Armadillo-inspired active morphing skeletons for soft machines Authors: Jianyu Zhou, Weixin Zhou, Seol-Yee (Jennifer) Lee, Shuang Wu, Ali Akbari, Yong Zhu Published: 2026 ABSTRACT Armadillos can rapidly reconfigure their body into a rigid, enclosed sphere in active response to external threats, combining adaptive shape change with robust mechanical protection, which is rarely achieved in engineered systems. This rapid reconfiguration is achieved through real-time sensing and muscle activation, while the synergistic coordination of the exoskeleton and endoskeleton provides structural stiffness and impact resistance. Inspired by this natural strategy, we present an active morphing skeleton called morpho-interlocking protective module (MIPM). The architecture integrates curvature-conforming segmented exoskeleton with a spine-inspired interlocking endoskeleton framework, forming a coordinated load-bearing skeleton. A muscle-like liquid crystal elastomer composite layer drives structural transformation, while an embedded sensing layer detects external threats and autonomously triggers localized Joule heating for actuation of the liquid crystal elastomer layer. This enables multimodal morphing behaviors such as curling, rolling, and grasping, without compromising the structural integrity. The MIPM withstands impact, puncture, and concentrated loading while carrying and protecting fragile payloads in harsh conditions. An integrated Bluetooth module facilitates wireless, untethered operation, broadening its applicability to hazardous or inaccessible environments. By combining adaptive morphology, real-time sensory feedback, and on-demand mechanical stiffening within a unified internal-external skeletal framework, this work presents a previously unidentified paradigm for concurrent morphing and protection, with broad applicability across adaptive systems ranging from soft robotics to next-generation flexible electronics. INTRODUCTION Soft machines—spanning flexible electronics (1–3), soft robotics (4–6), and morphable metamaterials (7–9)—pave the way for intelligent systems that can deform, conform, and reconfigure. However, the very compliance that enables their large, reversible deformations is at odds with the ability to provide mechanical protection. As a result, most soft machines remain highly flexible but vulnerable when exposed to external threats. This trade-off becomes especially limiting in extreme environments such as space exploration, search-and-rescue missions, and personal protective wearables, where adaptability must be accompanied by resistance against impact, puncture, or abrasion (1, 10–13). Reconciling this conflict is therefore essential for advancing soft machines into real-world applications. In both living organisms and engineered systems, defensive mechanisms are critical for safeguarding functionality against external disturbances (14–16). Throughout evolution, species have engaged in a continuous arms race of offense and defense, driving the development of increasingly protective skins, shells, and scales to resist predation and environmental hazards (17–19). Inspired by these strategies, researchers have developed a wide range of structural materials with outstanding mechanical properties: for example, nacre-inspired “brick-and-mortar” composites with improved strength and toughness (20–22), tooth- and enamel-inspired ceramics with interlocking nanorod architectures for extreme hardness (23, 24), and Bouligand helicoidal structures that enhance impact tolerance (16, 25). While these rigid designs provide excellent protection, their intrinsic lack of flexibility restricts their use in dynamic or complex environments. To address this limitation, researchers have turned to hybrid strategies designed to balance flexibility with protection. For example, bioinspired structures based on segmented armor of living organisms have shown how embedding rigid components in compliant matrix enables a balance between adaptability and mechanical strength (26–28). However, most engineered bioinspired protective systems primarily rely on passive structural reinforcement (29). In complex and dynamically changing environments, such systems exhibit limited multimodal morphing and lack fully integrated sensing-actuation loops. By contrast, natural organisms integrate structural features with dynamic shape transformations triggered by functional demands or environmental changes, both indispensable for survival. For instance, fish adjust body curvature to navigate turbulent flows (30–32), and snakes modulate posture and friction to traverse rough terrain (33, 34), both relying on flexible bodies covered with rigid scales to balance adaptability and protection. Particularly notable is the volvation defense of three-banded armadillo (Tolypeutes), which rapidly transits from an extended posture to a fully enclosed, protective sphere when threatened (35, 36). During volvation, the overlapping exoskeletal armor conforms to the body’s curved surface, while the interlocking endoskeletal spine imparts structural rigidity; together, they ensure robust protection against external threats (37, 38). This coordinated response exemplifies how living organisms couple structural design with stimulus-responsive morphing to seamlessly transition between mobility and defense. Drawing inspiration from the armadillo’s volvation defensive strategy, we developed a morpho-interlocking protective module (MIPM). Distinct from protective strategies that primarily rely on rigid scale coverage (39), MIPM integrates a curvature-conforming exoskeleton with a xenarthrous spine–inspired segmented endoskeleton, whose cooperative interaction provides both conformal coverage and deformation-activated internal load-bearing support. The muscle-like compliant composite layer, based on flexible sensing elements and liquid crystal elastomer (LCE) actuation, is not implemented as an independent component; instead, it is integrally coupled with the exo- and endoskeletons, thereby enabling coordinated structural responses to external hazards. Localized Joule heating drives targeted bending, supporting spatially and temporally programmable morphing behaviors—including curling, rolling, and grasping—without compromising structural integrity. Beyond passive shielding, the MIPM not only protects fragile payloads from puncture, impact, and concentrated loads but also supports controlled locomotion across hazardous terrains. Through sequential actuation, the module dynamically reconfigures its geometry in response to external hazards, allowing it to adapt to or evade environmental threats. By seamlessly integrating adaptive morphology, real-time sensory feedback, and coordinated exo- and endoskeletal protection within a single architecture, the MIPM establishes a biologically informed paradigm for next-generation soft protective systems. RESULTS Bioinspired active morphing exo- and endoskeletons In nature, animals such as turtles, crocodiles, armadillos, and snakes have evolved rigid shells or overlapping scales that provide exceptional mechanical protection to the underlying soft tissues (18, 35, 40–42). These biological structures exhibit substantially higher stiffness and impact resistance than the soft materials commonly used in soft machines such as LCE and polydimethylsiloxane (PDMS), making them compelling models for bioinspired protective design (fig. S1A). Notably, many armored organisms also perform volvation, a defensive behavior in which the body curls into a protective sphere, providing an additional layer of protection for vulnerable tissues inside (fig. S1, B to G). The three-banded armadillo (Tolypeutes) exemplifies this strategy as it curls into a near-spherical configuration when threatened, thereby minimizing exposed surface area and shielding soft ventral tissues with a rigid exterior (Fig. 1A). From natural armadillo armor to bioinspired MIPM. (A) Defensive volvation process of the three-banded armadillo (Tolypeutes) curling into a protective sphere. (B) Thermally induced morphing of the MIPM from planar to curled state, demonstrating shape reconfiguration (scale bars, 10 mm). (C) Schematic of the MIPM showing the layered configuration with exoskeletal, endoskeletal, and sensing-actuation components (exploded view). (D) Interlocking vertebrae and overlapping osteodermal bands of the (a) three-banded armadillo (Tolypeutes), compared with the endoskeleton–exoskeleton architecture and (b) curled morphology of the MIPM. (E) (a) Muscle tissues and (c) neural underlying the armadillo armor, responsible for actuation and sensing during curling, compared with the (b) thermally contractile LCE layer and (d) conductive silver nanowire (AgNW)/PDMS network integrated in the MIPM. Inspired by this natural defensive strategy, we developed an MIPM, which exhibits an analogous response: When the sensing layer detects external stimuli/threats, Joule heating in the LCE composite layer drives a controlled morphing response, transitioning the structure from flat to curled, providing effective protection (Fig. 1B and movie S1). In the flat state, the MIPM achieves functional integration through its layered architecture, as illustrated in the exploded view, which reveals the exoskeletal, endoskeletal, and sensing-actuation layers (Fig. 1C). The detailed fabrication process is shown in fig. S2. In the curled state, the exoskeletal segments conform to the curved surface to form a rigid outer shield, while the intermediate sensing-actuation layer enables both stimulus detection and active actuation. Beneath these layers, the segmented endoskeleton interlocks to provide structural support (fig. S3A). The three-banded armadillo defends itself through a sensing-processing-actuating loop, triggering muscular contraction for rapid curling into a protective sphere. The MIPM follows the same logic loop; external stimuli are detected by the sensing layer and processed by a control circuit, which turns on Joule heating and activates the LCE composite actuator, driving the transition from the flat state to the curled, protective state (fig. S3B). A key feature in the MIPM is the seamless integration of rigid and compliant components within a unified framework. The rigid components are inspired by the skeletal architecture of the three-banded armadillo, in which shield-like osteodermal bands (carapace) form a robust external barrier, while the vertebral spine, reinforced by xenarthrous articulations, provides the internal structural support [Fig. 1D(a)]. A magnified view of the armadillo skeleton further highlights the protective role of the osteodermal shield against penetration by sharp teeth and claws, together with the interlocking function of the xenarthrous spine that enhances the overall structural rigidity. Analogously, the curled configuration of the MIPM illustrates a comparable strategy in which rigid modules cooperate with soft constituents to realize morphable protection [Fig. 1D(b)]. The MIPM incorporates an exoskeletal layer that conforms to the curved surface during curling to form a continuous protective barrier, while the endoskeletal framework is structurally designed to achieve geometric self-locking; their cooperative interaction enhances stiffness and improves resistance against external threats. Soft components endow the MIPM with the capacity for deformation and adaptive function. Abundant and flexible muscle tissues are distributed beneath the osteoderms of the armadillo (43), enabling agile motion and supporting diverse behavioral responses [Fig. 1E(a)]. In the MIPM, this function is achieved through a multilayered soft composite consisting of a prestrained LCE layer (44, 45), a silver nanowire (AgNW)/PDMS composite layer (46), and a Kapton [polyimide (PI)] tape, together with a conductive fabric–based serpentine trace placed beneath the PI tape [Fig. 1E(b)]. Upon Joule heating of the conductive trace, the multilayered soft composite bends following the thermal bimorph mechanism (47–49); more specifically, the LCE layer shrinks while the AgNW/PDMS and PI layers expand. Note that the AgNW/PDMS layer serves additionally as the embedded sensor to detect external stimuli, analogous to the sensory skin of the armadillo [Fig. 1E(c)]. AgNWs create a conductive percolation network embedded in the PDMS matrix, acting as a deformable resistor whose electric resistance varies with applied mechanical pressure [Fig. 1E(d)]. Patterning the AgNWs enables formation of distributed sensors (fig. S4) (50). By integrating the soft functional layers with the rigid protective modules, the MIPM unites adaptability with mechanical robustness, emulating the armadillo’s natural defense strategy. Structure design and actuation As mentioned above, actuation of the MIPM relies on the thermal bimorph effect with the LCE layer contracting while the PI and PDMS layers expanding under heat. This mismatch in the thermal expansion coefficient induces bending of the composite structure. As the MIPM bends, the barbed features on the endoskeleton interlock, enhancing structural stiffness (Fig. 2A and fig. S5). The dimensions of the segmental endoskeletal components and the origami-inspired interface geometry (linterface = n × latt + n × lfree) between the endoskeleton and the LCE layer must be optimized to ensure reliable interlocking after actuation. Corresponding lines of latt and lfree are shown in Fig. 2B. Here, the origami concept provides hinge-like joints between adjacent endoskeletal segments and establishes a flexible connection between the rigid endoskeleton and the soft LCE layer, thereby enabling coordinated deformation and robust interlocking. Endoskeletal structure and thermally actuated morphing of the MIPM. (A) Schematic illustration of the segmented endoskeletal structure of the MIPM and its interlocking transformation before and after deformation. (B) Side-view illustration of the assembled MIPM. (C) Design variations of internal geometries with tunable interlocking dimensions. (D) Joule heating performance of the laser-cut conductive pathway (heating layer). (E) Heat transfer between multilayered components during Joule heating. (F) Comparison of deformation amplitudes of the MIPM under different applied currents. (G) Quantified curvature of the MIPM under different applied currents. (H) Actuation efficiency under different heating conditions. (I) Reconfigurability and stability over repeated thermal actuation cycles. An endoskeleton with n = 10 segments is selected as an illustrative example (Fig. 2B), because it yields the best mechanical performance in the following evaluations. The interlocked endoskeletons form an enclosed polygonal space within the structure, highlighting the potential of the MIPM for gripping and protecting payloads. Increasing the number of endoskeletal segments provides longer interface length and greater design redundancy for the LCE-endoskeleton interface, enabling a broader range of internal cavity sizes (Fig. 2C). In contrast, when linterface is less than the length of LCE (90 mm), the interface length cannot accommodate the full flat-to-curved transition (fig. S6). As a representative case with n = 10, the configuration of the full-length origami interface is illustrated (fig. S7A). In practice, the endoskeletal components cannot be fully bonded along the origami-inspired interface, because curling requires space between adjacent segments to accommodate their protrusions. Thus, the interface design must be adjusted to ensure both strong bonding and reliable interlocking (fig. S7B). The actuated configuration is shown in fig. S7C, where each endoskeletal segment with carefully designed length achieves complete interlocking. The functional relationship of the origami interface length as a function of the bonding of the internal skeleton and the origami interface (latt) and the radius of the circumscribed circle of the internal polygonal space (Rin), as defined in Fig. 2B, is presented in fig. S8. When n > 6, the full-length origami interface is sufficiently long to provide ample design redundancy for adjusting the interface dimensions. The final design dimensions corresponding to different n values are summarized in table S2. Morphing of the MIPM structure is actuated by Joule heating of the conductive trace owing to the thermal bimorph effect (Fig. 2D). Curvature of the bimorph as a function of temperature change can be given by the classical Timoshenko equation (51)k=6(α1−α2)(T−T0)(1+m)2h[3(1+m)2+(1+mn)(m2+1mn)]where m = t1/t2 with t1 and t2 as the thicknesses of the two layers (AgNW/PDMS/PI layer and LCE layer, respectively), h = t1 + t2, n = E1/E2 with E1 and E2 as the Young’s moduli of the two layers, T0 is the initial temperature, T is the temperature of the actuator, and α1 and α2 are the coefficients of thermal expansion of the two layers, respectively. The resulting curvature is governed by the thermally induced strain mismatch between the two layers and dimensionless parameters including the thickness ratio and modulus ratio between the passive and active layers. The temperature rise was measured layer by layer using a thermocouple. No substantial heat loss occurred across different layers (Fig. 2E). When heated, the LCE layer exhibited a maximum negative strain of up to 42.1% (fig. S9, A and B). When cooled, the LCE recovers its original length. As shown in fig. S9C, the LCE recovered 95.2% of the original length after 60 s of cooling. The recovering speed of LCE during natural cooling is also an important consideration for actuator design as it influences the temporal response of the actuator and the period of the actuation cycles. An LCE strip was heated to 150°C and then allowed to cool naturally in ambient air at room temperature. As the current applied to the conductive trace increases, the bending curvature increases. The ultimate shape of the LCE/PDMS/PI composite material under different currents is shown in Fig. 2F. The experimental results agree well with the finite element method (FEM) simulations. Notably, when the applied current is too high (e.g., 0.4 A), excessive contraction of the LCE layer may cause interfacial delamination between LCE and PDMS, preventing the structure from achieving the bending curvature predicted by the FEA (Fig. 2G). For a given current input, the bending curvature increases over time. Once the target curvature is achieved, two small magnets (2 mm in diameter) located at both ends of the MIPM attract each other, locking the ring configuration and allowing the Joule heating to be turned off. At a low current (e.g., 0.1 A), the driving force generated by the LCE is insufficient to induce noticeable deformation, making the actuation barely observable. A current of 0.3 A was found to provide a relatively fast actuation without interfacial failure. Therefore, 0.3 A was selected for actuation in all subsequent experiments (Fig. 2H). The MIPM demonstrated stable cyclic performance over 100 repeated deformation cycles (Fig. 2I). The high repeatability and stability of the thermally driven actuation establish a solid foundation for the reliable mechanical performance discussed in the following section. Mechanical properties of MIPM In nature, after curling into a ball, an armadillo may face crushing or puncturing loads from predators, mechanical challenges that are often encountered in engineering applications (52, 53). In this work, MIPMs with different numbers of segments were fabricated and assembled (Fig. 3, A and B), and their cycling compressive performance was evaluated under displacement control from 1 to 5 mm (Fig. 3, C to G). The endoskeleton was fabricated by stereolithography (SLA) three-dimensional (3D) printing using a rigid resin (Precision Model V1 resin). The resulting force-displacement curves show that the structural stiffness increases with the number of internal segments, indicating enhanced mechanical reinforcement through finer segmentation. Both FEM simulations and experimental results reveal that the compressive reaction force increases with the number of endoskeletal segments, although the increase gradually saturates (Fig. 3H). Although the MIPM-12 (with 12 segments) exhibits slightly higher load-bearing capacity than the MIPM-10 (10 segments), the improvement is only 2.9%. Considering the higher fabrication cost and self-weight of the MIPM-12, the MIPM-10 was chosen as the optimized configuration for subsequent experiments. Evaluation of load-bearing and anti-puncture capabilities in the MIPM. (A) Photographs of the MIPM with varying numbers of segmental endoskeleton. (B) Side-view schematics of the MIPM with different numbers of endoskeletal segments. (C to G) Cycling compression force-displacement curves of the MIPM with different numbers of endoskeletal segments (n = 4, 6, 8, and 10) in the vertical orientation. (H) Comparison between experimental and FEM results for load-bearing capacity with varying numbers of endoskeletal segments. a.u., arbitrary units. (I) Schematic of different MIPM orientations during the rotation. (J) Experimental comparison of load-bearing capacity under different rotation angles. (K) Demonstration of the puncture-resistant area on an armadillo compared with the MIPM under puncture loads. (L) Needle loading force-displacement curves. (M) Puncture and compressing process of the MIPM without and with exoskeletons. Mechanical tests were conducted on three different states of the MIPM: flat state, curled state without the segmented endoskeleton, and curled state with the segmented endoskeleton (fig. S10A). Simply curling the composite layer to a circle shape without the segmented endoskeleton does not offer stiffness increase compared to the flat state (fig. S10, B and C). However, the MIPM, in the curled state and with the segmented endoskeleton, leads to a pronounced increase in stiffness (fig. S10D). As an example, with 10 segments, the structural stiffness (KS) increase was 667.5 times. The MIPM also exhibits direction-dependent mechanical behavior, with varying structural responses at different rotational orientations. Figure 3I shows two states: aligned and rotated. As n increases, the angle between the aligned and rotated states decreases (fig. S11A). In both aligned and rotated states, the maximum forces that the MIPM can sustain increase with the segment number, but the improvement gradually saturates (Fig. 3J and fig. S11, B to F). When compared with other reported variable-stiffness materials and structures, the MIPM exhibits a distinctly superior stiffness change ratio while maintaining an exceptionally low initial stiffness (fig. S12) (54–64). Figure S13 shows the FEM results of the deformation and stress distribution under compressive loading in both aligned and rotated states. In the aligned state, once a compressive load is applied, the segmented endoskeleton interlocks and collectively bears the load. In contrast, in the rotated state, when n is relatively small (e.g., n = 4 or 6), compression may cause buckling of the soft layer between the endpoints of adjacent endoskeletal segments, leading to a degradation in load-bearing capacity. As n increases, this weakness diminishes; for n = 10 and 12, the difference between the aligned and rotated states becomes negligible in terms of load bearing. This is another reason why n = 10 was selected for subsequent experiments. We further established a geometry-based mechanics model for endoskeleton-supported interlocking, which demonstrates that increasing endoskeleton density leads to enhanced overall stiffness of the MIPM with diminishing marginal gains; this trend is consistent with our experimental and FEM results, thereby justifying the selection of n = 10 for subsequent analyses and enabling predictive optimization of mechanical responses across different structural configurations and loading conditions (fig. S14) (65, 66). Beyond structural rigidity from interlocking vertebrae, the armadillo also uses overlapping bony bands, particularly the movable transverse bands of the three-banded armadillo, for conformal coverage during bending. This layered exoskeletal arrangement ensures close surface contact and offers protection against sharp puncture threats, such as predator teeth (Fig. 3K). Inspired by this biological architecture, the overlapping exoskeletal scales (made of the same resin as the endoskeleton) can conform closely to the curvature of the MIPM during bending. To quantitatively evaluate puncture resistance, controlled needle puncture tests were conducted to assess the mechanical response of the MIPM-10 under localized penetration (67). Without the protection of the exoskeleton, the soft composite layer was easily pierced by the needle. With the exoskeletal protection, needle penetration was prevented even when the overall structure collapsed (Fig. 3, L and M). At the same prescribed needle displacement, the measured resistance force increased by more than 180% compared to configurations without the exoskeleton. These quantitative puncture-resistance metrics provide a performance benchmark that supports subsequent demonstrations of reliable MIPM operation in environments containing sharp objects and puncture hazards. Sensing and reaction loop design The armadillo’s curling behavior in response to external threats represents a typical form of biological feedback and active defense (Fig. 4A). Here, the MIPM realizes an analogous active protective response, enabling autonomous reconfiguration upon sensing impact or puncture loads (Fig. 4B). The AgNW/PDMS composite serves as the pressure sensing element. To evaluate its response for dynamic mechanical stimuli, an impact testing platform was constructed to apply dynamic loads, releasing a steel ball from varying heights (Fig. 4C). A voltage divider circuit was used to convert the resistance change to a voltage change. The sensing layer was placed at the impact zone and connected to an oscilloscope for recording the impact signals (fig. S15A). During the impact process, stress is concentrated in the localized region beneath the impactor, where the AgNW-embedded sensing zone is located. Figure 4D shows the axisymmetric distribution of the principal stress in the vertical direction. Impact sensing and stimulus-responsive protection in the MIPM. (A) Schematic of an armadillo curling into a protective posture upon threat perception. (B) Diagram of the MIPM deformation in response to external stimuli. (C) Impact signal acquisition system for real-time sensing. (D) FEM simulation result showing stress distribution in the impact zone. (E) Relationship between relative resistance change (ΔR/R0) and impact force. (F and G) Sensing signals at 20- and 60-mm heights, respectively. (H) Reaction loop of the MIPM under tactile stimulation. The impact force was also recorded by a commercial force sensor, with the peak value versus the height shown in fig. S15B. The relative resistance change of the AgNW/PDMS sensor versus the measured peak impact force, as shown in Fig. 4E, provides the calibration. Representative electrical signals at low (20 mm) and high (60 mm) impact heights are shown in Fig. 4 (F and G). In both cases, the sensor successfully captured not only the initial impact signal but also the rebound signal with high spatial resolution. Compared to the case of the 20-mm height, the 60-mm case exhibited a longer time interval between the primary and rebound peaks, reflecting a greater impact energy and extended rebound duration. Connecting the sensing layer to a controller chip and the heater establishes a closed-loop feedback system that enables impact-triggered actuation (fig. S16). The sensitivity of the feedback system can be tuned by adjusting the triggering threshold. Both gentle touch (Fig. 4H and movie S2) and severe impact (fig. S17 and movie S3) can trigger the MIPM’s defensive response. This contact-triggered curling demonstrates a representative mode of stimulus-responsive morphing, in which external touch autonomously drives the structure into a protective configuration. Defensive response of MIPM In biological organisms, shape reconfiguration often plays a critical role in protective strategies, such as curling, shielding, or encapsulation in response to threats. The MIPM is designed not only to deform actively but also to leverage its reconfigured morphology to perform defensive functions, including resisting puncture, alleviating impact, and safeguarding fragile objects inside after deployment. The MIPM was initially in its natural flat state [Fig. 5A(a)] and a soft raspberry was placed on top to simulate a fragile payload [Fig. 5A(b)]. Upon receiving a triggering signal, the MIPM actively curled into a closed-loop configuration, enclosing the raspberry and providing physical protection against external threats [Fig. 5A(c) and movie S4]. To further demonstrate its protective capability during locomotion, the MIPM was placed on an inclined platform lined with sharp thumbtacks to simulate hazardous terrain [Fig. 5B(a)]. Upon activation, the structure rolled down the slope while maintaining its curled state [Fig. 5B(b)]. The overlapping exoskeletal structure provided effective penetration resistance against the sharp tacks, protecting the enclosed raspberry from damage [Fig. 5B(c) and movie S5]. During rolling on an inclined plane with angle θ, the normal force can be estimated as N=mgcosθ. A detailed rolling-motion analysis is provided in the Supplementary Materials. Protective manipulation capabilities of the MIPM. (A) Controlled bending of the MIPM is used to grasp a fragile object (raspberry). (a) Origin status of MIPM; (b) MIPM can be triggered by putting light weight on it; (c) triggered protective status of MIPM. (B) The MIPM rolls across a needle-covered surface without structural damage. (a) A soft berry is wrapped in MIPM; (b) during the rolling process, MIPM protects the berry from being harmed by tack; (c) MIPM escaped away from the dangerous area. (C) Setup for free-fall tests of a glass sphere and the MIPM (from the wing of a small aircraft, ~1.2 m above the ground). An unprotected (D) glass sphere and (F) raspberry are shown after the free fall. (E) A glass sphere and (G) a raspberry protected by the MIPM remain intact after the free fall. To quantitatively evaluate the impact-protection performance of the MIPM under dynamic loading, controlled drop-impact experiments were conducted using pressure-sensitive papers and an accelerometer. A pressure-sensitive paper wrapped around the falling cylinder and another placed on the rigid substrate enable direct visualization of the transmitted impact force and contact area with and without MIPM protection (fig. S18, A and B). Without protection, pronounced coloration is observed on both pressure-sensitive papers, indicating severe and highly localized impact loading. In contrast, with MIPM protection, the pressure-sensitive paper on the weight exhibits negligible coloration, while the paper on the rigid substrate shows weaker coloration distributed over a larger area, reflecting effective load redistribution and damping enabled by structural deformation during impact (fig. S18C). In addition, quantitative measurements of peak acceleration were obtained for drop heights ranging from 100 to 400 mm. Repeated trials were conducted at each height, and the results are reported with error bars. The MIPM-protected configuration consistently exhibits reduced peak acceleration, reaching 80.6% decrease at the height of 400 mm (fig. S18D). These measurements provide quantitative evidence that the MIPM effectively attenuates impact loading and reduces the force experienced by the protected object. To further demonstrate the effectiveness of the MIPM in practical impact situations, a glass sphere and the MIPM were released from a height of ~1.2 m (Fig. 5C). Photos were taken during the free-fall process. Without protection, the glass sphere shattered upon hitting the brick ground (Fig. 5D). In contrast, when enclosed by the MIPM, the same glass sphere remained intact upon hitting the same ground, demonstrating the structure’s ability to absorb and dissipate impact energy (Fig. 5E). With the MIPM protection, the mechanical response during the impact was dominated by structural deformation and reconfiguration of the module, while no noticeable deformation or fracture of the exoskeleton was observed. The similar tests were conducted with a raspberry, which was dropped onto a bed of sharp tacks from the same height (Fig. 5F). The unprotected fruit was readily punctured and crushed, whereas the MIPM-encapsulated raspberry remained undamaged, indicating that the structure effectively resists penetration and provides mechanical cushioning (Fig. 5G). Active escaping of MIPM As previously described, the armadillo uses an active defensive strategy by curling into a sphere, a configuration that shields vital body parts and hinders predators from gripping or attacking, sometimes even allowing escape from danger. Adult armadillos have been observed transporting their curled juveniles by rolling them; however, the rolling is generally passive, relying on the downslope terrain (as shown in Fig. 5B as an example). To the best of our knowledge, active rolling locomotion by the armadillo in the curled state has not been reported. By leveraging programmable, spatially distributed heating as demonstrated in this work, the MIPM could potentially achieve controlled rolling motion. This capability would further enhance the MIPM’s active defense strategy for operation in extreme environments. The internal architecture of the MIPM consists of 10 segmented endoskeletal units, dividing the composite layer into 10 equally spaced regions. When the distributed heaters in two diagonally opposite regions are activated simultaneously, the corresponding arc lengths bend locally, distorting the initially circular structure (Fig. 6A). The induced curvature asymmetry causes the MIPM to roll toward the direction of the diagonal (relative to the vertical), as shown in the time-sequenced infrared images during the actuation (Fig. 6B and movie S6). FEM simulations further reveal the underlying mechanism with channels 1 and 6 activated as an example. The two channels are diagonally opposed (Fig. 6C), and the local heating increases the local curvature in the two activated regions, deforming the initially circular structure into an elliptical shape. This asymmetry shifts the center of mass and generates a torque for rolling, thereby enabling rolling locomotion via asymmetric thermal actuation (Fig. 6D). Programmable thermal actuation and locomotion in the MIPM. (A) Infrared image demonstrates region-specific heating of the MIPM. (B) Localized deformation shifts the center of mass. The channels are numbered incrementally clockwise (C) Schematic showing the channel-separated layout of the heating layer. (D) Asymmetric thermal input triggers induced rolling; U denotes displacement magnitude in the finite element model. (E) Wireless control is implemented via a Bluetooth interface. (F) Schematic illustration of the programmable actuation with wireless control via Bluetooth. (G) Temperature distribution is mapped during sequential heating; MCU, microcontroller unit. (H) Forward rolling of the MIPM is driven by thermal actuation. The internal protective cavity of the MIPM provides an ideal compartment for incorporating electronic components (Fig. 6E). To drive the MIPM remotely, we are integrating it with a Bluetooth module and a transistor-based power switch circuit (Fig. 6F). The multichannel property of the control system enables spatially selective heating and actuation (movie S7). Area 1 and area 6 are first activated to initiate rolling, followed by area 2 and area 7, maintaining a diagonal heating pattern throughout the rolling motion. The corresponding thermal image demonstrates the sequential regional heating across the composite structure, illustrating the programmable distributed actuation strategy (Fig. 6G and movie S8). With this sequential heating strategy, the MIPM ideally rotates by one segment (36°) per actuation cycle, completing a full 360° rotation after ten steps (Fig. 6H). In practice, however, factors such as payload weight and terrain surface conditions could introduce variability, with each actuation cycle resulting in a rolling displacement of one to three structural segments (movie S9). The MIPM perimeter is discretized into 10 segments, such that each rolling event advances the module by ~1/10 to 3/10 of its circumference, corresponding to a mean locomotion speed of 2.5 mm s−1. Together, these results demonstrate that the MIPM can actively curl into a closed configuration to resist puncture and impact threats, like the armadillo, and additionally deploy rolling as an active escape strategy beyond its natural counterpart. DISCUSSION Since the dawn of human civilization, protective strategies have been essential for survival in hazardous environments. Early protection, such as animal hides and bones, offered limited resistance, while the advent of metallurgy introduced metallic armors and shields that markedly enhanced strength and impact resistance, though often at the expense of mobility. Today, the development of advanced protective structures and materials continues to be a key frontier in materials science and engineering (53, 54, 68). Over the past decade, the rise of soft machines—including soft robotics, flexible electronics, and morphable metamaterials—has highlighted their advantages in compliance, adaptability, and multifunctionality. Yet, as soft machines move into extreme environments, such as soft wearable devices for workers in hazardous settings (e.g., disaster response and underground mining) or soft robotic explorers in harsh terrains (e.g., extraterrestrial, desert, or polar environments), their intrinsically low modulus and limited load-bearing capacity may constrain both performance and survivability (69–72). These limitations hinder their ability to safeguard humans or fragile payloads in practical applications. To overcome this, researchers have proposed flexible protective structures, including bioinspired designs such as scale architectures (73, 74), pangolin dermal armor (75, 76), cuttlebone (77), and chiton girdle (28). However, most remain essentially passive, trading mechanical protection for flexibility and lacking the ability to actively sense or adapt to external threats. In contrast, our work draws inspiration from the armadillo, a unique organism that exhibits stimulus-triggered morphing. Agile in its natural state, the armadillo can rapidly curl into a rigid protective sphere under threats, where overlapping osteoderms form an external shield and interlocking vertebrae enhance rigidity from within. Guided by this principle, we developed the MIPM that integrates segmented exoskeletal scales, a spine-inspired interlocking endoskeleton, and embedded sensing-actuation layers. Unlike previous flexible protective systems, the MIPM perceives external stimuli, including dynamic impacts, and autonomously reconfigures, seamlessly transitioning between flexible and rigid states with orders-of-magnitude changes in stiffness. By uniting structural design with active behavioral logic, this work bridges the long-standing conflict between softness and protection, establishing a versatile platform for adaptive soft machines capable of safe and reliable operation in extreme environments. Nevertheless, several challenges and opportunities should be noted. The contact-triggered curling demonstrated here illustrates the stimulus-responsive and protective capabilities of the MIPM. Nevertheless, practical deployment may involve additional considerations, including triggering conditions and Bluetooth communication robustness. In particular, in extreme environments, such as high or low temperatures, high humidity, dust or particulate exposure, and airflow variations, environmental stability of the system represents an important direction for future studies. These considerations motivate application-specific tuning of sensing thresholds, actuation strategies, structural architecture, and control schemes. Furthermore, the rigid scales have primarily served as structural supports, but their material choices and design space remain largely unexplored, presenting opportunities for further innovation. Beyond mechanical reinforcement, the scales could be further engineered as multifunctional platforms. For instance, optical functionalities could be integrated into selected scales to enable localized optical sensing, while thermally or electrically conductive scales could facilitate distributed heat dissipation or power routing during actuation. Such extensions are structurally compatible with the modular, segmented architecture of the exoskeleton. These opportunities highlight a clear pathway toward intelligent, multifunctional protective systems built upon the same architectural framework demonstrated here. MATERIALS AND METHODS LCE ribbon fabrication LCE ribbons were synthesized via a Michael addition reaction between thiol and acrylate functional groups, following a network-forming protocol (78). To prepare the LCE precursor solution, 2 g of the mesogenic monomer RM257 (1,4-bis-[4-(3-acryloyloxypropyloxy)benzoyloxy]-2-methylbenzene) was first dissolved in 0.7 g of toluene at 85°C, followed by cooling to room temperature. Subsequently, 0.42 g of the chain extender 2,2′-(ethylenedioxy)diethanethiol (Sigma-Aldrich), 0.18 g of the cross-linker pentaerythritol tetrakis(3-mercaptopropionate) (Sigma-Aldrich), and 0.012 g of the photoinitiator (2-hydroxyethoxy)-2-methylpropiophenone (Sigma-Aldrich) were added. The mixture was reheated to 85°C to ensure full dissolution. A catalyst solution-dipropyl amine (2 wt % in toluene, 0.288 g) was then introduced into the system. After thorough mixing and degassing, the solution was transferred into a mold and thermally cured for 24 hours. The resulting LCE film was uniaxially stretched to 100% uniaxial strain and subsequently exposed to ultraviolet light (365 nm, 20 mJ/cm2) for 10 min to complete the alignment fixation. Synthesis of AgNWs To synthesize AgNWs, 60 ml of a 0.147 M polyvinylpyrrolidone (PVP; molecular weight ~40,000; Sigma-Aldrich) solution in ethylene glycol (EG) was prepared in a flask equipped with a magnetic stir bar. The mixture was placed in an oil bath maintained at 151.5°C and stirred at 260 rpm for 1 hour. Following this, 200 μl of a 24 mM CuCl2 solution (prepared from CuCl2·2H2O, 99.999+%, Sigma-Aldrich) in EG was injected into the reaction mixture. Subsequently, 60 ml of a 0.094 M AgNO3 solution in EG (99+%, Sigma-Aldrich) was added to initiate nanowire formation. Following formation of the nanowires after 2 hours, the reaction was stopped and the AgNW solution was cooled to room temperature. The AgNW was first washed with acetone and then with ethanol to remove EG and PVP. Last, the nanowires were dispersed in ethanol for further use (79). Fabrication of the sensing layer A patterned mask was fabricated by laser-cutting a thin PI film affixed to a glass substrate. The synthesized AgNW dispersion was then drop cast onto the masked glass slide and placed on a hot plate at 50°C to facilitate solvent evaporation. Once dried, the PI mask, along with the AgNW deposited on its surface, was peeled off, leaving behind the desired conductive pattern. Liquid PDMS (SYLGARD 184, Dow Inc.) was subsequently applied over the patterned AgNW, followed by spin coating to achieve a uniform film thickness. The AgNW/PDMS composite was thermally cured at 70°C for 1 hour. To enable strong interfacial bonding, both the PDMS surface and the prepared LCE ribbon were treated with air plasma for 20 s before being laminated together under pressure. Electrical connections were established by attaching copper wires to the ends of the conductive patterns using silver epoxy (MG Chemicals). Fabrication of the endoskeleton, exoskeleton, and origami interface The endoskeleton and exoskeleton were fabricated using a Formlabs Form 4 SLA 3D printer to construct rigid structural components for integration with soft functional layers. While 3D printing was used in this study, similar structures can be fabricated using other methods and with different rigid materials, depending on specific design requirements. The origami interface was prepared by laser engraving structural paper to locally reduce thickness and create hinge-like regions. These compliant folds reduce bending resistance and enable controlled deformation under actuation. Assembly of the MIPM The LCE actuator and the AgNW/PDMS sensing layer were first bonded together. The PDMS surface of the composite film and the surface of the LCE ribbon were plasma treated for 20 s and then laminated under pressure to ensure strong adhesion. The origami interface, prefolded into its target configuration, was then bonded onto the LCE surface. Segmented endoskeletal components were aligned and adhered to designated regions on the origami interface. On the opposite side of the AgNW/PDMS composite, a PI tape layer was applied to enhance structural rigidity. Last, a laser-cut heater pattern made of conductive fabric (Angkeel) was attached onto the surface of the PI tape to complete the multilayer assembly. Closed-loop feedback and wireless control For the feedback-driven response, the AgNW/PDMS sensing layer responds to external mechanical stimuli by exhibiting a change in resistance. This change is monitored in real time by a sensing circuit, digitized by an analog-to-digital converter, and processed by an ESP32-S3 microcontroller (Espressif Systems). When the signal exceeds a predefined threshold, the microcontroller activates a heating channel via a MOSFET, inducing structural reconfiguration of the MIPM. Additionally, the microcontroller’s built-in Bluetooth enables direct user control, allowing for programmable heating of specific regions through a mobile interface to achieve custom deformation and locomotion. Characterizations The microstructure of the AgNW network was observed using scanning electron microscopy. The multilayered configuration of the MIPM was examined under an optical microscope. Mechanical performance under compression was evaluated using a universal testing machine at a constant loading rate of 0.2 mm/s, whereas puncture testing was performed using a conical indenter with a tip diameter of 200 μm to apply localized penetration loads. Electrical signals generated during impact were collected using a digital oscilloscope. The MIPM was powered by a programmable DC power supply, and actuation was controlled via a microcontroller-based circuit. Drop-impact experiments were conducted using a free-fall setup, in which the test object was released from predefined heights. For quantitative impact measurements, drop heights ranging from 100 to 400 mm were investigated with repeated trials at each height. Pressure-sensitive paper (Fujifilm Prescale Low Pressure) was used to characterize the contact area and contact force, while an accelerometer was used to measure the peak acceleration. The results are reported with error bars. Unless otherwise specified, all mechanical testing, sensing calibration, and actuation experiments were conducted under controlled laboratory conditions at room temperature (22° to 24°C) and ambient humidity (~40 to 50% relative humidity). FEM simulation All geometries were reconstructed from experimental measurements and modeled in Abaqus/CAE. The LCE/AgNW/PDMS/PI composite layer was represented as a multilayer shell and meshed with quadrilateral reduced-integration elements (S4R). The LCE was assigned orthotropic thermal expansion coefficients (−0.1, 0, 0 along the x, y, and z directions), so that thermal loading induced in-plane contraction and bending of the composite layer. Material properties of PDMS and PI were taken from datasheets. The contribution of AgNW was neglected in both compression and thermally driven deformation simulations, due to its extremely small thickness ratio (<1:20). Segmented endoskeletons were modeled as components with a modulus much higher than that of the surrounding composite layer. Their interlocking behavior was captured by introducing nonlinear spring elements between adjacent endpoints, which constrained relative sliding after interlocking and represented the mechanical interaction during locking. All simulations were performed within a nonlinear finite element framework with large-deformation kinematics enabled. Compression was simulated under displacement control. Thermal actuation was modeled by prescribing a spatially uniform temperature field within the thermally responsive layer, representing Joule-heating-induced actuation under electrical input. The temperature field was treated as quasistatic within each loading increment and converted into thermal strain through the material’s coefficient of thermal expansion, driving LCE contraction and curvature evolution. For the impact simulation, the dimensions of the AgNW pattern were identical to those of the experimental specimens. A partitioned material assignment was used, in which the PDMS regions containing the AgNW patterns were assigned a higher elastic modulus than the pure PDMS regions, consistent with the experimental characterization. The model was discretized using eight-node solid elements. The impact process was simulated using an explicit dynamic finite-element formulation. The impacting sphere was modeled as a rigid body. The bottom surface of the AgNW/PDMS composite was constrained in the thickness direction.