed MILS simulations to simulate their controller’s performance. The MILS block diagram is depicted in Figure 3.

After achieving desirable results in MILS, the next step is to transition to Software-in-the-Loop simulation (SILS). Instead of using a controller model, SILS relies on the C-code of the controller block as depicted in Figure 4. The plant model is unaltered (same as MILS) whereas the controller model is emulated in C code for SILS. SIL simulation is typically performed to mimic the behavior of the embedded processor on which the developed algorithm will eventually be deployed. SILS’s main objective is to check whether the designed algorithm is software implementable or not? In this regard, Mathisen et al. [90] used a non-linear MPC controller to achieve a precision deep-stall landing of a fixed-wing UAV, and then validated the controller’s performance with Software-in-the-Loop simulations. Authors of [37,127,128] used SILS to validate their designed algorithms.

Processor-in-the-Loop simulations (PILS) follow SILS. Similar to SILS, PILS involves burning a controller model onto an embedded processor and then running a closed-loop simulation of the simulated plant. Figure 5 depicts the PILS setup, with the controller model code executing on hardware and plant operating in a simulated environment. PIL simulation is an effective tool for system design and development that permits engineers to assess and improve both the software as well as the hardware aspects of a system. When testing whether a processor can reliably run the controller code without errors or delays, PILS simulations prove invaluable. In this regard, You et al. [108] confirmed the controller’s effectiveness during the take-off and landing of a fixed-wing UAV. Likewise, Ulker et al. [129] conducted PILS for a fixed-wing UAV in windy conditions in various flight scenarios, including straight-and-level flight, level climb, and turn. Similarly, authors of [130] used PILS to validate their flight control algorithm and microcontroller’s performance.

Hardware-in-the-loop simulation comes after PILS has been effectively implemented. Incorporating some physical components of the actual plant into the simulation loop is the basic idea behind HILS. Actual plant hardware (servos, actuators) is employed in the simulation loop instead of a mathematical model of the plant to verify the control algorithm [131]. An example of adding a feedback servo motor in the feedback control loop of a plant during HIL simulation is shown in Figure 6. HIL simulation can be utilized to assess and verify the integration of a system’s software and hardware. This can assist in identifying potential issues like different noises in the sensors and synchronization issues between the components. Figure 7 depicts the HILS system’s fundamental block diagram. The autonomous takeoff and landing of a UAV were simulated using Hardware-in-the-Loop by Carnes [21]. Johnson et al. [132] utilized SILS and HILS for the verification of both hardware and software changes before the flight and after gain adjustments for system performance verification. Similarly, authors of [96,97,127,128,133,134] also performed HIL simulations.

5.2. User Adaptation Techniques

In the aviation industry, visualization software is used for training, research, and development. However, actual flight testing is the most accurate method for tuning the controller gain and validating the developed algorithms. Nevertheless, it is costly and UAV flight testing can pose safety risks. In articles [86,88,95,103,108,134] the researchers validated the performance of their controllers through an actual flight of fixed-wing UAVs. However, in the majority of cases, flight testing is not a viable option. Flight testing in a 3D virtual environment is considered a safer and more cost-effective method for evaluating aircraft models and control systems. Flight simulators were initially utilized for entertainment/gaming purposes before being put to use in the realm of virtual flight testing. Matlab/Simulink has been interfaced with numerous flight simulators, including Xplane, FlightGear, and Gazebo-ROS, in order to validate the GNC algorithms of aircraft in a virtual test environment. Co-simulation combines the strengths of MATLAB/Simulink and other flight visualization software to evaluate UAV models, their GNC algorithm, and other algorithms during the design phase. These co-simulation techniques are currently used as a safe substitute for actual flight evaluations. In this regard, Arif et al. [135] proposed a simulation system using X-Plane 10, Python, and MATLAB to assess the controllability of an aircraft. Using the X-Plane simulator, numerous aircraft configurations, system failures, and environmental variables were simulated and tested. Similarly, Ribeiro et al. [136] proposed a test platform to assess the autopilot control system. MATLAB/Simulink was used to test the autopilot whereas X-Plane software was used to visualize the aircraft model. Moreover, the authors also integrated a microcontroller to drive the aircraft actuator/servos. Bulka and Nahon [134] implemented a flight controller for an agile fixed-wing UAV. The controller was initially evaluated with HILS, and the results were then displayed with an X-Plane simulator. Afterward, a successful actual flight test was carried out. Nugroho [137] compared modern and conventional landing controllers for a fixed-wing UAV. X-Plane simulator was utilized to visualize the landing of the UAV based on different controllers. Similarly, Kaviyarasu et al. [37] verified their control and navigation algorithms using SILS and X-Plane simulator.

Like Xplane, FlightGear is another extensively utilized flight simulator. The formation flight of multiple fixed-wing UAVs was presented by Yang et al. [127]. SILS and HILS are used to test the designed guidance and formation algorithms, along with Gazebo-ROS as a 3D visualization software. Priyambodo and Majid [138] build UX-6 fixed-wing flight models using analytical and empirical approaches. In this work, Simulink and FlightGear simulators are used as visualization tools. Prabowo et al. [128] designed an image-based flight control system for a fixed-wing UAV. UAV tracks the target based on the features in the image field captures by the camera. SILS and HILS were performed to validate the control algorithm. In addition, the FlightGear software is utilized to show the UAV camera view. Similarly, Sorton and Hammaker [133] tested the avionics and control system of a small fixed-wing UAV using HILS and FlightGear simulator. Using FlightGear, Zhang et al. [139] validated their designed control and navigation law. Several weather conditions and types of terrain were utilized to test the algorithm. It was found that the difference between the virtual flight and the real flight was negligible. Concluding from the above discussion, the detailed comparison between the most commonly used flight simulators is given in Table 4.

6. Existing Challenges and Way Forward

Unmanned aerial vehicles (UAVs) have been the subject of significant research for more than a century, however, fully autonomous UAV flight because of its wide spectrum and increasing complexities remains a challenging task. A lot of research work has been carried out to theoretically examine UAV performance through 6-DOF/3-DOF simulations. However, they lack the crucial analysis element of determining the algorithm’s hardware implementation’s feasibility and viability for usage. This is crucial since simulation-based work sometimes yields wildly different outcomes than that produced through actual hardware implementation. Software in Loop (SILS), Process in Loop (PILS), and Hardware in Loop (HILS) simulations are examples of near-real simulation techniques that help bridge the performance gap between simulated and actual controller outputs. Additionally, the use of visualization techniques such as Xplane, FlightGear and Gazebo-ROS as alternative means of flight simulators is also non-existent.

The state estimator provides an estimate of the internal state of a plant based on its input and output measurements. They’re crucial in cases where the system is sensor-deficient, the sensor readings are noisy and to cater to sensor failures. Based on a review of the relevant literature, it is apparent that only a small number of studies have consolidated the estimation strategies available to the research community.

There have been several research contributions and control algorithms for autonomous UAV flight, but very little comparative analysis of controller performance for the entire flight regime has been published. This is crucial because each controller has its own pros and cons. While the PID controller is intuitive to build, it lacks robustness; the LQR approach, on the other hand, is an optimum and engineering-friendly controller but requires full-state feedback. Likewise, the feedback-linearization controller can deal with the non-linearities of the system, but at a considerable computational cost. Choosing the right controller for a UAV is a challenging and time-consuming process that requires thorough knowledge of UAV dynamics. This research shows that there is a dearth of performance comparisons in the existing literature, especially when taking into account a similar baseline in which multiple controllers are used to attain the same objective.

Based on the literature reviewed, it is evident that little research has been conducted on the entire flight regime of a UAV (takeoff, mission, and landing). Despite the fact that landing and takeoff are the most crucial phases of any autonomous fixed-wing UAV operation, these phases have been the subject of limited research. Furthermore, examination ‘of’ the impact of wind disturbances and other exogenous inputs is most critical in these flight phases.

Future Work Directions

This section provides a brief summary of recommendations and potential future research directions. The employment of realistic simulation techniques must be emphasized as they bring different components of the architecture within the framework of actual development. This will help validate the performance of the designed algorithm on an actual onboard processor, evaluate propagation delays, and verify the computational aspects of flight controllers [17]. Furthermore, many different flight simulators are available with user-friendly interfaces that can be used for co-simulation. Saving resources and improving flying performance are two benefits of using a flight simulator during the UAV’s design phase. With these tools, researchers can analyze the UAV’s performance in a number of scenarios, and they help to detect potential issues well before the development of the actual prototype. Therefore, researchers engaged in the design and development of UAVs should take advantage of available visualization software. Moreover, state estimators/observers should be employed whenever designing the GNC algorithm for a UAV. As the majority of sensors produce noisy output data in practical scenarios. Therefore, the UAV’s GNC algorithm must be robust and include state estimators to handle sensor failure and noisy measurements. In this regard future work should be focused on the advancement of both model-based and data-driven estimation algorithms.

It is also recommended to investigate multiple control algorithms and compare the efficacy of various controllers. It is important to investigate the computational cost and the benefits of various control algorithms in various flight phases. In addition, the control algorithm must be sufficiently generic and robust to handle the entire flight regime of a UAV. In this regard, the exploration of artificial intelligence and machine learning algorithms will enable aerial vehicles to make complex decisions in a variety of environments. Moreover, the utilization of feedback servos for UAV applications is the most recent trend and presents major advantages over traditional servos. In comparison to traditional servos, feedback servos offer better accuracy, stability, and reliability, making them the superior choice for UAV applications.

To enable autonomous situational awareness in UAVs, more study of emerging technologies like edge computing and drone swarm data communication is required. Autonomous situational awareness is crucial for the safe operation of unmanned aerial vehicles (UAVs), especially in environments with limited or nonexistent network coverage.

Illustration for Further Work

Figure 8 depicts a graphical representation of the probable direction for future research catering to the highlighted aspects. The authors are developing an HILS-based GNC algorithm for a fixed-wing UAV covering its all flight phases in order to address the missing links. The prototype fixed-wing UAV model used has a wing span and chord length of 3 and 1.5 m, respectively. The UAV dynamics model is simulated in MATLAB/SIMULINK. As a first step, a nonlinear total energy-based guidance and control algorithm has been developed for autonomous take-off, loiter, and landing. The block diagram of the current work is shown in Figure 9.

MILS, SILS and PILS were performed to check the efficacy of the designed control algorithm. The performance of the designed controller in MILS and HILS (only feedback servos were incorporated) is shown in Figure 10. The selection of feedback servo motors depends upon the UAV dynamics/type. High-fidelity feedback servo motors were integrated with the microcontroller. Figure 10 shows the feedback servo lag that was found during the aforementioned work, providing further evidence for the necessity of employing SILS, PILS, and HILS before actual implementation/flight. In the next step, we will integrate sensors and check the performance of the designed algorithm using a 3D axis rotation table. After the completion of the work, the complete results along with the findings will be shared with the research community.

7. Conclusions

All flying phases, but especially take-off and landing, are crucial for a fully autonomous UAV. The literature review and cited papers necessitated a centralized platform connecting relevant studies and filling in the missing link. Several crucial aspects need additional consideration, including the absence of a comparative analysis between controllers that can be used in different phases, a review of estimating methodologies, and a near-actual implementation of the suggested control algorithms. This research seeks to address these gaps. This review paper provides a comprehensive analysis of current research on control algorithms, estimate techniques, and simulation methodologies. The limitations are presented by consolidating the details and analyzing them separately. Additionally, we offered future research areas that can be implemented for realistic simulations utilizing SILS, PILS, and HILS before actual hardware implementation. In addition, the use of visualization software during different design phases of an unmanned aerial vehicle (UAV) can expedite the development process by allowing researchers to evaluate the UAV’s behavior under various kinds of environmental effects and diagnose problems before constructing the actual prototype. Insight into future research directions for autonomous flight of fixed-wing UAVs will be provided by this study. This study will serve as a foundation for future research on UAV autonomous flight.