Hierarchical laser-programmed soft actuators for designing bionic robots with freeform morphing shapes Authors: Yuhan Guo, Mingguang Han, Weixiong Yang, Meihong He, Haibin Duan, Xilun Ding, Sida Luo Published: 2026 ABSTRACT Bio-inspired shape-morphing structures, essential for next-generation soft robotics with unprecedented adaptability, demand actuators capable of complex and configurable three-dimensional motions. By overcoming traditional stimulus-responsive strategies facing the trade-off between manufacturing simplicity and kinematic sophistication, here, we introduce a spatially differentiated laser-programming technology for digital manufacturing laser-induced graphene-based soft actuators (LIG-SAs) with freeform morphing capabilities. Via cross-scale control of lasing energy and scribing direction, material heterogeneity and structural hierarchy can be tuned simultaneously for introducing decoupled electrothermal distribution and stiffness anisotropy, thus encoding LIG-SAs with four typical motion units: straight bending, directional curling, rigid supporting, and soft connecting. By arbitrarily grouping multimodal morphing units into concretized devices, this approach further empowers freeform design of bionic robots including octopus-like tentacles and inchworm/seal–like crawlers toward multitask locomotion of conformal grasping, path navigation, and obstacle avoidance. This framework bridges digital design with physical intelligence, unlocking previously unidentified avenues of soft robots for creating sophisticated and programmable morphologies. INTRODUCTION Biological morphology—from octopus tentacle articulation to vine circumnutation—inspires revolutionary developments of shape-morphing structures, with the capability to reconfigure three-dimensional (3D) architectures for performing diverse mechanical motions (1–3). Benefited from such shape-programming principles, soft robots become one of the most promising intelligent devices with unprecedented multifunctionality and environmental adaptability to accomplish multimodal and multiscenario tasks, e.g., for realizing wriggling and leaping movements in robotic insects (4–6), swimming poses in robotic fishes (7, 8), and flying postures in bionic birds (9, 10). In such smart devices, the core actuation component with complex, sophisticated, and precise self-morphing behaviors becomes critically important. Accordingly, stimulus-responsive materials, such as hydrogels (11, 12), shape-memory polymers (13), liquid-crystalline elastomers (14), and nanocomposites (15), have been basically applied. Although stimulable by fields of electricity, magnetism, light, and humidity, these materials with homogeneous structures or compositions could only unlock elementary morphing shapes such as swelling, contracting, and uniform bending. Considering apparent constraints of these actuators when confronting tasks requiring advanced kinematic coordination and morphological reconfiguration, novel strategies need to be proposed for implementing diversified morphing capabilities through programmable and customizable material/structural design. For a film actuator as a critical morphing unit, local programming/modification/treatment has emerged as the pivotal strategy to transform uniform bending into complex motions such as torsional twisting (16, 17), helical coiling (18), local folding (19, 20), and buckling (21) by either introducing anisotropic structures or heterogeneous materials. For example, 3D printing, laser engraving, laminating, or mechanical gluing processes were applied to alter the actuator with kirigami (22, 23), origami (24, 25), lattice (26, 27), or thickness-gradient structures (28, 29), thereby creating localized stiffness mismatch for guiding specialized morphing. Similarly, inkjet printing (30), photomask patterning (31, 32), electrospinning (33, 34), or laser etching (35–37) methods were used to spatially allocate distribution, orientation, alignment, mixing ratio, and patterning style of active materials, thus introducing complex morphing on the basis of asymmetric stimulus responsiveness. Although these local-programming approaches were successfully explored for improving shape-morphing complexity, critical limitations still exist. First, conventionally applied molding, lithography, and printing approaches fundamentally compromise between fabrication resolution and material diversity, for example, photolithography could achieve micrometer-scale precision yet face stringent material-selection standard, while direct ink writing enables multimaterial integration at the expense of reduced resolution and interfacial reliability. Second, although a variety of morphing methodologies have been proposed (38–40), integrating multiple, distinct actuation modes into a single, monolithic structure or device often necessitates complex, multistep fabrication processes that combine disparate materials and techniques. For example, Wang et al. (41) fabricated asymmetric elastoplastic graphene/piezoelectric actuators with bimodal performance of bending and curling by sequentially introducing fabrication processes of spray-coating, adsorption condensing, interface transferring, local stacking, and hot compression. Certainly, this reliance on heterogeneous integration would compromise manufacturing efficiency and flexibility and device scalability and integrity and eventually impede the freedom of shape-morphing programmability. Accordingly, a unique manufacturing strategy for simultaneously supporting high-precision processing, highly flexible material/structural programming, and multimodal shape-morphing control toward successful assembly of soft robotics with various biologically inspired morphing structures is highly anticipated. As a promising solution, laser-induced graphene (LIG) was recently emergent for offering a unique one-step laser-writing process to highly efficiently and highly reliably assemble various smart devices ranging from sensors (42), heaters (43), and electric generators (44) to soft actuators (45). Relying on digitalized programmability for simultaneously tuning processing routes and parameters, the assembled LIG devices reflected great uniqueness for presenting a variety of structures [e.g., surface patterns and freestanding fibers, films, and foams (46)] and a series of functions [e.g., electrical conducting, surface wetting, Joule heating, and energy harvesting (47)]. The above features thus facilitated the appearance of pioneering works for proving LIG-enabled soft actuators with multiple morphing capabilities by consistently using strategies of either (i) homogeneous processing of LIG for achieving uniform bending (48, 49), (ii) localized processing or directional scribing of LIG for achieving nonuniform bending or curling (50–52), or (iii) overall geometric design or multiunit combination of LIG for achieving special morphing structures at the expense of bulky sizes or multiple sources (53–55). Although swift design and manufacture could definitely be reflected, current LIG actuators persistently fell in the trade-off between structural integration and morphing complexity, engendering limited actuating performances. To best exploit processing and programming uniqueness of LIG, here, we proposed a spatially differentiated laser-programming technology for enabling single-step manufacturing of soft actuators with freeform morphing capabilities through cross-scale control of both material heterogeneity and structural hierarchy. In the basic level, a local-scale process with multimodal tuning of lasing energy and scribing direction together dictates the LIG with decoupled electrothermal and stiffness anisotropies for exhibiting specific curvature, curling angle, and stiffness, thus encoding four types of basic motions, i.e., straight bending, directional curling, rigid supporting, and soft connecting. In the advanced level, component-scale assembly then programs and bridges all local regions into an actuator unit with distributed multiphysical characteristics toward freeform morphing behaviors. In the premium level, device-scale design is further applicable to alter the original shape of actuators or assemble multiple units into a concretized device for assembling diversified bionic soft robots, spanning from octopus-like tentacles for extracting targets in confined spaces or conformally grasping varied targets to inchworm- and seal-like robots with varied crawling and steering capabilities for path-planning navigation and multitask obstacle avoidance. As briefing in Fig. 1, it is highly anticipated that the proposed LIG technology with a cross-leveled designing concept is able to dream and build various plant/insect/animal–inspired soft robots from spatially differentiated laser manufacturing, multimodal property control, and configurable structure design toward freeform morphing performance. Schematic illustration showing spatially differentiated laser manufacturing of LIG-SA with cross-leveled design for fabricating bioinspired soft robots with multimodal property control, configurable structures, and freeform morphing performance. The illustrations are created by graphics software Blender and Adobe Photoshop. RESULTS Manufacturing and programming strategies of LIG-SA units toward freeform morphing behaviors To study the cross-leveled strategy of LIG-based soft actuators (LIG-SAs), a U-shaped actuator unit was first applied as the basic programmable structure with strip-like shapes. Figure S1 illustrates the fabrication sequence of LIG-SAs. Briefly, a three-layered Kapton-LIG-polydimethylsiloxane (PDMS) structure was designed for enabling electrothermally triggered actuation, in which the LIG layer was formed by irradiating a thin film of polyimide (PI) paper (90-μm thickness) through computer-controlled direct laser writing with a specifically programmed pattern to serve as both Joule-heating and mechanical-constraining functions (fig. S2). The Kapton film (50-μm thickness) was taped at the bottom for sealing purpose, and PDMS was spin-coated and cured on top for rendering thermal activeness. Last, the assembled LIG-SA was controlled by a direct current (dc) power source to implement precise electrical-thermal–induced morphing behaviors, attributed to a differential thermal-expansion effect from the multilayered structural design (fig. S3). To guarantee accurate control of multimodal shape morphing, fig. S4 proves that the resistance of the LIG layer reflects rather low drift effects under both mechanical bending (0.004%/°) and temperature (0.03%/°C). To realize freeform morphing, spatially differentiated laser irradiation is the key to engineer the LIG layer with distributed material heterogeneity and structural hierarchy. Specifically, two types of processing strategies were raised for designing LIG-SA units, namely, differentiated lasing energy (DLE) and directional laser scribing (DLS). For DLE, site-selective irradiation by applying switchable lasing energies determines the LIG with different levels of carbonization and resistance (56), thereby engineering every local region with certain Joule-heating temperature and bending curvature. As demonstrated in Fig. 2A, by choosing lasing conditions between focus (low laser fluence) and defocus (high laser fluence), the focused regions (high resistance) markedly show a higher temperature and bending curvature than the defocused regions (low resistance) when heated by the same current source. Thus, by offering enough temperature difference, the focused region could display knuckle-like motion to connect two neighboring defocused regions with comparatively flat shapes, which is absolutely distinguishable to uniformly bending behaviors. More convincingly, by assembling varied regions in arbitrary orders, specialized motions of the LIG-SA unit can be customized by allocating knuckle-like structures with varied positions (from left to right), lengths (from 15 to 40 mm), and quantities (from one to five), as shown in both optical and infrared pictures. Compared with DLE, DLS offers another strategy to irradiate LIG with programmable scribing orientations, resulting in anisotropic stiffness and thermal-induced helical effects. It is believed that DLS effectively creates the microstripe pattern with alternating soft and rigid regions, thus constraining (or facilitating) the bending motion along (or across) the scribing direction (52, 53). As demonstrated in Fig. 2B, by varying the scribing angle from −90° to 90°, all LIG-SA units universally tend to bend with a specific curling angle that is faithfully registered to the scribing direction. Attributed to the dimensional effect of the U-shape with a big aspect ratio, it is also noticed that the 90° unit is almost not deformable, which thus can be used as a rigid supporter. Representative processing strategy for freeform morphing of LIG-SAs. (A) Strategy 1: DLE for engineering the joint deformation of the actuator to achieve module partitions of varied positions, lengths, and quantities. (B) Strategy 2: DLS for engineering the morphing direction of the actuator to achieve the control effect from straight bending to directional curling at different angles. (C) Strategy 3: Combined DLE and DLS strategy for programing complex deformation effects of multiple joints. Black scale bars are 1 cm, and the red scale bar is 200 μm. By combining DLE with unique control of local bending and DLS with unique control of curling angle and stiffness, a more creative strategy can be offered to program every local region following either one of the three morphing types, i.e., regular bending, directional curling, or immobilizing. By successfully applying this spatially programmable and multimodally morphable strategy, freeform morphing shapes can possibly be designed. In Fig. 2C, a series of LIG-SA ribbons were demonstrated with three independent regions to be selectively processed under 0°, 45°/−45°, or 90°. Thus, with arbitrary combinations, a series of specialized morphing shapes were presented. For example, the 90°/0°/90° unit displays a flat-bent-flat shape, and the 45°/0°/90° unit displays a left curl-bent-flat shape, while the 45°/90°/−45° unit displays a left curl-flat-right curl shape, etc. Precise and quantitative shape-morphing control of LIG-SA units under DLE and DLS strategies By raising DLE and DLS strategies, it is then critical to understand the processing mechanisms for quantitatively controlling various morphing shapes. First, to avoid undesirable wrinkles, PI paper was selected as the optimized precursor for processing LIG with the best flatness, as compared in Fig. 3A. It was investigated that the PI paper with randomly distributed fibers creates considerable voids to effectively dissipate thermally induced residual stresses (57). For DLE, the tuning of lasing energy governs the formation of the LIG layer with varied microstructures and electrical properties. Accordingly, by exploiting the Gaussian-distributed laser-energy profile for generating both melted and heat-affected zones (fig. S5), the defocus level (DL) was applied as the critical parameter (Fig. 3B). When DL = 0, its microstructure (fig. S6) periodically performs the ridge-groove microarray, which can be precisely controlled by both laser power and scanning velocity for determining the appropriate processing window (6.25 to 28.75 J/cm2). With the increase in DL from 0 to 3.5 mm, figs. S7 to S10 systematically explore the evolution of LIG with an enhanced spot size (95 to 210 μm), reduced surface roughness (20.1 to 6.5 μm), and tunable porous morphology, chemical composition, and Raman characteristics. With precise control of microstructures, Fig. 3C further summarizes the synergistic modulation of DL and laser power for achieving broad-range tuning of sheet resistance from 8 to 261 Ω/sq, thus providing prerequisite conditions for designing Joule heating–determined shape morphing. Then, with three-layered U-shaped structures (fig. S11), Fig. 3D proves that the uniformly processed LIG-SA (122 Ω/sq) could reversibly bend and recover when electrical current is turned on and off for 100 cycles. Specifically, when a 60-mA current is on (or off), the curvature of unit starts to increase (or decrease) from 0.007 to 0.63 cm−1, which is exactly synchronous as the ramping (or cooling) of temperature from 29° to 194°C. By enlarging the current level from 10 to 80 mA, Fig. 3E and fig. S12 further support the finding that the bending curvature of LIG-SA is linearly augmented from 0.09 to 1.1 cm−1 as the heating temperature is accordingly enhanced from 25° to 255°C. In addition to the current level, by altering the processing recipe for controlling the unit with varied sheet resistance, the bending curvature could be further adjusted. As summarized in Fig. 3F and fig. S13, when sheet resistance is tuned from 62.6 to 148.4 Ω/sq, the bending curvature is unanimously enlarged under varied levels of excitation current. For example, with an 80-mA current, the 148.4-Ω/sq unit shows markedly higher curvature (1.40 versus 0.66 cm−1) than the 62.6-Ω/sq unit. In addition to bending curvature, key performances including actuation force, energy efficiency, environmental stability, actuation frequency, cyclic durability, material fatigue, and actuation performance after minor damage have also been measured and summarized in texts S1 and S2 and figs. S14 to S19. Characterization of the processing mechanism, material properties, and shape morphing of DLE-based LIG-SAs. (A) Comparison of the flatness between PI paper and the Kapton film after DLE processing. (B) Schematic diagram of the forming principle and SEM images of the LIG microstructure in focus and defocus modes. (C) Effect of square resistance of LIG under different defocus levels and laser powers. (D) Stability of the deformation curvature under 100 cyclic excitations with a 60-mA current. (E) Quantitative relationship between the maximum curvature and the temperature of the actuator under an 80-mA current. (F) Relationship between the maximum curvature and the sheet resistance of the actuator under different currents. (G) Quantitative relationship between the resistance ratio and the curvature ratio of regionally differentiated actuators under different currents. Design and morphing characteristics of knuckled LIG-SAs with different resistance levels (H), lengths (I), locations (J), and quantities (K) under a 35-mA current. Black scale bars are 2 cm, and white scale bars are 50 μm. Benefiting from accurate process-dependent resistance and curvature control, by designing the LIG-SA unit with conductively differentiated regions connected in series, the curvature is also expected to be controlled independently from region to region. Figure 3G then raises and proves the concept that with a two-region allocation, the resistance ratio (1.3 to 19.1) from part A to part B faithfully reflects a linear dependence on its curvature ratio (1.2 to 12.7), regardless of the applied current level. Notably, although material stiffness determined by specific processing parameters could introduce an adverse effect on actuating performance, the level of resistance is still the dominant factor for governing the Joule heating–induced actuation (as shown in text S3 and fig. S20). Moreover, the laser energy should be carefully optimized within a specific, reliable processing window to reliably balance high actuation performance with high processing robustness (as shown in text S4 and fig. S21). Accordingly, by combining different high- and low-resistance regions, LIG-SAs with various curvature profiles can be specifically processed. Following this idea, Fig. 3 (H to K) systematically designs the morphing shape of 16-Ω/sq-resistance and 85-mm-long actuators by inserting high-resistance regions with different resistance levels, lengths, locations, or quantities. To specify, by increasing the sheet resistance (59 to 173 Ω/sq) of the central region (15-mm width), Fig. 3H shows that the central-area curvature continuously rises to higher levels (0.18 to 0.72 cm−1), while the edge regions remain with smaller curvatures (0 to 0.059 cm−1). Correspondingly, the bending shape is clearer like a rotating knuckle to connect two flat parts, as a higher temperature is concentrated at the center. Similarly, by applying the high-resistance region (122 Ω/sq) with varied lengths (from 10 to 40 mm) or positions (15 to 50 cm from the left edge), both the curvature profile and knuckle part are correspondingly shifted, as shown in Fig. 3 (I and J) and fig. S22. In addition, by introducing two high-resistance regions with different locations (Fig. 3K) or resistance levels (fig. S23), the profile accordingly transforms into two peaks, reflecting the doubled knuckles. For DLS, it is noticed that although the laser-scribing angle determines the bending motion with a consistent curling direction, both the number of curling turns and deviation distance are varied from different processing conditions. To understand this effect, polar coordinates were introduced in Fig. 4 (A and B) to study the process-dependent motions. Specifically, with a fixed current (e.g., for 80 mA), the curling number of turns (Fig. 4A) increases monotonically like a mirror image from 0 to 1.1 when the scribing angle is altered both from −90° to 0° and from 90° to 0°. Comparatively, the deviation distance (Fig. 4B) is changing nonmonotonically. As the scribing angle alters from 90° to 0° (or from −90° to 0°), the deviation distance enlarges first from 0 to 38 mm when the scribing angle is approaching 45° (or −45°), while a further decrease in processing angle introduces an adverse effect to gradually reduce the deviation distance back to 0 mm. Figures S24 and S25 display detailed observations from multiple viewing angles, and fig. S26 shows the 100-cycled durability test. To explain the phenomena, the sheet resistance and bending modulus were compared simultaneously. As shown in Fig. 4C, it is clear that by exerting three-point-bending stress (fig. S27) along the longitudinal direction of LIG-SA specimens, the bending modulus notably shows a linear decrease from 103.4 to 33.4 MPa with the change of scribing angle from 90° to 0° (or from −90° to 0°). This scribing angle–dependent modulus confirms the mechanical anisotropy of LIG-SAs with alternating soft-and-stiff microstripe patterns for advising the curling motion toward the least mechanical constraint, i.e., along the cross-direction of scribing. In addition to anisotropic stiffness, electrical resistance is another critical factor for determining the bending behavior. By testing along the longitudinal direction, the sheet resistance of LIG-SAs reflects a linear increase from 76.3 to 118.2 Ω/sq as the scribing angle changes from 90° to 0° (or from −90° to 0°). It is believed that the ridge-groove microarray introduces highly conductive and low-conductive regions with distinct electron-transport efficiency along or across the scribing direction. Accordingly, the coupled electrical performance and mechanical performance together dictate the curling motion with accurate controls of bending turns and deviations. For the 0° specimen with the highest resistance (i.e., the highest actuation temperature) and the least bending constraint, it shows straight bending with the largest bending turns. For specimens scribed from 15° to 75° (or from −15° to −75°) with a moderate resistance/temperature and bending constraint, they reflect reduced bending turns but dominant directional-curling effects. For the 90° specimen, it is subjected to the largest bending constraint with the lowest resistance/temperature. Thus, it has no space to curl and is visually immobilized. Characterization of the DLS-based LIG-SA and various morphing shapes achieved by a combined DLE and DLS strategy. Number of curling turns (A) and deviation distance (B) of the DLS-based LIG-SA under different currents and processing conditions. (C) Sheet resistance and bending modulus of the DLS-based LIG-SA under different processing conditions. (D) Schematic diagram of the microstructure and optical photographs of the actuator under the two typical processing strategies. M-shaped (E) and S-shaped (F) five-region bridges with differentiated stiffness developed by the combined processing strategy. Various morphing postures of single-knuckle strip (G), double-knuckle strip (H), and multiknuckle cross-actuators (I) by randomly grouping multiple local regions with independent morphing capabilities. (J) Closed-loop structures for diverse planar-to-solid morphing. (K) Deployable structures for 2D-to-3D shape transformations. Scale bars are 1 cm. To better evaluate the combined DLE and DLS protocol with various local-programming strategies, four types of morphing capabilities have been categorized, namely, straight bending (e.g., focused lasing and 0° scribing), directional curling (e.g., focused lasing and 45° scribing), rigid supporting (e.g., focused lasing and 90° scribing), and soft connecting (e.g., defocused lasing and 0° scribing). Figure 4D compares these morphing postures before and after actuation. Before actuation, the two 0° specimens are naturally hanging down because of its soft nature, while the 45° and 90° specimens are clearly stiffer. After actuation, both 0°-focused and 45°-focused specimens are showing regulated straight-bending and directional-curling motions, while 0°-defocused and 90°-focused specimens are showing two styles of immobilizations. For the former, the specimen remains reclinate, manifesting its softness and high conductivity, which is immune to electrothermal excitation. For the latter, the specimen becomes even straighter, proving the 90° inward curling for further stiffening of the unit. Beyond these distinct morphing modes, the four actuator types also exhibit pronounced differences in response speed and load capacity (fig. S28), which directly govern their performance in integrated structures. To better illuminate these capabilities, Fig. 4E compares a five-region bridge in which three planes are connected by two knuckles. When planes are processed by the 0°-defocused condition, the bridge itself is not standing up. When processed by the 90°-focused condition, as a clear comparison, the 80-mA actuated bridge is able to hold a 6.8-g block that is 9.1 times heavier. Similarly, Fig. 4F shows another design in which two edges of the five-region bridge are replaced by the 45°-focused condition for exhibiting an S-shaped morphing structure. To further use the multicapabilities for enabling various complex morphing structures, finite element analyses (FEAs) were performed to accurately predict every local-programming design on the basis of the combined DLE and DLS protocol (fig. S29). Specifically, by randomly grouping multiple local regions with independent morphing capabilities, a series of strip and cross-actuators have been programmed to exhibit a variety of special motions with single, double, or multiple knuckle-like components. For example, Fig. 4G uses a straight-bending or directional-curling knuckle to connect two rigid planes for revealing half-fold or angle-fold effects. Figure 4H then introduces two knuckles to connect three planes for revealing more complex shapes, including asymmetric straight-and-angled fold, equidirectional fold, and antidirectional fold. Figure 4I and movie S1 further group four strip units to assemble cross-actuators, in which every tentacle includes one or two knuckles for revealing motions of radial inward crimping, unidirectional rotational pleating, and double-reverse compound folding. In addition to strip and cross-structures, the shape of LIG-SAs also can be designed specifically such as closed-loop and deployable structures with fancy 2D-to-3D shape transformations (movie S2). Accordingly, Fig. 4J demonstrates diverse planar-to-solid morphing from circular, square, hexagonal, and star-shaped (fig. S30) 2D precursors for reaching various 3D geometries with localized bulging and buckling deformations. It is deducible that the hybrid soft-connecting parts sandwiched in between the bending or directional-curling parts are critical for triggering these out-of-plane deformations. Similarly, with specific distribution of rotating and rigid-supporting components, Fig. 4K demonstrates other three structures that are shiftable between compact and deployed states, in which a crossed sheet is self-folded into a closed cuboid, a hexapod membrane is synchronously inward folded into a lanternoid skeletal configuration, and a furled six-petal flower is able to bloom. Bioinspired design of LIG-SA–based soft-robotic devices toward multifunctional and multiscenario applications On the basis of successful exploitation of LIG-SA units for achieving precise and freeform morphing capabilities, it is then promising to stride into device level for designing soft robots toward diversified locomotion functionalities. Accordingly, by exploiting local-scale and multimodal programmability, different biomimetic concepts have been applied to design a series of bioinspired grasping robots and crawling robots. For grasping robots, an octopus tentacle is first inspired with orthogonal muscular hydrostat kinematics and viscoelastic energy management for achieving unconstrained morphological reconfiguration during prey capture. Then, single-tentacle robots with three modes of locomotion were presented, including forward curling for swarm collection, distal reflexing for confined-space probing, and helical winding for slender-object extraction. In the first mode, 0°-focused condition was applied to enable straight bending capability. With a high sheet resistance (122 Ω/sq), the tentacle is able to coil at ~720° under 80-mA excitation. On the basis of this feature, by mimicking the adhesion-coiling synergy of an octopus tentacle, Fig. 5A and movie S3 demonstrate the locomotion of a robot for capturing tiny particles by continuously anchoring several tens of the item through surface sticking. After full collection, the device is then coiled to transport the prey to the final destination. In the second mode, 90°-focused condition was primarily applied to lock the main body stiff, while only the distal segment was 0° processed for presenting a buckling motion. By mimicking the octopus tentacle for tip-initiated food exploration, Fig. 5B and movie S4 demonstrate the complete locomotion of a robot for inserting into a 3-mm-wide crevice for retrieving an annular item through localized distal buckling that enables interlock hooking. In the third mode, 45°-focused condition was applied to present an entire-body helical-winding motion. By mimicking surface wrapping of foods via muscular hydrostat reconfiguration, Fig. 5C and movie S5 demonstrate the complete locomotion of the bionic tentacle. Specifically, it uses helical-winding deformation to freely navigate through a 4-mm-diameter crevice in a long-tube container to extract a slender object via conformal surface wrapping followed by progressive circumferential constriction. Octopus-inspired tentacle grasping robots for conformally grasping multimorphic targets. The illustrations are created by Blender. (A) Adhesion-coiling synergistic grasping robots performing forward curling for swarm collection. (B) Hook-grasping robots performing distal reflexing for confined-space probing. (C) Helical-winding grasping robots for slender-object extraction in confined spaces. (D) GR-I for conformally capturing sphere targets. (E) GR-II for conformally capturing rectangular targets. (F) GR-III for conformally capturing cone targets. (G) GR-IV for conformally capturing combined targets. Scale bars are 1 cm. In addition to a single-tentacle device, multitentacle robots were further demonstrated for grasping products with all kinds of shapes. Inspired by an octopus preying through coordinating multiple tentacles with specialized arm heterogeneity for highly efficient food grasping via conformal envelopment, the cross-architecture by spreading four tentacles was designed to fulfill topography-conformal grasping through arm-specific curvature profiling. Accordingly, Fig. 5 (D to G) and movie S6 present four cross-architected grippers, in which the type I grasping robot (GR-I; 0°/0°/0°/0° configuration) is able to capture sphere/square targets by uniformly straight-bending all four tentacles (Fig. 5D and fig. S31), the type II grasping robot (GR-II; 45°/−45°/45°/−45° configuration) is able to capture rectangular targets by inward-curling all tentacles to hug the object through a narrow side (Fig. 5E), the type III grasping robot (GR-III; 0°/−45°/45°/0° configuration) is able to capture cone targets by straight-bending left-two tentacles to hug the sphere head and inward-curling the other two to fix the cone tail (Fig. 5F), and the type IV grasping robot (GR-IV; −45°/−45°/45°/45° configuration) is specialized to capture a combined rectangular-cylinder structure, in which the left two are outward curled to push against the board, while the other two are inward curled to hug the cylinder part (Fig. 5G). Along with grasping robots, prototypes of multiple crawling robots were also developed by implementing varied biomimetic motion principles. Initially, although uniformly processed strip LIG-SAs could achieve crawling or climbing through periodic bending motions (fig. S32), the locomotion can only function properly with the assistance of specialized surface protrusions. To avoid this limitation, coordinated multisegment morphological programming and adaptive gait synchronization were applied to enable soft crawling robots with varied types of bioinspired multimodal locomotion. In Fig. 6A, by mimicking the peristaltic movement of inchworms with a discrete anchoring-release mechanism, a type I crawling robot (CR-I) was first developed by designing a dual-strip device with a bilateral architecture for achieving bidirectional locomotion. In detail, two 6-cm-long strip units are aligned end to end to be driven independently by two separate sources. Thus, by coordinating two independent bending motions (0° processed), the on and off states of excitation sources can be specifically programmed to actuate each strip with an alternating sequence for moving forward or backward (as shown in fig. S33A and text S5). For example, when the right source is on, the right part of CR-I is buckled and anchored, which generates a dragging force to let the left strip move rightward. Then, by shifting the on state to the left source, the left CR-I becomes buckled, while the right strip turns relaxed. By lastly turning off the left source, the anchored left strip is gradually relaxing while pushing the right strip to migrate forward. Thus, by sequentially providing these buckling-relaxing cycles, the inchworm-inspired peristaltic movement has been achieved and further confirmed by FEA results, as shown in fig. S33B. As displayed in Fig. 6A and movie S7, the crawling velocity of CR-I reaches 36.5 mm/min by migrating a total distance of 198 mm (99 mm forward and 99 mm backward) under 642-s operation. Furthermore, quantitative benchmarking is summarized in fig. S34, table S1 and text S6, revealing the competitive performance of CR-I among thermally actuated robots with bidirectional crawling capabilities. To further accelerate the locomotion, an active cooling strategy is tentatively implemented (as shown in fig. S35 and text S7), suggesting substantial potential for performance enhancement. Inchworm- and seal-inspired crawling robots for multimodal locomotion. The illustrations are created by graphics software Adobe Photoshop. (A) Inchworm-inspired CR-I for bidirectional linear movement. (B) Seal-inspired CR-V for circumferential steering movement. (C) Seal-inspired CR-Y for exhibiting both steering and wriggling movements. (D) CR-Y traversing a maze from entrance to exit following a specified route. Scale bars are 1 cm. In addition to CR-I with straight-line movements, a type V crawling robot (CR-V) and type Y crawling robot (CR-Y) were further designed to mimic the terrestrial gaits of seals by using multiple flippers. First, through tripodal bracing, seals are able to achieve pivot steering with two foreflippers for providing anchored propulsion and the hindflipper for maintaining balance. Inspired by this movement, CR-V comprises a pair of discrete limb units for presenting alternated curling motions (45° processed, 5 cm long), while a rear unit is for stabilization (90° processed, 4.5 cm long), as depicted in Fig. 6B. Similar to CR-I, CR-V could achieve leftward and rightward rotational locomotion by alternately actuating the two limbs for activating the anchoring-curling-steering sequences. Movie S8 displays that CR-V costs 29 s to rotate 27° in every actuation cycle. Accordingly, a 360° omnidirectional left/right rotation has been achieved with a total time of 344 s. On the basis of CR-V, CR-Y was further designed by exhibiting both steering and wriggling movements. As demonstrated in Fig. 6C, with bionic inspiration of seals for using coordinated fore/hindflippers to accomplish undulatory crawling, a three-channel power source is applied to control the curling motion of two anterior units (45° processed, 2.5 cm long) and the bending motion of the rear unit (90° processed-0.5 cm long and 0° processed-4 cm long). In addition to rotational movement by merely controlling anterior units, CR-Y is also able to crawl with synergistic actuation of all the three units. Specifically, when all sources are initially turned on, CR-Y lies in a buckled state, while the two curled limbs anchor the ground to force the bending tail to migrate. Subsequently, after the anterior sources are off, the two limbs accordingly relax to shift the anchor point to the rear. When the rear source lastly goes off, the relaxing tail then pushes the whole body to move forward. By combining steering and crawling movements, CR-Y could ideally relocate to anywhere in a 2D map. Movie S9 gives an example that CR-Y is crawling a linear distance of 172 mm in 637 s before exhibiting leftward turning by 128° for 305 s, followed by another 34.3-mm straight crawling for 179 s. To further demonstrate the path-adaptive capability with goal-oriented locomotion, Fig. 6D and movie S10 present a complete process of CR-Y to accomplish a maze obstacle to migrate from its entrance to exit by using the multimodal movement of straight crawling and left/right steering. Similar to bidirectional crawling robots, quantitative benchmarking among multilegged angle-turning robots is also summarized in fig. S36, table S2, and text S6, revealing the competitive performance of CR-Y for achieving omnidirectional movement. This advantage stems from the unique design of two front legs in CR-Y, which can undergo large, inward-spiral deformations, allowing for a substantial turning angle per actuation cycle while preserving efficient crawling. With path-planning navigation, it is also expected that the biomimetic crawling robot could perform multiscenario tasks by additionally leveraging its intrinsic electrical conductivity, Joule-heating capability, and adaptive locomotion modalities (Fig. 7A). Accordingly, Fig. 7 (B to D) proves that the CR-Y–type robot could perform obstacle negotiation, electrothermal ice ablation, and autonomous circuit restoration. In Fig. 7B and fig. S37, the soft robot with polymorphic locomotion is able to implement a slit-crossing strategy via adaptive tripodal gait sequencing, in which its dual forelegs could alternatively traverse the 1-cm-wide fissure through rotational motions, while the hindleg subsequently executes high-amplitude bending to propel the body over the gap (movie S11). In Fig. 7C, the robot is able to execute an autonomous deicing strategy by providing tripodal envelopment grip to conformally contact the ice and release high-temperature Joule-heating energy. As evidenced in movie S12, the activated robot is self-heated to 272°C, which is able to highly efficiently melt an irregular ice block (21 g) within 26 min. In Fig. 7D, the robot is also able to provide a strategy of in situ circuit restoration by autonomously bridging fractured conductors. As displayed in movie S13, after perching across a damaged circuit, the robot itself could supply an ohmic path on the basis of its silver-coated belly with a high conductivity of 3.2 × 107 S/m, thus ensuring immediate recovery of light-emitting diode (LED) lights. With multimodal functionalities, Fig. 7E and movie S14 lastly demonstrate that the soft robot could comprehensively accomplish various challenging tasks. Specifically, when the mini city meets power failure, the biomimetic robot autonomously navigates in a confined pipeline, including undulatory crawling, pivot steering, and slit crossing through adaptive gait shifting. After it melts and breaks through the ice wall through electrothermal heating (figs. S38 and S39), it eventually arrives and detects the fault site to reconnect the circuit for swiftly recovering electrical power. Multifunctional performance of CR-Y for multitask obstacle avoidance. (A) Schematic illustration of multifunctional robot performances under harsh outdoor condition (the illustrations are created by graphics software Adobe Photoshop). (B) Slit crossing by adjusting the locomotor gait and moving route. (C) Ice cube melting through electrothermal deformation. (D) Disconnected circuits restoring by equipping with conductive silver paste. (E) Multitask performance including melting ice walls, crossing slits, changing moving direction, and recovering power systems in pipelines under cold environments. Scale bar is 1 cm. DISCUSSION In conclusion, this work addresses the critical demand for soft actuators capable of complex and reconfigurable 3D shape morphing in next-generation bioinspired soft robotics. Moving beyond conventional stimulus-responsive strategies, which often sacrifice kinematic sophistication for fabrication simplicity, we have developed a spatially differentiated laser-programming technique that enables digital manufacturing of LIG-SAs with freeform morphing capabilities. By implementing cross-scale control of lasing energy and scribing direction, we simultaneously tailor material heterogeneity and structural hierarchy, thereby introducing a decoupled electrothermal response and stiffness anisotropy. This allows the encoding of four fundamental motion primitives: straight bending, directional curling, rigid supporting, and soft connecting. Arbitrary combination of these multimodal units facilitates the freeform design of integrated bioinspired robots, such as octopus-like tentacles and inchworm- or seal-inspired crawlers, capable of performing multitask locomotion including conformal grasping, path navigation, and obstacle avoidance. This methodology not only overcomes the traditional trade-offs between fabrication flexibility and actuation complexity but also establishes a seamless bridge between digital design and physical intelligence. As such, it opens previously unidentified pathways for creating soft robots with sophisticated, programmable morphologies, substantially advancing the frontiers of adaptive and intelligent robotic systems. MATERIALS AND METHODS Fabrication of the LIG-SA The LIG-SA fabrication process commenced with laminating a 50-μm Kapton tape onto 90-μm PI paper (PolyKing, cat. Yilun-P-90) to form a composite substrate. A 10.6-μm-wavelength CO2 laser system (Universal Laser Systems DLS 2.3) with a fixed-focus spot diameter of ~100 μm was used to selectively graphitize the PI surface following computer-aided design paths. Laser parameters including defocus level (0 to 3.5 mm) and power (0 to 25 W) were programmatically controlled through integrated motion software. Post–laser processing, conductive silver paste (Jingling Co.) was applied to predefined electrode regions, forming ohmic contacts with copper wires. A PDMS solution (10:1 base-to-curing agent ratio; SYLGARD 184, Dow Inc.) was spin-coated (1000 rpm, 60 s) onto the graphene-patterned surface, followed by thermal curing (80°C, 1 hour) to create a 100-μm active layer atop the 140-μm LIG-Kapton substrate. Final actuator geometries were obtained through precision laser cutting along predefined contours. This four-stage process, comprising substrate preparation, laser graphitization, electrode integration, and encapsulation, enabled rapid fabrication of customized actuators within 2 hours and allows batch manufacturing of multiple samples in one substrate. Fabrication of the multijoint LIG-SA The 85-mm-long parameterized prototype integrates two functional domains: high-resistance joint regions [laser parameters: 2 inch (5.08 cm)/s, 1 W, scribing direction = 0°, DL = 0 mm; red lines in design schematic] and low-resistance connection regions [2 inch (5.08 cm)/s, 2 W, scribing direction = 0°, DL = 2.5 mm; black zones in design schematic]; the remaining fixed parameters of the design are velocity = 2 inch (5.08 cm)/s and pixels per inch = 1000. The assembly of the components is the same as the general processing steps of LIG-SAs. Fabrication of various LIG-SA–based soft-robotic devices The lantern actuator precursor was fabricated by stacking two preprepared hexagonal strip actuators in a face-contact configuration using Kapton tape, with the ends of each strip bonded using additional tape. Both hexagonal strip actuators were connected to one source; upon activation, the lower actuator lifted the upper unit via electrothermal expansion. The flower actuator precursor was formed by folding the prepared array structure into a cylindrical form, with its initial and terminal ends bonded by Kapton tape to achieve a self-standing 3D structure. All grasping robots were adhered to a 10-cm plastic rod using adhesive postfabrication to enable operator-assisted spatial transfer. The circuit-repair functionality of CR-Y was implemented by brushing liquid silver paste onto the bottom surface of the Kapton tape with a soft-bristle brush, followed by natural air-drying to solidify the conductive layer. The adhesion-coiling synergistic grasping robots were integrated by attaching 0.05-mm double-sided tape (3M, 55256) to the PDMS side of the robot at equal intervals. Construction of the interactive environment of the soft robots All targets manipulated by the grasping robots were fabricated from foam with a density of 25 kg m−3, shaped into specific geometries using customized cutting tools. For the ice-melting experiments of CR-Y, ice blocks were prepared by freezing aqueous solutions containing 0.1% (w/v) methylene blue in custom molds. The thermochromic properties of methylene blue induced a purple-to-blue transition during melting, which served as a visual phase-change indicator. In the circuit-repair experiments of CR-Y, disconnection sites were modified with copper foil tape to establish robotic contact points. Structural and material characterizations A dc power source with three independent output channels (IT6300) was used to provide input current, and a FLIR IR camera (T420) was used to monitor the electric heating temperature. A self-developed LabVIEW program was used to monitor the resistance change of LIG-SAs in real time. The response of the actuator under different currents was analyzed by curve fitting using the graphics processing software ImageJ to obtain the curvature information. Scanning electron microscopy (SEM; Helios G4 CX) was performed to characterize the microstructure and thickness size of LIG-SAs. X-ray diffraction (Empyrean) and Raman spectra (LabRAM HR Evolution) were used to test the chemical composition and Raman spectrum of LIG, respectively. The quality of graphene was studied by calculating the integrated intensity ratio of the D peak to the G peak (ID/IG) and the integrated intensity ratio of the 2D peak to the G peak (I2D/IG). The resistance of the LIG was measured by a digital multimeter (Keithley 2110) using the four-probe method (Universal Probe). The force-deflection curves of LIG-SAs prepared under different processing parameters were tested using a dynamic mechanical analysis system (DMS Q850, TA Instruments, US). The dimensions of the samples were cut to 2 by 5 cm and subjected to three-point bending tests in a fixture. FEA simulations The computer-aided engineering module in the commercial software ABAQUS (2022 version, Dassault Systems) was used for FEA to predict the deformation effect of LIG-SAs with different design forms. The temperature distribution field and strain field of the model were calculated by inputting the electric power value by using the electric-thermal-mechanical coupling calculation method. To simplify the modeling and calculation cost while ensuring the calculation accuracy, the PI film composed of Kapton, PI paper, and LIG was assigned according to the material properties measured experimentally. The LIG substrate and PDMS were geometrically modeled according to the microstructure morphology in the SEM image to ensure the anisotropic characteristics under different scratching orientations. Given that the materials of LIG-SAs do not involve large deformation, the linear elastic constitutive model is used for the calculation of the substrate and PDMS. The material parameters of PDMS include Young’s modulus of 300 kPa, Poisson’s ratio of 0.45, an electrical conductivity of 1 × 10−7 (S/m), a thermal conductivity of 0.18 W/(m2·K), and a thermal expansion coefficient of 3.1 × 10−4. The material parameters of the LIG film include Poisson’s ratio of 0.2, a thermal conductivity of 50 W/(m2·K), and a thermal expansion coefficient of 1 × 10−5. Its electrical conductivity was assigned according to the test results depending on the scratching direction. In the kinematic simulation calculations of friction, the CR-I was modeled as a bilayer 3D deformable part, with its geometry and dimensions adopted from experimental measurements. The model was meshed with quadratic tetrahedral elements (C3D10H). The ground was defined as a rigid body. A uniform friction coefficient of 0.3 was assigned to the bottom and side contacts between the CR-I and the ground. The variation in frictional force induced by structural deformation and movement was computed using a dynamic explicit analysis and normalized by the self-weight of the soft robot.