Acoustic resonators as wireless actuators in air for small-scale robots Authors: Junsun Hwang, Quentin Angéloz, Ashwin Subramanian Murugan, Hervé Lissek, Mahmut Selman Sakar Published: 2026 ABSTRACT While acoustic resonators have been extensively studied in liquids for the realization of untethered microrobots, their implementation in air remains largely unexplored. Here, we revisit the classical principle of Helmholtz resonance and show that 3D printed cavities with sizes ranging from centimeter to micrometer generate spatiotemporally controlled thrust ranging from millinewton to micronewton upon resonance excitation. We provide a comprehensive design framework to efficiently harness acoustic and aerodynamic forces that are validated through analytical modeling, numerical simulations, and experiments. We show robotic navigation of small-scale boats powered by either air-borne or structure-borne sound, highlighting the versatility of the actuation scheme. We present one type of microflier propelled by vertical thrust that weighs only 150 micrograms with a thrust-to-weight ratio of 4.9 and another type of microflier propelled by a rotor that weighs only 184 micrograms and rotates as fast as 13,000 revolutions per minute. The actuation scheme is compatible with even further miniaturization, pushing the boundaries of robotics and aeronautics. INTRODUCTION Acoustic actuation holds great promise for robotics and automation by offering contactless, minimally invasive, and gentle manipulation capabilities. Bulk and surface acoustic waves exert radiation forces on cells, microbubbles, and particles submerged in physiological liquids, enabling transport inside diverse environments including microfluidic devices and living organisms (1–9). Beyond acoustic trapping, acoustic waves can be harnessed to wirelessly power resonant actuators through fluid-structure interactions. Traveling acoustic waves strongly vibrate air bubbles (10, 11) and solid structures with sharp edges (12–15) at their resonance frequencies, generating acoustic streaming and thrust. Leveraging advanced manufacturing techniques such as two-photon lithography, complex polymer microstructures can be printed with nanometer resolution (16–18). A variety of three-dimensional (3D) printed rigid and soft robotic microsystems have been presented that are actuated by the acoustically driven vibrations (19–24). Acoustic radiation forces can levitate lightweight objects in air through formation of standing waves or beam shaping using phased array transducers (25–30). As the manipulation scheme relies on the formation of acoustic traps rather than structural resonance, it does not allow powering of robotic actuators. Moreover, acoustic levitation is constrained by the acoustic contrast factor, which depends on the density and compressibility differences between the object and the surrounding medium (31), providing limited force for manipulation. The design space is also limited, as levitation performance is highly dependent on the geometry of the levitated object. Airborne sound can be amplified through Helmholtz resonance that arises from the oscillation of air through the neck of an enclosed cavity (32). The seminal work by Helmholtz (33, 34) and theoretical formulations of Rayleigh (35) revealed that the air within the cavity acts as a mass-spring system, with the resonance frequency determined by the cavity volume and neck geometry. Oscillations of air within the neck can reach amplitudes high enough to induce nonlinear flows, producing directional jets through momentum exchange (36–38). Everyday objects such as empty bottles and Christmas ornaments can serve as resonators and generate thrust upon excitation by sound waves. In this work, we present an integrated design, fabrication, and control methodology that transforms basic polymer structures with cavities into untethered mobile robots. We experimentally investigated how geometric and mechanical properties such as cavity shape, neck geometry, wall thickness, and stiffness influence thrust generation. Rigid and soft resonators spanning centimeter to micrometer scales were fabricated using a range of additive manufacturing techniques, including fused filament fabrication (FFF), stereolithography (SLA), digital light processing (DLP), and two-photon polymerization (TPP). We built various high-power acoustic control systems that generate sound waves from 1 Hz to 40 kHz to excite the resonators across scales and transform them into wirelessly powered actuators. We extend the principle of air resonance beyond airborne excitation, demonstrating that thrust can be generated via structure-borne sound as well, where a transducer directly excites the cavity. Last, we demonstrate the potential of our actuation technique for small-scale robotics via telerobotic navigation of miniaturized boats and takeoff of microfliers harnessing either vertical thrust or aerodynamic lift. RESULTS Resonator design and fabrication The Helmholtz resonator consists of an air cavity and a neck, which together form a mass-spring system (Fig. 1A). The resonance frequency depends on the resonator geometry and is analytically given by (32)f=c2πAVLeff(1)where c is the speed of sound in air (∼343 m s−1 at 20°C), A is the cross-sectional area of the neck, and V is the cavity volume. The effective neck length, Leff, is the sum of the physical neck length Ln and the end correction ΔL (i.e., Leff = Ln + ΔL) that accounts for the additional inertial length due to sound radiation at the neck opening. The end correction is typically approximated as ΔLn ≈ 0.6r for a neck with the radius r (39). Helmholtz resonators as wireless actuators in air. (A) Mass-spring model of a Helmholtz resonator, highlighting the geometric parameters that determine the resonance frequency. In this model, the neck air is analogous to the mass and the cavity air to the spring. (B) Schematic showing oscillation of the air mass in the neck in response to airborne sound excitation, generating a nonzero time-averaged thrust. (C) Thrust generated by the air jet emitted from the resonator excited at its resonance frequency. High-speed imaging reveals the formation of a strong jet at resonance. (D) 3D-printed monolithic resonators fabricated across a range of dimensions and resonance frequencies. The cavities of the resonators are scaled isotropically. The inset shows a magnified image of a resonator microfabricated using two-photon lithography. The resonance frequency of a Helmholtz resonator is inversely related to the square root of the cavity volume, increasing as the overall size of the resonator is reduced, as formulated in Eq. 1. We calculated the resonance frequency of a resonator using finite element (FE) simulations (fig. S1A and Materials and Methods), which is in line with the solution of Eq. 1 (fig. S1B). The first resonance mode corresponds to the Helmholtz resonance, whereas the higher-order modes correspond to standing wave resonances (fig. S1C). When excited by airborne sound at its resonance frequency, the air mass in the neck oscillates with large amplitude (Fig. 1B). As the flow entering and exiting the neck does not remain perfectly symmetric over a cycle, a nonzero time-averaged momentum exchange produces a directional jet and net thrust (37). The jet formation can be visualized from the motion of smoke injected into the resonator (Fig. 1C and movie S1). We 3D printed resonators with dimensions ranging from centimeter to micrometer scales, corresponding to resonance frequencies between 200 Hz to 40 kHz (Fig. 1D). The final dimensions of each resonator’s neck and cavity were determined by choosing a resonance frequency, predicted using numerical simulations in combination with the analytical model. The experimentally measured resonance frequencies agreed well with the theoretical predictions, with a maximum deviation of 5% (fig. S1E). Impact of design parameters on thrust The thrust generated by the resonator was measured using a precision scale from the apparent mass increase under acoustic excitation at resonance (Fig. 2A). The sound pressure in the workspace was continuously monitored using a microphone to correlate output thrust with input voltage. All measurements were conducted inside an anechoic chamber equipped with sound absorbers to minimize acoustic reflections and external interference (Fig. 2B and fig. S2). The impact of design parameters on thrust. (A) Schematic illustration of the overall experimental setup for thrust characterization inside the anechoic chamber. (B) Photograph of the measurement platform inside the anechoic chamber. (C) Plot of generated thrust against cavity volume for resonators with varying dimensions at 130 dB (n = 3). The R2 is 0.999. (D) Thrust measurement for different cavity geometries (n = 3). The spherical resonator is taken as the normalization factor for comparison. (E) Photographs of a soft resonator made of elastomer before and during external loading. (F) Effect of neck geometry on thrust generation. The resonator with a straight neck is taken as the normalization factor for comparison (n = 3). (G) Effect of wall thickness on the thrust generation. The thrust generated by the thickest neck (3.5 mm) is taken as the normalization factor (n = 3). n.s., not significant. First, we measured the thrust generated by resonators with different cavity volumes and observed that the thrust monotonically increased with resonator volume across the full measurement range. The results exhibit a clear linear scaling between thrust and cavity volume of the resonator [y = 0.044x, coefficient of determination (R2) = 0.999]. The fitted slope indicates that the thrust increases by ∼44 μN for every 1 cm3 increase in cavity volume (Fig. 2C). These results highlight the strong dependence of thrust on resonator size and provide a basis for estimating force output across a wide range of device dimensions. Next, we explored the influence of cavity shape on the generated thrust. Three resonators were fabricated with spherical, cubic, and cylindrical cavities, respectively. They produced similar thrust levels regardless of their cavity shape (Fig. 2D). This result is in accordance with the fact that when the resonator size is substantially smaller than the acoustic wavelength (i.e., kL = 2πL/λ ≪ 1), where k is the wave number, L is the characteristic size of the resonator, and λ is the wavelength of the applied sound), the cavity shape has a negligible influence on the generated thrust. This shape independence could facilitate the manufacturing of resonator arrays to increase the overall thrust output of a device. As a representative example, we fabricated arrays of resonators with cubic cavity, as this shape allows maximal packing efficiency (fig. S3A). Experimental results showed that thrust increases linearly with the number of resonators in an array, providing a design framework for scaling onboard thrust generation (fig. S3B). We measured the quality (Q) factor of two representative resonators fabricated using FFF and SLA techniques that were designed to operate at 300 Hz (fig. S4). The Q factor was calculated as Q = f0/(f2 − f1), where f0 is the center frequency and f1 and f2 are the lower and upper cutoff frequencies, respectively. SLA-fabricated resonator exhibited both higher thrust (1.4 times more) and higher Q factor (8.9 versus 12.2). These results show that the quality of printing is among the important factors influencing resonator performance and the resulting Q factor. To explore the impact of elastic modulus, E, on thrust, we fabricated a soft resonator from elastomer material (Fig. 2E). Resonators made of borosilicate glass with E = 64 GPa, polylactic acid (PLA) with E = 2 GPa, and elastomer with E = 5 MPa generated comparable thrust (fig. S5). This result opens the path to the realization of highly deformable soft microactuators powered by sound. Moreover, PLA and elastomer resonators are noticeably lighter than glass resonators, offering a higher thrust-to-weight ratio (fig. S5). The neck profile plays a critical role in thrust generation. Resonators without a neck produced substantially lower thrust, indicating that the presence of a neck is essential for effective thrust generation in Helmholtz resonators (Fig. 2F). Among the tested geometries, resonators with a straight neck generated higher thrust than those with converging or diverging neck profiles. These observations are consistent with thrust generation arising from nonlinear oscillatory flow at the resonator neck that produces a directional air jet, which is expected to be influenced by the neck geometry. We speculate that the air mass is free to move back and forth most effectively in a straight neck. Smoke visualization showed that straight and converging necks produce directional jets, whereas no-neck and diverging-neck configurations generate broader and more diffuse plumes (fig. S6). We investigated the effect of wall thickness and found that the generated thrust decreased with increasing wall thickness. Specifically, resonators with a wall thickness of 0.75 mm produced ∼35% more thrust than those with a wall thickness of 3.5 mm (Fig. 2G). To further study the cause of this effect, we fabricated resonators in which the cavity and neck thicknesses were varied independently while maintaining the same internal volume. Resonators with different cavity thicknesses but identical neck thicknesses generated comparable thrust, whereas variations in neck thickness at fixed cavity thickness led to significant changes in thrust (fig. S7). Together, these results indicate that neck thickness and geometry are the dominant parameters governing thrust generation. We postulate that having a thinner neck facilitates flow separation and vortex shedding at the neck edge, which can enhance momentum transfer and thrust generation. Thinner neck creates stronger shear layers and therefore more focused vortex rings, which would be expected to travel longer before being diffused. Further studies that involve FE simulations are required to reveal the exact mechanism. Acoustically powered rotors As a foundational machine element, we developed a rotor comprising four millimeter-scale resonators with a resonance frequency of fr = 1.2 kHz mounted on a central 3D printed rotor hub (Fig. 3A). The rotor hub was suspended from the top wall using a thread. One end of the thread was tied to the ceiling, while the other end was fixed to the bottom of the hub through a small flat washer. When excited by airborne sound, each resonator emits a localized acoustic jet that generates thrust, which is transduced by the structure into continuous angular motion (Fig. 3B). Following excitation at the resonance frequency, the rotor accelerates from rest and approaches a steady-state angular velocity within several seconds due to air drag (Fig. 3C). At the steady state, the thrust produced by a single resonator (Fthrust) is balanced by the combined aerodynamic drag acting on the rotor arm and the resonator (Fdrag,arm + Fdrag,resonator), which depend on the arm length l, thickness t, and resonator radius Rc (Fig. 3D). Fthrust was calculated to be 70.4 μN from the simulated total drag (Supplementary Methods), which is in agreement with the measured thrust of 71.9 μN (Fig. 2C). Thus, using our methodology, rotary machines can be designed from the first principles using the measured thrust and calculated drag. Acoustically powered rotors. (A) Photograph of a rotor showing four millimeter-scale resonators mounted on a hub that is suspended from the ceiling using a thread. (B) Schematic illustration of the rotation principle: the rotor rotates at an angular velocity ω in response to airborne sound, as a jet is emitted to generate thrust at the resonance frequency. (C) Angular velocity of the rotor shown in (A) over time. The maximum velocity is used as the normalization factor. (D) Schematic of the force balance on one rotor arm for aerodynamic force analysis. Thrust is counteracted by the sum of aerodynamic drag from the arm and the resonator. The arm has a square cross section with width w, arm length l, and resonator radius Rc. (E) Steady-state rotational speed as a function of sound pressure for three different rotors (n = 3). (F) Snapshots showing frequency-selective actuation of the three rotors under excitation at 1.2, 5.4, and 1.8 kHz. We built three different rotors that were designed to operate at different frequencies to demonstrate frequency-selective actuation. Steady-state angular velocity increased linearly with input sound pressure for all the three rotors driven at the resonance frequencies of 1.2, 1.8, and 5.4 kHz, respectively (Fig. 3E). The rotor driven at 5.4 kHz started to rotate at a higher sound pressure compared to the other two because thrust scales inversely proportional to the resonator size. On the other hand, this rotor accelerated more rapidly than the lower-frequency versions due to the reduced aerodynamic drag associated with its smaller resonator size. When excited sequentially at their respective resonance frequencies, the three rotors exhibited distinct frequency-selective rotation without noticeable interference from primary acoustic radiation forces or from one another, demonstrating the potential of the resonators for frequency-controlled control at small scales (Fig. 3F and movie S2). As a proof of concept, we fabricated an entirely soft rotor from an elastomeric material that operates at 1.2 kHz (fig. S8). The soft rotor achieved a rotational speed comparable to a rigid rotor, indicating its potential for rotary systems that require large deformation, collision resilience, and long life span under large cyclic loading (40–42). Miniature boats as acoustically powered mobile robots We printed boats that can house acoustic resonators to demonstrate the feasibility of wirelessly controlling the navigation of small-scale mobile robots using airborne sound. The first prototype had a single resonator with fr = 1.2 kHz; thus, it could sail in one direction (Fig. 4A). The boat carried a nut as a payload and progressively moved only when it was excited at 1.2 kHz (Fig. 4B). The propulsive thrust could be tuned by adjusting the input voltage to the speaker (Fig. 4C). The displacement curves became increasingly nonlinear at higher voltages, indicating higher acceleration under stronger acoustic excitation. Miniature boats as untethered mobile robots. (A) Photograph of a boat equipped with a single resonator for propulsion. (B) Time-lapse images of a boat moving forward under acoustic excitation at its resonance frequency. (C) Displacement of the boat as a function of the input voltage applied to the speaker. (D) Two identical boats crossing each other while moving in opposite directions under excitation at a single resonance frequency. (E) Photograph of a boat equipped with three resonators: a 1.2 kHz resonator for forward propulsion and 1.8- and 2.4-kHz resonators for steering. (F) Angular displacement of the boat in the clockwise (CW) and counterclockwise (CCW) directions. (G) Teleoperation of the boat along an infinity-shaped trajectory. The boat follows a predefined reference trajectory. (H) Time-lapse images demonstrating obstacle avoidance during teleoperation. The green arrowed curve indicates the paths followed, with the start and end positions labeled. We performed additional experiments to characterize the impact of primary acoustic radiation force. As a first demonstration, two identical boats with identical resonators (fr = 1.2 kHz) were positioned in opposing orientations (Fig. 4D). When excited at 1.2 kHz, the boats moved in opposite directions with the same velocity. Next, we assembled two boats, one equipped with a resonator and the other without. The boat without the resonator remained stationary, whereas the boat with the resonator exhibited controlled motion and was able to pass in front of the stationary boat (movie S3). These results indicate that the observed propulsion originates from resonator-mediated acoustic thrust and the primary radiation forces generated by the speaker are negligible under the operating conditions. To demonstrate fine control over translation and potential for assembly tasks, we decorated the two boats with complementary interlocking components and aligned them facing each other. The boats moved in opposite directions in a straight trajectory and assembled through overlap of the components to form the Swiss flag, without exhibiting noticeable lateral drift or unintended deflection during motion (movie S3). These results also show the feasibility of wirelessly powering mobile robots without prescribing a direction of motion enabled by the proposed resonator-based actuation scheme in air using a single broadband wave, which is often difficult to achieve with other actuation schemes, such as magnetic actuation. In magnetic actuation, magnetic dipoles exposed to the same applied magnetic field tend to align with the field direction, making independent control of motion extremely challenging, an issue that can be only partially addressed through the use of geometrically or magnetically distinct microrobots (43, 44) or implementing complex external magnetic field generators (45). In our actuation scheme, the orientation of the resonator determines the direction of motion and not the acoustic source. The second prototype was equipped with two resonators, exhibiting resonance at 1.2 and 2.4 kHz, positioned at the front and rear of the boat, respectively. This machine presented reversible, bidirectional motion through frequency selective excitation (movie S3). This result demonstrates that the thrust generated by each resonator is intrinsically directional and can be independently addressed through frequency modulation. To demonstrate teleoperated navigation of a miniature mobile machine, we constructed a boat equipped with acoustic resonators operating at three distinct resonance frequencies: 1.2 kHz for forward propulsion, 1.8 kHz for clockwise (CW) rotation, and 2.4 kHz for counterclockwise rotation (Fig. 4E). The boat rotated with different angular velocities of 17° s−1 and 15° s−1 (Fig. 4F) where slower motion in the CW direction can be attributed to the smaller thrust generated by the 2.4-kHz resonator (∼5 μN) compared to the 1.8-kHz resonator (∼15 μN). Using the calibration data, the boat was teleoperated to follow a predefined infinity-shaped trajectory, visualized by a reference line attached to the bottom of the water tank (movie S4). Accuracy was quantified by calculating the tracking error between the reference trajectory and the measured path of the boat (Fig. 4G). The mean and maximum tracking errors were 5.70 and 17.39 mm, respectively, indicating that the boat can be controlled with good positional accuracy along a prescribed path under visual feedback. In addition to trajectory tracking, the boat demonstrated a high level of maneuverability in constrained environments with obstacles. This frequency-selective actuation enabled the boat to navigate around multiple obstacles without contact, demonstrating effective obstacle avoidance driven by directional thrust control (Fig. 4H and movie S4). Structure-borne resonance for onboard acoustic actuation We hypothesized that structure-borne sound, which is generated when an acoustic transducer is directly attached to the resonator surface, could generate an equivalent acoustic jet (Fig. 5A). If true, then the machine could be actuated by on-board transducer(s), relaxing the dependence on external speakers. Experiments with differently shaped resonators revealed that resonance could be achieved under structure-borne excitation when the resonator was placed on a flat surface perpendicular to the longitudinal axis of the neck (fig. S9). On the basis of this finding, we designed a bell-shaped resonator featuring an ellipsoidal cavity with a flat bottom surface. This geometry was selected to approximate a spherical cavity while providing a planar interface for transducer attachment. When a transducer was attached such that its oscillation direction was aligned with the longitudinal axis of the neck, high-amplitude oscillations were observed, and a pronounced jet was emitted, consistent with Helmholtz resonance (Fig. 5B). Structure-borne excitation for onboard actuation. (A) Schematic illustration of Helmholtz resonance achieved through contact vibration and sound propagation. A transducer is directly attached to a flat surface of the resonator, and the resonator emits acoustic jets when driven at resonance. (B) Photograph of a bell-shaped resonator emitting a jet under resonance through structure-borne excitation. (C) Thrust generated by three different resonators at varying driving voltages (n = 5). (D) Exploded schematic view of an untethered boat with onboard electronics and transducer. The system consists of three resonators for propulsion and steering, two transducers for excitation, a battery, control electronics, and a boat hull. (E) Photograph of the fully assembled untethered boat. (F) Path-following demonstration of the boat, tracing the letters “EPFL” through frequency-selective actuation. The yellow curves indicate the paths followed, with the start and end positions labeled for each letter. (G) Time-lapse images demonstrating obstacle avoidance. Representative time stamps and corresponding excitation frequencies are highlighted. We explored several types of commercially available transducers including coin vibration motors, piezoelectric transducers, magnetic voice coils, and linear resonant actuators (LRAs). Among these options (table S1), we selected an LRA for our application based on a balanced trade-off between operating frequency range, size, vibration amplitude, and cost. The effective operating frequency window of the selected LRA spans from 230 to 700 Hz. On the basis of this range and considering the size variations of resonators at different frequencies, we 3D-printed three resonators with resonance frequencies of 455, 540, and 630 Hz for the teleoperated navigation of the boat. Before assembling the boat, we characterized the thrust generated by each resonator. In accordance with earlier results (Fig. 2), the generated thrust increased proportionally with the input driving voltage and the largest resonator produced the highest thrust (Fig. 5C). Notably, the achieved thrust levels were comparable to those obtained using airborne acoustic excitation, demonstrating that structure-borne excitation can generate similar propulsion performance with all functional components integrated onboard. Furthermore, structure-borne excitation opens opportunities for actuating multiple resonators with a single transducer. We constructed a prototype consisting of four resonators interconnected by tree-like branches that act as waveguides (fig. S10). When excited by a single transducer, the system generated a net thrust comparable to the sum of the thrust produced by the four resonators when they were excited individually. Next, we developed a fully integrated remotely controlled boat with onboard power and actuation systems. The boat consists of a 3D-printed hull that accommodates three resonators (540 Hz for propulsion and 455 and 630 Hz for steering), two LRAs serving as transducers, a battery, and control electronics (Fig. 5, D and E). For steering, a single transducer is shared between the 455- and 630-Hz resonators. A key component of the system is the onboard electronics, which are implemented on a custom printed circuit board (PCB) powered by the battery (fig. S11). A boost converter steps up the battery voltage to 5 V to supply two dual H-bridge drivers, ensuring sufficient power and maximizing current delivery to the transducers. A microcontroller manages the signal chain by generating the drive signals and processing control commands received via a Bluetooth module. The system achieves a total communication latency of ∼30 ms, thereby enabling seamless, near real-time control. To showcase navigation capabilities of the boat, we commanded the boat to follow the letters of “EPFL” as a trajectory (Fig. 5F and movie S5). The boat demonstrated recognizable maneuverability with sharp turns, indicating a good level of precision and controllability of our control scheme for successfully completing given tasks. Beyond simple shape tracing, the reliability of the boat in a constrained environment was assessed by performing an obstacle avoidance task. Two stationary balloons were placed in the pool, and the vehicle was operated to perform an infinity-shaped trajectory around them (Fig. 5G and movie S5). This maneuver was repeated for four consecutive cycles to assess the repeatability of the locomotion. The prototype successfully navigated the narrow corridors between obstacles without collision in all trials. To test for robustness against disturbances, we challenged the teleoperated boat by blowing air or physical push. We could command the boat to reorient itself and complete the prescribed course (movie S5). These results show that the structure-borne excitation achieves precise and robust navigation of a mobile machine under teleoperated control. The boat can also be programmed for automated navigation using the onboard microcontroller. Automation was achieved through a sequential playlist of commands that specified excitation frequencies and durations for both forward motion and steering. By executing these commands in sequence, the boat was able to navigate a predefined “S”-shaped trajectory in an open-loop fashion while avoiding obstacles (fig. S12 and movie S5). Acoustically powered microflier via vertical thrust The unique capabilities of our actuation scheme are highlighted with further miniaturization. Acoustic focusing follows diffraction principles, with the achievable focal size scaling on the order of the acoustic wavelength. At ultrasonic frequencies, wavelengths are reduced to millimeter or submillimeter scale, enabling the fine confinement of acoustic power to spatially address individual actuators and machines. Moreover, as ultrasound lies beyond the human audible range, high-amplitude acoustic fields can be generated efficiently and safely. These characteristics have been exploited in applications such as acoustic levitation, which rely on precisely focused, high-intensity acoustic fields. We hypothesized that a microscale resonator can generate enough thrust to hover a micro aerial vehicle considering the following two points (Fig. 6A). First, advanced microfabrication techniques allow the fabrication of ultralight resonators with extremely thin walls and minimal structural mass, providing high thrust-to-weight ratio for vertical take-off. Second, focused ultrasound can generate high-pressure sound waves within a constrained and well-controlled workspace. We designed a proof-of-concept microflier composed of three microresonators operating at fr = 40 kHz and supported by a lightweight frame incorporating dual-disk central mounting holes, resulting in a total mass of ∼150 μg or 1.47 μN in weight (Fig. 6B). The mounting hole constrains lateral motion during characterization, ensuring stable vertical actuation. A resonance frequency of 40 kHz was selected because ultrasonic transducers at this frequency are commercially available, cost-effective, and well characterized based on prior work in airborne acoustic levitation. The microflier was fabricated monolithically using TPP of a photocurable polymer (Fig. 6C and fig. S13). Acoustically powered microflier via vertical thrust. (A) Schematic illustration showing the necessary dynamic conditions for a microscale resonator to take off. (B) Schematic of the downward-facing microflier with a central dual-disk structure consisting of three microscale resonators connected by supporting frames. (C) Optical image of an upward-facing microflier after microfabrication. (D) Simulated airflow around a resonator at resonance, showing pressure distribution within the cavity, and the velocity field near the neck. (E) Time-dependent thrust output at 40 kHz for different sound pressure levels. Asymmetry between the positive and negative force amplitudes results in a nonzero time-averaged thrust. (F) Time-averaged thrust as a function of sound pressure. (G) Photograph of the experimental platform, consisting of an ultrasonic phased array and a holding pin supporting the microflier. (H) Snapshots showing the microflier at the initial position, maximum height, and equilibrium height during flight, with the corresponding sound pressures. The focal region of the acoustic field remains stationary near the pin head. (I) Measured height and acceleration of the microflier as a function of time during take-off. (J) Schematic illustrating the force balance acting on the microflier during flight. As the microscale resonators generate forces on the order of micronewton, the direct experimental measurement of thrust is challenging. Instead, we performed numerical analysis to estimate the net force produced by a single resonator considering low Reynolds number flow. The simulated pressure field reveals a pronounced pressure buildup within the cavity in response to the incident acoustic oscillations (Fig. 6D and fig. S14), establishing a strong pressure gradient across the neck. This pressure gradient drives a localized air jet through the neck, as shown by the corresponding velocity field, which exhibits a concentrated high-velocity outflow at the neck exit (Fig. 6D). From the simulated flow field, the time-dependent force output was extracted over a range of sound pressure levels. While the instantaneous force is approximately sinusoidal, asymmetry between the positive and negative force amplitudes results in a nonzero time-averaged force on the order of micronewton (Fig. 6E). This net force increases linearly with increasing sound pressure amplitude (Fig. 6F) and can exceed the weight of the structure, supporting the feasibility of acoustically driven vertical lift at the microscale. In our demonstration, the sound pressure at the focal region was measured as ∼300 Pa, corresponding to a thrust of 2.4 μN per resonator. With three resonators, this yields a thrust-to-weight ratio of ∼4.9. We built and programmed a 16 by 16 ultrasonic phased array comprising 256 transducers to provide sufficient acoustic power to the microflier through constructive interference of acoustic waves (Fig. 6G). The microflier was positioned at the center of the phased array, at the focus 3 cm above its surface, supported on a thin pin with the resonator necks facing downward to generate take-off under acoustic excitation. Upon acoustic excitation, the microflier moves up, reaches a maximum height, and subsequently settles into a stable equilibrium position (Fig. 6H and movie S6). The focus of the acoustic field remained fixed throughout the experiment; therefore, the observed motion arises from the intrinsic force balance of the microflier rather than direct primary acoustic radiation from the sound. At resonance, the vertical force balance on the microflier consists primarily of the thrust force generated by the resonant cavity–induced jet (Fthrust), gravitational force (Fweight), and aerodynamic drag (Fdrag), with a possible contribution from acoustic radiation force (Frad). For a spherical object of comparable mass under the same acoustic excitation, the radiation force is calculated to be on the order of a few nanonewtons (Supplementary Methods)—three orders of magnitude smaller than the thrust generated by the microflier. Consistent with this estimate, we experimentally demonstrated that spherical objects made of the same polymer remain unresponsive to the applied acoustic field and cannot be levitated, indicating that acoustic radiation forces are unlikely to play a major role in the observed take-off. To further strengthen this statement, we positioned the microflier upside down and subjected it to the same acoustic excitation (movie S8). In this configuration, no upward motion was observed. Together, we conclude that the contribution of Frad can be considered negligible in this system. Upon activation, the microflier undergoes rapid acceleration, reaching a peak value of ∼30 m s−2 within the first 10 ms (Fig. 6I), during which Fthrust > Fweight + Fdrag (Fig. 6J). As the microflier ascends, the effective thrust decreases due to increasing distance from the acoustic focal point and increased aerodynamic drag, leading to a gradual reduction in acceleration. The microflier reaches a maximum height of ∼12 mm before descending slightly and settling into a stable equilibrium position at ∼7 mm, where Fthrust = Fweight + Fdrag. At the equilibrium height, the measured acoustic pressure is 85 Pa. The total calculated Fthrust generated by three resonators at this pressure is 1.9 μN, which is close to the measured Fweight of 1.47 μN, and thus validating our thrust estimation. Acoustically powered microflier via aerodynamic lift Building upon the rotor study presented in Fig. 3, we designed a microflier capable of high-speed rotation for aerodynamic lift (Flift) generation. The combined use of a circular arc airfoil, Gurney flap augmentation (46), and precision microfabrication enables efficient lift generation at microscale, supporting stable aerodynamic operation. The microflier consists of three Helmholtz resonators with attached blades, symmetrically arranged around a central dual-disk hub, similar to the configuration used in the previous prototype (Fig. 7A). This dual-disk architecture enhances structural rigidity and suppresses wobbling, thereby minimizing unwanted lateral motion during rotation and enabling stable aerodynamic operation. The total mass is 184 μg, and the radius of the nanoprinted rotor is 3.36 mm (Fig. 7B). Acoustically powered microflier via aerodynamic lift. (A) Schematic of the microflier consisting of three resonators equipped with rotor blades and mounted on a central dual-disk frame with holes. A circular arc airfoil and a trailing-edge Gurney flap are implemented on each blade to enhance lift generation. (B) Optical microscope images of the fabricated microflier taken from the top and side views. (C) Lift and drag coefficients (CL and CD) as functions of angle of attack for blades with and without Gurney flaps. Peak lift-to-drag performance is achieved near an angle of attack of 20°. The inset shows a magnified microscopic view of the Gurney flap on a fabricated blade. (D) Simulated airflow under 40-kHz excitation. The velocity contour indicates the formation of a high-speed jet emitted from the resonator at resonance. (E) Simulated airflow around the microflier during operation, shown as a velocity contour. The spatially distributed flow indicates aerodynamic interaction between the three blade-resonator units, contributing to the overall lift generation. (F) Schematic illustrating the force components acting on the microflier. Aerodynamic forces on a single blade and a resonator are shown for brevity. (G) Snapshots showing the microflier at the initial position, maximum height, and equilibrium height during operation. The focal region of the acoustic field remains stationary. (H) Measured rotational speed of the microflier as a function of time. (I) Equilibrium height and corresponding rotational speed reached by the microflier at different input voltages (n = 3). (J) Vertical displacement and acceleration of a representative microflier during flight. The blades were engineered to maximize aerodynamic lift under low Re conditions. Each blade adopts a circular arc airfoil profile and incorporates a small trailing-edge Gurney flap, as shown in Fig. 7A. The circular arc airfoil was selected due to its favorable pressure distribution and delayed flow separation in this regime compared with conventional airfoil profiles (47). Each blade tapers toward its tip and is mounted on top of a Helmholtz resonator, forming an integrated blade-resonator unit. The Gurney alters the near-wake shear layer (47, 48), resulting in a systematic increase in lift coefficient (CL) across a broad range of angles of attack with only a modest increase in drag coefficient (CD), and the peak lift-to-drag performance is observed at an angle of attack of ∼20° (Fig. 7C). Numerical simulations of the microflier revealed pronounced flow acceleration at the resonator neck, where acoustic resonance produces a high-speed jet. The velocity contour also shows the radially varying velocity of the blade due to its rotation (Fig. 7D). The air flow field exhibits a distributed wake structure beneath the device, formed by the combined action of the three resonator-blade units (Fig. 7E and movie S7), and the superposition of these flows generate a net downward momentum flux in the surrounding air. This system-level airflow pattern is consistent with the generation of an upward aerodynamic force acting on the microflier during operation. It also shows reduced pressure on the upper surface of the blades relative to the lower surface, indicating aerodynamic lift generation (fig. S15). Simulations were performed at various advance ratios (J) to evaluate the effect of forward velocity on the lift produced by the microflier (Materials and Methods and fig. S16). The results indicate that lift decreases as forward velocity increases. This trend can be attributed to a change in the blades’ effective angle of attack, which reduces aerodynamic efficiency and leads to premature stall and lower lift generation (49). This behavior is consistent with the blades having been optimized for maximum lift at J = 0 (static-thrust/takeoff conditions). When the microflier is under operation with a rapid rotational speed ω (Fig. 7F), each propeller blade element generates aerodynamic lift, which collectively produces a net thrust/lift force. For a blade element, the lift can be expressed as (50)Flift=12CLρV2A(2)where CL is the lift coefficient, ρ is the density of air, V is the local relative flow velocity, and A is the reference surface area. Frad is coming from the acoustic field, and Fweight is acting downward due to gravity. There exists also drag from the blades, but it acts as a moment due to symmetry. For the microflier to take off, Flift + Frad > Fweight should be satisfied, and since Frad is substantially smaller than Fweight, Flift must be larger than Fweight for the microflier to take off. The high-speed imaging of the rotary microflier showed similar flight dynamics as the one powered by vertical thrust where the microflier reached a maximum height shortly after take-off and descended to stably rotate at a final equilibrium height (Fig. 7G and movie S8). Throughout the flight, the focal region of the acoustic field remains stationary, indicating that the vertical motion of the microflier is governed by the balance of forces rather than by changes in the acoustic focal position. We recorded a rapid increase in rotational speed upon acoustic excitation, reaching a peak of ∼13,000 rpm before gradually decreasing to a lower steady-state value as the microflier approaches its equilibrium height (Fig. 7H). During take-off, the blades accelerate and generate increasing aerodynamic lift. Once the lift-to-weight ratio reaches unity, the microflier begins to ascend and continues accelerating over a finite distance, reaching a peak lift-to-weight ratio of ∼1.2 (Supplementary Methods). As the microflier rises, the acoustic pressure experienced by the resonators decreases, and the system ultimately settles into equilibrium, as the resonators produce less thrust at lower acoustic pressure. The equilibrium height increases with the input driving voltage applied to the phased array, indicating that higher voltages generate stronger acoustic pressure distributions, resulting in higher equilibrium heights (Fig. 7I). At each equilibrium height, the steady-state rotational speed remains nearly constant (∼6500 rpm) across different driving voltages, indicating that the aerodynamic lift balances the weight of the microflier. Notably, when the microflier was oriented upside down, it did not ascend (movie S8), providing experimental evidence that aerodynamic lift and not the acoustic radiation force governs the observed motion. The vertical dynamics of the microflier during flight are shown in Fig. 7J. Immediately after acoustic excitation, the microflier exhibits an increase in upward acceleration, reaching a peak within the first ∼25 ms. This initial acceleration corresponds to the onset of take-off, when the combined upward forces exceed the microflier’s weight. As the microflier ascends, the acceleration decreases and exhibits oscillatory behavior while remaining predominantly positive until ∼80 to 100 ms. During this interval, the height increases rapidly, indicating continued upward motion under a net positive force. The subsequent reduction in acceleration reflects increasing aerodynamic drag and the redistribution of forces as the microflier gains height, but the sound pressure level is lower, leading to less thrust from resonators. At later times (>150 ms), the acceleration fluctuates around zero and the height stabilizes, indicating a steady-state hovering condition at equilibrium. This dynamic behavior differs from the vertical thrust-driven case, in which the initial acceleration is substantially higher (Fig. 6I), resulting in a more impulsive take-off over a shorter timescale. As a maneuverability demonstration, we generated a nontrivial trajectory for the microflier using an ultrathin curved wire with a diameter 30 μm. We made the microflier follow the prescribed path by moving the focal region of the acoustic field along the wire (movie S9). This experiment shows that the microflier can be actively powered to fly within the full workspace of the transducer array provided that its 3D position can be traced. Last, we demonstrated the unconstrained flight of microfliers in air following take-off (movie S10). The microfliers propelled by vertical thrust exhibited rapid take-off and ascent, but they occasionally experienced flipping motions during flight (Fig. 8A). In contrast, the generation of aerodynamic lift through rotation consistently led to a stable flight (Fig. 8B). We postulate that the rapid spinning of the microflier with blades, which generates gyroscopic stabilization, helps to maintain flight orientation and resists rotational disturbances, similar to the stabilization mechanisms observed in helicopters and spinning disc toys. Unconstrained flight of microfliers. (A) Time-lapse images during the flight of a microflier propelled by direct vertical thrust. (B) Time-lapse images during the flight of a microflier propelled by aerodynamic lift. DISCUSSION Previous work in acoustic levitation have demonstrated that airborne sound can trap small, lightweight objects, allowing their suspension in mid-air (29). These studies have provided valuable insights into acoustic radiation forces and pressure gradients, forming the physical basis for using sound to directly manipulate objects in air. Our work extends the concept of acoustic manipulation toward active propulsion by harnessing cavity resonance to generate directional thrust and controlled motion. This approach broadens the scope of airborne acoustic actuation toward small-scale robots capable of autonomous operation in open environments as demonstrated by the navigation of a miniature boat. The presented onboard actuation system is constrained by the fixed geometry and predetermined frequency response of commercially available transducers, which offer limited flexibility in tuning operating frequency or frequency-selective controllability at small scale. Miniature vibration motors are typically centimeter to millimeter scale because they require a proof mass to generate sufficient force with a submillimeter stroke. Moreover, small lithium polymer batteries exhibit rising internal resistance and voltage sag when supplying currents on the order of 100 to 300 mA through a boost converter, as required by these actuators, thus defining a lower bound on battery size. Existing microelectromechanical systems (MEMS) technologies support small transducers such as piezoelectric micromachined ultrasonic transducers, thin-film lead zirconate titanate membranes, and lithium niobate–based microultrasonic structures, demonstrating that resonance frequencies and coupling characteristics can be precisely engineered for specific applications (51–53). As actuation systems scale down, the resonance frequencies of acoustic cavities and structural elements naturally increase, often reaching tens of kilohertz or higher, and MEMS fabrication processes are well suited for producing transducers matched to these regimes. Recent work has shown that very small batteries and wirelessly powered systems can now be built and integrated with small-scale robots actuated by piezoelectric devices (54–56). Pairing microfabricated transducers with these compact power sources can greatly improve frequency control, energy transfer, power efficiency, and overall system scalability. Airborne excitation is particularly attractive for the actuation of extremely small and light-weight microfliers as they are only expected to contain cavities for actuation. However, this technique relies on the power and precision of the external hardware to generate sufficiently strong sound field continuously at the position of the microflier, and the overall workspace will be constrained by the actuation and imaging system. Structure-borne excitation is fully compatible with onboard actuation and autonomous navigation, but it comes with substantial practical challenges associated with integrating compact transducers, power supplies, and sensors while maintaining a weight that is smaller than the lift. We introduced two alternative acoustic propulsion mechanisms for microscale flight: vertical thrust and rotary lift. Vertical thrust-driven propulsion enables rapid and robust take-off through direct momentum transfer while aerodynamic lift supports gradual ascent, stable equilibrium flight, and tunable hovering behavior. The distinct acceleration profiles and kinematics highlight the importance of matching propulsion strategy to the desired flight regime. The designs of both machines were carefully tuned for minimizing weight and drag and maximizing propulsion. The first microflier reached a thrust-to-weight ratio of 4.9 and the second one rotated as fast as 13,000 rpm. Unlike magnetic actuation, which requires ferromagnetic or magnetically doped materials that add considerable weight, our approach is largely material independent. Similarly sized magnetic fliers weigh three orders of magnitude larger than ours (57). Nevertheless, there is still room for improvement by considering lower density materials and topological optimization of structural elements. Our acoustic powering system operated at a single frequency of 40 kHz, which limited the degrees of freedom in actuation. A multifrequency phased array built using transducers with multiple effective operating frequencies could enable selective excitation of different resonators and dynamic control of thrust direction for steering and coordinated motion during flight. Furthermore, Langevin transducers (58), composed of a bolt-clamped stack of piezoelectric rings, represent one of the most powerful and efficient ultrasonic actuator designs, capable of sustaining high acoustic intensities and continuous operation. The development of an acoustic powering platform incorporating multiple Langevin transducers could provide strong, multifrequency sound fields for actuating small-scale robotic machines in air through traveling waves. Considering further miniaturization of the microflier, state-of-the-art 3D nanoprinting systems typically offer around ∼700-nm resolution. Therefore, the fabrication of Helmholtz resonators at several micrometer scale is feasible. However, the Reynolds number will become smaller, and viscous effects will start to dominate over inertial forces, diminishing thrust generation through energy dissipation within thermoviscous boundary layers. The linear scaling of thrust with respect to cavity volume shown in Fig. 2C may not hold in this regime. Inspired by the locomotion of small-scale acoustic microrobots operating in low Reynolds number regimes (59), alternative propulsion strategies may need to be explored for further miniaturization of micro aerial vehicles. We experimentally demonstrated that our actuation scheme is highly versatile in material selection, functioning effectively across a wide range of elastic moduli from a few megapascals to several gigapascals, provided that an internal air cavity is maintained to sustain resonance. Dense arrays of microbubble cavities can be integrated into flexible substrates for distributed actuation (60, 61). We can envision similar strategies to achieve analogous capabilities in air. By embedding multiple Helmholtz resonators of varying geometries within a monolithic flexible structure, it may be possible to realize spatially programmable acoustic responses, enabling localized actuation, deformation, or vibration modes tuned by frequency. These configurations could facilitate aerodynamic shape-morphing robotic devices that respond to externally applied sound fields. Moreover, acoustic resonators may become instrumental in the realization of small-scale mechanical metamaterials (62). Compared to reversible magnetic switching of physical binary elements (bits) one at a time (63), sound waves can actuate many bits simultaneously through frequency modulation. Beyond conventional materials, sustainable microrobots may also be realized by repurposing recyclable or biofriendly materials. For example, researchers have demonstrated small-scale robots fabricated by 3D printing of sugar (64), which is a natural material that can be readily dissolved. In addition, biodegradable microrobots have been fabricated using TPP (65), and recyclable photoresins have been developed for sustainable microscale additive manufacturing (66). MATERIALS AND METHODS Fabrication of resonators All polymer resonators were 3D-printed in house, while the borosilicate glass resonator was fabricated via glassblowing by a glass manufacturer (Schmizo) according to our specifications. The polymer resonators were designed using computer-aided design software (SolidWorks 2024, Dassault Systèmes) and produced using the following additive-manufacturing techniques. Centimeter-scale PLA resonators were printed using the FFF method (Prusa MK4S); rigid, millimeter-scale resonators were printed using DLP with a high-resolution photopolymer resin (RC70, EnvisionTEC); and soft resonators were fabricated using SLA with an elastomeric polymer (BioMed Flex 80A, Formlabs). The microfliers were microfabricated from a photocurable resin (IPX-Q) using a TPP nanoprinting system (Photonic Professional GT+, Nanoscribe) with a 10× objective. The microfabrication protocol involved cleaning of a silicon substrate (25 mm by 25 mm) with acetone and isopropyl alcohol (IPA), drop-casting the resin onto the substrate, printing of the structures, and developing the printed structures in propylene glycol methyl ether acetate for 2 hours, followed by a 30-min rinse in IPA. Acoustic instrumentation and imaging inside the anechoic chamber All experiments involving airborne sound at audible frequencies were conducted in an anechoic chamber equipped with 1-m-thick sound-absorbing wedges, providing effective absorption above 80 Hz according to the quarter-wavelength criterion (fig. S2). For measurements between 100 and 1000 Hz, a bass loudspeaker (MB15X351, RCF) housed in a custom-built enclosure was used. For frequencies above 1 kHz, a horn speaker (MRD-480 N, Monacor) was used. The speakers were driven by an amplifier (PA-9100D, Monacor) that received its input from a signal generator (JDS6600, Joy-IT). A free-field microphone (Nor-1225, Norsonic) connected to a calibrated sound level meter (Nor-118, Norsonic) continuously monitored the sound pressure level inside the chamber. A precision scale (EMB 200-3, Kern) measured the thrust generated by the resonator. The measured mass change was converted to force (N) by multiplying the mass (kg) by the gravitational acceleration of 9.81 m s−2. A digital single-lens reflex camera (D7200, Nikon) was used to record all experiments conducted inside the chamber. For high-speed imaging, we used a high-speed camera (FASTCAM NOVA S20, Photron) equipped with a macro lens (105 mm F2.8 EX DG OS HSM, Sigma). An additional light source (SL100D, Godox) provided sufficient illumination during recording. Videos were recorded using the accompanying imaging control software (PFV4, Photron). To measure the resonance frequency of resonators, a small microphone (smartLav+, RØDE) was inserted into the cavity. A frequency sweep was then performed over the predicted resonance band. At the resonance frequency, the pressure amplitude inside the cavity reached a maximum, which appeared as a distinct peak in the sound pressure level spectrum obtained from the microphone measurements. All equipment inside the chamber was monitored and controlled by an operator outside. The sound pressure level was converted to sound pressure using Eq. 3 p=p010SPL20(3)where p is the root mean square sound pressure, SPL is the sound pressure level in dB, and p0 is the reference sound pressure in air (20 μPa). Flow visualization Smoke was generated by thermal vaporization of a glycerin-based fluid using a commercial smoke machine (The Fog Meister, Wasteland Industries) for the visualization of the fluid flow at the neck of the resonators. High-speed imaging was conducted at a frame rate of 5000 fps to capture the jet dynamics under resonance. To isolate the evolving jet from background illumination, an image subtraction approach was used for the magnified view of the neck region. A background reference image, corresponding to a frame acquired before the onset of jet ejection, was subtracted from each frame using the Image Calculator tool. Experimental setup for rotors The resonators were mounted on the arms of a 3D-printed rotor hub through an interference fit. A thread (diameter: 0.06 mm; Daiwa J-Braid) was passed through the center of the hub and tied to the ceiling to suspend the rotor. Downward motion was constrained by securing the opposite end of the thread to a small washer. The horn speaker was positioned beneath the rotor to ensure uniform acoustic pressure at each resonator. Image processing To characterize the boat’s linear displacement under different driving voltages, the boat trajectory was manually digitized to ensure high accuracy by identifying the centroid of the boat (center of the nut) at discrete time intervals. The cumulative displacement was calculated as the Euclidean distance from the initial position. To suppress manual tracking noise and small fluctuations during stationary periods, a monotonicity constraint was applied to ensure that the displacement reflected only forward motion. Spatial calibration was performed using a factor of 0.23 mm pixel−1. To analyze steering performance, video recordings were acquired for each turning direction, and rotation angles were extracted using a custom Python script. Foreground objects were isolated by subtracting a median background computed from 100 uniformly sampled frames. The boat was tracked by detecting the two steering resonators via color thresholding, followed by morphological noise reduction and contour filtering based on circularity (>0.2) and radius (5 to 80 pixels). The boat orientation was defined as the vector perpendicular to the line connecting the resonator centroids, and the orientation angle was computed for each frame using a four-quadrant arctangent. The angular trajectory was normalized t = 0 and angle θ = 0, converted to absolute angular displacement to compare clockwise and counterclockwise rotations, and smoothed using a Savitzky-Golay filter (window size 11, polynomial order 3). For precision analysis in Fig. 4G, a reference navigation path was binarized and processed using morphological opening and closing and then skeletonized to obtain a one-pixel-wide centerline. The skeleton coordinates were used as reference points for deviation analysis. Boat tracking was performed on recorded videos by subtracting a median background computed from 200 evenly spaced frames, followed by foreground extraction using hue, saturation, and value (HSV) color segmentation and contour detection. HSV thresholds for the two resonators were manually selected from the first frame. Contours corresponding to the circular markers were filtered based on area, radius, and circularity, and the two most probable contours were identified as the front and rear markers. The boat position was defined as the midpoint between the two markers, with orientation given by the connecting line. Temporal continuity was enforced by rejecting contour pairs with abrupt changes in intermarker distance, radius, or orientation. Deviation from the reference path was calculated as the minimum Euclidean distance between the boat centroid and the nearest point on the skeletonized trajectory and converted to physical units using a spatial calibration factor of 0.213 mm pixel−1. Onboard electronics for boat navigation A microcontroller (Arduino Nano Every, ATmega4809, 16 MHz) was selected for its small form factor and sufficient input and outputs to implement multiplexed motor control. Wireless teleoperation was implemented using an HC-06 Bluetooth module (serial port profile), connected via universal asynchronous receiver/transmitter (UART). Power was provided by a single-cell lithium polymer battery (Amewi LiPo 3.7 V, 500 mAh, 25C). A regulated 5-V system rail was generated from the battery using a synchronous boost converter (TPS61088RHLR, Texas Instruments) for 3.7 V → 5 V conversion. The boost converter output was routed directly to the Arduino 5-V pin while also powering the Bluetooth module and two motor driver integrated circuits (DRV8833PW, Texas Instruments). A custom PCB served as the electrical hub of the boat and was designed using CAD software (KiCad). The board footprint was 33 mm by 60 mm and used a lightweight manufacturing stack: 0.8-mm substrate thickness, electroless nickel immersion gold surface finish (for oxidation resistance in humid/water environments), and a standard two-layer stack-up. A routing oversight in prototype V1.0 (missing 5-V connection from boost output to the DRV8833 VM pins) was corrected on the physical board using external wire jumpers, and the fix was later integrated into the design files. Fine-pitch surface mount devices (SMDs) were assembled using manual solder paste deposition and pick-and-place and then reflowed in a vapor phase oven; larger SMD and through-hole components were soldered manually using conventional stations. Firmware on the microcontroller generated the actuator drive signals and processed Bluetooth commands in real time. To maintain stable frequency generation while concurrently servicing UART, the firmware avoided blocking delays and instead used a nonblocking scheduler based on the internal microsecond counter, toggling H-bridge polarity by comparing the current time against a target timestamp that advanced by the desired half-period after each toggle. Actuation was performed using two LRAs as contact transducers (VLV221007E, Vybronics). For power budgeting, the LRA impedance was measured at ∼17 Ω, and the current draw was experimentally observed to be ∼200 mA per active channel at the fixed 5-V drive. A conservative worst-case estimate assumed simultaneous activation of four channels across the drivers (I ≈ 1.6 A). The LRAs were driven by two dual H-bridge motor driver integrated circuits (DRV8833PW, Texas Instruments). As the microcontroller offers a limited number of hardware pulse width modulation (PWM) pins, the control board implemented a shared-bus topology for the H-bridge (DRV8833, Texas Instruments) inputs. Both H-bridge chips shared the same four PWM lines (D3, D5, D6, and D10), and the active driver was selected using the nSLEEP pins as chip-select lines [Actuator/driver enable on D7 and D15 (MISO/A1)]. This approach avoided the PWM pin bottleneck. Teleoperation was conducted from an external computer using keyboard commands transmitted over the HC-06 Bluetooth link. The command map on the keyboard was: “w” toggled the rear thruster at 540 Hz (forward thrust), “s” toggled the steering actuator, “a” set steering to 455 Hz (left turn), and “d” set steering to 630 Hz (right turn). The total system latency (Ttotal) was estimated as Ttotal=TUART+Tdead+Tair where TUART is the UART transmission time, Tdead is the firmware safety deadtime, and Tair is the Bluetooth air-interface latency. For a baud rate of 9600 baud and a 10-bit UART frame, the transmission time for a single command byte is TUART = 10/9600 ≈ 1.04 ms. The firmware introduces a fixed deadtime of Tdead = 5 ms during state transitions, while reported Bluetooth serial port profile transmission latencies range from 15 to 25 ms under nominal conditions. For automated navigation, a custom graphical user interface was developed in Python using the tkinter and pyserial libraries. Navigation trajectories were defined as text command sequences and transmitted to the onboard microcontroller via the Bluetooth module. Ultrasonic phased array setup The 16 by 16 phased-array system consisting of 256 ultrasonic transducers (40 kHz) was constructed following a well-established protocol for ultrasonic phased arrays (67). The transducers (MA40S4S, Murata Manufacturing) were mounted on a custom two-layer PCB and were individually driven using high-voltage driver-integrated circuits together with cascaded shift-register control lines. A field-programmable gate array module (CoreEP4CE6) provided phase-resolved control of all channels, which enabled precise generation of programmable phase delays required to form an acoustic focal region through constructive interference of acoustic fields. The input voltage was controlled using an adjustable power supply. To estimate the acoustic pressure within the workspace of the phased array, a receiver transducer (MA40S4R, Murata Manufacturing) was used as a pressure probe. The transducer was connected to a computer-based oscilloscope (PicoScope 5443D, Pico Technology) to record the voltage signal generated by the incident acoustic field. The recorded voltage signal was converted to acoustic pressure using a sensitivity of 7.08 mV Pa−1. The microfliers were mounted on an insect pin (200 μm in diameter), which was attached to a 3D-printed holder containing an embedded magnet to secure the pin. The pin position was precisely adjusted in a 2D plane using two linear stages with 500-μm travel resolution (XR25C/M, Thorlabs). For the maneuverability experiment, a 30-μm gold-plated copper wire was used. Numerical simulations Acoustic resonance analysis of resonators Acoustic analysis was performed using COMSOL Multiphysics (v6.2) with the Acoustics Module. The resonance frequency of each single-neck Helmholtz resonator was first estimated using a classical lumped-parameter formulation implemented in a 2D axisymmetric acoustics configuration. In this approach, the air in the neck is treated as an acoustic inertance and the air in the cavity as an acoustic compliance, an approximation valid when the acoustic wavelength greatly exceeds the resonator dimensions, so that the cavity pressure remains spatially uniform. Neck radius, neck length, and cavity volume were parameterized directly in the axisymmetric model to obtain an analytical-numerical estimate of the expected resonance. To validate this estimate, a frequency-domain FE simulation was carried out by sweeping the excitation frequency across the anticipated resonance band, and the fundamental resonance was identified from the peak acoustic pressure response in the cavity. This modeling strategy is consistent with established work in the literature (68), which reports excellent agreement between lumped-element predictions and FE simulations for Helmholtz resonators, with deviations below 1% in standard geometries. Computational fluid dynamics simulations for resonators Computational fluid dynamics simulations were carried out in ANSYS Fluent using a finite-volume formulation. The 3D, unsteady Reynolds-averaged Navier-Stokes (RANS) equations were solved with the k-ω shear-stress-transport (SST) model including the low Reynolds number correction. The 40-kHz resonator was placed at the center of a cylindrical air domain. A time-harmonic pressure inlet operating at 40 kHz was prescribed at the bottom face to excite the resonator, while all remaining external boundaries were treated as nonreflecting pressure outlets. The 3D, unsteady, compressible Navier-Stokes equations were solved using the pressure-based coupled solver in transient mode. A uniform time step of Δt = 2.5 × 10−7 s (100 timesteps per acoustic cycle) was used. The computational domain was discretized using an unstructured polyhedral mesh containing ∼6 million cells. The net force on the resonator was obtained by surface integration of pressure and viscous stresses over the entire resonator surface. The acoustic amplitude was varied between 130 and 142 dB to investigate its effect on the generated thrust. Simulations were advanced for 15 acoustic cycles to ensure convergence to a repeatable periodic state before data analysis (Supplementary Methods). Computational fluid dynamics simulations for microfliers The 2D steady-state RANS simulations were conducted to evaluate the aerodynamic performance of the circular arc airfoil with and without a Gurney flap. The pressure-based solver was used in conjunction with the k-ω SST model with low Reynolds number correction. The angle of attack was varied from 0° to 22° to obtain lift and drag polars for both configurations. The microflier was positioned at the center of a cylindrical fluid domain. Rotation was modeled using a rotating mesh approach, with the rotor driven at a constant speed of 6500 rpm. The top boundary of the domain was prescribed as a velocity inlet, while all remaining external boundaries were specified as pressure outlets. Simulations were performed in transient mode using the pressure-based coupled solver. A time step of Δt = 9.23 × 10−7 s was used. An unstructured polyhedral mesh consisting of ∼7.5 million cells was generated, with local refinement near the microflier to resolve rotor blade and near-body flow features. Simulations were conducted for advance ratios J ranging from 0 to 0.5, defined as J = v/(sD), where v is the freestream velocity, s is the rotational speed, and D is the propeller diameter; the freestream velocity was adjusted accordingly (fig. S16). Aerodynamic forces and moments were computed by surface integration of pressure and shear stresses over the microflier surfaces. In a separate set of simulation, the rotating microflier was modeled in the same cylindrical domain under acoustic excitation. The bottom surface was prescribed as a time-varying pressure inlet, while all other boundaries were treated as nonreflecting pressure outlets. The inlet pressure waveform was defined asP(x,y,z,t)=75cos(2π⋅40000t+2π⋅40000⋅x2+y2+9×10−4340)(4)where x, y, and z are spatial coordinates and t is time. The platform produces an acoustic focus ∼3 cm above the inlet, corresponding to the position of the microflier. Acoustic focusing is achieved by phase-modulating the inlet pressure distribution, such that waves emitted from different inlet locations arrive at the focal point constructively. Specifically, by subtracting the initial phase delay 2πd/λ, where d = x2+y2+9×10−4 is the propagation distance from the inlet to the focus and λ is the acoustic wavelength, all waves reach the focal location in phase. This results in constructive interference and a localized pressure maximum. The excitation frequency was set to 40 kHz and the speed of sound was taken to be 340 m s−1 (Supplementary Methods). Statistical analysis Statistical analyses were performed using OriginPro 2023b (OriginLab). Comparisons between two independent groups were conducted using unpaired two-sample t tests with Welch’s correction, which does not assume equal variances. All tests were two-tailed, and a significance level of P < 0.05 was applied. Experimental measurements are reported as means ± standard deviation (SD) over n independent experimental repeats. Error bars in all graphs represent the SD across these n independent replicates. Statistical significance is indicated as follows: *P < 0.05; **P < 0.01; ***P < 0.001; and not significant, P ≥ 0.05.