MANZOOR Tayyab,XIA Yuan-qing,ALI Yasir,HUSSAIN Khurram
(School of Automation,Beijing Institute of Technology,Beijing 100081,China)
Abstract:Recent progress in the flight control design has inspired sophisticated autonomous aerial vehicles to accomplish various tasks.Among different vertical take-off and landing airborne vehicles,the ducted fan aerial vehicles(DFAVs)are a significantly important type of tail-sitter aircraft because their moving and actuated parts are shielded by annular fuselage called duct,which is crucial for aircraft protection as well as operator safety with the vehicle’s ability to fly in confined,cluttered,and hazardous environments.Moreover,they exhibit the characteristics of both fixed-wing airplanes and helicopters.Furthermore,these vehicles inherit several other qualities such as long perch,stare capability and large payload capacity.This first review paper on DFAVs aims to overview the latest advances while focusing on several aspects related to these vehicles.First,a comprehensive history and a novel categorization are provided with available flying platforms developed worldwide,and their features with pros and cons to other airborne configurations are summarized.Next,different flight control approaches employed to control these aircraft are explained.Finally,concluding remarks,key observations,existing challenges,and new trends in the conducted survey are discussed.
Key words:flight control;flight dynamics;tail-sitter;VTOL unmanned aerial vehicle
Since the last few decades,aircrafts have been increasingly utilized in different fields due to their broader applications in surveillance,target acquisition,remote sensing,path planning,border monitoring,imaging,inspection,and transportation,etc.These flying vehicles either are designed to operate under partial human supervision or are fully autonomous to fulfil their designated task.According to the recent work,UAVs can be categorized mainly into four major types based on their flight mechanism,i.e.,horizontal takeoff and landing,vertical takeoff and landing (VTOL),hybrid,and bio-based drones[1].Many VTOL aerial vehicles with conventional structures have been developed over the years.The open-rotor aerial vehicle (ORAV) such as a helicopter,is one example that has a swash-plate in the vertical direction and can hover but has low endurance in cruise flight.The ducted fan aerial vehicle (DFAV)is a special kind of airborne platform in which the propeller is protected by a duct,and this intuitive configuration has several advantages.It generates more thrust and provides higher efficiency at low speed [2].Moreover,the aerodynamic properties of the duct can be optimized to enhance the thrust production in helicopter mode[3].These airborne platforms also generate considerable lift force being capable of handling the high-speed flight[4].Compared to ORAV,they are more suitable to operate in a cluttered and hazardous environment since their blades are protected by the duct.Inside the duct,the fan acts like a thruster to minimize the rotation of the body of the DFAV by the resultant torque [5].DFAVs also differ from tilt-rotors and tilt-wing aircraft because changing the configuration is only possible by altering the attitude of the complete airframe.Then the actuators stabilize the flight mode in DFAV which can be achieved by changing the configuration through tilting the rotors or the wings among tilt-rotors and tilt-wing platforms[6].Furthermore,these vehicles use vectored thrust,which provides the required feature to control the movement of the DFAV by correcting the thrust utilizing a combination of ailerons,wings,and control surfaces [7].Moreover,this vectored thrust capability makes them end-effector-like devices [8].DFAV with low aspect ratio produces a space for acoustic liners that results in noise reduction [9].Also,thrust generation capability not only avoids the contraction of the airstream but tip loss reduction [10-11].DFAVs have three flight operating modes:hovering,transition,and cruise modes [12].Hovering and cruise flight modes can be referred to as helicopter and airplane modes,respectively.Indeed,inheriting these qualities of fixedwing aircraft and helicopter have been in high demand in aerospace and aviation industries [13].The flight of the DFAV covering the entire flight envelop can be classified into three flight phases:vertical,transition,and horizontal flight[12,14-16].
Over the years,some researchers have analyzed various DFAVs.Nonetheless,not a single paper has made any attempts to give a complete account of the available drones around the globe.Moreover,classifications of various UAVs have been carried out based on different aspects[1,5,13].Nevertheless,none of these works ever specifically categorized DFAVs.Furthermore,there is hardly any existence of a review article on the recent advances of flight control methodologies for DFAVs.
Motivated by the above discussion,we intend to provide the recent advances of DFAVs concentrating on early history,various platform,classification,and flight control techniques in this article.To the best of authors’knowledge,this is the first review article which gives a comprehensive overview of the DFAVs that can act as a baseline in a holistic way for future researches and development in this area.First,a comprehensive history,and a novel classification of the DFAVs are provided.Then,available platforms around the globe from the late 80’s to present are discussed.Afterward,DFAVs advantages and disadvantages are summarized.Thereafter,a concise summary of the various flight control methodologies on DFAVs are explained.Finally,summary,key observations,and concluding remarks are given.
More specifically,the main contributions of this first survey paper on flight control techniques and classification of DFAVs are summarized as follows:
1) In contrast with few existing literature,which covers some aircraft,an overview of the available DFAVs is provided.
2) To overcome the deficiency in the existing works,a novel categorization of the DFAVs is proposed,which divides these airborne platforms into three different classes.
3) To the best of Authors’ knowledge,this is the first article which attempts to describe the existing control techniques for the DFAVs in a concise manner.
The remaining paper is organized as follows.Section 2 presents platform design,which encapsulates the early history,a complete list of recent airborne vehicles with specifications,classification of DFAVs,advantages,and disadvantages of DFAV over other VTOL configurations.In Section 3,an overview of flight dynamics and control methodologies involving linear,nonlinear,optimal,and learning-based control methodologies are provided.Finally,summary,key observations,and concluding remarks on the review are drawn in Section 4.
In general,DFAVs have similar design features.The control principle in most of the configuration is identical,where control surfaces are located in the fan slipstream that is responsible for providing control,forces,and moments.These vehicles consist of a different number of control vanes and flaps.Control surfaces are located at the end of the duct,which is designed to provide lift.Some of these configurations utilize electric motors,and others use the gasoline engine.These flying platforms consist of various landing mechanisms involving either the ring or legs[14].
The main idea behind the transition flight started with the development of tail-sitter aerial vehicles (TSAVs).TSAVs were first produced in the early 1950s,and they were able to change the flight configuration from hover to level flight and vice-versa.Early fixedwing TSAVs such as XFY-1,X-13,and XFV-1 are shown in Fig.1.
Fig.1 Early TSAVs
During the 1950s,many manned prototypes based on ducted fan (DF) configurations were tested,namely Hiller 1031 flying platform,VZ-1E Pawnee,Piacseki 59H Air-Jeep,VZ-8P,Chrysler VZ-6 [20].Among them,VZ-1 was a vital development shown in Fig.2(a),mainly built as direct-lift aircraft involving a contrarotating DF design for providing lift.
Fig.2 Manned DFAV designs
Although the flight experiments didn’t achieve the required objectives as the US army couldn’t see the combat potential in these configurations.However,it inspired the development of many later vehicles like the Hummingbird flying platform shown in Fig.2(b).
The first successful flight with a DF configuration able to transition from hovering to high speed forward flight was conducted by Bell X-22 in 1965 shown in Fig.3(c).Bell X-22 consisted of four DFs’and a fuselage that had a different structure than most recent DFAVs shown in Fig.5,since it had planner tandem wings and a fuselage that remained level during the actual flight.One of the first aircraft used with an annular fuselage was French-built SNECMA Col´eop`etre in the 1950s shown in Fig.3(a).It was capable of transition from hovering flight mode without using a ducted propeller but a single jet engine.Next,the successful flight was performed of a less known DFAV known as PEEK shown in Fig.3(b),developed by General Dynamics in the 1960s.This vehicle could perform all the maneuvers from hovering to cruise flight mode [23].Further improvements in the DFAVs were made in the 1970s onward to meet the growing need.
Fig.3 Early DF configurations for transition flight
There are various standards (i.e.,such as weight,range,endurance,and wingspan,etc.)to classify the aircraft both in civilian and military fields [5,26].In this paper,we categorize the DFAVs into three categories:1)Class I:conventional ducted fan tail-sitter aerial vehicle(DFTSAV);2)Class II:shrouded DFAV;3)Class III:spherical DFAV.These classes can be further divided into various subtypes based on their weights,proposed in[5].The spectrum of the platforms are shown in Fig.4.The DFAVs lie in the miniature aerial vehicle(MAV),micro aerial vehicle,and unmanned aerial vehicle(UAV)type in this spectrum based on their weights.
Fig.4 Spectrum of airborne vehicles[5]
Several programs involving DFAVs have been terminated in the past,but many of them are still underway with many variants of the original configurations that are in use.The characteristics of the shrouded rotary wing(SRW)vehicles are provided in Table 1.
Table 1 Different SRW based DFAV’s characteristics with manufacturers
2.2.1 Class I:conventional DFTSAV
One of the earliest configurations of DFTSAV is iSTAR (intelligence,surveillance,target acquisition,and reconnaissance)shown in Fig.5(t),built by Micro Craft,one of the divisions of Allied Aerospace under the Defense Advanced Research Projects Agency (DARPA)MAV program in many different sizes ranges from 9 inches to 29 inches.The original design was probably inspired by RTV(1975)and RWV(1993)DFTSAV.Afterward,Micro craft acquired a patent in 1999 for iSTAR[23].The main aim was to provide the military an airborne-based mobile platform that can be launched,recovered,and refueled to extend the autonomous flying vehicle.Several flight tests were conducted where the vehicle had undergone high-speed flight,and the aircraft was tilted at an extreme pitch angle.A comprehensive overview of launch,recovery,and refueling can be found in [27],and the brief procedure to estimate the aerodynamic properties were reported in[3,28].
Sandia National Laboratories built a DFTSAV for the US marine corps [29],known as airborne remote operated device (AROD),which provided real-time surveillance to the operator shown in Fig.5(s).AROD could perform hovering and tilted forward flight.A 26 hp gasoline engine powers the vehicle,and a fixedpitch fan capable of generating the lift.However,there were some flaws such as it could only achieve a steady speed of 28 km/h in a 20-degree tilt angle forward flight with zero rates of climb,which is considered too slow from the desired speed of 56 km/h at this tilt angle.Furthermore,there were some stability issues during the actual flight,and the vehicle was too heavy (36 kg) to be deployed as a surveillance platform.
Fig.5 Different DFTSAV structures(a)Royal Melbourne Institute of Technology DFAV[14](b)ASIO B[33](c)Beijing Institute of Technology DFTSAV[34](d)Nanjing University of Aeronautics and Astronautics DFTSAV[35](e)Hovereye[30](f)Spyball[33](g)University of Bologna DFMAV[36](h)Fantail[14](i)SLADe[37](j)Chungnam National University DFTSAV[38](k)Korea Institute of Industrial Technology-DFAV[39](l)South China University of Technology DFAV[40](m)Goldeneye[41](n)Helispy[42](o)GTSpy[4](p)Kestrel[43](q)T-Hawk RQ-16[44](r)Moller Aerobot[45](s)AROD[46](t)iSTAR9[31](u)DFMAV of Universit´e Libre de Bruxelles[47](v)DFTSAV[48](w)ASIO[49](x)BAE IAV2[50](y)Honeywell MAV[51](z)EDF-8[52](I)University of Bologna DF-Aerial Robot[53](ii)A15-5[45](iii)A24-50[45](iv)Harbin Institute of Technology DFAV[54]
Bertin technologies developed Hoevereye for a very short-range combat intelligence shown in Fig.5(e)[30].With a very compact design,that makes it suitable to operate in a congested environment.An electric motor powered the main configuration,and due to this,it performed as a low noise flying platform with a flight endurance and range of up to 10 mins and 1.5 km,respectively.Hovereye can achieve a maximum speed of 48 km/h in the presence of 32 km/h wind speed.While it has some advantages,it possesses an unstable dynamic,and cannot be flown without an adequately designed artificial stabilization system.Aerodynamic moments are susceptible to flight conditions.Relative wind is difficult to measure and estimate.Few control strategies were employed to develop a control system for Hovereye[31-32].
Fantail developed by ST Aerospace(a subsidiary of ST Engineering)can attain a high-speed forward flight at almost horizontal orientation shown in Fig.5(h).Two variants of 3 kg and 5.5 kg have been built,providing several surveillance solutions.
Honeywell’s T-Hawk program is arguably one of the best due to its deployment in the real conflict zone in 2007,shown in Fig.5(q).With two small gasoline engines,T-Hawk has 40 mins flight endurance and a maximum speed of 90 km/h in the presence of 37 km/h wind speed.A concise review of aerodynamics characteristics of T-Hawk RQ-16 can be found in[10].
BAE Systems successfully attained the first untethered flight of its 3rd generation DFAV,known as IAV2,completed a course of 10 waypoints in September 2005.DFAV has a 22-inch diameter fan with a nominal outside diameter of 37 inches,5 ft tall shown in Fig.5(x).The IAV2 is more capable than its predecessor IAV1,allowing for an increase in flight endurance,range and service ceiling.Moreover,there is a decrease in acoustic signature and enhanced payload capacity.It was developed to operate in hover and forward wing flight modes.Wings can be removed to have better maneuverability,small airborne signature,and safer operations in highly populated areas.DF design makes it suitable for company-level operations in close proximity.
GTSpy is a modified form of Helispy,developed by Micro autonomous systems shown in Fig.5(n).Although both DFTSAVs have similar exterior,the changes were made in the onboard electronics of GTSpy.It has a maximum weight of 2.26 kg and is powered by an engine driving a two-bladed fixed propeller.Six vanes are located at the end of the duct shown in Fig.5(o).These vanes move together in such a way that they enable the aircraft to rotate around its centerline.The GTSpy was successfully controlled through a neural-network-based adaptive control technique[4].
Goldeneye-50 (see Fig.5(m)) DFTSAV is the design developed for DARPA’s effort for small clandestine VTOL UAV,which can deliver a payload of 1 to 10 kg into hostile territory.The fan unit is equipped with a mounted rotary engine turning a 4-bladed,high solitary propeller near the duct’s opening.Control of the Goldeneye in the hovering and cruise flight mode can be achieved utilizing two rows of control surfaces in the slipstream at the nozzle’s exit.The first row is a set of four rudder vanes control yaw angle,and dual split tailerons control roll and pitch angle.During the hovering flight,the nozzle is opened to utilize the mass flow and static thrust.In forward flight,the flaps near the nozzle area are closed,which forces the rotor at its maximum advance ratio and the engine at its peak value of power output value.Further details about the functioning of Goldeneye can be found in [41].Wings can also be removed like IAV2 to have similar benefits.Other variants GoldenEye-80 and GoldenEye-100,possess MTOW of 68 kg and 74.4 kg respectively[55].
Kestrel DFTSAV,designed by Aerovironment for Honeywell and DARPA,is a fully autonomous VTOL aircraft capable of transition flight with lift augmented ducted propulsion system shown in Fig.5(p).The two prototypes were developed by Aerovironment,each weighing below 5.5 kg.
Two DFTSAV ASIO-B and Spyball-B were designed by Selex Galileo(based in Italy and the UK)and UTRI for the Italian army.ASIO-B is a completely autonomous VTOL DFTSAV with the primary objective of “hover and stare” with a flight endurance of 0.5 hr,and an MTOW of 8 kg.On the other hand,Spyball-B was developed for over hill intelligence with an MTOW of 2 kg and a similar flight endurance like ASIO-B UAV.Furthermore,both vehicles are equipped with a stable,standardized gimbal and electro-optical/infra-red cameras [65].Another mini electric DFTSAV named ASIO was developed by Selex,which has a slightly similar purpose like Spyball for providing tactical forces“over the hill” and “around the corner” intelligence,giving them real-time situational awareness shown in Fig.5(w).AVID-EDF 8 is an electric DFAV shown in Fig.5(z),which is capable of VTOL and hovering flight that can carry up to 0.45 kg of payload [63].Different DFAV design patents(see among others[66-67])related to Moller’s Aerobot series(A15-5,A15-15,A24-50)of the airborne platform are also significant,which utilize either electric or fueled powered fans.
A few other DFTSAVs were developed in different research institutes around the world.At Royal Melbourne Institute of Technology,a DFTSAV was designed with a cruciform tail configuration and with two rectangular wings shown in Fig.5(a)).Few other institutions involved in the development of the DFAVs are given as follows:Beijing Institute of Technology (see Fig.5(c)),Nanjing University of Aeronautics and Astronautics (Fig.5(d)),Chungnam National University(Fig.5(j)),Stellenbosch University’s DFAV named as Surface Launched Aerial Decoy (SLADe) (Fig.5(i)),Korea Institute of Technology(Fig.5(k)),South China University of Technology (Fig.5(l)),Universit´e Libre de Bruxelles (Fig.5(u)),National University of Defence Technology(Fig.5(v)),Harbin Institute of Technology (Fig.5(iv)),and the Amirkabir University of Technology.Two different DFTSAVs with a fixed-pitch propeller were developed at the University of Bologna,shown in Fig.5(g)and 5(I).
In 2018,an enormously capable DFTSAV was flown at a US army base called V-bat designed by Martin UAV.Using a gasoline engine,long-endurance,high altitude flight capability,and is suitable to operate in confined spaces,unlike most fixed-wing surveillance UAVs (See Fig.6(a)).A similar DFTSAV but with delta wings was developed at KAIST,South Korea [15],shown in Fig.6(b).In recent years,some patents have been published with double ducted fan design(e.g.,[68]).However,this paper only deals with the single DF configurations.
Fig.6 DFTSAV designs
2.2.2 Class II:shrouded DFAV
Entecho developed two airborne platforms based on the SRW configuration.The smaller one,known as Mupod,is shown in Fig.7(a).The bigger hoverpod is a manned version with a seating capacity of up to 3 people.Both aircrafts have a speed of 120 km/h.AESIR originated from Geoff Hatton GFS projects,which is also known as Geoff Flying Saucer,is shown in Fig.7(b).Its main product utilizes the Coanda effect to produce a stable VTOL UAV capable of flying in hovering mode and can navigate inside the buildings.AESIR has produced many prototypes that have several benefits over conventional DFTSAV (i.e.,such as T-Hawk) because AESIR used a fan on top of the vehicle by exploiting the Coanda effect,allows the downstream to the cone,which generates lift for aircraft to fly.It exhibits stable flight because of its design,which floats over a cone of air compared to a column in the traditional DFTSAV[70].Also,the Coanda effect has been investigated in a similar design in [62,71].
In the late 80s,Sikorsky aircraft tested its first flight of DFAV named Cypher for military and civilian applications shown in Fig.7(d).This aircraft consists of coaxial rotors inside a torus-shaped air-frame with a fly-by-wire control,integrated avionics,and onboard computer,including a sensor payload [72].The torusshaped shroud enhanced safety and lift.Although the initial proof of concept flew in 1988,the first real prototype Cypher started operation in 1992,and the untethered flight test was performed in 1993.It has a hover capability and a two to three hours long flight endurance powered by a 53 hp rotary engine.This also led to developing the second prototype called Cypher II,famously known as Dragon Warrior.Some analogies can be made with another DFAV(see Fig.7(c))developed at the Polytechnic University of Turin,where a Wankel rotary engine powers two coaxial rotors.On the converse,several differences with conventional DFTSAV(i.e.,iSTAR)can be observed in size and configuration,which uses a fixed-pitch propeller to generate thrust.In this configuration,the forward flight was realized through vector deflection[60],with a maximum takeoff weight of 900 N and was powered by three strokes engines,each rated at 14 hp.The vehicle had an external diameter of 1.8 m and a rotor radius of 0.5 m.This DFAV employed collective and cyclic blade-pitch variations to control thrust force and moment generated by the rigid rotors.Flight Control System (FCS) design,through hardware-in-the-loop simulation for the complete range of flight operating conditions,was assessed in[60].
Fig.7 Shrouded DF configurations
A doughnut-shaped drone called Cleo drone is a DFMAV that offers safety,ease of use,portability,which is shown in Fig.7(e)[73].Its compact size is also useful to operate in confined spaces[57].Cleo drone is the only MAV included in this class because of the exterior design.
2.2.3 Class III:spherical DFAV
Spherical DF designs such as Fleye and Flyball are shown in Fig.8.
Fig.8 Spherical DFAV
Bertin technologies developed flyball MAV with the primary objectives of intelligence gathering and detecting hazardous and non-conventional weapons.Another recent configuration is the Fleye ducted developed by Aerobot SA with a total weight of 0.45 kg and 10 mins flight endurance.The Fleye is enormously robust and claimed to be one of the safest drones by its inventors [77].It is noteworthy to mention that this paper only deals with single DF configurations.Some work on tandem DFAVs can be found in [2,78-80].
Table 2 summarizes advantages&disadvantages of DFAV over other aerial vehicle configurations.
Table 2 DFAV strength&weaknesses as compared to other VTOL aerial vehicles[75-76]
Among a few other benefits,the ducted propulsion provides better overall efficiency at low speed as compared to conventional rotor configurations.One reason is the reduction in the blade tip losses,and the other one is the effect the duct has on wake produced[81].An issue with the DFAV is associated with a high angle of attack on the vertical flight.During the low-speed tilted flight,airflow into the duct is asymmetric as opposed to symmetric in hover mode.In this situation,the vehicle generates more lift,which in turn creates a higher pitching moment.Momentum drag and ram drag is another challenge.In general,aerodynamic thrust is produced from varying the momentum of the free-stream airflow.This force needed to turn the momentum vector of air mass flow through the fan is called ram drag force[3].At the exit of the duct,control vanes may not perform effectively due to some flight conditions that require knowledge of skew characteristics for different flight ranges to design exit control vanes.Other issues arise due to the interaction of the duct and rotor in various flight conditions when creating the complete rotor model.
A suitable flight dynamical model over the whole flight envelop is essential for the development of an autonomous FCS for a DFAV.Generally,DFAV can be expressed by the Euler angle or its alternative quaternion formulation to avoid gimbal lock singularity.To the best of our knowledge,most researches on DFTSAV employed Newton-Euler representation for rigid body dynamics of DFTSAV[6,12,14,31-32,36,38,48,59,82-101],and on rare occasions Euler-Lagrange representation was utilized[102-105].
Reference frames and rotation matrices are crucial for modeling and control of a DFAV.To explain the dynamics and kinematics of the DFAV,two coordinate systems are considered in Fig.9:inertial frameI={O,xI,yI,zI}and body-fixed frameB={G,xb,yb,zb}attached to the centre of the body.Positionξ=[ξxξyξz]Tand Euler anglesΘ=[φ θ ψ]Tare defined in the inertial frame,whereφ,θ,ψdenote roll,pitch and yaw angle,respectively.A rotation matrix is required to map the attitude of a vector fromBtoIframe,and is given as follows:(see among others[12])
Fig.9 DFAV frames
whereC(·)=cos(·),S(·)=sin(·).The propeller is responsible for producing the thrust forceTwhich in turn creates the translation motion and a torqueΓfor attitude monitoring.Applying the Newton-Euler formulation,the equation of motion of the vehicle is given as follows:
wherem,Fe,d(t),Ω(∗),J,Γe,τg,RBIrepresent airborne vehicle mass,external forces,angular velocities,external disturbances,Skew-symmetric matrix,inertia matrix,external torque,gyroscopic torque,and rotation matrix fromBtoIframe of DFAV,respectively.e3=(0 0 0)Tis the 3rd vector of the canonical basis and also the vector of coordinates in body-fixed frame of the vectordenote the constant coupling matrices.represents thatTmay not apply exactly at the DFAVs centre of mass.In most VTOL aerial vehicles such as DFAVs,this term is relatively small and can be compensated for by the control actions[91,106].The other term is negligibly small(e.g.,as in Hovereye DFAV),i.e.,
whereLdenotes the distance between the plane of the controlled fins and DFAVs centre of mass.Due to various complexities of fluid dynamics and interaction between actuator and the vehicle,it is difficult to modelFeandΓeacting on the DFAV.Existing DFAV models are based on the superposition principle that constitutes the following procedures:1)Decomposition of the vehicle into duct,rudder,and propellers.2) Calculating force/torque by neglecting the aerodynamic effects.3) Adding all the forces/torques to evaluate the resultant force/torque.These aerodynamic forces consist of lift,drag(i.e.,momentum drag,ram drag).In the existing literature,calculating the forces separately is hardly justified[91].A simplified form of(1)is defined as
In practice,torque generation is produced via rudder or flaps in DFAV,whereas in multi-copters through propellers,and in helicopters,it is usually generated by swash-plate mechanism [91].1) The coupling matrixin Eq.(1) is also one of the different aspect,which is negligible for quadrotors/shrouded quadrotors[107].Nonetheless,it can be significantly large among DFAVs and helicopter due to rudder system and swash-plate,respectively.2)Among fixed-wing aerial vehicles,the direction of the thrust vector and air velocity is within few degrees,whereas this difference is quite large among DFAVs[3].3)The flow field distribution among DFAVs is more complex as compared to traditional fixed-wing UAVs,which makes the modeling more difficult [54].The foremost significant challenge in designing a reliable FCS is to choose the framework possessing adequate robustness in the presence of wind gust or other modeling errors.The control of DFAV is not an easy task since the dynamics is highly nonlinear and complex,which can be unstable because the platform operates like both fixed-wing and VTOL UAV.Irrespective,if operating modes are studied separately,it still constitutes many nonlinearities.
According to recent literature,the FCS techniques can be classified into three sub-categories[5,108]:linear flight control,model-based nonlinear,and learningbased flight control approach.The linear controllers can be employed through the linearization by applying relative equilibrium near the operating point.Although the linear controllers are easy to implement and require less computational resources,the control performance of the DFAV like any hybrid UAV deteriorates,when the linear controllers are applied during the transition flight [13,109].On the other hand,model-based approaches can overcome this problem.However,model-based FCS depends on the accuracy of the dynamical model of the vehicle.To solve this issue,learning-based flight control methods are beneficial [110].To utilize the characteristics of different types of control techniques in a single FCS,compound flight control techniques are also significant[111].Linear flight control techniques constitute Proportional-Integral-Derivative(PID)[23,31,35,94-95],H-infinity control [54,112],gain scheduling,and linear quadratic regulator (LQR) along with other LQR based controllers [29,38-39,48,89,104,113-114].Learningbased flight control strategies consist of Fuzzy logic [14,115],Human-based learning,and Neural Network[4,40,116].Model-based nonlinear controllers involve feedback linearization,Model predictive control(MPC) [12,86,88-89,92-93,100,114],adaptive control [35],Backstepping [32,90,117],Adaptive control(model reference adaptive control (MRAC) [118],L1adaptive control[15]),and Sliding Mode Control(SMC)[119-120].Other nonlinear control techniques were proposed in [7,36,53,82-83,91,98-99,105,121].Composite control methodologies were reported in [4,12,32,35,39,48,54,86,88-90,96,100,104,119-120,122-123].In general,a comparison between different flight control schemes is provided in Table 3.
Table 3 Comparison between difference flight control approaches for DFAV[40]
Pflimlin et al.presented a PID based attitude stabilization strategy with experiments for Hovereye DFAV[94].A similar attitude controller was designed for a DFAV based on the PID controller subject to crosswind [31].The nonlinear dynamical model was linearized around the hovering flight equilibrium.Pitch and roll angle had a similar expression due to axissymmetry.Therefore,the pitch angle controller was chosen as the main controller.In the end,simulation and flight experiments revealed similar control performance.A dual-loop control strategy based on the PID controller for a hovering and forward flight control was developed in[95].Also,the PID controller was adopted to achieve stability augmentation control of iSTAR DFAV in hovering and low-speed flight[23].Another dual loop control technique based on PID was investigated for a DFAV in[101].Our review indicates that PID controllers’gains were mostly evaluated through empirical tuning,which gave an excellent tracking response under simple uncertainties.I.K.Peddle et al.introduced the successive loop closure (SLC) approach with linear decoupled estimators employed as feedback signals to the SLADe DFAV.First,the vehicle model was linearized at hover flight trim condition to achieve longitudinal,lateral,directional,heave,and navigational decoupled small perturbation dynamics[37].Then,the SLC controller was employed to regulate these modes of motion of flying vehicle.Simulation results demonstrated that the linear estimations were very closer to the coupled,nonlinear estimation in terms of performance.Moreover,the flight test also showed the ease of tuning associated with the SLC technique.Furthermore,SLC was adopted as a simple autopilot design,which eases the practical testing of vehicles in different stability augmentation modes.Nonetheless,the methods lack optimal solutions of the flight control problem.An H-infinity D-stabilization control strategy for linear uncertainty was presented in[54].By calculating the Linear Matrix Inequality (LMI),the controller was developed.Simulation results showed adequate robustness for the attitude controller.In [112],an H-infinity control technique was utilized to solve the attitude control problem of a linearized DFAV model in hovering flight condition.The control performance was validated through simulations and flight experiments.
Optimal control strategies are also crucial in flight control.The classical controller synthesis approach and LQR controller were employed to AROD DFTSAV in[29].After linearization around the hovering flight condition,the roll behavior was uncoupled from the rest of the dynamics and formulated as a Single-Input Single-Output(SISO)control problem.The pitch and yaw control problem due to Multi-Input Multi-Output(MIMO)in nature is more complicated to compute but can be solved in a similar manner as a SISO control loop.Simulation result and singular value analysis validate the performance of the LQR controller.An experimental study based on an LQR controller for a DFAV was researched in [113].The control design constituted an on-board control (inner-loop) that tracks the reference angular rates,which were obtained from the off-board computer.On the other hand,assuming that the innerloop to be faster than the outer-loop,the angular rates were considered as an input to control position,attitude,and velocity of the airborne platform.The LQR strategy was validated through experiments.A hovering control system for a DFAV was presented in[39].The proportional derivative (PD) controller was employed to stabilize the Euler angles and angular rates.To effectively regulate the position of the DFAV,linear quadratic integration(LQI)was employed with a Kalman Filter(KF).Moreover,an ultrasonic positioning system was used for the 3D target position.Finally,the performance was validated by comparing simulation results and experiments.An attitude controller based on the combination of LQR and internal model principle known as robust servomechanism LQR was designed in[48].The basic idea was to add an integral action to the LQR controller for tracking a constant command.The linearized model obtained around hovering flight was divided into two uncoupled dynamics:1)roll-pitch dynamics;2)yaw dynamics.Lastly,a comparative analysis with the PID controller indicated better tracking performance and robustness.In [104],a pendulum-like oscillation control strategy involving LQR and Particle Swarm Optimization(PSO)was proposed for a DFAV in the presence of external disturbances.A linearized DFAV model chosen at hovering flight conditions derived from wind tunnel test was employed.Afterwards,the compound strategy was designed to stabilize the oscillations,whereby PSO was utilized for parameter optimization.A number of comparative results revealed the effectiveness of the designed composite technique.Jeong et al.developed two optimal control strategies based on LQR,namely as linear quadratic tracker(LQT)and Linear Quadratic tracker with integrator (LQTI) under hover,transition,and cruise flight operating modes (see Fig.10) [38].First,the LQT was designed by augmenting the system with a tracking error term.Then control input was determined by Riccati equation.However,LQT couldn’t mitigate the steady-state error,and for that,LQTI was replaced by adding an integral term of the tracking state into the system state.First,a comparative analysis between LQT and LQTI was given for hovering flight then a dual loop strategy based on the PID for trajectory tracking and LQTI for attitude control in the presence of Dryden wind turbulence was provided for operating modes.For capturing the flight from ascending to landing,a set of controllers usingL1adaptive control for horizontal,vertical and transition flight regimes were presented in [15].Both simulations and flight experiments validated the control performance of the developed framework.Two control techniques,namely linear controller and MPC,were utilized for control verification of a DFAV [92].At first,the nonlinear model was linearized near its equilibrium point.Then longitudinal and lateral disturbances were applied for getting the closed-loop response of both controllers.Simulation results conclusively indicated that MPC performed better in terms of tracking and robustness.A similar comparison between linear MPC and adaptive MPC was proposed in [93].A near hover attitude control strategy for a DFAV built upon online parameter estimation for handling parametric uncertainties,and an adaptive gain scheduling algorithm for adequate tracking performance was presented in [122].The efficiency of the designed adaptive technique was verified by simulations and flight testing.A constrained receding horizon control for Caltech DF model was developed in[114].Its primary purpose was to stabilize the DF at one operating point and to exhibit disturbance rejection characteristics.Finally,gain scheduled LQR was employed to validate the performance of the receding horizon approach.Also,various problems related to non-zero computation times,selection of horizon length,and terminal cost were investigated.Another study on the receding horizon approach for Caltech DF model was reported in[88].The designed strategy was based on receding horizon control and Control Lyapunov function (CLF).In the beginning,a CLF was determined,and then by selecting the CLF as the terminal cost,stability was ensured.By choosing a long horizon length and with a simple CLF,good control performance was guaranteed.On the other hand,if the long horizon length is not possible because of computational cost,then stability can be proved with the quasi-linear Parameter Varying(LPV)method used to evaluate CLF.Numerical simulations were provided for corroborating the effectiveness of both methodologies.Moreover,a comparative analysis based on various methodologies was researched in [89] on Caltech DF model.These techniques are as follows:MPC without CLF,Jacobian linearization with LQR,Frozen Riccati equation,Global linearization with LQR using LMI,Quasi-LPV,CLF from LPV with Sontag’s formula,CLF from LPV with MPC(different time horizons),optimal control.Simulation results were executed for a certain initial condition that incorporated the nonlinear nature of the control problem.Finally,it was shown that the composite technique based on CLF and MPC provides better control performance as compared to other control strategies.
Fig.10 DFAV flight modes(hover,transition and cruise)
T.Manzoor et al.presented a trajectory tracking control scheme based on offset-free tracking MPC for a DFAV [12].Initially,the vehicle was linearized at different airspeeds denoting hover,transition,and cruise flight modes.The Dryden wind turbulence model was added to the translational and angular velocity,along with sensor delays and parametric uncertainties in the dynamical model.LQTI controller with KF was employed to estimate the external disturbances.Finally,a comparative analysis was provided for a fully autonomous flight covering the entire flight envelope of the DFAV.The fully autonomous flight of DFAV can be divided into three flight phases,vertical flight,transition flight,and horizontal flight.1) Vertical flight involves Ascend,Descend,and Hover (ADH),Pirouette,Low Speed Tilted Flight(LSTF);2)Transition flight constitutes of Level to Hover(LtoH)flight and Hover to Level(HtoL)flight;3)Horizontal flight contains Straight and Level Flight(SLF),bank,turn.During the ADH flight,the roll angle remains at zero with north and east position at 0 m.While altitude changes during different time intervals.In the vertical orientation,the DFAV initiates a flight motion known as LSTF,which is further classified into forward-backward and left-right motion.The pirouette motion is needed when the aircraft changes it facing direction,which exactly works like a roll motion in DFAV.The vehicle hover at certain altitude and aileron controller is given a reference signal to stabilize the roll angle.During this phase,DFAV north and east position remain at zero.For HtoL transition flight,pitch angle should change from 90°to 0°.On the other hand,the pitch angle should vary from 0°to 90°during the LtoH transition flight.Horizontal flight is configured for the aircraft to fly in high speed forward flight.The schematic diagram of complete flight motions is shown in Fig.11.Different control surfaces play a key role in controlling the vehicle during these flights.Throttle plays the dominant role during ADH and SF.Pirouette,and bank motion are controlled by aileron.Elevator contributes as the main control during LSTF,LtoH,HtoL,and Climb.During the turn,rudder is the primary controller in the DFAV.Adaptive MPC and Robust MPC (RMPC) techniques based on nonlinear prediction and trajectory linearization to solve the trajectory tracking and attitude control problem of DFAV while having non-minimum phase behavior were presented in [86].The control algorithms were validated in the presence of input delays,actuator deflection,etc.In the adaptive MPC framework,two Disturbance Observers(DOB)were adopted as oppose to one DOB in robust MPC design.Moreover,the Widrow-Hoff algorithm was substituted to reduce the computational time in the adaptive MPC approach.Finally,simulation results revealed that adaptive MPC has a better control performance than robust MPC scheme.Nevertheless,adaptive MPC incorporated two DOBs and Widrow-Hoff algorithm,which results in design complexity.A disturbance rejection MPC based compound control scheme to address the position and attitude control problem of the DFAV with non-minimum phase behavior was presented in [100].Recursive feasibility was ensured,and the closed loop stability was proven with ISS paradigm.Finally,a comparison with RMPC technique revealed better control performance.Zhao et al.presented an attitude control strategy for a DFMAV’s decoupled dynamical model based on PID and adaptive control[35].The nonlinear DFAV was linearized at trimmed points.The simulation results and real-time flight experiments demonstrated better tracking response and improved stability margin for the designed approach.A layered state-feedbackcontrol technique for a DFAV was presented in [103].Initially,the three virtual controllers(forward pitch motion,forward roll motion,and yaw motion)were developed based on linearization and model decomposition.Then the virtual controllers’outputs were reconfigured.In the end,the DFAV was stabilized by adjusting the angular velocities of both propellers and cylinder.A hierarchical strategy for the position control of DFAV was proposed in [117],which used the attitude and thrust as the control variable to stabilize the DFAV’s position and then employed the backstepping approach to evaluate the torque-input capable of stabilizing the desired attitude.A control allocation technique was proposed for a DFAV to enhance the accuracy of the generation mechanism that can simplify the feedback control approaches based on vectored thrust approximation [6].For this case,a simplified dynamical form of DFAV in the body-fixed frame is given as
Fig.11 DFAV flight from ascending to landing
The schematic diagram of control structure of DFAV is provided in Fig.12,whereα,vare angle of attack,and control vector,respectively.
Fig.12 Control structure for DFAV based on vectored thrust approximations[6]
M.Xu et al.proposed a compound control law by unifying prediction control technique and adaptive control scheme for attitude control problem of DFAV[123].The simulation results demonstrated acceptable robustness of the presented strategy.Also,an adaptive decoupling control law based on MRAC controller was presented to solve the problem of inertia coupling between the pitch and roll control in [118].The simulation results revealed good control performance.A prioritized control allocation technique which obtains the admissible control of a DFAV with redundant control surfaces [124].The simulation and real-time experiments showed that the designed framework can return admissible control for a large range of desired moments as opposed to pseudo-inverse scheme.A control technique based on adaptive backstepping combined with a filter was presented for a fully dynamic image-based servo control for a DFAV [125].The designed strategy was validated through simulations and experiments.
Aruneshwaran et al.presented a neural adaptive control technique for DFTSAV performing bop-up maneuver [96].A decoupled nonlinear DFAV model was used,and individual controllers were developed for Euler angles with a separate controller for altitude tracking.Simulation results revealed the effectiveness of the designed framework.Another approach that can cover the full flight envelope without gain scheduling is Nonlinear Dynamic Inversion(NDI).Also,the NDI control strategy was employed on the DFMAV in the lateral motion[59].In[97],the nonlinear inversion was incorporated with full state feedback with a wind tunnel test.This approach was based on the outer-loop regulator design for linear and nonlinear matrix for making the entire system robust.A neural network adaptive controller was employed to estimate the modeling error in a DFAV utilized for feedback linearization through NDI[4].The dynamics of GTSpy DFAV is defined using quaternion vector and are given as follows:
whereQ,Mare quaternion vector and total moment of DFAV.Pseudo Control Hedging forced the vehicle to adapt to aggressive maneuvers,particularly in actuator saturation.Simulation and flight experiments revealed that the controller performed better in a near-hover regime than the more aggressive flight maneuvers.An SMC technique with a scheduled observer was employed on a DFAV to control as frequency domain synthesis strategy[119].Through comparison with NDI,it was observed that NDI had a better stability margin.For a DFAV,a saturated estimator based on equivalent output injection Sliding Mode-DOB and SMC controller was designed to overcome the disturbances in translational and rotational dynamics,respectively [120].Pflimlin et al.proposed a position control scheme based on adaptive Backstepping for a DFAV in the presence of crosswind/wind gust[32,90].Backstepping was used to stabilize the DFAV’s position while unknown aerodynamics effects were considered as unknown perturbations,estimated online by the controller.A fuzzy logic controller for the autonomous flight was researched in [14,115].Fuzzy logic controllers can intelligently regulate systems like humans.Avanzini et al.proposed a full-envelop FCS utilizing a structured-singular-value technique involving two multivariable linear robust controllers [60].Cheng et al.introduced a neural network controller to achieve a steady transition from hover to high-speed flight.The performance was validated with both simulation and real-time experiments [40].Also,a similar control framework was presented to address the hover-to-transition cruise flight control problem of a DFAV in[116].
A standard Lyapunov-based control technique was presented for a class of aircraft such as DFAV to stabilize the desired trajectories in velocity,Euler angle,position,and thrust direction for mitigating modeling errors and robustly overcome the effect of wind gust[106].Adaptive position tracking for a DFAV was designed for a set of bounded disturbances [7].Two flight control laws that ensured the following conditions:1) For any initial conditions,the position-tracking task is satisfied.2) For a set of initial conditions on the control gains,the tracking objective is satisfied.In the end,simulations conclusively demonstrated the effectiveness of the designed approach.Nonlinear robust control for a DFMAV to track desired vertical,lateral,longitudinal,and yaw attitude subject to parametric uncertainties was presented in [83].Moreover,an experimental study based on the nonlinear control law applied to the DFMAV for tracking arbitrary longitudinal,lateral,heading,and vertical references was presented in [36].This cascaded control system was realized by implementing two different time scales under the nested saturation functions,shown in Fig.13.The advantage of using such a mechanism is to enable the inner loop(attitude control)to have the role of faster dynamics and longitudinal and lateral loop of the slower dynamics.An experimental study for solving the output feedback control stabilization of DFAV was presented in[126].Naldi et al.presented a nonlinear robust control law used for take-off and landing of DFMAV under model uncertainties[98].The free flight dynamics of DFAV is given as
Fig.13 Cascaded control structure with inner and outer loop[36]
whereux,uyare lateral and vertical control input,respectively.ax,ayrepresent first order aerodynamic coefficients.Kgis a constant parameter.By representingPH1,PH2are the two constants which can contact the horizontal ground surface.Consider the DFAV interacts with the horizontal surface in Fig.14(a).The gravity force is given byFg=Kfuxuy,whereKfis also a constant coefficient.Consider the point of application of aforementioned force w.r.t.PH1,the dynamical model of the DFAV interacts with a horizontal surface can be written as[98]
Marco et al.presented a robust control technique based on the path following application,that was verified through simulations involving DFMAV to dock and undock from a vertical surface[82].The primary challenge of performing this control task is to model the DFAV accurately.The free flight dynamics of the DFAV will remain the same(5),assuming that ram drag forces are negligibly small.To model the vertical interaction in which the DFAV slides along a vertical surfacePV1,shown in Fig.14(b).Consider the case in which DFAV impacts at.Representingz=ξz-lvsin(θ+)the position ofPV1along thezIaxis.The generalized forcesFθ(ux,uy)andFz(ξz,,θ,,ux,uy)acting on the DFAV w.r.t.coordinates and are given as follows:
Fig.14 DFAV interacting with environment
whereλvrepresents the viscous friction of the vertical surface,andandare the angles betweenxbandlvand,respectively,dv.It was observed that matrixG(θ)is invertible for anyθsatisfying
The lagrangian function of the system is given by
and it is governed by the lagrangian equation
which yields
and in terms of new inputsF1andF2.The dynamics of the DFAV(11)is given by
To perform mid-air operations,control of a DFAV able to interact with the vertical environment was reported in[53].The controller design was based on hybrid force and position feedback control technique that relied upon feedback linearization strategies.Similarly,the vehicle was modelled as in(14),shown in Fig.14(c).There are differences in terms of selection of control input which wereFandT.The angleθcharacterized asθ <-γg,and the preliminary choice of control inputs
The resultant control inputs are similar in nature as(13),and are given as follows:
Cichella et al.utilized the nonlinear control framework for a 3D path following control of a DFAV on SO(3)[99].Some perspective work related to DFAV equipped with a robotic arm interacting with the environment can be found in [105,121].The robotic arm employed in the experiment consisted of a parallel Delta configuration.It is important to mention that the manipulator is a generic device with 3 DOF.Consider the dynamics of the approximated model by assuming that mass of the manipulatormris negligibly small compared to the end effector.
whereis the position of the base of the manipulator.Our review also found that the existing work on single DFAVs deals with lightweight manipulators.The concept of aerial manipulators is further explained using medium weight manipulators for tandem DFAVs [80].As earlier explained,we only discussed the single DF configurations in this review.A comprehensive summary of the flight control schemes is given in Table 5.
A comprehensive survey of the DFAVs has been provided in this manuscript.A short but concise early history of the DFAV followed by classification of DF configurations,various platforms around the world,and an overview of the different flight control strategies implemented on these airborne vehicles and their respective dynamics have been discussed.Various key observations of the review have been concluded as follows:1) The DFAVs have been investigated in the industry since the 1950s.The idea of transition flight emerged with the development of TSAVs (i.e.,such as XFY-1,X-13,and XFV-1 shown in Fig.1).The initial DF configurations were developed for the military,but the flight test of several different manned prototypes (i.e.,such as Hiller VZ-1 shown in Fig.2(a)) were unsuccessful.For transition flight,early DFAVs were tested during the late 1950s (i.e.,Col´eop`etre,PEEK,X-22 shown in Fig.3).However,most of them were unsuccessful due to several reasons.2)It was observed that there are several procedures (i.e.,such as weight,endurance,wingspan and weight,range and weight)that exist to classify these aerial vehicles.In this paper,the single DF configurations have been classified into three main categories:Class I (conventional DFTSAV),Class II (Shrouded DFAV),Class III (Spherical DFAV).According to the weights of the vehicle,these classes can be further divided into sub-classes.The different models are shown in Figs.5-8.3) There are various advantages and disadvantages of using DFAVs.On the bright side,for operator safety,perch and stare capability,ability to operate in confined spaces,superior payload capacity,make them more suitable as compared to other flying vehicles.However,DFAVs have complex aerodynamic model,so it is cumbersome to construct a reliable FCS.Similar to other kinds of VTOL UAVs,the DFAVs consume more power during vertical flight,which discharges the batteries quickly and therefore reduces the flight endurance of these airborne vehicles.So,low flight endurance is another current issue among DFAVs,which has been partially addressed in some configurations(i.e.,such as V-bat,and Demipod,etc.).In short,to carry a heavy payload in a confined space with an adequate stare capability,DFAV is a better choice.4) The existing work on modeling these airborne vehicles have also been one of the challenging parts.Although enormous researches were conducted,but mostly they adopted simplified models neglecting various aerodynamic effects.On the other hand,these methods mitigate the workload of designing a complex flight control design,but simplified models in some ways are questionable due to the unavailability of the model validation of these designs.Moreover,this oversimplification may result in deterioration of the overall performance of the flight control design.5) The autonomous flight in the DFAV can be categorized into three flight phases:Horizontal,transition,and vertical flight.The DFAV is divided into three flight modes (i.e.,hover,transition,and cruise shown in Fig.10).However,it has been observed that most of the work covers only a single flight mode(i.e.,hover,etc.).The autonomous flight of the DFAV covering the entire flight envelop,including all the flight modes,has only been addressed in [12,14],which remains a promising trend for future works.6) Flight control techniques can be divided into three main types:linear flight control,model-based nonlinear control,and learning-based flight control.It has been observed that most of the work based on linear flight control due to easy real-time implementation,ease of tuning,and lower computational time.However,it is not useful for transition flight.For that purpose,model-based nonlinear control techniques are preferable.If the complete dynamical information is unavailable,the learning-based flight control methodologies are suitable.Compound flight control strategies are also crucial for incorporating various techniques into a single FCS.Although various flight control techniques have been adopted,as provided in Table 5,different control problems can still be explored.Some of the future prospects can be to address the non-minimum phase behavior of the DFAVs,which has been rarely addressed in the literature[86,100,126].In complex nonlinear optimal control based techniques,designing an adequate controller with acceptable computational burden is another issue.7) The DFAV can also interact with the environment more efficiently as compared to other ORAVs due to duct[53,98].Another exciting trend is the role of DFAV as an aerial manipulator with the environment to perform mid-air operations[105,121].Nonetheless,the existing literature only deals with lightweight manipulators for single DF configurations[80].
Due to their promising features,this first review paper presented on the classifications and flight control techniques of DFAVs is expected to be informative for the researchers who are interested in the DFAVs.