Modular reconfiguration and actuation of microrobotic assemblies Authors: Prashant Kishor Sharma, Tsung-Yen Lu, Chia-Yuan Chen Published: 2026 ABSTRACT SummaryTransport in microfluidic systems occurs under low-Reynolds-number conditions, where localized particle manipulation requires controlled fluid-structure interaction. Here, a modular magnetic microfluidic robotic assembly was developed using soft-bodied, hard-magnetic, and arc-shaped modules that self-assembled under external magnetic fields, underwent oscillatory actuation, and reversibly disassembled within the same platform. Field modulation enabled transitions between force- and torque-dominated regimes, allowing periodic deformation to be transmitted from the actuated legs to the functional head region. During oscillatory actuation, particles were redistributed via contact-assisted sweeping, accompanied by localized fluid disturbance. Forward and backward transport efficiencies of 91.1% ± 3.6% and 92.1% ± 2.6% were achieved, with a round-trip retention efficiency of 83.8% ± 4.8%. Frequency-dependent analysis identified an actuation condition that balanced deformation transmission and actuation rate. This work provides a basis for programmable microrobotic transport and localized manipulation in confined microfluidic environments.Keywords: fluid-structure interaction, FSI, magnetic microrobots, microfluidics, microrobots, particle transport, oscillatory actuation, reconfigurable systems Introduction Microfluidic technologies have emerged as powerful platforms for biomedical diagnostics, chemical synthesis, and lab-on-a-chip systems due to their ability to manipulate small volumes of fluid within confined environments with high precision.1,2,3,4,5,6,7 These systems offer advantages such as reduced reagent consumption, enhanced analytical sensitivity, and improved process control, making them highly suitable for integrated and portable applications.8,9 However, despite these benefits, fluid transport at the microscale remains fundamentally constrained by low-Reynolds-number conditions, where viscous forces dominate and inertial effects are negligible.10,11,12 Under such conditions, mixing and mass transport are primarily governed by molecular diffusion, which is inherently slow and inefficient over practical length scales.13,14,15 As a result, localized particle collection, concentration gradients, and residual contaminants frequently develop within microchannels, particularly in geometrically confined or stagnant regions.16,17,18 These limitations not only reduce operational efficiency but also compromise the reliability and repeatability of microfluidic processes.19,20 Conventional processes based on bulk flow or global flushing lack spatial selectivity and often disrupt the surrounding environment, making them unsuitable for targeted manipulation.21,22,23 Thus, there is a growing need for active and localized transport mechanisms that can operate effectively within diffusion-dominated regimes while preserving system stability. Building on this need, magnetically actuated microrobots have been gradually studied as active tools for microscale manipulation and fluid control.24,25 Magnetic actuation enables wireless, untethered operation within enclosed microfluidic environments, eliminating the need for onboard power sources and complex integration while maintaining precise external control.26,27,28,29 By carefully designing their geometry, material composition, and magnetization profiles, microrobots can achieve programmable motion, deformation, and interaction with surrounding fluids.30,31,32,33 In particular, bioinspired actuation strategies have demonstrated strong potential for enhancing fluid-structure interactions (FSIs) beyond passive transport limitations. Periodic deformation modes, inspired by natural actuation mechanisms, generate unsteady flow fields characterized by vortical structures and localized shear gradients.34,35,36 These flow features promote convective transport and significantly improve mixing efficiency in diffusion-dominated environments.37,38,39,40,41 As a result, magnetically driven microrobots provide a promising approach for enabling localized fluid manipulation, targeted transport, and active cleaning in confined microscale systems.42,43,44 However, despite these advantages, their effectiveness remains constrained by limited structural adaptability and functional versatility. Despite these advances, most existing microrobotic platforms rely on monolithic architectures with fixed geometries and predefined functionalities.45,46,47,48 Such designs naturally limit adaptability, making it difficult to perform multiple tasks that require different structural or functional configurations within the same environment.49,50 In practical microfluidic applications, operations such as navigation, localized actuation, and recovery often occur sequentially and require different modes of operation that cannot be efficiently implemented with a single fixed structure.51,52 Furthermore, controlled assembly and disassembly of microrobotic components within confined channels remain challenging, particularly when reversible and repeatable operation is required for cyclic tasks.53,54,55 In addition, the relationship between structural reconfiguration, induced hydrodynamic flow fields, and resulting functional performance has not been systematically explored.56,57,58,59,60 This lack of understanding limits the ability to design systems that effectively couple programmable deformation with optimized fluid transport and manipulation.61,62,63,64,65 Therefore, to thoroughly realize the potential of microrobotic systems in microfluidic environments, there is a clear need for modular, reconfigurable designs that enable adaptive operation, precise control, and efficient transport under low-Reynolds-number conditions.45,66,67 To address these limitations, a modular magnetic microfluidic robotic assembly was developed to achieve programmable assembly, oscillatory actuation, and controlled disassembly within a unified microfluidic platform. The system consisted of soft-bodied, hard-magnetic, and arc-shaped modules that self-assembled under externally applied magnetic fields and served as reconfigurable building blocks. Modulation of the magnetic field enabled transitions between force-driven assembly, torque-induced deformation, and field-controlled disassembly. The assembled structure underwent periodic deformation, and the deformation generated at the actuated regions was transmitted along the structure to the functional head region, where particle interaction occurred. Particle redistribution was interpreted as contact-assisted particle sweeping accompanied by localized fluid disturbance, rather than purely flow-mediated transport. Experimental results demonstrated repeatable bidirectional particle transport, with a frequency-dependent response that revealed an optimal actuation condition for deformation transmission and particle redistribution. This work provides a reversible microrobotic framework that integrates magnetic self-assembly with programmable actuation. In contrast to conventional monolithic microrobots and previously reported modular systems that primarily focused on either structural reconfiguration or functional manipulation, the proposed approach enabled the same modular units to perform assembly, localized oscillatory actuation, particle transport, and controlled disassembly within a single microfluidic workspace. The primary contribution of this study was the demonstration of this reversible operational cycle together with an experimental analysis linking deformation transmission, transport performance, and localized fluid disturbance. These results provide design principles for multifunctional and reconfigurable microrobotic systems operating under diffusion-limited conditions. Results Design and magnetic actuation of the arc-shaped modular robotic assembly The design and actuation concept of the arc-shaped modular magnetic robotic assembly are presented in Figure 1. Each module consisted of a flexible polydimethylsiloxane (PDMS) body containing magnetized neodymium-iron-boron (NdFeB) microparticles, forming a soft-bodied, hard-magnetic composite structure that deformed under externally applied magnetic fields (Figure 1A). The structure was configured as a one-quarter arc, where magnetized regions near the leg/head area generated torque during actuation. The geometrical configuration and key dimensions of the module are shown in Figure 1A(i). A compliant PDMS buffer layer was incorporated to enhance structural flexibility and support reversible separation during disassembly. The assembled robotic body had an overall width of 10.2 mm and height of 5.0 mm, with a single-leg length of 2.0 mm and a body/trace width of 0.3 mm. Therefore, the assembled structure was at the millimeter scale, although it was operated inside a microfluidic channel for sub-millimeter particle manipulation. The transported particles had a diameter of 500 μm, and the working fluid viscosity was 1.0 × 10−3 Pa s. Accordingly, the term “microrobotic” in this work refers to the microfluidic robotic manipulation platform and manipulation of sub-millimeter particles, rather than implying that the entire assembled robotic structure was micron-sized. The actuation mechanism is illustrated in Figure 1B. When an external magnetic field (Bext) was applied, the programmed remanent magnetic moments in the magnetized NdFeB-PDMS composite generated magnetic torque due to moment-field misalignment. This interaction induced bending deformation of the arc-shaped structure, enabling controlled shape change and reorientation. An electromagnetic actuation platform was developed to generate programmable magnetic fields (Figure 1C). The system consisted of four orthogonally arranged coils surrounding a central workspace, allowing spatial and temporal control of the magnetic field. By modulating the input currents, the modular robotic assembly was actuated in a controlled manner. An overhead imaging system was used to monitor the motion and deformation during experiments. Design, structure, and magnetic actuation system of the soft arc-shaped robot. Design concept, magnetic actuation principle, and experimental workflow of the modular microrobotic system (A) Structural design and operational sequence of the modular microrobot. (i) Geometrical dimensions of a single quarter-arc robot and the PDMS buffer layer. The quarter-arc module had an arc length of 5.0 mm, a leg length of 2.0 mm, and a structural width of 0.3 mm, while the PDMS buffer layer had a height of 2.8 mm and a thickness of 0.2 mm. (ii) Three-dimensional schematic of the assembled modular microrobot, showing two arc-shaped modules connected through the head area. Each module consisted of a PDMS body and magnetic material at the leg/head region. (iii) Schematic workflow of the reversible modular operation, including magnetic self-assembly of the modular units, translation to the mission zone, oscillatory particle transport by head sweeping, and return/disassembly after the transport task. (B) Schematic representation of the magnetic actuation mechanism under an external magnetic field (Bext). The applied field guided the orientation and oscillatory deformation of the magnetic modules. (C) Experimental setup used for magnetic actuation and optical observation of the modular microrobot in the microfluidic channel. The microrobot was actuated inside the channel using external electromagnetic coils and was recorded using a microscope-camera system. Magnetic field programming enables reversible assembly and disassembly The programmable magnetic control strategy and corresponding robotic behaviors are presented in Figures 2A–2E. By modulating externally applied magnetic fields, a complete assembly-actuation-disassembly cycle was achieved. Three distinct magnetic-field modes were generated through programmed coil-current sequences (Figure 2A). In stage I, the programmed field sequence guided the initially separated modules toward each other and promoted magnetic self-assembly. The translational motion observed during assembly was not attributed solely to an ideal, perfectly uniform magnetic field. Instead, module convergence was associated with the combined effects of the programmed magnetic-field sequence, weak residual magnetic-field gradients within the workspace, substrate interaction, and increasing dipole-dipole attraction as the distance between adjacent magnetized modules decreased. During stage I, magnetic dipole interactions promoted the assembly of two quarter-arc modules into a stable semicircular structure. Representative time-lapse images of the assembly process are shown in Figure S2A. In the representative sequence, two initially separated modules approached each other and formed an assembled structure within approximately 3 s. In stage II, phase-shifted alternating currents generated a time-varying magnetic field, producing cyclic deformation of the assembled structure. In stage III, a rotating magnetic field induced progressive interfacial misalignment between the two modules, followed by detachment and complete separation. Representative time-lapse images of reversible disassembly are shown in Figure S2B. In the representative sequence, complete separation was achieved within approximately 9 s. The microfluidic platform used for functional evaluation is illustrated in Figure 2C. The system consisted of an assembly zone, a connecting channel, and a mission zone with geometrical confinement. The assembled structure was guided from the assembly region into the mission zone, where oscillatory actuation was applied for localized particle manipulation. After the operation, the structure was returned and disassembled, completing the reversible functional cycle. The repeatability of assembly and disassembly was evaluated by over ten independent trials. Assembly was successful in 9 of 10 trials, corresponding to a 90% success rate. Disassembly with the buffer layer was successful in 8 of 10 trials, corresponding to an 80% success rate. In contrast, disassembly without the buffer layer was successful in only 2 of 10 trials, corresponding to a 20% success rate (Figure S3). The operation times were quantified from successful trials. The assembly time was 4.2 ± 2.5 s across nine successful assembly trials, and the disassembly time was 11.1 ± 2.9 s across eight successful disassembly trials with the buffer layer (Figure S4). These results indicated that the modular robotic assembly was repeatably assembled and reversibly disassembled under the tested magnetic actuation conditions. The modular units remained functional during the repeated assembly, actuation, and disassembly experiments performed under the tested operating conditions. However, systematic long-term cyclic fatigue testing under prolonged oscillatory actuation was not performed in the present study. Magnetic control strategy, robot behavior, and particle transport performance of the modular arc-shaped microrobot (A) Coil current control sequences are used to generate distinct magnetic field modes. (i) Stage I—converging: single-coil excitation generates a programmed magnetic-field condition that promotes module convergence and assembly. (ii) Stage II—oscillatory actuation: three-coil excitation with phase-shifted sinusoidal currents generates a time-varying magnetic field that induces periodic deformation. (iii) Stage III—rotating field: four-coil excitation produces a rotating magnetic field for controlled reorientation and disassembly. (B) Observed robot behaviors corresponding to each control mode. (i) Converging motion and self-assembly of modular units. (ii) Periodic deformation of the assembled structure under oscillatory actuation. (iii) Structural separation is induced by the rotating magnetic field. (C) Schematic of the microfluidic workspace consisting of an assembly zone, connecting channel, and mission zone. The assembled structure is guided into the mission zone for particle manipulation. Dimensions are given in millimeters. Experiments were conducted with ∼30 particles of 500 μm in diameter. (D) Particle transport efficiency as a function of the normalized driving parameter (f/fopt). Maximum efficiency is achieved near the optimal condition, with the dashed line indicating the high-performance threshold (η ≥ 0.95 ηmax). (E) Force and torque analysis of the microrobot during magnetic actuation. Magnetic forces (Fm,x and Fm,y), magnetic torque (Mm,z), and viscous drag (Fd) are shown, with the center of mass (CM) indicating the reference point for deformation and motion. Oscillatory actuation and particle transport performance The bidirectional particle transport performance of the assembled modular robotic structure was evaluated under identical oscillatory actuation conditions (Figure 3A). To improve the statistical assessment of transport performance, six independent transport experiments were conducted under identical operating conditions. Transport efficiency was defined as the fraction of particles successfully relocated relative to the initial particle count. Round-trip retention efficiency was defined as the fraction of particles recovered at their original location after completion of one forward-backward transport cycle. The displacement transmission ratio was defined as the ratio between the oscillation amplitudes measured at the head and leg regions of the assembled structure. Detailed mathematical formulations are provided in the supplemental information. The forward transport efficiency was 91.1% ± 3.6%, while the backward transport efficiency was 92.1% ± 2.6% (mean ± SD, n = 6). These results indicated comparable particle transport performance in both directions. The round-trip retention efficiency was 83.8% ± 4.8%, indicating that most particles remained recoverable after one complete forward-backward transport cycle. The reduction in round-trip retention efficiency relative to the single-direction transport efficiencies was attributed to cumulative particle redistribution during repeated oscillatory operation. Although most particles followed the intended transport pathway during both forward and backward transport, a small fraction experienced localized displacement caused by variations in particle-robot interactions and local fluid disturbance near the oscillating structure. Therefore, the apparent loss from the original region was mainly associated with gradual spatial redistribution rather than complete particle removal from the working region. The oscillatory deformation behavior of the assembled structure was further analyzed as a function of driving frequency (Figure 3B). The normalized tip displacement was measured at both the leg region, corresponding to the actuation site, and the head region, corresponding to the functional particle-manipulation interface. For both regions, the oscillation amplitude decreased with increasing frequency from 2 to 4 Hz. The normalized leg-tip displacement decreased from approximately 7.0 at 2 Hz to 2.5 at 4 Hz, while the normalized head-tip displacement decreased from approximately 5.8 at 2 Hz to 2.0 at 4 Hz. The displacement transmission ratio showed a non-monotonic dependence on driving frequency, reaching a maximum of approximately 0.85 at 3 Hz. Although the largest leg and head displacement amplitudes were observed at 2 Hz, the maximum particle transport performance occurred at 3 Hz. This result indicated that particle transport was not governed solely by absolute structural deformation. Instead, particle transport depended on the combined effect of head-region sweeping displacement, deformation transmission, and actuation frequency. At 2 Hz, the structure generated a larger displacement, but the lower actuation frequency reduced the number of effective sweeping events per unit time. At 3 Hz, the absolute displacement was slightly smaller. However, the deformation transmission ratio reached its maximum value, indicating more effective propagation of motion from the legs to the functional head region. The higher actuation frequency also increased the frequency of repeated head-particle interactions. Therefore, 3 Hz provided a favorable balance between transmitted head motion and actuation rate, resulting in the highest particle transport performance. The characteristic velocities of the oscillatory motion were 0.004682, 0.0081061, and 0.0076445 m/s at 2, 3, and 4 Hz, respectively. Based on the measured operating conditions, the corresponding Reynolds numbers were 10.21, 18.12, and 17.42, while the Strouhal numbers were 2.99, 1.21, and 0.68, respectively. These dimensionless parameters further described the operating regime of the system and supported the interpretation that the 3 Hz condition provided a favorable balance between oscillatory motion, deformation transmission, and particle redistribution. Representative time-lapse images of bidirectional particle transport are shown in Figure S2C. The time-lapse observations suggested that particle relocation was mainly achieved through direct sweeping and contact-assisted interaction between the particles and the oscillating head region of the assembled structure. Localized fluid disturbance generated by oscillatory deformation may have assisted particle redistribution near the moving structure. Therefore, the transport mechanism was interpreted as contact-assisted particle sweeping accompanied by localized fluid disturbance, rather than purely flow-mediated transport. However, dedicated control experiments were not performed to fully separate the relative contributions of direct particle-robot contact and deformation-induced fluid motion. Quantitative evaluation of particle-transport performance and frequency-dependent actuation response of the assembled modular microrobot (A) Statistical summary of bidirectional particle-transport efficiency. The forward and backward transport efficiencies were 91.1% ± 3.6% and 92.1% ± 2.6%, respectively, showing comparable transport capability in both directions. The round-trip transport efficiency was 83.8% ± 4.8%, indicating that most particles were retained after sequential forward and backward transport. Bars represent mean values, error bars indicate standard deviation, and circular markers show individual results from six independent trials (n = 6). (B) Frequency-dependent normalized tip displacement and transportation ratio of the assembled modular microrobot. The normalized leg-tip displacement decreased from approximately 7.0 at 2 Hz to 2.5 at 4 Hz, while the normalized head-tip displacement decreased from approximately 5.8 at 2 Hz to 2.0 at 4 Hz. The transport ratio reached a maximum of approximately 0.85 at 3 Hz, indicating the optimal actuation condition for particle transport. The shaded region highlights the optimal transportation range around 3 Hz. Error bars indicate standard deviation. Ensemble-averaged vorticity and shear-rate fields induced by oscillatory actuation To examine the physical link between oscillatory deformation and localized particle redistribution, the fluid response generated by the assembled robotic structure was characterized using two-dimensional particle image velocimetry (PIV) (Figure 4). While the transport experiments demonstrated the functional capability of the system, they did not directly show how structural deformation influenced the surrounding fluid. Therefore, PIV analysis was performed to examine whether oscillatory actuation generated localized flow disturbances that could contribute to particle redistribution. The PIV results were interpreted as supporting evidence for FSI during actuation, rather than as a complete explanation of the transport mechanism. The PIV measurement was performed at the 3 Hz operating condition, which corresponded to the highest particle transport performance. To capture the system’s steady-state behavior, ensemble averaging was performed under periodic actuation. The initial transient response was excluded, and only the steady oscillatory regime was considered. Three consecutive oscillation cycles were selected at 3 Hz (T ≈ 0.33 s). Velocity fields corresponding to identical phases within each cycle were extracted and averaged to obtain a phase-consistent representation of the flow field. The resulting ensemble-averaged vorticity field indicated localized regions of alternating positive and negative vorticity near the oscillating legs (Figure 4A). The vorticity magnitude ranged from approximately −0.3 to 0.5 s−1, with the highest values concentrated near the distal ends of the legs. These regions corresponded to locations where structural displacement and velocity were largest during oscillation. The vorticity distribution showed regions of opposite sign adjacent to each other, indicating the formation of localized counter-rotating flow structures induced by cyclic deformation. In contrast, the central body region showed weak rotational activity, and the vorticity magnitude decreased rapidly away from the oscillating structure. The ensemble-averaged shear-rate field further showed localized shear structures around the oscillating legs and adjacent wake regions (Figure 4B). The shear-rate field ranged from approximately −0.6 to 0.4 s−1. Peak positive shear rate reached approximately 0.3–0.4 s−1, while peak negative shear rate reached approximately −0.5 to −0.6 s−1. These localized shear-rate gradients indicated that oscillatory deformation actively perturbed the surrounding fluid near the moving structure. The velocity vectors overlaid on the flow fields showed that in-plane fluid motion was concentrated around the oscillating legs, while the surrounding domain remained comparatively weakly disturbed. Although the measured vortical and shear structures demonstrated localized fluid disturbance, the flow field should not be interpreted as definitive proof that particle transport was driven solely by hydrodynamic effects. Considering the relatively weak vorticity magnitude and the large particle size used in the transport experiments, particle relocation was mainly attributed to contact-assisted sweeping by the oscillating head region, accompanied by localized fluid disturbance. In addition, PIV measurements were performed only at the 3 Hz operating condition. Therefore, the present result should be interpreted as a representative flow-field measurement under the optimal tested condition, rather than as a complete frequency-dependent flow characterization. Details of vorticity calculation and PIV processing are provided in the supplemental information. Ensemble-averaged vorticity and shear-rate fields induced by oscillatory actuation of the assembled microrobotic structure (A) Ensemble-averaged vorticity field around the assembled microrobot. Vorticity was computed from the PIV-measured velocity fields and ensemble-averaged over three steady-state oscillation cycles at 3 Hz (T ≈ 0.33 s) using phase-aligned averaging. The color map represents the ensemble-averaged vorticity, ranging from −0.3 to 0.5 s−1, and the black vectors indicate the local in-plane flow direction. Localized regions of positive and negative vorticity were observed near the oscillating legs, indicating the formation of counter-rotating flow structures. Peak vorticity was concentrated near the distal ends of the legs, whereas the central body region showed weak rotational activity. The vorticity decayed away from the microrobotic structure, indicating that the fluid disturbance was spatially confined under oscillatory actuation. (B) Ensemble-averaged shear-rate field around the assembled microrobot under the same actuation condition. The shear rate was computed from the PIV-measured velocity gradients and ensemble-averaged over three steady-state oscillation cycles at 3 Hz using phase-aligned averaging. The color map represents the ensemble-averaged shear rate, ranging from −0.6 to 0.4 s−1, and the black vectors indicate the local in-plane flow direction. Localized positive and negative shear-rate regions were observed around the oscillating legs and adjacent wake regions. Peak positive shear rate reached approximately 0.3–0.4 s−1, while peak negative shear rate reached approximately −0.5 to −0.6 s−1. These shear-rate gradients indicated that the oscillating legs perturbed the surrounding fluid and generated localized shear structures during oscillatory actuation. Discussion The results demonstrated that the coupling between magnetic actuation, structural deformation, and localized fluid response governed the functional performance of the modular magnetic robotic assembly. By integrating magnetized NdFeB-PDMS composite regions within a compliant arc-shaped architecture, controlled deformation was achieved under externally applied magnetic fields. Therefore, the robotic modules were interpreted as soft-bodied, hard-magnetic composite structures with programmed remanent magnetization, rather than induced soft-magnetic structures. This design enabled reversible transitions between assembly, oscillation, and disassembly without onboard energy or physical connectors. A key feature of the system was the selective activation of force- and torque-dominated regimes through magnetic-field programming. During assembly, the applied magnetic field aligned the programmed remanent magnetic moments of the modules, generating attractive dipole-dipole interactions between adjacent modules. As the separation distance decreased, the attractive interaction increased, promoting the stable assembly of the semicircular structure. The translational motion observed during assembly was not attributed solely to an ideal, uniform magnetic field. Instead, module convergence was associated with the combined effects of the programmed magnetic-field sequence, weak residual magnetic-field gradients in the workspace, substrate interaction, and increasing dipole-dipole attraction between adjacent magnetized modules. In contrast, oscillatory actuation was primarily driven by magnetic torque, resulting in cyclic deformation of the assembled structure. Under the rotating magnetic field used for disassembly, the magnetic moments attempted to follow the changing field direction. This generated progressive interfacial misalignment and relative sliding between the two modules. As the interfacial misalignment increased, the effective magnetic attraction decreased, leading to complete separation. Thus, assembly and disassembly were interpreted as two distinct magnetic-interaction regimes activated by programmed field modulation. The selection of an arc-shaped geometry played an important role in the observed reconfigurable behavior. The arc-shaped geometry provided a unified structural framework capable of supporting assembly, deformation-based actuation, and reversible disassembly. When two quarter-arc modules were assembled, a semicircular configuration was formed, creating a functional interaction region suitable for localized particle manipulation. In addition to facilitating assembly, the curved architecture promoted distributed deformation under magnetic actuation, allowing motion generated at the actuated segments to be transmitted toward the functional head region. Furthermore, the curved interface supported controlled rotational misalignment during disassembly, enabling reversible structural reconfiguration. These characteristics made the arc-shaped geometry suitable for coupling modular transformation with functional actuation within confined microfluidic environments. Particle transport behavior further showed that system performance was not determined solely by deformation amplitude. Instead, transport performance depended on the effective transmission of deformation to the functional head region, where particle-robot interaction occurred. Although larger leg and head displacements were observed at 2 Hz, the maximum particle transport performance occurred at 3 Hz. This indicated that absolute deformation amplitude alone did not govern transport. At 2 Hz, the structure generated a larger deformation, but the lower actuation frequency reduced the number of effective sweeping interactions per unit time. At 3 Hz, the displacement amplitude was slightly lower, but deformation transmission from the leg region to the head region was improved, and repeated head-particle interactions occurred more frequently. At higher frequencies, viscous damping and reduced structural response suppressed deformation amplitude and transmission. Therefore, the optimal transport observed at 3 Hz was attributed to a balance among deformation amplitude, deformation transmission, and actuation frequency. The transport mechanism was interpreted cautiously. The observed particle relocation was mainly associated with contact-assisted sweeping by the oscillating head region, accompanied by localized fluid disturbance. Flow visualization provided supporting evidence of localized fluid disturbance during oscillatory actuation. The vorticity field showed confined counter-rotating structures near the oscillating legs, while the central body region showed weak rotational activity. The shear-rate field also showed localized shear structures near the oscillating regions and adjacent wake areas. These results indicated that mechanical energy introduced through magnetic actuation was transferred to the surrounding fluid, generating spatially confined hydrodynamic activity. However, the PIV results were treated as supporting evidence of localized FSI, rather than as definitive proof that particle transport was driven solely by hydrodynamic effects. This interpretation was necessary because the transported particles were relatively large (500 μm) and the measured vorticity magnitude was modest. Therefore, direct particle-robot interaction likely contributed substantially to particle redistribution. The PIV measurements were performed only at the 3 Hz operating condition. Therefore, the measured flow field should be interpreted as a representative flow response under the optimal tested condition, rather than as a complete frequency-dependent flow characterization. Since deformation amplitude, deformation transmission, and transport efficiency varied with actuation frequency, future work should include frequency-resolved and phase-resolved PIV measurements at additional frequencies to clarify how structural deformation, local flow generation, and particle transport are coupled. The present study employed 500 μm spherical particles as a controlled model debris system to establish the fundamental transport capability of the proposed microrobotic platform. The main objective of this work was to demonstrate a new microrobot operation concept based on reversible assembly, oscillatory actuation, and disassembly. Since the demonstrated transport mechanism was governed by oscillatory sweeping motion and localized fluid disturbances, the transport behavior was expected to vary depending on particle size, shape, and particle-robot interaction characteristics. Smaller particles may be more strongly influenced by localized fluid disturbances generated during oscillatory actuation, whereas larger particles may require greater direct mechanical interaction with the oscillating structure. In addition, irregularly shaped particles may exhibit orientation-dependent motion, increased frictional interactions, or heterogeneous transport behavior compared with spherical particles. Therefore, while the present results demonstrated the feasibility of localized particle transport using the proposed assembly-actuation-disassembly strategy, systematic evaluation using irregular and multiscale particles will be addressed in future work to further assess the versatility and robustness of this modular microrobotic system in practical applications. The present study was conducted in a simplified microfluidic channel to establish the fundamental relationship between modular reconfiguration, oscillatory deformation, and particle transport. Although the results demonstrated the feasibility of reversible assembly, actuation, and localized manipulation within a controlled environment, the system’s performance in geometrically complex or highly confined microfluidic architectures was not investigated. Practical applications such as microscale cleaning, localized cargo delivery, and operation within heterogeneous microenvironments may introduce additional challenges related to navigation, obstacle avoidance, and transport robustness. Therefore, evaluation of the proposed modular platform in more complex microfluidic geometries and application-specific environments will be an important direction for future studies. Overall, this study demonstrated a relationship between modular reconfiguration, deformation transmission, and localized particle manipulation in a magnetically actuated robotic assembly. The primary contribution was the realization of a reversible operational cycle in which the same modular units were used for assembly, localized oscillatory actuation, particle redistribution, and controlled disassembly within a single microfluidic workspace. By combining programmed magnetic control with a compliant, arc-shaped modular design, the system provided a framework for adaptive, reconfigurable operation in diffusion-limited microfluidic environments. In this work, a modular magnetic microrobotic system was developed to achieve programmable assembly, oscillatory actuation, and reversible disassembly within confined microfluidic environments. Soft-bodied, hard-magnetic, and arc-shaped modules enabled field-driven structural reconfiguration without onboard energy or physical connections. The system demonstrated repeatable bidirectional particle redistribution, associated with contact-assisted sweeping and a localized fluid disturbance, during oscillatory actuation. By integrating magnetic control with a compliant structural design, the platform enabled coordinated transitions between functional states within a single framework. These results provide a reconfigurable microrobotic approach for localized manipulation and a design strategy for adaptive operation in diffusion-limited microfluidic systems. Limitations of the study This study has several limitations. First, the particle-transport mechanism was interpreted as contact-assisted sweeping accompanied by localized fluid disturbance; however, the relative contributions of direct particle-robot contact and fluid-mediated transport were not fully separated. Dedicated controls, including magnetic-field-only, robot-present/no-oscillation, and reduced-deformation conditions, were not performed in the present study. Therefore, future work should include these controls, along with particle-trajectory analysis, to clarify the transport mechanism more rigorously. Second, the PIV analysis was performed only at 3 Hz because this frequency produced the highest and most repeatable particle-transport performance among the tested conditions. Although this measurement provided representative evidence of localized fluid disturbance under optimal actuation conditions, it did not capture frequency-dependent flow variations or three-dimensional near-wall effects. Future studies should include frequency-resolved and three-dimensional flow measurements. Finally, the present experiments demonstrated short-term repeatability using 500 μm spherical particles in a simplified microfluidic channel. Device-to-device variability, long-term cyclic fatigue, and transport performance with particles of different sizes or shapes were not systematically evaluated. In addition, the present conclusions were validated at the current millimeter-scale robotic assembly size, and further scaling down would require additional evaluation of magnetic force scaling, viscous resistance, fabrication tolerance, structural compliance, and particle-robot interactions. These aspects should be investigated in future studies to assess robustness and broader applicability. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Chia-Yuan Chen (chiayuan@mx.nthu.edu.tw). Materials availability This study did not generate new unique reagents or materials. The magnetic modular robotic units were fabricated from commercially available PDMS and NdFeB microparticles using the procedure described below. Data and code availability •All data supporting the findings of this study are available within this article and its supplemental information. All data supporting the findings of this study are available within this article and its supplemental information. •No code was generated in this study. No code was generated in this study. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request. Acknowledgments This study was supported through the National Science and Technology Council of Taiwan under contract no. NSTC 108-2221-E-006-221-MY4 and NSTC 108-2221-E-007-107-MY4 (to C.-Y.C.). Author contributions Conceptualization, C.-Y.C.; investigation, P.K.S., T.-Y.L., and C.-Y.C.; writing – original draft, P.K.S.; writing – review and editing, C.-Y.C.; funding acquisition, C.-Y.C.; project administration, C.-Y.C. Declaration of interests The authors declare no competing interest. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERChemicals, peptides, and recombinant proteinsPolydimethylsiloxane (PDMS) base and curing agentDow Corning, USASylgard 184Neodymium-iron-boron (NdFeB) magnetic microparticles, 5 μmMagnequench, SingaporeMQP-15-7Deionized waterLaboratory supplyN/ASpherical model particles, 500 μm diametersStarek Scientific, TaiwanBZ-06Tracer particles for PIV, 39 μm diametersMicrogenics, USA35–7Experimental models: Organisms/strainsThis study did not use human participants, animals, cell lines, microorganisms, or biological experimental models.N/AN/ASoftware and algorithmsCross-correlation-based PIV analysis softwareDantec DynamicsDynamic Studio 2015Image analysis software used for particle counting and displacement trackingNational Institutes of HealthImageJ 1.54pStatistical analysis softwareTecplot, Inc.Tecplot 360 EX 2022R1OtherCustom four-coil electromagnetic actuation systemThis paperN/ATransparent microfluidic channel with assembly zone, transport pathway, and working zoneThis paperN/AOverhead microscope-camera imaging systemWatec, Japan; Nikon, JapanCCD camera: WAT-902H Ultimate; AF Micro-NIKKOR: 60 mm f/2.8DMicrofabricated quadrant-arc moldsThis paperN/A Experimental model and study participant details This study did not involve human participants, animals, cell lines, microorganisms, or biological experimental models. All experiments were performed in an in vitro transparent microfluidic channel using fabricated magnetic soft modular robots, 500 μm spherical model particles, and PIV tracer particles. Method details Fabrication of modular magnetic microrobots Modular magnetic robotic modules were fabricated using a composite of PDMS and NdFeB microparticles. The NdFeB particle size was 5 μm. The PDMS base and curing agent were mixed at a 10:1 ratio with NdFeB microparticles to obtain a homogeneous magnetic precursor, which was subsequently cast into microfabricated molds defining a quadrant-arc geometry. The structures were cured at 22°C, demolded, and used as individual robotic modules. To establish a programmed magnetization profile, each module was mechanically straightened and aligned in a predefined direction before magnetization. An external magnetic field of 15.037 mT was then applied for 5s to impart a remanent magnetic moment along the longitudinal axis of the straightened structure. After magnetization, the modules were released and allowed to recover their original curved geometry while preserving the internal magnetic alignment. Therefore, the modules behaved as soft-bodied hard-magnetic composite structures during actuation. Here, “soft-bodied” refers to the compliant PDMS-based structure, while “hard-magnetic” refers to the programmed remanent magnetization provided by the embedded NdFeB particles. This fabrication and magnetization process produced programmed magnetic interactions that governed assembly, oscillatory deformation, and disassembly under external magnetic fields. Electromagnetic actuation system Magnetic actuation was implemented using a custom-designed electromagnetic system consisting of four independently addressable coils arranged symmetrically around the workspace. This configuration enabled programmable magnetic-field generation within the central operating region. The field strength, orientation, and temporal evolution were regulated through programmable current inputs. For assembly, programmed magnetic-field sequences were applied to guide individual modules and promote their convergence into a stable semicircular configuration. This convergence was associated with the combined effects of the programmed field sequence, weak residual field gradients, substrate interaction, and increasing dipole–dipole attraction between adjacent magnetized modules. Time-varying magnetic fields with controlled phase differences were used to induce oscillatory deformation, generating magnetic torque that resulted in periodic bending and recovery of the assembled structure. The oscillation characteristics were tuned by adjusting the amplitude and frequency of the input current. For disassembly, rotating magnetic fields were applied to generate asymmetric torques at the module interface, enabling controlled separation without physical contact. The complete operational sequence, including assembly, actuation, and disassembly, was achieved solely through external magnetic-field programming, demonstrating a fully reversible and contactless control strategy. Microfluidic experimental setup Experiments were conducted in a transparent microfluidic channel. The channel design ensured optical accessibility and well-defined confinement of the working fluid. It comprised three functional regions: an assembly zone, a transport pathway, and a working zone for localized manipulation. The working zone incorporated microstructure features, including cavities and grooves, to simulate confined environments and particle-trapping conditions. Prior to each experiment, the channel was filled with deionized water with a viscosity of 1.0 × 10−3 Pa s to maintain consistent initial conditions. Imaging parameters were kept constant throughout all experiments to ensure reproducibility of observations. The individual robotic modules were initially positioned in the channel’s upstream region. Under programmed magnetic-field actuation, the modules were guided to the assembly zone, where they self-assembled into a stable semicircular structure via magnetic alignment and dipole–dipole interactions. The assembled robotic structure was subsequently guided to the working zone for particle-manipulation experiments. After completion of the task, the structure was returned and disassembled into individual modules, demonstrating repeatable and reversible functionality. Assembly and disassembly assessment The repeatability of assembly and disassembly was evaluated using ten independent trials for each condition. Assembly was considered successful when two initially separated quarter-arc modules formed a stable semicircular structure. Disassembly was considered successful when the assembled structure separated into two independent modules after rotating-field actuation. Assembly was successful in 9 of 10 trials, corresponding to a 90% success rate. Disassembly with the buffer layer was successful in 8 of 10 trials, corresponding to an 80% success rate. In contrast, disassembly without the buffer layer was successful in 2 of 10 trials, corresponding to a 20% success rate. Operation times were calculated from successful trials. The assembly time was 4.2 ± 2.5 s across nine successful assembly trials, and the disassembly time with the buffer layer was 11.1 ± 2.9 s across eight successful disassembly trials. Particle manipulation experiments Particle manipulation experiments were performed to evaluate the capability for localized transport. Spherical particles with a diameter of 500 μm were introduced into the working zone to simulate debris. The assembled robotic structure was positioned adjacent to the particles, and oscillatory magnetic actuation was applied to induce periodic structural deformation. The resulting cyclic deformation produced contact-assisted particle sweeping by the oscillating structure, accompanied by localized fluid disturbance near the moving body. Particle relocation was observed during oscillatory actuation of the assembled structure. The transport behavior was associated with cyclic structural deformation and localized particle–robot interaction, with possible assistance from deformation-induced fluid disturbance. Optical images were acquired before and after actuation. Particle distributions were quantified through image analysis. Transport efficiency was defined as the ratio of successfully relocated particles to the initial particle count. Round-trip retention efficiency was defined as the ratio of particles recovered at the original location after one complete forward–backward transport cycle to the initial particle count. At the same time, the residual fraction represented the proportion of particles remaining in the original region after actuation. Particle image velocimetry processing Particle image velocimetry (PIV) analysis was performed to characterize the localized flow disturbance generated by oscillatory actuation of the assembled modular robotic structure. The working fluid was seeded with tracer particles of 39 μm diameter. Image sequences were acquired from a top-view perspective during the stable oscillation stage at a camera frame rate of 100 fps. Startup and transient frames were excluded before analysis so that the processed velocity fields represented the steady oscillatory regime. The PIV velocity vectors were cross-correlated using an interrogation window size of 32 × 32 pixels. The recorded image sequences were processed using a cross-correlation-based PIV algorithm. Spurious vectors were removed using a median filter, and missing vectors were interpolated to reconstruct continuous velocity fields. Three complete oscillation cycles at the 3 Hz operating condition were selected for ensemble averaging. Velocity fields corresponding to equivalent phase positions within each cycle were extracted and averaged to obtain phase-consistent flow fields. Because the PIV analysis was based on top-view two-dimensional imaging, only in-plane velocity components were resolved. The ensemble-averaged velocity fields were used to calculate vorticity and shear-rate distributions. The out-of-plane vorticity component was calculated from the in-plane velocity gradients as: where u and v represent the velocity components in the x and y directions, respectively. The spatial gradients were evaluated using central-difference schemes applied to the measured velocity field. The PIV fields were used as supporting evidence of localized fluid disturbance generated by oscillatory actuation, rather than as a complete three-dimensional flow characterization. Deformation measurement and transmission analysis The oscillatory deformation of the assembled robot was quantified using image-based tracking. The displacement amplitude was measured from the maximum and minimum positions within one oscillation cycle. The normalized tip displacement was evaluated at the leg region, corresponding to the actuation site, and at the head region, corresponding to the functional particle-manipulation interface. The displacement transmission ratio was defined as the ratio between the oscillation amplitude measured at the head region and the oscillation amplitude measured at the leg region. This ratio was used to evaluate the propagation of deformation from the actuation region to the functional region. Magnetic force and torque model The microrobot was actuated through magnetic interactions between its internal magnetization and the externally applied magnetic field. The magnetic force and torque acting on the structure were defined as: where m represents the magnetic moment of the microrobot and B is the external magnetic field. Under the tested operating conditions, magnetic forces, magnetic torque, elastic restoring forces, and viscous drag determined the deformation and motion of the structure. The magnetic field used in this study was estimated using analytical expressions derived for finite-length solenoids with a ferromagnetic core. The model assumed axial symmetry and uniform magnetization along the coil and core. This model was used to estimate the applied magnetic field and to interpret the magnetic forces and torques acting on the microrobot. Quantification and statistical analysis Quantitative analysis and force interpretation System performance was evaluated using quantitative metrics obtained from the assembly, disassembly, actuation, and particle-transport experiments. Quantitative values were reported as mean ± standard deviation unless otherwise stated. Particle-transport performance was evaluated using six independent experiments. Assembly success rate was evaluated in 10 independent trials, while disassembly success rates under the buffer-layer and no-buffer-layer conditions were evaluated in 10 independent trials each. Assembly and disassembly times were calculated only from successful trials. Particle-transport performance was quantified using forward, backward, and round-trip transport efficiencies. Forward transport efficiency was defined as the percentage of particles successfully transported from the initial region to the target region and was calculated as where (N0) was the initial number of particles and (NF) was the number of particles transported in the forward direction. Backward transport efficiency was defined as the percentage of particles transported from the target region toward the original region during the backward motion and was calculated as where (NT) was the number of particles present in the target region before backward transport and (NB) was the number of particles transported in the backward direction. Round-trip retention efficiency was defined as the percentage of the initial particles recovered at the original location after one complete forward–backward transport cycle and was calculated as where (NR) was the number of particles recovered at the original location after the complete transport cycle. The actuation response of the assembled microrobotic structure was evaluated using normalized tip displacement and displacement transmission ratio. The normalized tip displacement was used to compare the oscillation amplitude relative to the characteristic structural dimension. The displacement transmission ratio was defined as the ratio of the oscillation amplitude measured at the leg region to that measured at the head region. This parameter was used to evaluate the transmission of magnetically induced deformation from the head to the particle-interacting leg region. A qualitative force analysis was conducted to interpret the observed actuation and particle-transport behavior. Magnetic torque was generated through the interaction between the programmed magnetic moment of the structure and the applied magnetic field. This torque induced rotational deformation of the assembled microrobot. Magnetic field gradients also generated magnetic forces that contributed to translational motion. Elastic restoring forces within the structure opposed the induced deformation, while hydrodynamic drag exerted by the surrounding fluid resisted both rotational and translational motion. Therefore, the dynamic response of the microrobot was governed by the balance among magnetic torque, magnetic force, elastic restoring force, and hydrodynamic resistance. At lower actuation frequencies, a longer response time was available during each actuation cycle, which allowed greater structural deformation and larger leg-tip displacement. However, the lower number of oscillation cycles per unit time reduced the frequency of effective particle-sweeping interactions. At higher actuation frequencies, hydrodynamic resistance increased, and the available structural response time decreased. Consequently, the deformation amplitude and displacement transmission from the head to the leg were reduced. The combined effects of limited deformation and reduced deformation transmission decreased the particle-transport performance at higher frequencies. The highest transport performance was therefore obtained at an intermediate frequency, at which sufficient deformation and an adequate number of sweeping interactions were simultaneously achieved. The Reynolds and Strouhal numbers were evaluated to characterize the flow regime and the microrobot’s oscillatory response. The Reynolds number was calculated as where ρ is the fluid density, U is the characteristic velocity obtained from PIV measurements, L is the characteristic length of the microrobot leg, and μ is the dynamic viscosity. Strouhal number was calculated using: where (f) was the actuation frequency, (A) was the measured oscillation amplitude of the leg tip, and (U) was the characteristic fluid velocity. The characteristic velocities measured at actuation frequencies of 2, 3, and 4 Hz were 0.004682, 0.0081061, and 0.0076445 m/s, respectively. The corresponding Reynolds numbers were 10.21, 18.12, and 17.42, while the corresponding Strouhal numbers were 2.99, 1.21, and 0.68, respectively. These results indicated that the flow remained within a low-to-moderate Reynolds-number regime. The reduction in the Strouhal number with increasing actuation frequency indicated a change in the relationship among oscillation frequency, deformation amplitude, and fluid velocity. Further details regarding image processing, magnetic modeling, and dimensionless parameter calculations were provided in the supplemental information, including Equations S9 and S10 and Figure S1. Magnetic field characterization The magnetic field (B) used in this study was estimated using analytical expressions derived for finite-length solenoids with a ferromagnetic core. The axial magnetic field distribution can be expressed as: The first term represents the magnetic field generated by the current-carrying coil, while the second term accounts for the contribution of the magnetized iron core. This expression assumes axial symmetry and uniform magnetization along the coil and core. Where: z is the axial distance from the end of the electromagnetic coil, μ0 is the vacuum permeability, I is the input current, n is the number of coils turns per unit length, L1 is the length of the coil winding, R is the coil radius, L2 is the length of the iron core, b is the radius of the iron core, and M is the magnetization of the iron core. This model was used to estimate the magnetic field applied during actuation and to interpret the resulting magnetic forces and torques acting on the microrobot. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.117208. Supplemental information Document S1. Figures S1–S4 Video S1. Collaborative assembly and targeted particle delivery of a reconfigurable magnetic microrobot The video shows the coordinated behavior of modular magnetic microrobot units under external magnetic-field control. Initially, individual arc-shaped units are guided toward one another and assembled into a stable semicircular structure via magnetic interactions. Upon application of an oscillatory magnetic field, the assembled structure undergoes periodic deformation, generating localized fluid disturbances that enable particle displacement within the microfluidic environment. The structure is then directed to perform targeted particle delivery across the workspace. Eventually, a rotating magnetic field is applied to induce controlled disassembly, restoring the individual modules. The sequence demonstrates reversible assembly, programmable actuation, and consistent particle manipulation. References Associated Data Supplementary Materials Document S1. Figures S1–S4 Video S1. Collaborative assembly and targeted particle delivery of a reconfigurable magnetic microrobot The video shows the coordinated behavior of modular magnetic microrobot units under external magnetic-field control. Initially, individual arc-shaped units are guided toward one another and assembled into a stable semicircular structure via magnetic interactions. Upon application of an oscillatory magnetic field, the assembled structure undergoes periodic deformation, generating localized fluid disturbances that enable particle displacement within the microfluidic environment. The structure is then directed to perform targeted particle delivery across the workspace. Eventually, a rotating magnetic field is applied to induce controlled disassembly, restoring the individual modules. The sequence demonstrates reversible assembly, programmable actuation, and consistent particle manipulation. Data Availability Statement •All data supporting the findings of this study are available within this article and its supplemental information. All data supporting the findings of this study are available within this article and its supplemental information. •No code was generated in this study. No code was generated in this study. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request.