Aerobat: Bat-Inspired Morphing-Wing Robotics

• Aerobat is a bat-inspired, morphing-wing robotic platform that integrates compliant mechanisms to replicate bat-like flight dynamics.
• It combines monolithic rigid–flexible fabrication, reduced-order aerodynamic models, and linkage-level actuation to optimize performance during wing expansion and retraction.
• The research demonstrates practical control strategies and observer-based disturbance estimation to enhance stability and maneuverability in dynamic flight.

Aerobat is a bat-inspired flapping-wing robotic platform developed at Northeastern University to investigate dynamic morphing flight, especially the coupling of wing plunging and armwing flexion–extension that expands the wing during downstroke and retracts it during upstroke (Lessieur et al., 2021). In the published literature, the name designates not only a particular morphing-wing robot with a maximum wingspan of about 30 cm, but also a broader research program spanning monolithic armwing fabrication, reduced-order and multibody modeling, unsteady aerodynamic and wake-based gait design, observer-based disturbance estimation, high-DOF actuation, guarded hovering, and linkage-level thrust regulation (Lessieur et al., 2021).

1. Research lineage and platform scope

Aerobat first appears as a morphing aerial co-robot whose core lifting and maneuvering capability comes from a bat-like armwing fabricated as a monolithic rigid–flexible mechanism (Sihite et al., 2020). Subsequent work extends that concept into a physical flapping robot, a wake-optimized numerical design, a high-DOF embodied-actuation testbed, a guard-stabilized hovering platform, an arm-mounted fluid–structure-interaction calibration rig, and a single-wing linkage-regulation test platform (Lessieur et al., 2021).

Published descriptions therefore cover several closely related embodiments rather than one frozen hardware instance. Reported scales and masses vary by prototype and modeling context: the 2021 mechanical design describes a robot with a maximum wingspan of about 30 cm; the 2022 wake-design study uses a numerical regime with 34 cm wingspan and 15 cm mean chord; the 2022 unsteady-aerodynamics model reports a total mass of 19.5 g; later force-estimation and guard-hovering studies report, respectively, a body weight of about 30 g and a 40 g vehicle with battery and microcontroller plus 20 g payload capacity (Lessieur et al., 2021, Sihite et al., 2022, Sihite et al., 2022, Gupta et al., 5 Aug 2025, Dhole et al., 2023).

Variant or context	Defining role	Representative paper
Armwing mechanism	Monolithically fabricated bat-inspired morphing armwing	(Sihite et al., 2020)
KS flapping robot	Physical testbed for plunging and flexion–extension	(Lessieur et al., 2021)
Wake-design Aerobat	Reduced-order gait and wake optimization platform	(Sihite et al., 2022)
High-DOF actuation platform	Tail-less 14-joint embodied-locomotion testbed	(Ramezani et al., 2022)
Guard–Aerobat	Hovering platform stabilized by six small thrusters	(Dhole et al., 2023)
Aerobat-β / Delta	Kinova-mounted FSI rig and linkage-regulation testbed	(Gupta et al., 2024, Ciampaglia, 20 Apr 2026)

Taken together, these papers indicate a unified objective: to compress bat-like articulation, compliance, and wake manipulation into a small aerial robot that can exploit morphology as part of the control architecture.

2. Mechanical architecture and morphing wing

The core mechanical element is the “kinetic sculpture” (KS), a morphing armwing built from several four-bar mechanisms that act as mechanical amplifiers, converting a single crank input into coupled shoulder plunging and elbow flexion–extension (Lessieur et al., 2021). At the robot level, the 2021 prototype uses a single brushless DC motor plus gearbox, a 75:1 reduction, and a geartrain of 7 spur gears to drive both wings symmetrically. Each wing is realized in that prototype with rigid acrylic linkages and metal hinges, but these are described as mechanically equivalent to an earlier PolyJet-fabricated monolithic KS (Lessieur et al., 2021).

The KS is organized as a chain of rigid “bones” corresponding to humerus, radius/ulna, and distal wing sections. Its principal output coordinates are the shoulder joint angle $\theta_s$ and elbow joint angle $\theta_e$, which together generate the defining bat-like pattern: wing expansion during downstroke and retraction during upstroke. In the design formulation, 38 geometric parameters are reduced to 30 after symmetry and fabrication constraints are imposed (Lessieur et al., 2021).

The fabrication logic is as important as the kinematics. The armwing was designed for monolithic PolyJet printing using rigid Vero White and flexible Agilus Black photopolymers, thereby integrating rigid beams and compliant flexural hinges into one part without discrete assembly (Lessieur et al., 2021, Sihite et al., 2020). Hinge candidates were tested at 1.3 mm and 2 mm thickness and at 50A, 70A, and 85A hardness. The reported best trade-off was FLX 9870 at 70A with 1.3 mm hinge thickness: 50A was too soft, while 85A at 2 mm was too stiff and brittle (Lessieur et al., 2021).

Aerobat’s morphing is not limited to bare linkage motion. The platform literature also describes a flexible printed-circuit-board wing membrane that reduces weight and supports onboard electronics; later work explicitly places sensors, motor drivers, communication modules, and a microcontroller on flexible wing structures, framing the wing not only as an aerodynamic surface but also as an embedded electromechanical substrate (Sihite et al., 2020, Lessieur et al., 2021).

3. Kinematics, dynamics, and aerodynamic formulations

The kinematic target used in the mechanical-design literature is explicitly bat-like. The desired shoulder trajectory is sinusoidal,

$\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$

while the elbow trajectory is a skewed sinusoidal function chosen so that the wing expands faster than it retracts and reaches full span near mid-downstroke (Lessieur et al., 2021). Linkage geometry is then optimized by minimizing the sampled tracking error

$\min_{\mathbf q} \; (\mathbf y^\top \mathbf y)/N$

subject to parameter bounds and geometric constraints, with a first stage for the humerus mechanism and a second for the radius mechanism (Lessieur et al., 2021).

Later dynamic models reduce Aerobat to five rigid bodies: a body, left and right proximal wings, and left and right distal wings. One representative generalized coordinate set is

$$q_d = \begin{bmatrix} p_B^\top & q_s & q_e \end{bmatrix}^\top,$$

with dynamics written in manipulator form as

$$M_d(q_d,R_B)\,a_d = h_d(q_d,\dot q_d,R_B,\omega^B) + u_a + u_m + u_f,$$

where $u_a$ denotes aerodynamic forces, $u_m$ motor torques mapped through the KS, and $u_f$ unknown external forces such as gusts or impacts (Gupta et al., 2024). Earlier roosting and perching studies use a richer joint set, including plunge, mediolateral shoulder motion, elbow extension, and feathering on each wing, and explicitly compute whole-system angular momentum to analyze body reorientation and upside-down maneuvers (Sihite et al., 2020).

Aerodynamic modeling evolves from quasi-steady blade-element formulations to reduced-order unsteady state-space models. The 2022 unsteady-aerodynamics paper combines Prandtl lifting-line theory with Wagner’s function, representing spanwise circulation by truncated Fourier series and introducing aerodynamic memory states so that lift buildup and wake effects can be embedded in a compact state vector $\boldsymbol{\zeta}$ (Sihite et al., 2022). The wake-design paper then elevates the wake itself to the definition of gait, introducing a desired wake structure $\theta_e$0 and an optimization

$\theta_e$1

over the zero-dynamics manifold, with wake evolution computed from reduced-order mechanics, unsteady aerodynamics, and Biot–Savart reconstruction (Sihite et al., 2022).

A further step is the explicit fluid–structure-interaction calibration pipeline. In Aerobat-β, the robot is mounted to a Kinova Gen3 arm through an ATI Nano17 load cell, and an unsteady lifting-line/Wagner model is tuned against 6-axis force–moment data gathered during banked trajectories. The tuned model is then embedded in a collocation controller for simulated 3D banking turns (Gupta et al., 2024).

4. Control, estimation, and embodied actuation

A recurring theme in the Aerobat literature is that the structure is part of the controller. The 2021 MIMIC framework—“Morphing via Integrated Mechanical Intelligence and Control”—treats morphology as a computational resource and uses small Feedback-Driven Components (FDCs) to change selected link lengths in the KS (Sihite et al., 2021). In that formulation, a baseline morphology

$\theta_e$2

is optimized for nearly stable forward flight, and pitch stabilization is obtained by

$\theta_e$3

with

$\theta_e$4

The result is simulation-level pitch stabilization without adding new large actuators (Sihite et al., 2021).

The 2022 high-DOF actuation paper generalizes that idea into a tail-less 14-joint platform actuated by “primers,” small SMA-based actuators embedded into a computational structure (Ramezani et al., 2022). A 40 mm-wide rhombic primer driven by 11 Flexinol loops is reported to yield about 1.04 mm stroke, and, when combined with an elbow mean angular sensitivity of $\theta_e$5, about 80° change in elbow angle. The same paper reports a reaction time of approximately 0.25 s at 55 mA for the SMA wires, which clarifies both the promise and the bandwidth limits of that actuation method (Ramezani et al., 2022).

Because Aerobat is tail-less and open-loop unstable, some papers move stabilization to an external structure. The Guard–Aerobat platform suspends the flapping robot inside a protective carbon-fiber guard carrying six small thrusters. The guard controller treats Aerobat’s unsteady aerodynamic output as an extended disturbance state $\theta_e$6 and uses an observer together with a feedback-linearizing law

$\theta_e$7

to stabilize hovering in the presence of flapping-induced disturbances (Dhole et al., 2023).

Force estimation became a separate line of work. A conjugate-momentum observer was adapted to Aerobat’s morphing-wing dynamics, separating known motor torque $\theta_e$8 and modeled aerodynamic loads $\theta_e$9 from unknown generalized forces $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$0 (Gupta et al., 2024). In simulation, this observer achieved

$\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$1

under Gaussian noise and step-like disturbances (Gupta et al., 2024). A later tethered study compared that physics-based observer with a multi-layer perceptron trained on 40 datasets collected over flapping frequencies of 2.5, 3, 3.5, and 4 Hz, wind speeds of 0.5, 1.0, 1.5, and 2.0 m/s, and pitch angles of $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$2. The reported RMSEs were 0.0401, 0.0696, and 0.1155 for the MLP on $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$3, versus 0.0467, 0.1059, and 0.1322 for the conjugate-momentum observer (Gupta et al., 5 Aug 2025).

5. Experimental evidence and maneuver studies

The empirical record is heterogeneous but technically coherent. Structural analysis of the morphing armwing used a hyperelastic Mooney–Rivlin model for FLX 9870 and found maximum strain of about 43% at onset of downstroke and about 30% at onset of upstroke, both comfortably below the reported elongation at break of 120–140% (Lessieur et al., 2021). Sensitivity analysis in the same work showed that some geometric parameters are low-sensitivity while others are high-sensitivity leverage points, supporting later regulator and primer strategies (Lessieur et al., 2021).

Wake-based gait design compared one-axis and three-axes morphing wings under common conditions of 1 m/s forward speed and 2 Hz flapping (Sihite et al., 2022). The Aerobat three-axes design produced smaller, weaker vortices during upstroke and larger, stronger vortices during downstroke, consistent with the design goal of reducing negative lift during upstroke while concentrating useful momentum flux during downstroke (Sihite et al., 2022). Experimental wake validation was not yet available in that paper, but the optimized gait was prototyped as a physical robot and tested with external power (Sihite et al., 2022).

The 3D banking-turn studies extend this line from wake optimization to trajectory control. In Aerobat-β, the robot was mounted on a Kinova arm and tested at a fixed pitch angle of $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$4, with roll angles of 10°, 15°, 20°, and 25° in one set of datasets and forward speeds of 0.7, 0.8, and 0.9 m/s at 15° roll in another (Gupta et al., 2024). The resulting force–moment data were used to tune the FSI model, and a collocation-based controller then maintained approximately 15° roll during a simulated bank while the pitch oscillated around $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$5; over 5 s, the simulated vehicle traveled about 4.1 m in $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$6 and about 5.0 m in $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$7 (Gupta et al., 2024).

A separate mechanical-regulation thesis addressed the absence of independent left/right thrust control in the single-motor platform by modifying the effective length of the first radius link $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$8 (Ciampaglia, 20 Apr 2026). Static tests using 3D-printed $\hat{\theta}_s = 35^\circ \sin(\phi) - 10^\circ,$9 links at 28.58, 29.33, and 30.08 mm and flapping frequencies of 3, 4, and 5 Hz showed that a 1.5 mm length increase produced a 37% increase in peak lift force and shifted peak force timing within the downstroke (Ciampaglia, 20 Apr 2026). After string-tension and micro-servo regulators failed, a TULA-50 piezoelectric slip-stick actuator was integrated into a direct-drive variable-length mechanism, which was demonstrated in a preliminary bench-top test but did not yet produce sufficient force for dynamic flapping tests (Ciampaglia, 20 Apr 2026).

6. Limitations, clarifications, and significance

A persistent clarification in the literature is that Aerobat is not yet uniformly presented as a fully autonomous, untethered, free-flying bat robot. The 2021 mechanical-design paper explicitly frames the platform primarily as a mechanical and structural design-and-analysis study and does not claim successful sustained free flight for that prototype (Lessieur et al., 2021). Later papers often rely on simulation, guarded hovering, arm-constrained FSI experiments, or tethered load-cell tests rather than unconstrained autonomous flight (Dhole et al., 2023, Gupta et al., 2024, Gupta et al., 5 Aug 2025).

The main technical obstacles are repeated across the corpus. Aerobat is tail-less and structurally unstable in pitch; active or passive stabilization is therefore required (Lessieur et al., 2021). High-DOF morphing increases control authority but also couples inertial and aerodynamic dynamics strongly, which complicates both identification and feedback design (Ramezani et al., 2022). Reduced-order aerodynamic models based on lifting-line theory and Wagner’s function are computationally attractive, but the same papers note omitted effects such as separation, leading-edge vortices, membrane deformation, and other fluid–structure interactions outside the reduced model class (Sihite et al., 2022, Gupta et al., 2024). The SMA-primer actuation framework offers large morphological leverage but a response time of approximately 0.25 s, which limits within-wingbeat control bandwidth (Ramezani et al., 2022).

Another recurring misconception is that morphing is treated merely as geometric imitation of bats. The published work instead uses morphing in several distinct technical roles: to minimize negative lift during upstroke, to define wake-based gaits, to alter the momentum balance for banking turns, to support observer-based disturbance estimation, and to regulate wing thrust by embedded linkage modulation (Lessieur et al., 2021, Sihite et al., 2022, Gupta et al., 2024, Ciampaglia, 20 Apr 2026). This suggests that, within the Aerobat program, morphology is not a decorative biological analogy but the primary locus where actuation, computation, and aerodynamics are intentionally coupled.

In that sense, Aerobat occupies a distinctive place in flapping-wing robotics. It combines monolithic compliant-mechanism design, bat-inspired armwing synergies, reduced-order unsteady aerodynamics, and control strategies that repeatedly shift complexity from high-power actuators toward linkage geometry, embedded compliance, and low-power regulators. The platform’s significance lies less in a single finished vehicle than in a coherent methodology for morphing-wing aerial systems: define biologically motivated kinematic objectives, realize them through computational structure, model the resulting fluid–structure coupling in low-dimensional form, and use that structure—rather than oppose it—as the basis for control.