Abstract

Soft artificial muscles offer transformative potential in robotics, wearable electronics, and biomedical devices due to their light weight, mechanical compliance, and multidirectional actuation. However, their broader utility is hindered by an intrinsic trade-off between stretchability and energy output, often resulting in limited work densities. Here, a high-performance magnetic composite actuator is presented that addresses this limitation through an optimized dual cross-linked polymer network comprising covalent bonds and dynamic physical interactions. The actuator incorporates a stiffness-tunable polymer matrix embedded with surface-functionalized magnetic microparticles, enabling reversible, on-demand stiffness modulation and reprogrammable actuation. This composite architecture achieves exceptional deformability (elongation at break of 1274%) and programmable stiffness switching from 213 kPa to 292 MPa (switching ratio of 1.37 × 103), with shape fixation exceeding 99%. Together, these properties yield a work density of 1150 kJ m−3 and an actuation strain of 86.4%, representing one of the highest values reported for soft artificial muscles. It also supports loads exceeding 4000 times its own weight, demonstrating a powerful and reconfigurable platform for next-generation soft actuation.

1 Introduction

Artificial muscles—soft actuators that emulate the dynamic functionality of biological muscles—have emerged as key components for next-generation robotics,[1] wearable systems,[2] and biomedical devices.[3] Compared to conventional rigid actuators such as electric motors or pneumatic pistons, artificial muscles offer a lightweight form factor, superior mechanical compliance, and the capability for complex, multi-degree-of-freedom motion.[4] These advantages not only enable safe and adaptive interaction with humans and dynamically changing environments but also facilitate compact integration into mobile and untethered robotic platforms.[5]

To date, a wide range of artificial muscles has been developed, including dielectric elastomer actuators (DEAs),[6] fiber actuators (e.g., carbon nanotube yarns, twisted coils),[7] liquid crystal elastomers (LCEs),[8] and phase-change material (PCM)-based systems.[9] (Table S1, Supporting Information). These technologies offer high compliance, low weight, and diverse actuation modes, making them attractive for soft robotic applications. Moreover, each system brings unique advantages: for instance, DEAs achieve large strains (≈250%) with fast response times (≈100 ms). Fiber actuators are extremely lightweight (≈tens of mg), and LCEs enable diverse actuation modes, including bending, twisting, and contraction, with strains of ≈45%. PCM-based systems, meanwhile, demonstrate notable stiffness switching capability (≈200-fold) with moderate strain (≈60%). Nevertheless, many still suffer from limited work density—a key performance metric defined as the mechanical energy delivered per unit volume.[10] This limitation arises from the inherently low modulus and limited stress generation of soft materials, which restrict output force, load-bearing capacity, and mechanical repeatability.[11] These limitations present a critical challenge to realizing compact, high-performance actuators capable of executing complex, high-load motions in soft robotic systems.

To address this challenge, stiffness-variable polymeric (SVP) systems have emerged as a promising strategy for adaptive actuation. Materials such as shape memory polymers (SMPs) allow for dynamic modulation of mechanical stiffness via thermal stimuli, enabling actuators to remain compliant during motion while achieving mechanical stability and shape fixation upon stiffening.[12] Despite these advantages, most SVP-based actuators still exhibit limited work density,[13] due to a fundamental trade-off between stretchability and stress generation. Materials with high extensibility tend to produce low force, while stiffer materials constrain strain, ultimately limiting mechanical energy output per cycle.[14] Overcoming this trade-off is essential for advancing artificial muscles toward real-world applications. Since work density (W = ∫σ dε) depends on both the stress (σ) an actuator can generate and the strain (ε) it can sustain, a simultaneous enhancement in both is critical.[1, 8] To this end, rational polymer network design—through control of cross-linking density and incorporation of soft–hard segment interactions—offers a promising route to enhance strain capacity and energy return without compromising mechanical integrity.[15] Nevertheless, most SVP-based actuators remain limited by this trade-off, highlighting the need for new material strategies that combine high stretchability, force output, and work density.

Here, we introduce a high-performance magnetic composite actuator that overcomes the long-standing trade-off between stretchability and force generation, resulting in markedly enhanced energy output in soft artificial muscles. This advancement is enabled by a dual cross-linking network combining covalent bonds with dynamic physical interactions, which together afford high deformability without compromising mechanical integrity. The actuator comprises a shape memory polymer matrix embedded with surface-functionalized magnetic microparticles. This architecture supports remote and programmable actuation via dual responsiveness to photothermal and magnetic stimuli, while enabling reversible stiffness modulation from 213 kPa to 292 MPa (switching ratio of 1.37 × 103) with shape fixation efficiency exceeding 99%. The actuator achieves a remarkable actuation strain of 86.4%, a work density of 1150 kJ m−3—among the highest reported for soft artificial muscles—and exhibits outstanding mechanical robustness, including an elongation at break of 1274% and a load-bearing capacity exceeding 4000 times its own weight.

2 Results and Discussion

2.1 Dual Cross-Linked Composite Design and Actuation Mechanism

To simultaneously achieve large actuation strain and high force output—two critical performance metrics for soft artificial muscles—we developed a magnetic composite actuator featuring a dual cross-linked polymer network (Figure 1a,b). This design integrates two distinct cross-linking mechanisms: a covalently bonded chemical network and a reversible, physically interacting network. The chemical network is established via covalent bonds between stearyl methacrylate (SMA) and the bifunctional cross-linker ethylene glycol dimethacrylate (EGDMA), which provides permanent mechanical integrity. Simultaneously, the physical network arises from the crystallizable long alkyl side chains of SMA, which mediate stiffness tuning via reversible phase transitions and elastic energy storage.[16] To further reinforce the physical network, we incorporated NdFeB microparticles surface-functionalized with octadecyltrichlorosilane (ODTS). These particles are uniformly dispersed throughout the polymer matrix (Figure 1b). Their long alkyl chains promote interfacial compatibility with the hydrophobic polymer via van der Waals interactions, hydrogen bonding, and chain entanglement, thereby strengthening the physical network without compromising compliance.[17] This dual cross-linking architecture effectively resolves the intrinsic trade-off between stiffness and extensibility, enabling a mechanically robust, highly stretchable actuator with tunable stiffness and programmable actuation capabilities.

Dual cross-linking strategy and thermomechanical actuation mechanism of the magnetic artificial muscles. a) Schematic illustration of the dual cross-linked polymer network, integrating chemical and physical cross-links to achieve high mechanical robustness and deformability. b) Structural composition of the magnetic composite, consisting of a thermoresponsive polymer matrix embedded with uniformly dispersed ODTS-treated NdFeB microparticles. c) Thermally and magnetically driven actuation mechanism based on reversible crystallization/melting of long alkyl side chains and magnetic-field-induced deformation, enabling programmable stiffness modulation and shape programming.

While dual-network strategies have been widely applied in hydrogels to enhance mechanical stability,[18] and magnetic actuators have typically relied on either single covalent[12, 19] or physical networks with limited mechanical performance,[20] the combined use of a dual cross-linked SMP matrix and magnetic particles for programmable actuation remains largely unexplored. The present system demonstrates that such integration can yield a mechanically robust actuator with high strain, tunable stiffness, and reprogrammable magnetic deformation (Table S2, Supporting Information).

The actuation mechanism of our dual-responsive artificial muscle combines reversible magnetic actuation with thermally induced shape-memory effects (Figure 1c). Below the crystallization temperature (Tc), the crystallizable side chains form ordered domains that stiffen the network and lock the programmed configuration. Heating above the melting temperature (Tm) disrupts these domains, producing a soft amorphous matrix that can undergo reversible magnetic actuation (e.g., stretching, bending, or twisting) under an applied field. This temperature-induced structural switching was confirmed by temperature-dependent Wide-Angle X-ray Scattering (WAXS), which showed the disappearance and reappearance of isotropic diffraction patterns upon heating and cooling, respectively (Figure S1 and Note S1, Supporting Information). Subsequent cooling induces crystallization, which locks the deformed configuration while storing elastic energy. Upon reheating, the muscle re-enters the soft state and becomes deformable again. If a magnetic field is applied at this stage, actuation can occur; however, when the field is removed, the actuator recovers its original configuration by releasing stored elastic energy, and subsequent cooling below Tc recrystallizes the side chains, fixing the recovered shape in the rigid state. In addition to this reversible magnetic actuation cycle, thermal triggering provides complementary functions such as shape programming, stiffness modulation, and one-way contraction characteristic of SMP.

2.2 Particle Surface Modification and Optimization of Physical Cross-Linking

To enhance interfacial compatibility and reinforce the physical network, NdFeB microparticles underwent a two-step surface functionalization (Figure 2a-i). First, a uniform silica shell (≈20 nm) was deposited via the Stöber process using tetraethyl orthosilicate (TEOS). Subsequently, ODTS was grafted onto the silica to form an alkyl-rich monolayer (≈5 nm), increasing hydrophobicity and interfacial affinity with the poly(SMA-co-EGDMA) matrix.[12, 21] This modification facilitated homogeneous dispersion and strong particle–matrix interactions through van der Waals forces, hydrogen bonding, and chain entanglement, thereby integrating the microparticles into the reversible physical network. The successful formation of both the silica and ODTS layers was verified by transmission electron microscopy (TEM) (Figure 2a-ii), energy-dispersive X-ray spectroscopy (EDX, Figure S2a, Supporting Information), and Fourier-transform infrared spectroscopy (FT-IR, Figure S2b and Note S2, Supporting Information). Scanning electron microscopy (SEM) and EDX mapping revealed that unmodified particles exhibited poor dispersion and pronounced agglomeration within the matrix (Figure S3a, Supporting Information), whereas surface-modified particles displayed uniformly distributed Fe and Nd signals, providing direct evidence of improved dispersion via enhanced interfacial compatibility. Uniaxial tensile testing further supported strong particle–matrix interactions: the surface-modified composites showed concurrent increases in elastic modulus and elongation at break relative to the unmodified counterparts, indicating more efficient interfacial stress transfer and robust particle–polymer coupling (Figure S3b, Supporting Information).

Surface modification of NdFeB microparticles and optimization of physical cross-linking. a) Schematic of the two-step surface treatment: (i) silica shell formation via the Stöber method followed by ODTS grafting for hydrophobic modification; (ii) TEM image confirming distinct silica (≈20 nm) and ODTS (≈5 nm) layers. b) TGA curves showing enhanced thermal stability with increasing NdFeB content. c) DSC curves demonstrating preserved thermal phase transition behavior after particle incorporation. d) Saturation magnetization (Ms) increases from 288 to 405 kA m−1 with higher NdFeB content, confirming retention of magnetic functionality. e) Stress–strain curves of composites with varying NdFeB content at 70 °C. f,g) Quantitative comparison of mechanical properties: elastic modulus, maximum stress, elongation at break, and toughness (n = 20 independent samples).

Thermogravimetric analysis (TGA) was performed to evaluate the thermal stability of composites containing varying amounts of NdFeB microparticles (0, 9, 11, and 13 g) (EGDMA concentrations: 0.1 wt.%) (Figure 2b). The pristine polymer exhibited rapid degradation above 190 °C with nearly complete weight loss (≈99%), whereas the composites retained ≈ 80% of their mass at 250 °C. This improvement is attributed to constrained chain mobility and enhanced interfacial bonding resulting from particle surface functionalization.[22]

Differential scanning calorimetry (DSC) confirmed that the thermal phase transition behavior of the SMP matrix was preserved after NdFeB incorporation (Figure 2c). The melting (Tm ≈ 37.3 °C) and crystallization (Tc ≈ 26.4 °C) temperatures remained nearly unchanged compared to the pristine polymer, while the latent heat of fusion (ΔH) decreased from 32.7 to 16.9–23.8 kJ m−3 (Figure S4, Supporting Information). This reduction is attributed to a lower fraction of phase-changeable matrix caused by the non-melting filler content, without compromising the reversibility or structural integrity of the switching mechanism. Magnetic hysteresis analysis confirmed that the composite exhibits typical ferromagnetic behavior, characterized by a pronounced hysteresis loop with remanence and coercivity (Figure 2d). The saturation magnetization (Ms) increased from 288 to 405 kA m−1 with increasing NdFeB content.

Uniaxial tensile tests at 70 °C (the thermally soft state) showed that the composite with 11 g of NdFeB microparticles exhibited optimal mechanical performance, achieving peak values in modulus, strength, elongation at break, and toughness (Figure 2e–g). This improvement is attributed to the well-balanced combination of uniform particle dispersion and strong interfacial reinforcement. In contrast, excessive loading (13 g) induced particle agglomeration and interfacial defects, leading to mechanical degradation (Figure S5, Supporting Information). This aggregation at high filler content arises from increased particle–particle interactions and an insufficient polymer matrix to prevent direct contact and clustering, which leads to interfacial weakening.[23] Accordingly, the 11 g formulation was selected as the optimal composition for subsequent experiments. This non-monotonic trend is consistent with percolation theory, which predicts that mechanical performance peaks near the percolation threshold due to enhanced network connectivity and stress transfer, but deteriorates at higher filler content due to particle aggregation and interfacial discontinuities.[24] In the sub-percolation regime, micromechanical models such as the Halpin–Tsai framework can quantitatively describe the modulus enhancement with increasing filler content, supporting the observed trends.[25] These findings are also in agreement with segregated composite studies, which show that well-dispersed fillers form efficient load-bearing networks, whereas excessive loading induces interfacial defects and local stress concentrations.

2.3 Optimization of Chemical Cross-Linking for Enhanced Actuation Performance

To elucidate the role of the dual cross-linking architecture, we compared mechanical and actuation performance across samples with (dual cross-linking) and without (chemical cross-linking only) surface-functionalized NdFeB particles, varying EGDMA content from 0.01 to 0.5 wt.% (Figure 3a; Figure S6, Supporting Information). At 70 °C, where the SMP matrix enters its soft state, the elastic modulus of both systems increased with EGDMA content due to restricted chain mobility from denser covalent networks (Figure 3b).[15, 26] Across all cross-linking densities, dual cross-linked composites exhibited significantly higher modulus values than their only chemically cross-linked counterparts, owing to additional physical reinforcement from SMA crystallites and polymer–particle entanglement with surface-functionalized NdFeB particles. These physical interactions supplemented the covalent network, enhancing stiffness even in the soft state.

Influence of chemical cross-linking density on mechanical and actuation properties of artificial muscles. a) Schematic illustration of single (chemical only) and dual cross-linked networks. b) Elastic modulus and c) elongation at break of chemically cross-linked (blue) and dual cross-linked (red) composites as a function of EGDMA content. d) Actuation strain and e) actuation stress during constrained thermal cycling. f) Representative loading–unloading curves used to calculate work density at maximum strain. g) Work density and h) actuation recovery ratio, both extracted from unloading curves at each EGDMA concentration (n = 5 independent samples).

In contrast, elongation at break decreased with higher EGDMA content in both systems (Figure 3c), reflecting the classic stiffness–ductility trade-off. Nonetheless, dual cross-linked samples consistently showed greater stretchability, indicating that the physical network helps maintain extensibility despite increased covalent constraints—a key feature for durable soft actuators. In addition to tensile properties, all dual cross-linked samples exhibited robust stiffness switching between the rigid (25 °C) and soft (70 °C) states across the tested EGDMA concentrations. Although the switching ratio declined from 1648.6 to 680.1 with increasing cross-linker content, all formulations retained excellent switching capability (Figure S7, Supporting Information).

We next assessed actuation performance under thermal cycling and fixed boundary conditions (see Experimental Section). As EGDMA concentration increased, actuation strain steadily decreased from 89.2% to 35.6% (Figure 3d), reflecting progressive network stiffening and reduced deformability. Notably, composites with 0.01% and 0.025% EGDMA exhibited over 85% strain with full recovery within 1 s, confirming rapid, large deformation—essential for dynamic actuation. Moreover, across various temperature and loading conditions, the actuator consistently demonstrated rapid recovery within 1 s even under static loads up to 100 g, demonstrating comparable or superior performance compared with other SMP-based systems (Figure S8, Supporting Information). Conversely, actuation stress increased from 0.104 to 0.488 MPa with EGDMA content (Figure 3e), indicating greater internal resistance to deformation with increasing covalent cross-linking density. This inverse relationship demonstrates a tunable strain-stress trade-off: low cross-linking supports a large stroke with low force, whereas high cross-linking enhances force output but limits deformability. This trend is consistent with classical rubber elasticity theory and recent findings that strain stiffening and extensibility are governed by the cross-link network architecture and the molecular weight between cross-links.[15] Meanwhile, the slight initial rise in stress followed by a decrease during heating is attributed to transient thermal expansion of the SMP matrix occurring below the melting temperature (Tm), prior to the onset of shape-memory recovery.[27]

To quantify energy delivery, we performed loading-unloading tests at 70 °C and computed work density from stress–strain areas (Figure 3f,g; Figure S9, Supporting Information). The composite with 0.025 wt.% EGDMA exhibited the highest work density of 1150 kJ m−3, achieving an optimal balance between energy storage and elastic recovery. Importantly, work density not only reflects the stored mechanical energy but also provides a proxy for dynamic actuation potential, as it integrates both actuation stress and strain. This coupling implies that actuators with high work density can deliver large deformations rapidly under load, supporting fast and forceful operation critical for robotic applications. Above 0.25 wt.%, work density declined sharply, correlating with a marked drop in actuation recovery ratio (Figure 3h). Up to 0.1 wt.%, recovery remained near 100%, indicating full shape reversibility. Beyond this threshold, excessive cross-linking led to irreversible deformation and hindered energy release. Collectively, these results identify 0.025 wt.% EGDMA as the optimal formulation, offering a synergistic combination of mechanical robustness, large actuation strain, high energy return, and >99% shape recovery. Shape fixation—defined as the ability to fix a temporary shape by cooling the deformed SMP below Tc—was quantitatively evaluated and found to exceed 99%, accompanied by excellent shape recovery (Figure S10, Supporting Information).[28] The optimal formulation also retained over 87% of both mechanical and actuation performance after 300 thermo-mechanical cycles (Figure S11, Supporting Information).

2.4 Thermomechanical Actuation and Performance Benchmarking

To quantitatively assess the actuator’s load-bearing and thermomechanical actuation capabilities, we conducted a series of weight-lifting experiments under thermal cycling using fixed-mass protocols on samples with varying EGDMA concentrations (0.01–0.5 wt.%) (Figure 4a; Movie S1, Supporting Information). In the soft state (T > Tm), the actuators underwent passive stretching under a constant 1 kg load. Low cross-linked samples exhibited exceptional thermal compliance, achieving elongation strains of 767% (0.01 wt.%), 442% (0.025 wt.%), and 281% (0.1 wt.%), whereas higher cross-linked samples showed reduced extensibility—reaching only 113% (0.25 wt.%) and 67% (0.5 wt.%)—due to significantly increased stiffness and network constraints (Figure 4a-i). Following mechanical deformation, the actuators were cooled to lock the stretched configuration (T < Tc) and subsequently reheated under a 50 g load to evaluate contractile actuation strain (Figure S12, Supporting Information for other loads). Among the tested formulations, the 0.025 wt.% sample exhibited the highest contraction strain of 71.7%, indicating robust and reversible thermomechanical actuation when the cross-linking density is optimally tuned (Figure 4a-ii).

Actuation demonstration and performance benchmarking. a) Thermally induced (i) elongation under a 1 kg load and (ii) contraction under a 50 g load for dual cross-linked actuators with varying EGDMA concentrations. b) Photograph of a single artificial muscle (1.25 g, EGDMA 0.025 wt.%) lifting a 5 kg load in the stiffened state, demonstrating an outstanding load-bearing capacity over 4000 times its own weight. c) Benchmarking of actuation strain versus work density in comparison with representative synthetic and biological muscle systems.

Notably, the 0.025 wt.% actuator also demonstrated outstanding mechanical robustness in the rigid (T < Tc) state, sustaining a static load of 5 kg—over 4000 times its own weight (1.25 g)—without mechanical failure (Figure 4b). Given its rectangular cross-section (≈10 mm × 3.2 mm) and length of 20 mm before loading, this corresponds to a normalized stress of ≈1562.5 kPa. This high load-to-weight ratio underscores the actuator’s potential for applications requiring lightweight yet mechanically resilient materials, addressing a common limitation of conventional elastomer- or hydrogel-based actuators that are often too compliant to support large static loads.[10, 29] To further demonstrate the material’s performance in its functional state, we also evaluated its load-bearing capacity in the soft state (T > Tm), where it successfully supported a 1 kg mass—equivalent to a stress of ≈312.5 kPa and over 800 times its own weight—without failure.

To benchmark the actuation performance of the dual cross-linked artificial muscles, we compared their work density and actuation strain with those of representative biological and synthetic actuator systems (Figure 4c). Natural skeletal muscle, while biologically optimized, delivers only ≈40% strain and <40 kJ m−3 of work density.[30] Hydrogel- and organogel-based actuators achieve larger strains (40%–65%) but exhibit limited energy densities (<400 kJ m−3) due to their inherent softness.[10, 31] Thermally responsive systems such as shape-memory polymers (SMPs) and liquid crystal elastomers (LCEs) demonstrate improved energy output (up to ≈800 kJ m−3), yet typically show strains below 45%.[8, 32] Carbon nanotube (CNT) yarns and thermoplastic polyurethane (TPU)-coated fabrics offer relatively balanced performance (≈62%–92% strain; ≈487–842 kJ m−3), but seldom combine high energy density with large deformation.[33] In contrast, our dual cross-linked actuator achieves a high actuation strain of 86.4% and a record-high work density of 1150 kJ m−3 in its soft actuating state, thereby defining a previously unpopulated high-performance regime. Beyond actuation strain and work density, our actuator achieves a specific power of ≈404 W kg−1, which is competitive with representative artificial muscle systems (Table S3, Supporting Information). This unique integration of large strain and energy output positions the actuator as a strong candidate for next-generation artificial muscle systems.

2.5 Multifunctional Actuation in Magnetic Composite Muscles

To demonstrate the actuator’s multifunctionality—including magnetic responsiveness, thermally driven high-strain deformation, and mechanical work output—we conducted a series of application-level demonstrations inspired by human-like motion tasks. These tests highlight not only the actuation potential but also its applicability in soft robotic implementations. The actuator was first magnetized in a pre-curled geometry under a high-strength magnetic field, aligning embedded NdFeB microparticles to induce stable magnetic anisotropy (Figure 5a). To evaluate the magnetic responsiveness of this aligned structure, the actuator was first softened via remote laser heating at T > Tm, followed by the application of an external magnetic field (Field ON). The actuator rapidly curled in response to magnetic torque (τ = M × B) and magnetic force (F = ∇(M·B)), where M is the magnetization per unit volume and B is the magnetic flux density.[34] Upon field removal (Field OFF), the actuator returned to its original state (Movie S2, Supporting Information). Simultaneously, the SMP matrix thermally softened the actuator: under tensile load and T > Tm, the actuator elongated vertically by up to 360% under magnetic-field–driven actuation, indicating magnetically driven, thermally enabled stretchability (Figure 5b). To quantify this response at a fixed thermal setpoint, we held temperature constant and swept the magnetic flux density (B); the extension strain reached 450% at 364 mT, and the bending angle increased monotonically to 83.2° at 19.3 mT (Note S3 and Figure S13, Supporting Information).

Demonstrations of multifunctional actuation in dual cross-linked magnetic composite muscles. a) Schematic of the magnetization process and resulting magnetically triggered deformation, illustrating alignment of NdFeB particles and actuation under applied field (Field ON/OFF). b) High-temperature stretchability of the actuator, showing uniaxial elongation of 360% under tensile loading above the thermal transition temperature, driven by a magnetic field. c) Robotic arm-pulldown demonstration: (i) schematic illustration of the target motion; (ii) magnetic gripping of a bar under field application; (iii) localized photothermal contraction lifting a 115 g weight with 39% strain recovery. d) Bilateral actuation task: actuators were pre-extended to 220%, then induced to contract via photothermal heating under a 77 g load, achieving 52% strain recovery.

In a robotic arm–pulldown demonstration (Figure 5c), the actuator was configured into an arm–hand geometry. Localized heating (808 nm laser, 2 W cm−2) softened the hand segment, enabling magnetic actuation for grasping. After shape fixation by cooling, selective reheating of the arm segment triggered contraction, lifting a 115 g weight with 39% strain recovery. The actuator could be re-extended either by applying a magnetic field or by passive stretching under the load of the suspended weight (Movie S3, Supporting Information). In a second task (Figure 5d) emulating bilateral lifting, actuators pre-extended to 220% strain magnetically gripped bars and lifted 77 g weights upon photothermal-induced contraction (52% strain recovery). These demonstrations highlight that the actuator platform integrates programmable magnetic morphing and thermally reconfigurable actuation within a single material system. With high work density (1150 kJ m−3), reversible strain exceeding 86%, and remote activation capability, this design offers a versatile foundation for next-generation soft robotics, bioresponsive devices, and multifunctional human–machine interfaces.

3 Conclusion

We have developed a high-performance soft artificial muscle that addresses the longstanding trade-off between stretchability and mechanical output in polymer actuators. By employing a dual cross-linking architecture that integrates covalent bonds with reversible physical interactions, the actuator achieves simultaneous improvements in extensibility, mechanical strength, and work density—a combination long considered difficult to achieve in polymer-based actuators. Central to this advancement is the synergistic interplay between a thermoresponsive shape memory polymer matrix and surface-functionalized NdFeB microparticles, which function as both mechanical reinforcers and magnetic domains. This architecture enables reversible stiffness switching (1.37 × 103), >99% shape fixation, exceptional stretchability (elongation at break: 1274%), and a load-bearing capacity exceeding 4000× its own weight. At an optimal cross-linking density, the actuator delivers a work density of 1150 kJ m−3 and a maximum actuation strain of 86.4%, placing it within an otherwise inaccessible regime of actuator performance characterized by both high power and large reversible deformation.

Beyond static performance metrics, the actuator’s multimodal responsiveness to magnetic and photothermal stimuli was validated through robotic demonstrations involving programmable grasping, load lifting, and axial extension under remote control. These functional tests confirm its operational reliability, rapid strain recovery, and scalability for soft robotic integration. Collectively, this study establishes a rational material design approach for next-generation soft actuators that unify programmable shape morphing, stiffness adaptation, and energy-dense actuation within a single soft matter platform. The programmable magnetic deformation of the composite further enables reconfigurable multi-modal actuation without physical re-fabrication, opening opportunities for adaptive magnetic grippers, deployable and shape-morphing structures, and minimally invasive biomedical tools.[12, 35] Integration with sensing modules can expand these functions toward real-time monitoring, targeted manipulation, and autonomous operation in environments where conventional actuators are limited or ineffective.[36] As such, the proposed system offers compelling potential for untethered robotics, biointegrated devices, and adaptive human-machine interfaces, setting a new benchmark for multifunctional artificial muscles.

4 Experimental Section

Materials

NdFeB microparticles (MQFP-15-7) were purchased from Neo Performance Materials (Toronto, Canada). Tetraethyl orthosilicate (TEOS, 98%) and N-octadecyltriethoxysilane (ODTS, 95%) were obtained from Alfa Aesar (Haverhill, MA, USA). Ammonium hydroxide solution, stearyl methacrylate (SMA), ethylene glycol dimethacrylate (EGDMA), and azobisisobutyronitrile (AIBN) were purchased from Sigma–Aldrich (St. Louis, MO, USA). AIBN was recrystallized from methanol before use, and all other reagents were used without further purification.

Surface Modification of NdFeB Microparticles

To improve compatibility with the polymer matrix, NdFeB microparticles were modified via a two-step silanization process. A silica shell was first deposited on the particle surface using the Stöber method. Briefly, 40 g of microparticles were dispersed in 900 mL of ethanol and stirred at 1000 rpm to prevent sedimentation. After stabilization, 90 mL of 25% ammonium hydroxide was slowly added, followed by 3 mL of TEOS. After 30 min of hydrolysis and condensation, ODTS was introduced to graft hydrophobic alkyl chains. The reaction mixture was stirred at room temperature for 12 h, and the resulting particles were washed with ethanol, filtered, and dried. Surface modification was verified using transmission electron microscopy (FE-TEM, Tecnai G2 F20 X-Twin, FEI, USA).

Synthesis of Magnetic Composites

To fabricate the magnetic actuator, 17.4 g of SMA was mixed with 80 mg of recrystallized AIBN and various amounts of EGDMA (0.01–0.5 wt.%) as the cross-linker. ODTS-functionalized NdFeB microparticles (9 , 11 , or 13 g) were added to 3 mL of this prepolymer mixture and homogenized in a water bath at 70 °C. Polymerization was performed in a convection oven at 70 °C for 12 h to produce the final composite.

Thermal Analysis

Thermogravimetric analysis (TGA) was performed using a Q500 analyzer (TA Instruments, USA) at a heating rate of 10 K min−1 under a nitrogen atmosphere. The resolution of the instrument was ±0.1 µg. Differential scanning calorimetry (DSC, Q200, TA Instruments, USA) was used to investigate thermal transitions. Approximately 10 mg of the sample was sealed in an aluminum pan and heated/cooled at 10 K min−1 under nitrogen flow (50 mL min−1). The latent heat of fusion (ΔH) was extracted from the endothermic peak during melting, and the volumetric thermal energy storage was calculated by normalizing ΔH to the material’s density (kJ m−3).

Magnetic Programming and Characterization

Permanent magnetization was performed using a 14 T superconducting magnet system (Teslatron PT, Oxford Instruments, UK). A magnetic field of up to 140 kOe was applied at a ramp rate of 100 Oe s−1 to induce internal magnetic anisotropy. Prior to magnetization, the actuator was softened by heating, manually shaped into the desired 3D geometry, and fixed by cooling below the SMP transition temperature. The shaped sample was then placed in a cylindrical mold (19 mm diameter × 25 mm height) to maintain its geometry during magnetization. This process aligned the embedded NdFeB microparticles along the applied field direction, resulting in stable and programmable magnetization within complex structures. Magnetic hysteresis loops were measured using MPMS7 (Quantum Design, USA) at 25 °C, with a temperature resolution of 0.3 K and system sensitivity better than 1 × 10−5 emu.

Mechanical Measurement

Tensile testing was conducted using a UniVert system (CellScale, Canada) with a 200 N load cell, under two temperature conditions: 25 and 70 °C. Samples were prepared as rectangular strips (25 mm × 10 mm × 3.2 mm) and mounted in a temperature-controlled water bath fixture. The initial gauge length was 5 mm. Samples were stretched to failure at a constant rate of 5 mm min−1 to evaluate their mechanical properties in the rigid and soft states.

Thermal Actuation Characterization

The actuation strain (εact) was determined using the equation:

Actu atio nstr ain , ε act % = L i − L f L i × 100 $$\begin{equation} \textit{Actu}\textit{atio}\textit{nstr}\textit{ain},{\varepsilon}_{\textit{act}}\left(%\right)=\frac{\left({L}_{i}-{L}_{\mathrm{f}}\right)}{{L}_{i}}\ensuremath{\times{}}100 \end{equation}$$ (1)

where Li is the initial length of the pre-stretched specimen and Lf is the final length after actuation. Actuation stress was measured under isometric (constrained recovery) conditions using the automated tensile testing system (UniVert, CellScale, Canada). The specimen was first stretched to 150% of its original 5 mm gauge length in a 70 °C water bath and then removed from the bath while remaining clamped in the tensile tester. It was cooled to room temperature to fix its deformed shape. Once the shape was fully fixed, the specimen was reheated by immersing it again in the 70 °C water bath. The recovery stress developed while the sample was held at the fixed length (150%) was recorded as the actuation stress. Work density was evaluated under displacement-controlled conditions following a similar prestrain–fix–reheat protocol. Samples were first stretched to their maximum strain (determined by EGDMA content) in a 70 °C water bath, then removed from the bath and cooled to fix their deformed shape while remaining clamped. Once fully fixed, they were reheated in the 70 °C water bath to fully release internal stress. The clamps were then returned to the original 5 mm gauge length at a constant rate of 5 mm min−1. The released mechanical energy was calculated by integrating the area under the unloading stress–strain curve (unit: kJ m−3), representing the actuator’s energy release capacity.

Shape Recovery and Fixity Ratio

The recovery ratio was quantified as the ratio of the recovered strain (εrecovered) to the maximum strain during loading (εmax), calculated using the following equation:

R e c o v e r y r a t i o , R r % = ε r e c o v e r e d ε m a x × 100 $$\begin{equation}Recovery\ ratio, {R_r} \left( {\mathrm{% }} \right) = \left( {\frac{{{\varepsilon _{recovered}}}}{{{\varepsilon _{max}}}}} \right) \times 100\end{equation}$$ (2)

The shape fixity ratio was calculated as the ratio of the retained strain (εfixed) after unloading to the maximum strain applied during loading:

F i x i t y r a t i o , R f % = ε f i x e d ε m a x × 100 $$\begin{equation}Fixity\ ratio,{R_f} \left( {\mathrm{% }} \right) = \left( {\frac{{{\varepsilon _{fixed}}}}{{{\varepsilon _{max}}}}} \right) \times 100\end{equation}$$ (3)

Statistical Analysis

Statistical analyses were conducted on independently repeated experiments. The number of samples (n) for each dataset is specified in the corresponding figure panels.

Acknowledgements

This work was supported by the National Research Foundation of Korea grant funded by the Korea government (MSIT) (NRF-2021R1A2C3006297/2022R1A2C3007963/RS-2025-02223634).

Conflict of Interest

The authors declare no conflict of interest.

Open Research

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.