A Review of Design and Fabrication of the Bionic Flapping Wing Micro Air Vehicles

Bionic flapping-wing micro air vehicles (FWMAVs) are promising for a variety of applications because of their flexibility and high mobility. This study reviews the state-of-the-art FWMAVs of various research institutes driven by electrical motor, mechanical transmission structure and “artificial muscle” material and then elaborates on the aerodynamic mechanism of micro-winged birds and insects. Owing to their low mass budget, FWMAVs require actuators with high power density from micrometer to centimeter scales. The selection and design of the mechanical transmission should be considered in parallel with the design of the power electronic interface required to drive it. Finally, power electronic topologies suitable for driving “artificial muscle” materials used in FWMAVs are stated.


Introduction
At present, there are many complex and cluttered environments that humans cannot survive in for a long time, such as glaciers, deserts, dense forests and caves. To explore these rigorous environments, flapping-wing micro air vehicles (FWMAVs) have been included in research by many scientific institutions as one of the feasible solutions. The advantages of FWMAVs are their more flexible maneuverability and more efficient aerodynamics compared with those of fixed or rotary wing air vehicles.
Remarkable achievements have been accomplished with regard to designing and optimizing the constituent subsystems of FWMAVs, including aerodynamic mechanism [1], mechanical transmission [2], actuator [3] and power electronic interface [4]. Nevertheless, several critical challenges have to be resolved urgently to increase the practical ability of FWMAVs. This manuscript presents the (a) current research progress of existing FWMAVs investigated by scientific institutions; (b) aerodynamic mechanism of FWMAVs, including birds and insects; (c) actuators composed of new material; and (d) related power electronic interface. The aforementioned literature reviews can be used as a reference by researchers.
The remainder of this paper is organized as follows: Section 2 introduces the latest achievements in FWMAVs conducted by several research institutes; Section 3 presents the aerodynamic mechanism of FWMAVs, mechanical transmissions and various types of actuators and related power electronic interfaces; and finally, Section 4 discusses the conclusions and the future research prospects for FWMAVs.

FWMAVs with Different Actuation Mechanisms
Research on FWMAVs can be divided into three types according to the driving methods, including electrical motor, mechanical transmission and "artificial muscle" material. Additionally, Sahai et al. [23] attempted to integrate flexural hinges into a four-bar compliant flapping transmission for FWMAV with approximately 3 g of weight, as shown in Figure 2. A distinguishing feature of the mechanism is using rubber-based flexures in two of its joints (joints 3 and 4). According to the experiment, not only did compliant mechanism save up to 20% of the input power and 1% of the weight but also produced more thrust.
Coil Springs have also been directly coupled with DC motors for directly driving flapping wings toward resonance. Campolo et al. [24] presented a proof-of-concept flapping-wing micro aerial, shown in Figure 3. Additionally, Sahai et al. [23] attempted to integrate flexural hinges into a four-bar compliant flapping transmission for FWMAV with approximately 3 g of weight, as shown in Figure 2.

Mechanical Transmission-Driven Method
Another means of realizing ultra-light FWMAV is by using mechanical transmission-driven method. The "artificial butterfly" [22] shown in Figure 1 was constructed by the University of Tokyo, Japan, in 2010 using a butterfly-like crank-driven wing body with a flight time of only a few seconds. The flight of "artificial butterfly" is realized with simple flapping, requiring little feedback control of the feathering angle. Even, the stable forward flight could be realized without active feathering or feedback control of the wing motion Additionally, Sahai et al. [23] attempted to integrate flexural hinges into a four-bar compliant flapping transmission for FWMAV with approximately 3 g of weight, as shown in Figure 2. A distinguishing feature of the mechanism is using rubber-based flexures in two of its joints (joints 3 and 4). According to the experiment, not only did compliant mechanism save up to 20% of the input power and 1% of the weight but also produced more thrust.
Coil Springs have also been directly coupled with DC motors for directly driving flapping wings toward resonance. Campolo et al. [24] presented a proof-of-concept flapping-wing micro aerial, shown in Figure 3. A distinguishing feature of the mechanism is using rubber-based flexures in two of its joints (joints 3 and 4). According to the experiment, not only did compliant mechanism save up to 20% of the input power and 1% of the weight but also produced more thrust.
Coil Springs have also been directly coupled with DC motors for directly driving flapping wings toward resonance. Campolo et al. [24] presented a proof-of-concept flapping-wing micro aerial, shown in Figure 3.

Mechanical Transmission-Driven Method
Another means of realizing ultra-light FWMAV is by using mechanical transmission-driven method. The "artificial butterfly" [22] shown in Figure 1 was constructed by the University of Tokyo, Japan, in 2010 using a butterfly-like crank-driven wing body with a flight time of only a few seconds. The flight of "artificial butterfly" is realized with simple flapping, requiring little feedback control of the feathering angle. Even, the stable forward flight could be realized without active feathering or feedback control of the wing motion Additionally, Sahai et al. [23] attempted to integrate flexural hinges into a four-bar compliant flapping transmission for FWMAV with approximately 3 g of weight, as shown in Figure 2. A distinguishing feature of the mechanism is using rubber-based flexures in two of its joints (joints 3 and 4). According to the experiment, not only did compliant mechanism save up to 20% of the input power and 1% of the weight but also produced more thrust.
Coil Springs have also been directly coupled with DC motors for directly driving flapping wings toward resonance. Campolo et al. [24] presented a proof-of-concept flapping-wing micro aerial, shown in Figure 3. The prototype consists of two brushed DC motors, two 7 cm length wings, two helical springs and two shaft-spring-wing couplers. The pair of small helical springs is treated as the compliant structures for energy storage and recovery. The two separated DC motors can drive the individual wing to resonate, respectively. Experiments demonstrated the prototype can successfully lift off and the maximum lift-to-weight ratio can be achieved at the flapping frequency 10 Hz using controlling torques.
The traditional electrical motor-and mechanical transmission-driven methods of FWMAVs have problems of low driving efficiency and large power losses at low Reynolds number during flight. Thus, initiative ideas using novel driving methods are proposed to resolve the described challenges.

"Artificial Muscle" Material-Driven Method
To improve flapping frequency and aerodynamic efficiency, some scientific research institutions suggested using "artificial muscles" as a new driving actuator for FWMAV instead of electrical motor and machinery. These new materials have also been applied to some present robotic applications, providing novel ideas for FWMAVs.
In 2010, the Korean Academy of Science and Technology developed a bionic crawler robot with thermal material based on the thermoelectric stretching effect [25]. In 2014, the Massachusetts Institute of Technology in the US developed a circular closed-chain robot with shape-memory material that depends on the temperature memory effect [26]. The exploration of these new materials provides alternative options and avenues for the actuator of FWMAVs. Although the mechanical deformation and energy conversion efficiency of new materials are excellent, the maximum response frequency is less than 5 Hz, which means it fails to meet the requirements of folding ratio of flapping wings. In 2013, the University of Maryland in the US developed the electrostatic material bounce robot [27] and in 2007, Sungkyunkwan University in Korea fabricated an insect animal robot [28] that has better performance in frequency response. However, due to the limitations of inherent electrical characteristics, mechanical deformation was too little to meet the large physical transformation requirement from FWMAVs.
In addition, the University of California, Berkeley, developed a 25 mm (wingtip-to-wingtip) FWMAV capable of realizing sustained autonomous flight in 2007 [29]. Figure 4 shows the mentioned micromechanical flying insect, with four degrees-of-freedom, weighing approximately 100 mg, excluding battery or electronic devices. The main mechanical transmission component, namely, thorax, consists of two four-bar mechanisms that amplify and convert the mechanical motion into wing flapping and rotation. The biologically inspired system architecture results in a hierarchical structure of different control methodologies, which give the possibility to plan complex missions from a sequence of simple flight modes and maneuvers.  The prototype consists of two brushed DC motors, two 7 cm length wings, two helical springs and two shaft-spring-wing couplers. The pair of small helical springs is treated as the compliant structures for energy storage and recovery. The two separated DC motors can drive the individual wing to resonate, respectively. Experiments demonstrated the prototype can successfully lift off and the maximum lift-to-weight ratio can be achieved at the flapping frequency 10 Hz using controlling torques.
The traditional electrical motor-and mechanical transmission-driven methods of FWMAVs have problems of low driving efficiency and large power losses at low Reynolds number during flight. Thus, initiative ideas using novel driving methods are proposed to resolve the described challenges.

"Artificial Muscle" Material-Driven Method
To improve flapping frequency and aerodynamic efficiency, some scientific research institutions suggested using "artificial muscles" as a new driving actuator for FWMAV instead of electrical motor and machinery. These new materials have also been applied to some present robotic applications, providing novel ideas for FWMAVs.
In 2010, the Korean Academy of Science and Technology developed a bionic crawler robot with thermal material based on the thermoelectric stretching effect [25]. In 2014, the Massachusetts Institute of Technology in the US developed a circular closed-chain robot with shape-memory material that depends on the temperature memory effect [26]. The exploration of these new materials provides alternative options and avenues for the actuator of FWMAVs. Although the mechanical deformation and energy conversion efficiency of new materials are excellent, the maximum response frequency is less than 5 Hz, which means it fails to meet the requirements of folding ratio of flapping wings. In 2013, the University of Maryland in the US developed the electrostatic material bounce robot [27] and in 2007, Sungkyunkwan University in Korea fabricated an insect animal robot [28] that has better performance in frequency response. However, due to the limitations of inherent electrical characteristics, mechanical deformation was too little to meet the large physical transformation requirement from FWMAVs.
In addition, the University of California, Berkeley, developed a 25 mm (wingtip-to-wingtip) FWMAV capable of realizing sustained autonomous flight in 2007 [29]. Figure 4 shows the mentioned micromechanical flying insect, with four degrees-of-freedom, weighing approximately 100 mg, excluding battery or electronic devices. The main mechanical transmission component, namely, thorax, consists of two four-bar mechanisms that amplify and convert the mechanical motion into wing flapping and rotation. The biologically inspired system architecture results in a hierarchical structure of different control methodologies, which give the possibility to plan complex missions from a sequence of simple flight modes and maneuvers. In 2013, Harvard University proposed the use of a piezoelectric bimorph material as actuator to generate mechanical deformation using inverse piezoelectric effect [30][31][32] (Figure 5). This liftenhancing design of mimicking the flapping mechanism of a fly's 2 cm wingspan enabled the 80 mg In 2013, Harvard University proposed the use of a piezoelectric bimorph material as actuator to generate mechanical deformation using inverse piezoelectric effect [30][31][32] (Figure 5). This liftenhancing design of mimicking the flapping mechanism of a fly's 2 cm wingspan enabled the 80 mg FWMAV to fly autonomously. The Robobee was fitted with various individual sensors for onboard feedback. Pitch and yaw control of the RoboBee using an onboard magnetometer was presented with the robot constrained to rotate only about its principal axes. The integration of a MEMS gyroscope onto the RoboBee to provide attitude feedback in flight. However, it only worked with the connection from an external battery power supply. FWMAV to fly autonomously. The Robobee was fitted with various individual sensors for onboard feedback. Pitch and yaw control of the RoboBee using an onboard magnetometer was presented with the robot constrained to rotate only about its principal axes. The integration of a MEMS gyroscope onto the RoboBee to provide attitude feedback in flight. However, it only worked with the connection from an external battery power supply. The characteristics of the above FWMAVs in terms of driving method are shown in Table 1. In conclusion, the FWMAV driven by electrical motor method is the most successful and widely used because of its high maturity, low cost and wide application in the field. Currently, the mechanical transmission-driven FWMAV is only utilized for the experimental verification of the aerodynamic model, which is not practical. Although "artificial muscle" material-driven FWMAV is in the preliminary stage, it has wide application prospects and important research significance.
In the near future, FWMAVs will evolve to become ultra-compact in size, super light and will have longer flight duration. Thus, the challenges of investigating aerodynamic mechanism, transmission mechanism and power electronic interface should be resolved. The characteristics of the above FWMAVs in terms of driving method are shown in Table 1. In conclusion, the FWMAV driven by electrical motor method is the most successful and widely used because of its high maturity, low cost and wide application in the field. Currently, the mechanical transmission-driven FWMAV is only utilized for the experimental verification of the aerodynamic model, which is not practical. Although "artificial muscle" material-driven FWMAV is in the preliminary stage, it has wide application prospects and important research significance.
In the near future, FWMAVs will evolve to become ultra-compact in size, super light and will have longer flight duration. Thus, the challenges of investigating aerodynamic mechanism, transmission mechanism and power electronic interface should be resolved.

Aerodynamic Mechanism Bases
Unlike the fixed-wing and rotary-wing aerial vehicles, the body of an FWMAV is mainly constructed based on bionics inspired by birds and insects. A flutter cycle can be divided into two stages: lower flap and upper flap. The wings are twisted quickly during the transition between the lower and upper flaps and start to flip over at the end of each stroke. The aerodynamic basis of insect flight can be divided into four types.

1.
Delayed stall mechanism. For an in-depth study on the aerodynamic mechanism of flapping wings of insects, see that conducted by biologists C. P. Ellington and C. van den Ber et al. on insect behavior [33]. They used scaled-up model of hawkmoth wings for experiments. The front edge of the hawkmoth wing was equipped with a smoke-releasing device and a high-speed camera to record the changing formation of the air flow of its wings during flapping. The study indicated that the large lift produced by the hawkmoth's wings during flapping is due to the presence of delayed stalls. The angle of attack is much larger than the conventionally critical angle of attack, a difference that cannot be explained by classically aerodynamic principles. However, the experiment revealed that the formation of a vortex of circulating air flow at the leading edge is caused by the rapid movement of the wings. A low-pressure area will be generated because the vortex is located above the wings. Thus, generating a large lift force is beneficial. The observed phenomenon is consistent with the basic theoretical calculation, which is in line with the study of Liu H. [34].
Clap-and-fling mechanism. Weis-Fogh discovered a mechanism of lift generation when he observed wasp flight, that is, the wings are folded back (clap) and then quickly opened (fling) before the next incitement [35] ( Figure 6). The aerodynamic mechanism creates a discrete vortex at the wingtip and results in a big lift. Weis-Fogh named it the clap-and-fling mechanism, which explains the generation of large lift coefficients by insects when hovering.

Aerodynamic Mechanism Bases
Unlike the fixed-wing and rotary-wing aerial vehicles, the body of an FWMAV is mainly constructed based on bionics inspired by birds and insects. A flutter cycle can be divided into two stages: lower flap and upper flap. The wings are twisted quickly during the transition between the lower and upper flaps and start to flip over at the end of each stroke. The aerodynamic basis of insect flight can be divided into four types.
1. Delayed stall mechanism. For an in-depth study on the aerodynamic mechanism of flapping wings of insects, see that conducted by biologists C. P. Ellington and C. van den Ber et al. on insect behavior [33]. They used scaled-up model of hawkmoth wings for experiments. The front edge of the hawkmoth wing was equipped with a smoke-releasing device and a high-speed camera to record the changing formation of the air flow of its wings during flapping. The study indicated that the large lift produced by the hawkmoth's wings during flapping is due to the presence of delayed stalls. The angle of attack is much larger than the conventionally critical angle of attack, a difference that cannot be explained by classically aerodynamic principles. However, the experiment revealed that the formation of a vortex of circulating air flow at the leading edge is caused by the rapid movement of the wings. A low-pressure area will be generated because the vortex is located above the wings. Thus, generating a large lift force is beneficial. The observed phenomenon is consistent with the basic theoretical calculation, which is in line with the study of Liu H. [34].
Clap-and-fling mechanism. Weis-Fogh discovered a mechanism of lift generation when he observed wasp flight, that is, the wings are folded back (clap) and then quickly opened (fling) before the next incitement [35] ( Figure 6). The aerodynamic mechanism creates a discrete vortex at the wingtip and results in a big lift. Weis-Fogh named it the clap-and-fling mechanism, which explains the generation of large lift coefficients by insects when hovering. Figure 6. Clap-and-fling mechanism.
As shown in Figure 4, the clap and fling mechanism consists of two phases: the first one, the leading edges of both wings are clapped together at the end of the upstroke (from (a) to (c)) and the second one, the wings rotate around their trailing edges, thus flinging apart (from (d) to (f)). During the first "fling" phase, the fling motion is produced by a rotation of the wings about the common trailing edge, a pair of large leading edge vortices are formed. During the second "fling" phase, air flows around the leading edge of each wing which creates a bound vortex on each wing acting as the starting vortex for the opposite wing. This allows a rapid buildup of circulation as well as an increase in total lift production.
2. Rotational circulation mechanism. Dickinson M. H. et al. completed the experiment using a mechanical device to obtain the equations of the wings' flapping motion captured by the camera [36][37][38]. They simulated the movement of insect wings by driving the model wings placed in the cylinder and utilized a sensor to measure the lift and drag acting on the airfoil. As a result, they As shown in Figure 4, the clap and fling mechanism consists of two phases: the first one, the leading edges of both wings are clapped together at the end of the upstroke (from (a) to (c)) and the second one, the wings rotate around their trailing edges, thus flinging apart (from (d) to (f)). During the first "fling" phase, the fling motion is produced by a rotation of the wings about the common trailing edge, a pair of large leading edge vortices are formed. During the second "fling" phase, air flows around the leading edge of each wing which creates a bound vortex on each wing acting as the starting vortex for the opposite wing. This allows a rapid buildup of circulation as well as an increase in total lift production.

2.
Rotational circulation mechanism. Dickinson M. H. et al. completed the experiment using a mechanical device to obtain the equations of the wings' flapping motion captured by the camera [36][37][38]. They simulated the movement of insect wings by driving the model wings placed in the cylinder and utilized a sensor to measure the lift and drag acting on the airfoil. As a result, they found that the translational force generated by the wing attack was not sufficient, whereas they discovered rotational circulation mechanism generated more lift, usually two to three times the chord length. The theory of rotational circulation mechanism is that the wing of the fruit fly generates a reverse vortex when the wings are flapping forward at the end. So the airflow velocity above the fly is faster than at the bottom, forming a pressure difference and producing enough lift.

3.
Added mass effect mechanism. This is known to play a substantial role in defining the hydrodynamic forces acting on moving bodies because the movement of the surrounding fluid requires an additional force over and above what is necessary to accelerate the body itself [39]. Moreover, the mechanism was later devoted mostly to fast oscillating motions in view of flutter and stability studies. Andro J. Y. and Jacqin L. recently analyzed the added mass effect on a harmonically heaving airfoil by using 2-D direct numerical simulations [40]. Basing on previous studies, Giesing J. P. developed an unsteady panel method for calculating the forces acting on an airfoil executing arbitrary motions and calculated the added mass coefficients [41]. A fairly good agreement was found between the numerical and analytical values of the coefficients. Although many researchers made some achievements, the theoretical model to explain a variety of complex parameters still requires further improvement.

Transmission Mechanism Policies
Mechanical transmissions, such as electrical motors and smart materials, are investigated and designed based on the former discovered aerodynamic mechanism bases. Electrical motors are reliable, versatile, low cost and easily purchasable in the commercial market. Most of the FWMAVs described in Section 2 are driven by electrical motors; the first one is Microbat in 2001, whose transmission structure is shown in Figure 7 [5]. The rotation of the electrical motor drives the gear to actuate FWMAVs and is the common transmission principle in electrical motor-driven micro aircraft. However, mechanical transmissions have individualized designs based on their characteristics. found that the translational force generated by the wing attack was not sufficient, whereas they discovered rotational circulation mechanism generated more lift, usually two to three times the chord length. The theory of rotational circulation mechanism is that the wing of the fruit fly generates a reverse vortex when the wings are flapping forward at the end. So the airflow velocity above the fly is faster than at the bottom, forming a pressure difference and producing enough lift. 3. Added mass effect mechanism. This is known to play a substantial role in defining the hydrodynamic forces acting on moving bodies because the movement of the surrounding fluid requires an additional force over and above what is necessary to accelerate the body itself [39]. Moreover, the mechanism was later devoted mostly to fast oscillating motions in view of flutter and stability studies. Andro J. Y. and Jacqin L. recently analyzed the added mass effect on a harmonically heaving airfoil by using 2-D direct numerical simulations [40]. Basing on previous studies, Giesing J. P. developed an unsteady panel method for calculating the forces acting on an airfoil executing arbitrary motions and calculated the added mass coefficients [41]. A fairly good agreement was found between the numerical and analytical values of the coefficients. Although many researchers made some achievements, the theoretical model to explain a variety of complex parameters still requires further improvement.

Transmission Mechanism Policies
Mechanical transmissions, such as electrical motors and smart materials, are investigated and designed based on the former discovered aerodynamic mechanism bases. Electrical motors are reliable, versatile, low cost and easily purchasable in the commercial market. Most of the FWMAVs described in Section 2 are driven by electrical motors; the first one is Microbat in 2001, whose transmission structure is shown in Figure 7 [5]. The rotation of the electrical motor drives the gear to actuate FWMAVs and is the common transmission principle in electrical motor-driven micro aircraft. However, mechanical transmissions have individualized designs based on their characteristics. New smart materials have also emerged that have attracted widespread attention. If the size of the flying robot is reduced to millimeter levels, then the efficiency of the conventional electrical motor will be reduced dramatically. Therefore, various smart actuators are an optimal alternative choice for FWMAVs. Smart actuators are micro-mechanical devices that use artificial materials to generate deformation [42]. Table 2 shows the overview characteristics of smart actuators, such as strain, stress, elastic energy density, efficiency and response speed [43]. New smart materials have also emerged that have attracted widespread attention. If the size of the flying robot is reduced to millimeter levels, then the efficiency of the conventional electrical motor will be reduced dramatically. Therefore, various smart actuators are an optimal alternative choice for FWMAVs. Smart actuators are micro-mechanical devices that use artificial materials to generate deformation [42]. Table 2 shows the overview characteristics of smart actuators, such as strain, stress, elastic energy density, efficiency and response speed [43].
As illustrated in the table, shape memory alloy (SMA), shape memory polymer (SMP), electro-chemo-mechanical conducting polymer (EMCP), thermal polymer and mechanochemical polymer (MCP) are capable of large free strain and high resistance but have slow response and limited efficiency, which make them unsuitable for driving FWMAV. By contrast, piezoelectric actuators exhibit relatively low free strain. They have the ability to produce very high blocking forces and more efficient sensitivity. Owing to speed requirements, piezoelectric, dielectric elastomers (DEAs), electrostatics and electromagnetic actuators are effective alternatives to micro bionic flapping wing aerial vehicles.

1.
Piezoelectric actuators: Piezoelectric actuators are devices that use inverse piezoelectric effects [44] ( Figure 8). The drive voltage of a piezoelectric actuator is typically in the range of a few tens to several hundreds of volts. The operating voltage of piezoceramic stack actuators is realized by stacking monolithic multilayer elements in the range of 60-200 V and a higher required voltage of approximately 1000 V for discrete stack actuators. When in conjunction with a mechanical transmission, the actuator is capable of enhanced stroke amplitude and reciprocating motion for flapping flight [45,46]. In addition, piezoelectric actuators have high displacement, fast response [47,48] and high efficiency at high deformation frequency [49]. Therefore, piezoelectric materials are an optimal choice for use as an actuator in FWMAVs.  At Harvard University, Wood et al. [30,45,[48][49][50][51][52][53] conducted an in-depth study on an insect-scale flutter robot called RoboBee that uses a piezoelectric actuator. RoboBee was the first insect-sized robot with the ability to fly.
2. Dielectric elastomers: DEA is polymer material with flexible electrodes that have a large electromechanical response to the applied electric field (Figure 9) [53,54]. DEA typically operates at very high voltages (about 1-10 kV) with an electric field of approximately 100 MV/m and produces large strain at high working density [55,56]. In reference [57], DEA was used to drive approximately 15 g of FWMAV that extends the limitation of the artificial muscle to the level of energy required for a heavyweight aerial vehicle. However, the application is limited by the challenge of a high electric field requirement in the development of DEA. At Harvard University, Wood et al. [30,45,[48][49][50][51][52][53] conducted an in-depth study on an insect-scale flutter robot called RoboBee that uses a piezoelectric actuator. RoboBee was the first insect-sized robot with the ability to fly.

2.
Dielectric elastomers: DEA is polymer material with flexible electrodes that have a large electromechanical response to the applied electric field (Figure 9) [53,54]. DEA typically operates at very high voltages (about 1-10 kV) with an electric field of approximately 100 MV/m and produces large strain at high working density [55,56]. In reference [57], DEA was used to drive approximately 15 g of FWMAV that extends the limitation of the artificial muscle to the level of energy required for a heavyweight aerial vehicle. However, the application is limited by the challenge of a high electric field requirement in the development of DEA.

3.
Electrostatic elastomer: Electrostatic and piezoelectric actuators both offer efficient compliant actuation and are capable of providing high working densities [48]. Piezoelectric bimorph actuators have been successfully implemented for centimeter-scale robots [47] but the performance of thin film required by millimeter-scale robots deteriorates [58]. To make up for this disadvantage, electrostatic actuators are generally fabricated in chip level with Microelectromechanical Systems (MEMS) technique, which provide an excellent choice for mobile microrobots ( Figure 10) [59].

4.
Electromagnetic actuators: Electromagnetic actuators convert electrical energy to mechanical energy and vice versa by using electromagnetic mechanical principles. Electromagnetic actuators exhibit good performance owing to their quick response, simple structure, easy control and low voltage requirement from 0 to 24 V [60,61]. Electromagnetic actuators mainly consist of an electromagnetic coil, a permanent magnet rotor and a "virtual spring" magnet pair. Deng et al. [62] recently used a 2.6 g electromagnetic actuator to drive a FWMAV with wing-beat frequency, as shown in Figure 11.
2. Dielectric elastomers: DEA is polymer material with flexible electrodes that have a large electromechanical response to the applied electric field (Figure 9) [53,54]. DEA typically operates at very high voltages (about 1-10 kV) with an electric field of approximately 100 MV/m and produces large strain at high working density [55,56]. In reference [57], DEA was used to drive approximately 15 g of FWMAV that extends the limitation of the artificial muscle to the level of energy required for a heavyweight aerial vehicle. However, the application is limited by the challenge of a high electric field requirement in the development of DEA. 3. Electrostatic elastomer: Electrostatic and piezoelectric actuators both offer efficient compliant actuation and are capable of providing high working densities [48]. Piezoelectric bimorph actuators have been successfully implemented for centimeter-scale robots [47] but the performance of thin film required by millimeter-scale robots deteriorates [58]. To make up for this disadvantage, electrostatic actuators are generally fabricated in chip level with Microelectromechanical Systems (MEMS) technique, which provide an excellent choice for mobile microrobots (Figure 10) [59]. 4. Electromagnetic actuators: Electromagnetic actuators convert electrical energy to mechanical energy and vice versa by using electromagnetic mechanical principles. Electromagnetic actuators exhibit good performance owing to their quick response, simple structure, easy control and low voltage requirement from 0 to 24 V [60,61]. Electromagnetic actuators mainly consist of an electromagnetic coil, a permanent magnet rotor and a "virtual spring" magnet pair. Deng et al. [62] recently used a 2.6 g electromagnetic actuator to drive a FWMAV with wing-beat frequency, as shown in Figure 11.  4. Electromagnetic actuators: Electromagnetic actuators convert electrical energy to mechanical energy and vice versa by using electromagnetic mechanical principles. Electromagnetic actuators exhibit good performance owing to their quick response, simple structure, easy control and low voltage requirement from 0 to 24 V [60,61]. Electromagnetic actuators mainly consist of an electromagnetic coil, a permanent magnet rotor and a "virtual spring" magnet pair. Deng et al. [62] recently used a 2.6 g electromagnetic actuator to drive a FWMAV with wing-beat frequency, as shown in Figure 11. At present, most insect-scale FWMAVs are driven by piezoelectric actuators [3,4,31,47]. Although attempts were made to use electromagnetic actuators and insulative elastomer actuators (dielectric and electrostatic actuators) for driving FWMAVs, no report indicates that aerial vehicle prototypes can be lifted successfully.

Power Electronic Interfaces
Most compact energy sources potentially suitable for FWMAV applications, such as supercapacitors [63], solar cells [64] and fuel cells [65], generate output lower than 5V. At present, conventional batteries are the only commercially available technology that is appropriate for FWMAV. The actuators mentioned earlier are classified into two actuation modes. The first is the current mode, which requires high current and relatively low voltages and corresponds to SMA, SMP, At present, most insect-scale FWMAVs are driven by piezoelectric actuators [3,4,31,47]. Although attempts were made to use electromagnetic actuators and insulative elastomer actuators (dielectric and electrostatic actuators) for driving FWMAVs, no report indicates that aerial vehicle prototypes can be lifted successfully.

Power Electronic Interfaces
Most compact energy sources potentially suitable for FWMAV applications, such as supercapacitors [63], solar cells [64] and fuel cells [65], generate output lower than 5V. At present, conventional batteries are the only commercially available technology that is appropriate for FWMAV. The actuators mentioned earlier are classified into two actuation modes. The first is the current mode, which requires high current and relatively low voltages and corresponds to SMA, SMP, EMCP, thermal actuator and MCP. The second is voltage mode, which requires high voltages and relatively low currents and corresponds to piezoelectric, DEA, electrostatic and electromagnetic actuators.
The use of the above actuators requires a power electronic interface with high power efficiency and density to transfer energy from power source to actuator. The power electronic interface generally consists of a power stage, which regulates the voltage of the energy sources to the required level and a drive stage, which uses the output voltage to generate a time-varying signal applied on the smart actuator. This section illustrates a potential solution (not currently used) for the power electronic interface of both current-and voltage-mode actuators.

Power Electronic Interfaces for Current-Mode Actuators
Current-mode actuators rely on high current to raise the temperature of the active material through resistive heating. Generally, the voltage delivered to the actuator is lower than the energy source voltage.
One of the simplest ways to convert the energy source voltage to the required low level is to use the conventional buck converter. Two alternative schemes are described to realize low voltage with compact package: n-stage cascade buck converter [66] and tapped inductor buck converter [67].
The scheme consisting of an n-stage cascade combination of buck converter with single active switch is shown in Figure 12.

Power Electronic Interfaces for Current-Mode Actuators
Current-mode actuators rely on high current to raise the temperature of the active material through resistive heating. Generally, the voltage delivered to the actuator is lower than the energy source voltage.
One of the simplest ways to convert the energy source voltage to the required low level is to use the conventional buck converter. Two alternative schemes are described to realize low voltage with compact package: n-stage cascade buck converter [66] and tapped inductor buck converter [67].
The scheme consisting of an n-stage cascade combination of buck converter with single active switch is shown in Figure 12. This kind of converter requires an active power MOSFET and 2n-1 passive diodes and can be utilized only when the required number of stages is not very large. Otherwise, the whole conversion efficiency will deteriorate due to the parasitic losses of components.
Another feasible topology named tapped inductor buck converter is presented in Figure 13. This circuit utilizes a tapped inductor operated by one active switch to achieve high step-up ratio with the square of turns ratio between the primary and secondary windings.
The circuit architecture of the two topologies reveals that the latter topology requires fewer components than the former but since no tapped inductor is commercially available on the market, the circuit manufacturing technology is the main challenge for the latter topology. For the former one, the larger the output power, the higher the efficiency is subject to the exponential distribution. In addition, the upper efficiency is limited by the number of stages, while the latter is more efficient. This kind of converter requires an active power MOSFET and 2n-1 passive diodes and can be utilized only when the required number of stages is not very large. Otherwise, the whole conversion efficiency will deteriorate due to the parasitic losses of components.
Another feasible topology named tapped inductor buck converter is presented in Figure 13.

Power Electronic Interfaces for Current-Mode Actuators
Current-mode actuators rely on high current to raise the temperature of the active material through resistive heating. Generally, the voltage delivered to the actuator is lower than the energy source voltage.
One of the simplest ways to convert the energy source voltage to the required low level is to use the conventional buck converter. Two alternative schemes are described to realize low voltage with compact package: n-stage cascade buck converter [66] and tapped inductor buck converter [67].
The scheme consisting of an n-stage cascade combination of buck converter with single active switch is shown in Figure 12. This kind of converter requires an active power MOSFET and 2n-1 passive diodes and can be utilized only when the required number of stages is not very large. Otherwise, the whole conversion efficiency will deteriorate due to the parasitic losses of components.
Another feasible topology named tapped inductor buck converter is presented in Figure 13. This circuit utilizes a tapped inductor operated by one active switch to achieve high step-up ratio with the square of turns ratio between the primary and secondary windings.
The circuit architecture of the two topologies reveals that the latter topology requires fewer components than the former but since no tapped inductor is commercially available on the market, the circuit manufacturing technology is the main challenge for the latter topology. For the former one, the larger the output power, the higher the efficiency is subject to the exponential distribution. In addition, the upper efficiency is limited by the number of stages, while the latter is more efficient. This circuit utilizes a tapped inductor operated by one active switch to achieve high step-up ratio with the square of turns ratio between the primary and secondary windings.
The circuit architecture of the two topologies reveals that the latter topology requires fewer components than the former but since no tapped inductor is commercially available on the market, the circuit manufacturing technology is the main challenge for the latter topology. For the former one, the larger the output power, the higher the efficiency is subject to the exponential distribution. In addition, the upper efficiency is limited by the number of stages, while the latter is more efficient.

Power Electronic Interfaces for Voltage-Mode Actuators
The power electronic interfaces should be able to convert the low input voltage of a lithium battery to a high voltage signal that drives the piezoelectric or dielectric actuator [68]. Unlike current-mode actuators, voltage-mode actuators require up to several hundred volts. Recovering unused energy from the actuators is also another challenge for power electronic interfaces because only a portion of the input electrical energy is converted into a mechanical deformation of the actuators. Owing to losses in the passive inductor and active switch, as well as a very high switching frequency, the conventional boost converter becomes impractical to resolve the above challenges. Five alternative electrical interfaces are presented to achieve high voltages in a compact package: hybrid voltage multiplier boost converter, tapped inductor boost converter, cascade boost converter, high conversion ratio boost converter and power amplifier using a piezoelectric transformer (PT) [69].
A hybrid topology consisting of a conventional boost converter cascaded with a switchedcapacitor charge pump circuit, as shown in Figure 14, has been considered previously to drive piezoelectric actuators [70] and electrostatic MEMS devices [71]. It is an n-level DC-DC converter using one switch, 2n + 1 diodes and 2n capacitors. Operating in a regime of high efficiency, the boost converter stage provides a moderate boost to the input voltage, while its pulsed output naturally charges up the capacitor ladder through the diodes. The charge pump circuit multiplies the boost converter's output voltage, ideally by a factor equal to the number of charge pump stages. The output power is limited by the size of the charge pump capacitors and the maximum output power of the boost converter. the conventional boost converter becomes impractical to resolve the above challenges. Five alternative electrical interfaces are presented to achieve high voltages in a compact package: hybrid voltage multiplier boost converter, tapped inductor boost converter, cascade boost converter, high conversion ratio boost converter and power amplifier using a piezoelectric transformer (PT) [69]. A hybrid topology consisting of a conventional boost converter cascaded with a switchedcapacitor charge pump circuit, as shown in Figure 14, has been considered previously to drive piezoelectric actuators [70] and electrostatic MEMS devices [71]. It is an n-level DC-DC converter using one switch, 2n + 1 diodes and 2n capacitors. Operating in a regime of high efficiency, the boost converter stage provides a moderate boost to the input voltage, while its pulsed output naturally charges up the capacitor ladder through the diodes. The charge pump circuit multiplies the boost converter's output voltage, ideally by a factor equal to the number of charge pump stages. The output power is limited by the size of the charge pump capacitors and the maximum output power of the boost converter. As shown in Figure 15, replacing the inductor in a classical boost converter with a tapped inductor results in a combination of boost and flyback converter topology, named the tapped inductor boost converter [72]. The voltage gain of this converter is greatly improved, which depends on the switching duty cycle and the transformer turns ratio. As shown in Figure 15, replacing the inductor in a classical boost converter with a tapped inductor results in a combination of boost and flyback converter topology, named the tapped inductor boost converter [72]. The voltage gain of this converter is greatly improved, which depends on the switching duty cycle and the transformer turns ratio. As shown in Figure 15, replacing the inductor in a classical boost converter with a tapped inductor results in a combination of boost and flyback converter topology, named the tapped inductor boost converter [72]. The voltage gain of this converter is greatly improved, which depends on the switching duty cycle and the transformer turns ratio. Super voltage boost technology is widely used in electronic converter design to increase the voltage transmission gain. Despite the high complexity of the converter, the super voltage boost converter can generate output voltages that are related to the geometric progression of the cascaded circuit [73][74][75]. Two-stage cascade boost converter is cascaded by conventional boost converters, as shown in Figure 16. It can sufficiently meet the high driving voltage requirement of piezoelectric actuators because of its high gain performance. Super voltage boost technology is widely used in electronic converter design to increase the voltage transmission gain. Despite the high complexity of the converter, the super voltage boost converter can generate output voltages that are related to the geometric progression of the cascaded circuit [73][74][75]. Two-stage cascade boost converter is cascaded by conventional boost converters, as shown in Figure 16. It can sufficiently meet the high driving voltage requirement of piezoelectric actuators because of its high gain performance. The cascade boost converter is suitable for driving piezoelectric actuators in hundred volt level. Nevertheless, DEAs are electrically actuated material devices that produce large deformation when a high driving voltage in a few thousand volts is applied to the electrode. The conversion between low voltage coming from lithium battery and high exciting voltage, which can drive DEA actuators, is not enough. Therefore, the conventional cascade boost converter becomes impractical. A feasible circuit that can achieve a few thousand volts with compact package [76] is presented in Figure 17. The two-stage cascade boost converter is derived from the two-stage boost converter by adding a double/enhanced circuit in each conversion stage.  Figure 18 shows the circuit configuration of high conversion ratio boost converter. This boost converter is used to transfer energy from the DC source Vin in the low-voltage side to the DC output Vo in the high-voltage side. When the proposed converter is operating in the boost mode, the circuit characteristic is cascaded by the boost converter and flyback converter with the voltage doubled [77,78]. The cascade boost converter is suitable for driving piezoelectric actuators in hundred volt level. Nevertheless, DEAs are electrically actuated material devices that produce large deformation when a high driving voltage in a few thousand volts is applied to the electrode. The conversion between low voltage coming from lithium battery and high exciting voltage, which can drive DEA actuators, is not enough. Therefore, the conventional cascade boost converter becomes impractical. A feasible circuit that can achieve a few thousand volts with compact package [76] is presented in Figure 17. The two-stage cascade boost converter is derived from the two-stage boost converter by adding a double/enhanced circuit in each conversion stage. The cascade boost converter is suitable for driving piezoelectric actuators in hundred volt level. Nevertheless, DEAs are electrically actuated material devices that produce large deformation when a high driving voltage in a few thousand volts is applied to the electrode. The conversion between low voltage coming from lithium battery and high exciting voltage, which can drive DEA actuators, is not enough. Therefore, the conventional cascade boost converter becomes impractical. A feasible circuit that can achieve a few thousand volts with compact package [76] is presented in Figure 17. The two-stage cascade boost converter is derived from the two-stage boost converter by adding a double/enhanced circuit in each conversion stage.  Figure 18 shows the circuit configuration of high conversion ratio boost converter. This boost converter is used to transfer energy from the DC source Vin in the low-voltage side to the DC output Vo in the high-voltage side. When the proposed converter is operating in the boost mode, the circuit characteristic is cascaded by the boost converter and flyback converter with the voltage doubled [77,78].  Figure 18 shows the circuit configuration of high conversion ratio boost converter. This boost converter is used to transfer energy from the DC source V in in the low-voltage side to the DC output V o in the high-voltage side. When the proposed converter is operating in the boost mode, the circuit characteristic is cascaded by the boost converter and flyback converter with the voltage doubled [77,78]. Figure 18 shows the circuit configuration of high conversion ratio boost converter. This boost converter is used to transfer energy from the DC source Vin in the low-voltage side to the DC output Vo in the high-voltage side. When the proposed converter is operating in the boost mode, the circuit characteristic is cascaded by the boost converter and flyback converter with the voltage doubled [77,78].  PTs have high voltage gain ratio and high power density (up to 40 W/cm 3 ) [79] and have been widely used in actuators and sensors. Generally, PTs have to operate close to the mechanical resonance frequency to obtain high voltage gain and power efficiency. The equivalent electrical circuit of a PT is shown in Figure 19a. The gain of a PT is high at low loads, making it a good candidate for the high-voltage, low-current requirements of voltage-mode actuators.  Figure 19a. The gain of a PT is high at low loads, making it a good candidate for the high-voltage, low-current requirements of voltage-mode actuators. Figure 19b shows the class "E" resonant topology has a low number of additional components. The inductor is selected to resonate with the input capacitance Cin of the PT at a frequency close to the mechanical resonance frequency [80]. The resonance transfers energy to the PT from the inductor when the switch is off. The switch is turned on again as soon as the voltage across Cin is zero. Regulation of the output voltage is achieved by varying the switching frequency.  Table 3 shows some valuable parameters for comparing and quantifying the five types of topology performance parameters. Note: n is the number of cascaded stages. N is the turns ratio between secondary side and primary side of the transformer.
According to Table 3, a hybrid voltage multiplier boost converter has n times gain in contrast to a conventional boost converter because it uses n times voltage multiplier. The disadvantage of this topology is its large size, high weight and low efficiency (caused by the multiplier). However, it is  Figure 19b shows the class "E" resonant topology has a low number of additional components. The inductor is selected to resonate with the input capacitance C in of the PT at a frequency close to the mechanical resonance frequency [80]. The resonance transfers energy to the PT from the inductor when the switch is off. The switch is turned on again as soon as the voltage across C in is zero. Regulation of the output voltage is achieved by varying the switching frequency. Table 3 shows some valuable parameters for comparing and quantifying the five types of topology performance parameters.
Note: n is the number of cascaded stages. N is the turns ratio between secondary side and primary side of the transformer. According to Table 3, a hybrid voltage multiplier boost converter has n times gain in contrast to a conventional boost converter because it uses n times voltage multiplier. The disadvantage of this topology is its large size, high weight and low efficiency (caused by the multiplier). However, it is commonly used in hundreds of voltage outputs because it is easily fabricated, with a tapped inductor boost converter capable of achieving the boosting capability without a high duty cycle. To achieve high voltage gains, this method has a considerably lesser number of parts than the hybrid voltage multiplier boost converter. However, the rectifier diode and output capacitor must be rated for the output voltage. Additionally, a custom transformer should be demanded to meet the low mass requirement in microrobotic applications because no commercial parts below 10 g can be purchased. As indicated in Table 3, an n-stage cascade boost converter has a higher gain than the traditional boost converter. The gain also depends on the switching duty cycle. With this simple converter, a high boosting function can be obtained. As illustrated in Table 3, a high conversion ratio boost converter can obtain high voltage gain by increasing the turns ratio of the coupled inductor. This converter has the same disadvantage as a tapped inductor boost converter. The customized transformer is a critical factor and difficult to fabricate. APT is better than a magnetic transformer because of its simple geometries, giving it potential in milligram-scale power actuator design.

Drive Stage
As mentioned earlier, these converters are DC-DC converters whose outputs are high enough to drive the piezoelectric actuator. If we want the wings of robotic insects to start vibrating, then an arbitrary unipolar drive voltage should be provided. Using an inductor, two additional switches and two self-timed shutdown diodes with capacitive loads, [81] proposed a highly remarkable energy recovery from the wing vibration. However, this proposed design only focuses on the charge recovery of piezoelectric actuators with quasi-square waves. Another feasible method is to use an LC resonance to obtain an arbitrary driving wave [4]. This topology is called the switching amplifier driver ( Figure 20). boost converter. The gain also depends on the switching duty cycle. With this simple converter, a high boosting function can be obtained. As illustrated in Table 3, a high conversion ratio boost converter can obtain high voltage gain by increasing the turns ratio of the coupled inductor. This converter has the same disadvantage as a tapped inductor boost converter. The customized transformer is a critical factor and difficult to fabricate. APT is better than a magnetic transformer because of its simple geometries, giving it potential in milligram-scale power actuator design.

Drive Stage
As mentioned earlier, these converters are DC-DC converters whose outputs are high enough to drive the piezoelectric actuator. If we want the wings of robotic insects to start vibrating, then an arbitrary unipolar drive voltage should be provided. Using an inductor, two additional switches and two self-timed shutdown diodes with capacitive loads, [81] proposed a highly remarkable energy recovery from the wing vibration. However, this proposed design only focuses on the charge recovery of piezoelectric actuators with quasi-square waves. Another feasible method is to use an LC resonance to obtain an arbitrary driving wave [4]. This topology is called the switching amplifier driver ( Figure 20). After a series of charge and discharge pulses is applied to Q1 and Q2 at the appropriate time, an arbitrary waveform can be generated at Va. Differently, only a small amount of energy is processed in each switching cycle, which can be used to minimize the size of the inductor. After a series of charge and discharge pulses is applied to Q 1 and Q 2 at the appropriate time, an arbitrary waveform can be generated at V a . Differently, only a small amount of energy is processed in each switching cycle, which can be used to minimize the size of the inductor.

Control of Proposed Power Electronic Interfaces
After this the design and implementation of the proposed power electronic interfaces, a control system for the diagnostic of the actuator is needed and selected to be evaluated and implemented. The control system using estimation of the feedback parameters is shown in Figure 21.
The problem with the implementation of the control system is that it does not use the feedback of actuator displacement directly, instead it comes from the driving voltage/current estimation. The indirect feedback takes some time before the controller reduces/increase the control parameters when an overshoot/undershot occurs, if the estimated feedback parameters is too high or too low when a change is requested. To improve the controllability, one of the following techniques can be used (but is not limited to these).

Control of Proposed Power Electronic Interfaces
After this the design and implementation of the proposed power electronic interfaces, a control system for the diagnostic of the actuator is needed and selected to be evaluated and implemented. The control system using estimation of the feedback parameters is shown in Figure 21. The problem with the implementation of the control system is that it does not use the feedback of actuator displacement directly, instead it comes from the driving voltage/current estimation. The indirect feedback takes some time before the controller reduces/increase the control parameters when an overshoot/undershot occurs, if the estimated feedback parameters is too high or too low when a change is requested. To improve the controllability, one of the following techniques can be used (but is not limited to these).

Controller
Firstly, the Gain Scheduling Controller uses different control parameters depending either on the error, the size of the step or the region of the requested feedback parameters [82]. This should reduce the overshoot that some time when big steps are taken. One advantage of this is that the gain scheduling controller can handle the different regions of the actuator better than a conventional controller (static PI or PID controller, et al). But, the disadvantage is that is still needs to be tuned properly.
In addition, the LQ Controller uses a state space model of the actuator and an observer that is used to create the control signal to the system [83]. This requires both an accurate description of the system and observer. For the observer a Kalman filter is often used. One advantage is that the LQ controller both can handle disturbances and follow the reference signal equally good or better than conventional controller. The disadvantage is that the LQ controller is more complex to implement and tune than classic controller. However, if a good model of the actuator is available the implementation of the LQ controller becomes less complex.
The Self Tuning Controller is based on a black box model of the actuator [84]. To get an estimation of the black box model, the control system estimation needs to be done online. Then the controller uses parameters from the control system estimation to calculate the new control strategy. This will allow the control system to adapt to changes that occur due to differences in the load force and other external changes. One advantage of this is that it is capable of handle changes to the control system without losing the simplicity of the PID controller. Nevertheless, it needs to have a control system estimator that makes it hard to guarantee the stability of the control system because of the dynamics of the estimation.
The general solutions of the three described controllers can be used to control the described power electronic interfaces for actuators and even can be used in other applications. Self-Tuning Controller was chosen to be used more generally. One other reason was that it uses the PID structure which is more understandable than the Gain Scheduling Controller and LQ controller.

Conclusions
This paper summarizes and discusses the system level of FWMAVs with a focus on state-of-theart FWMAVs, aerodynamic mechanisms, transmission mechanisms and power electronic interfaces. First, various FWMAVs driven by electrical motor, mechanical transmission structure and "artificial Firstly, the Gain Scheduling Controller uses different control parameters depending either on the error, the size of the step or the region of the requested feedback parameters [82]. This should reduce the overshoot that some time when big steps are taken. One advantage of this is that the gain scheduling controller can handle the different regions of the actuator better than a conventional controller (static PI or PID controller, et al). But, the disadvantage is that is still needs to be tuned properly.
In addition, the LQ Controller uses a state space model of the actuator and an observer that is used to create the control signal to the system [83]. This requires both an accurate description of the system and observer. For the observer a Kalman filter is often used. One advantage is that the LQ controller both can handle disturbances and follow the reference signal equally good or better than conventional controller. The disadvantage is that the LQ controller is more complex to implement and tune than classic controller. However, if a good model of the actuator is available the implementation of the LQ controller becomes less complex.
The Self Tuning Controller is based on a black box model of the actuator [84]. To get an estimation of the black box model, the control system estimation needs to be done online. Then the controller uses parameters from the control system estimation to calculate the new control strategy. This will allow the control system to adapt to changes that occur due to differences in the load force and other external changes. One advantage of this is that it is capable of handle changes to the control system without losing the simplicity of the PID controller. Nevertheless, it needs to have a control system estimator that makes it hard to guarantee the stability of the control system because of the dynamics of the estimation.
The general solutions of the three described controllers can be used to control the described power electronic interfaces for actuators and even can be used in other applications. Self-Tuning Controller was chosen to be used more generally. One other reason was that it uses the PID structure which is more understandable than the Gain Scheduling Controller and LQ controller.

Conclusions
This paper summarizes and discusses the system level of FWMAVs with a focus on state-of-the-art FWMAVs, aerodynamic mechanisms, transmission mechanisms and power electronic interfaces. First, various FWMAVs driven by electrical motor, mechanical transmission structure and "artificial muscles" material and investigated by research institutes are presented in detail. The unique aerodynamic modes of bird-mimetic flapping wing and insect-mimetic flapping wing aerial vehicles, which are unlike those of fixed-wing and rotary-wing aerial vehicles, are likewise elaborated. The selection and design of the mechanical transmission are considered based on the stringent requirement of physical and electrical performance in micrometer-to centimeter-scale level. Finally, power electronic topologies suitable for driving "artificial muscle" materials used in FWMAVs are stated. These results present some possible solutions for the creation of insect-sized FWMAVs and a substantial step toward the realization of flying microrobots.
Further size and weight reductions of FWMAVs are important issues for the future. MEMS technologies can be used to provide devices, such as lighter, smaller and less power consuming components than the current state of-the-art ones. Nanotechnology could play an important role also in aerodynamic improvements. FWMAVs will most likely be equipped with GPS and radar systems. Infrared and/or high-definition cameras could be included. Furthermore, trends could also include the development of sophisticated software that will enable the operation of future ultrasmall FWMAVs. Finally, with the improvements of artificial intelligence, some of them will have decision-making capabilities, opening the way to completely new mission profiles.