Electrostatic artificial muscles for cable-driven actuation of compliant mechanisms

npj Robotics volume 3, Article number: 12 (2025) Cite this article

Metrics details

Dielectric elastomer actuators (DEAs) are soft, electrically-driven artificial muscles with high energy density and high bandwidth. As work requirements increase, the actuator volume must also increase. Integrating sufficiently large DEAs within mechanical linkages for robotic applications can be challenging since the actuators can be of comparable size to the mechanism itself (as with human forearm muscles and hands). Here, we demonstrate a way to use DEAs to power cable-driven mechanisms, thus allowing the actuator to be separated from the mechanism, enabling modularity in design. We detail the manufacture and characterization of scaled-up rolled DEAs for increased work output via a sequential roll-on-roll method, and cable integration for remote actuation. This cable-DEA architecture is used to actuate a pinching gripper, a multi-degree-of-freedom mechanism, and a soft end-effector. This work illustrates a promising method to pair the unique actuation characteristics of DEAs to rigid-material mechanisms.

Dielectric elastomer actuators (DEAs) are a promising class of artificial muscles due to their combination of electromechanical transduction and compliant nature. Previous efforts in creating soft actuators1,2 include cable-driven structures3, fluidic or pneumatic inflatable actuators4,5, shape-memory actuators6, magnetically-responsive materials7,8, and liquid crystal elastomers9. DEAs (as well as other electrostatic actuators10,11,12) present advantageous properties including direct electrical-mechanical transduction, light weight, high bandwidth, and fast response time, as well as the ability to operate silently13. Their structure, characterized by thin elastomer layers sandwiched between compliant electrodes, enables efficient storage and release of mechanical energy which can generate substantial strains (on the order of 100%) and stresses (on the order of 1–10MPa) that can be used to power robots in a volumetrically-efficient manner14. These properties make DEAs an ideal candidate for applications in human-robot interaction, delicate manipulation, healthcare, haptic interfaces, and prosthetics15.

DEAs can be configured to create different actuation modes including biaxial (planar) expansion, uniaxial extension and contraction, and bending15,16. The early examples of DEAs involve a stretchable membrane consisting of an elastomeric film with conductive electrodes patterned on the top and bottom, prestrained into a frame17. When an electric field was applied, the portion of the membrane subjected to the field expanded in the plane17,18,19,20. Non-circular DEA membranes can also be integrated into flexible frames that are curled at rest, and un-bend when actuated21,22,23,24. Planar expansion has been harnessed to generate linear motion by mounting an unstrained DEA membrane in a rigid frame, and using a biasing element normal to the membrane to create pre-strain25,26, transforming in-plane extension to out-of-plane motion. To directly access uniaxial motion, researchers have built multilayer DEA stacks and rolled DEAs that contract and extend, respectively, when an electric field is applied. Multilayer DEA stacks typically consist of tens to hundreds of alternating layers of electrode and dielectric material accomplished either by layer-by-layer deposition27,28,29 or by successive folding steps30 with displacement proportional to the number of layers. Rolled DEAs commonly consist of long DEA strips that are rolled to form a cylindrical structure that extends when an electric field is applied31,32,33,34,35,36,37. In addition to free-standing rolls, DEAs can also be pre-strained (in-plane), then rolled around a spring core (or wrapped around a pre-compressed spring core) which serves to increase the electric field and increase the stiffness of the actuator38,39,40. In this work, we will focus on rolled DEAs because they transform a biaxial planar expansion into linear extension along the axis of the roll, which is more amenable to transferring loads and displacements. In addition, rolling is a promising method for scaling up the DEA volume compared to potentially labor-intensive stacking that could require hundreds or thousands of layers to create muscle-like volumes.

Given the diversity in the geometry and functionality of DEAs, researchers have sought to use these soft actuators for robotic applications15,41,42,43,44. DEAs commonly interact with the external environment or objects with the materials comprising the actuator itself – the body and the actuator are one and the same – whether in the form of a membrane22,23,45,46, or as rolled DEAs for grasping and locomotion tasks37,40,47,48. Rolled DEAs have also been used to drive the motion of linkages, such as those used for flapping wing robots49,50,51 where the ends of the rolls are directly attached to the wing transmissions. Alternately, rolled DEAs have often been configured with one end that moves and the other fixed rigidly to a frame. Applications include using rolled DEAs as fluidic pumps and valves34,35,52, wearable haptics53,54, or as antagonist bending units32,55. In these previous examples, the DEAs are co-located with the driven mechanism or are themselves the mechanism that interacts with the environment. Researchers have proposed (but have not yet realized) using spring-roll DEAs to pull on cables for an orthotic device to increase hand strength56.

Despite the impressive progress in the field of DEAs, their integration into more complex structures has been limited. The forces and displacements that a DEA can produce scale with the volume of the DEA. As larger forces and displacements are required, the size of the DEA may become prohibitively large for direct integration and actuation of complex mechanisms, matching or exceeding the size of the structures being actuated. In order to overcome this mismatch in size, we introduce a method to translate DEA actuation to cable actuation. This approach loosens the restriction of co-locating the DEA and the target mechanism as the cable is able to transmit the force and/or displacement generated by the actuator, and the placement and size of the DEA can thus be decoupled from design of the mechanism in a modular fashion.

In this work, we present a large-volume rolled DEA coupled with various compliant mechanisms, including a gripper, a multi-degree-of-freedom linkage, and a silicone end effector. Figure 1A, B shows the 4.42g DEA-powered gripper which is able to carry objects of various shapes up to 5.5g. To achieve this, we developed a method to integrate a cable inside a rolled DEA in order to transform the motion of the end-plate into tension in the cable (Fig. 1C, D). Additionally, we describe a novel approach to scale up the size of rolled DEAs, further increasing the output force and expanding the potential applications of this actuation method. The modular designs and methods we present expand the possibilities for practical implementation of rolled DEAs in robotics, adding to an otherwise limited palette of electrically-driven artificial muscles.

When the rolled DEA is actuated (A, B), the extension of the actuator (C) pulls on a cable running down its center (D). Here, actuation of the DEA causes the cable to pull on the distal linkage of a single-degree-of-freedom finger, causing the joint to bend (red arrows in (B)). A pair of these fingers can be actuated together to perform a pinch grasp.

Prior research on rolled DEAs has shown promising results from the millimeter to the meter scale33,57; however, these previous demonstrations use rolled DEAs integrated directly into the device, meaning that the volumetric requirements of the DEAs must be considered in parallel with the design of the overall structure. Here, we build on previous work by Xu et al. on millimeter-scale rolled DEAs (7mm active height, 3.2 mm diameter)35, toward larger volume DEAs for higher forces and displacements appropriate for larger-scale robotic applications such as interacting with everyday objects. The DEAs presented here are fabricated using materials and layer thicknesses identical to ref.35 for direct performance comparison. The silicone elastomer dielectric exhibits low viscous losses (compared with acrylic elastomers)58, and the CNT electrodes add minimal mechanical impedance and are capable of self-clearing electrical defects prior to use59. The overall design also remains similar, consisting of six CNT electrode layers (here, scaled up in area) stacked in an alternating pattern between silicone elastomer layers before being cut into a rectangular strip and rolled onto itself (Fig. 2).

Note that figure elements are not to scale. A Electrodes were manufactured via masked spray deposition of CNT ink and IPA onto filter paper. B The layering process consisted of spin coating an elastomer layer, followed by stamp-transferring CNT electrodes onto the cured elastomer, repeated six times. The resulting multilayer DEA was then cut out using a vinyl cutter. C The multilayer strips were then rolled to form a hollow cylinder. Note that the schematic simplifies the six-electrode DEA strip to a two-electrode DEA strip, illustrating the offset between the two sets of electrodes for voltage and ground connection (See Supplementary Fig. 1 for more details). Following the first roll, a second strip can be rolled on top to increase the cross-sectional area of the actuator, as shown in (D). E After rolling, the cable, tubing, and end-plates were integrated to transform the DEA extension to cable tension. Carbon conductive adhesive was used to form the electrical contact between the carbon fiber end-plates and the exposed edges of the electrodes.

To scale-up our rolled DEAs, we started by manufacturing larger electrodes using a spray deposition process60 and an electrode formulation presented by Cohen, et al.61. This process, illustrated in Fig. 2A, consisted of spraying a carbon nanotube ink onto a polytetrafluoroethylene filter paper through a Mylar mask into the desired shape, with each electrode having dimensions of 120 mm × 20 mm (Supplementary Fig. 1A). These electrodes were then stamp-transferred onto 32 μm thick layers of spin-coated elastomer, using a roller to ensure uniform transfer, as illustrated in Fig. 2B. The layering process was repeated six times, followed by encapsulation of the final electrode (Supplementary Fig. 1B). The DEA multilayer strips were then cut from the substrate using a vinyl cutter such that the longer edge of the electrodes was exposed, while the shorter ends remained “capped” with a 2 mm margin of silicone to aid with rolling and prevent shorting across opposing electrodes (Fig. 2C; Supplementary Figs. 1C, 2). Using these 124 mm × 20 mm × 224 μm strips, we first created single-rolled DEAs by rolling one strip onto itself, relying on the inherent tackiness of the silicone to keep the DEA from unrolling (Supplementary Fig. 1D). To increase the cross-sectional area of the resulting rolls (proportional to the peak force production), we also tested an iterative scaling method by rolling another six-layer DEA around the single-rolled DEA as shown in Fig. 2C and D. These rolls (“double-rolled DEAs”), are particularly interesting because they demonstrate a way to easily build larger rolled DEAs for generating higher forces without the need for larger areas of the individual strips (with the area practically limited by the spin coater). Working with multiple smaller strips is advantageous because they are more compatible with traditional lab equipment and because potentially defective strips can be replaced prior to further assembly. As the strip size increases, the likelihood of incurring a defect during manufacturing (e.g., non-uniformity in CNT deposition, debris, improper lamination) also increases. Discretizing a large active area into smaller strips allows faulty strips to be identified and discarded prior to assembly into the final larger volume. To quantify the effects of this novel roll-on-roll scaling method, we also fabricated double-half rolled DEAs, which had the equivalent active volume to the single-rolled DEAs, made from two strips of dimensions 64 mm × 20 mm × 224 μm rolled one after the other, for comparison with a single-rolled DEA. The electrode areas on these double-half rolled DEAs were 60 mm × 17 mm, half the length of the single-rolled DEAs.

After rolling, the actuators were capped with carbon fiber end-plates, which serve both as electrical conductors and structural elements for mechanical mounting (Fig. 2E). Finally, we cleared away potential defects within the rolls using a pre-clearing process developed by Cohen, et al.61. Indeed, as we operate these actuators up to several kV, structural defects in the elastomer can trigger dielectric breakdown which can severely damage the device and lead to premature failure. The protocol consists of applying a gradually increasing voltage to the DEA while monitoring the current to identify defect-indicating behaviors (typically prolonged periods of relatively high current) which are characteristic of defects. When a defect is detected, the voltage increase is immediately stopped, and an alternating current burst is sent to the device which causes the controlled destruction of the CNT particles located around the defect, isolating the defect and preventing further breakdown in the vicinity. This intentional destruction of problematic regions of the DEA also leads to a reduction in the active volume of the DEA and is reflected by a reduction in the DEA’s capacitance. By tracking the evolution of the capacitance retention Cre (and the predicted Maxwell stress σ) of several test devices using an LCR meter, we determined the operational limit of each rolled DEA61,62 (Supplementary Fig. 3). For all devices in this work, this limit, corresponding to the pre-clearing voltage that maximizes the force and displacement we can generate, was found to be 1250V and was, therefore, the peak voltage applied to our devices during operation.

Once the devices were pre-cleared to 1250 V, we measured their capacitance using an LCR meter (Keysight E4980A) and compared them to theoretical estimates obtained using the equation:

where nlayers is the number of layers in the DEA strip (here, five), ϵp is the permittivity of the elastomer, lDEA is the length of the unrolled DEA, hDEA is the active height of the DEA, and dlayer is the thickness of each elastomer layer (values listed in Table 1). The active volume of a DEA can be calculated as the product of lDEA, hDEA, nlayers, and dlayer. The single DEAs and double DEAs had an active volume increase by a factor of 5.83 and 11.66, respectively, compared to the devices developed by Xu et al.35; the capacitance scales similarly, increasing by a factor of 5 for the single DEA, and 9.1 for the double DEA.

We characterized the actuation of each DEA at voltages from 600V to 1250V using a ramp-and-hold input signal, holding at each voltage level for five seconds under blocked force and free displacement conditions (Fig. 3, additional replicates of single-rolled DEAs in Supplementary Information section S.5). In both tests, the DEAs’ bottom end-plate was mounted to a platform with a hole in the center so that the cable could move freely during actuation. To measure blocked force, the load cell (Transducer Techniques, MDB-2.5) was mounted rigidly such that it was in contact with the top end-plate without exerting any pressure on the DEA when unactuated (Fig. 3B). Free displacement was measured using image analysis (Adobe Illustrator) of the DEA actuating alongside a ruler aligned with the axis of the DEA (Fig. 3D). Figure 3A and C show the results of the blocked force and free displacement tests, respectively (n = 5 at each voltage for each device, error bars show standard deviation about the mean) and compare them to the following theoretical predictions.

Four DEAs were tested, two double rolls made from two 124mm long strips, a single roll made from one 124 mm strip, and a double half roll made from two 64 mm strips. The plots show the mean and standard deviation across five trials for each test condition. A Blocked force of the manufactured DEAs compared with the theoretical value obtained using Zhao’s model at different voltages33. B Experimental setup to measure the blocked force using an MDB-2.5 load cell. C Free displacement of the DEAs compared with the theoretical free displacement value (same for both single and double DEA designs) obtained using Zhao’s model at different voltages33. D Experimental setup to measure the free displacement of the rolled DEAs. Both photos in (B) and (D) show a double-rolled DEA.

Based on the rolled DEA model developed by Zhao et al.33, we estimated the expected blocked force as a function of the roll geometric parameters as follows:

where tDEA is the thickness of the DEA strip (= nlayersdlayer), and Vin is the voltage input. Similarly, we can estimate the free displacement as follows:

where Y is the elastomer’s Young’s modulus (147.6 ± 16.6 kPa, Supplementary Fig. 4).

We note here that Eqs. (2) and (3) are approximations for extension and force generation purely in the axial direction. These equations have simplifying assumptions that ignore the effects of gravity, the boundary conditions (i.e., at the cap locations), radial deflection of the DEA walls, and represent an idealized rolled DEA that is free of defects. The force due to gravity for the double DEA is approximately 0.01 N, much smaller than the blocked force of the actuator. When the boundary conditions are neglected, the DEA is assumed to have a constant radius along the length of the actuator. During actuation under free displacement conditions, we visually observed that the DEAs did not exhibit significant radial deformation (see Supplemental Video 1). However, we expect that in reality, non-ideal geometries are not realized and measured values will underperform these predictions.

Considering the scaling in these equations, we compared the performance of the scaled up DEAs, benchmarked against Xu, et al.35 in Table 1. The blocked force is expected to scale proportionally to the cross-sectional area of the rolled DEA (lDEAnlayersdlayer) (Eq. (2)); here, the thickness is constant, and so we expect force generation to scale with the length of the strip. The single rolls were made from a strip 2.4 × the length of Xu’s and produced a force 2.6 × higher, while the double rolls were 4.8 × longer and yielded a force 4.3 × higher. The free displacement is expected to scale with the height of the strip (Eq. (3)); we measured displacements up to 4.0 × higher with an active height only 2.4 × larger. Additionally, we calculated the energy density of the DEAs as half of the product of the blocked force and free displacement at the maximum voltage (to approximate the integral under the force-displacement curve, assuming a linear behavior) and then normalized by actuator mass. The energy densities of the scaled-up actuators were similar to the quasi-static result of previous work using the same materials and manufacturing methods35. Other works on DEAs have reported energy densities ranging from 1 to 35 J/kg. These DEAs have higher energy density due to higher dielectric strength materials which allow for application of larger electric fields (ref. 52: 35 J/kg63, 22 J/kg28, 19.8 J/kg), or by operating the DEAs at resonance (ref. 35: 4.6 J/kg,62: 3.6 J/kg51, 1.75 J/kg). Our devices’ energy density could similarly be improved by tuning material and geometry to adjust their resonance to an appropriate range, as well as using materials with higher dielectric strength to optimize energy output to drive the mechanism.

At 1250 V, the double DEAs generated a maximum blocked force of 0.69 N and a maximum displacement of 1.84 mm, while the single and double half DEA reached blocked forces of up to 0.41 N and 0.38 N, and a maximal free displacement a 2 mm and 1.56 mm, respectively. We observed that the single DEAs (Single and Double-Half) generally outperform the double DEAs, compared with their respective model predictions, or when considering their force output normalized by length of the active area. Indeed, at 1250V, the single and double-half DEAs were able to reach 96.9% and 90.0% of the theoretical force and 77.4% and 59.6% of the theoretical displacement, while the double DEAs only reached between 76.7% and 81.5% of the expected force and between 59.7% and 70.8% of the expected displacement.

The performance of the single DEA and double-half DEA, both of which theoretically have identical active volumes, is closely matched in the blocked force performance but the double-half DEA exhibits reduced free displacement. By partitioning the longer DEA strip into two shorter half segments, an additional margin volume at the end of each strip is incorporated into the actuator. In the blocked force configuration, no deformation occurs (in theory), and the effect of this additional passive volume is not observed. However, in the free displacement case, the additional passive silicone acts as a spring opposing the motion of the DEA, resulting in reduced displacement. Moreover, any misalignment between strips would further hinder displacement—active regions layered upon inactive regions will result in bending or radial deflection. Similar issues plague the free displacement performance of the larger DEAs.

For the larger double DEAs, we believe the performance deficit in both cases can be further attributed to defects incurred during assembly (e.g., slight misalignment of strips when rolling) that is amplified due to the increased length of the strip. Specifically, as the strips were visually aligned and cut using a vinyl cutter, the active area is not necessarily centered inside the cutout rectangle. As the strips’ sides have to be carefully aligned to form a flat surface once rolled to ensure a good electrical connection with the end plates, a part of the active area of both strips can overlap with an unactuated part of the other strip.

While the roll-on-roll fabrication method did indeed yield larger DEAs capable of higher force output, there remains a gap in performance relative to model predictions. In addition to assembly defects, another source of performance deficit is due to loss of active volume due to pre-clearing. A localized “dead” zone in the DEA not only reduces the active volume of the actuator but also restricts actuation of neighboring intact regions. Pre-clearing is intended to eliminate volumes of the actuator that are serving as short-circuit paths between the high-voltage and ground electrodes. In reality, however, a complex relationship exists between the voltage and current limit applied during the pre-clearing process and the capacitance and defect quality and density of the device under test. The result of this conditioning is a device that can sustain the assigned operating voltage but that has undergone irreparable damage, which can be measured via proxy as the capacitance of the device. We observe that the capacitance after pre-clearing for the double DEAs drops below the predicted capacitance, whereas the single and double-half DEAs do not (Table 1). This suggests that the pre-clearing process cleared away a larger fraction of the active volume of the double DEAs compared to the smaller DEAs, and that the amount of energy supplied during pre-clearing is superlinear relative to the active volume. The proportionately higher active volume reduction of the double DEAs, relative to the single or double-half DEAs, leads to a larger experimental-theoretical blocked force discrepancy. Further investigations into the scaling factors of active volume before and after the pre-clearing process would provide a more precise model prediction of the performance of DEAs with various dimensions and designs.

Future work to improve performance can include incorporating inextensible inclusions that limit the deformation to specific directions, increasing the precision of the roll-on-roll cutting and alignment techniques, and improving our understanding of the pre-clearing process to balance clearing of defects with loss in electrode area.

As a rolled DEA is actuated, it extends along its central axis. In many robotic mechanisms, such an extension (pushing) motion is impractical; moreover, with the rolled DEA configuration, it is important to prevent buckling of the structure during extensional actuation. A few solutions to this problem have been presented which require embedding the DEA in a compliant mechanism or structure that inverts the direction of actuation or converts actuation to a bending motion22,49. While these solutions solve the issue in the context of their applications, they are not broadly compatible with many types of mechanisms and often complicate actuator integration. Moreover, one key theme of the proposed DEA architecture is scaling to larger volumes, and thus any associated transmission mechanisms should be small relative to the active actuator volume.

To convert DEA extension to cable tension, we attached a cable to one end-plate and threaded it through the hollow center of the roll, exiting through the other end-plate as shown in Fig. 1D. As the cable is connected to the top end-plate of the DEA, actuating the DEA while constraining the bottom end-plate will result in an upward motion of the top end-plate, which translates to the cable also moving upwards. In an effort to minimize friction and avoid electrical shorts, we used a non-conductive 0.36 mm diameter Kevlar cable and added a 1.4 mm diameter PTFE tube as a protective sheath for the cable. The detailed manufacturing process for inserting this cable and tube into the DEA is shown in Fig. 2E. After creating the rolled DEA, we attached a first carbon fiber end-plate to the bottom of the DEA using carbon conductive adhesive and reinforced the connection using cyanoacrylate adhesive. We then inserted the PTFE tube inside the DEA through the hole in the end-plate and pushed it until it was nearly level with the top of the DEA roll before gluing it to the end-plate. We attach the cable to the other end-plate, feed the loose end through the tube, and repeat the connection process to secure the end-plate to the rolled DEA.

After fabricating cable-DEA versions of the three DEA types discussed previously, we assessed the force-displacement behavior of each device’s cable-driven actuation in standalone conditions and when integrated with a compliant finger (Fig. 4). For evaluating the standalone cable actuation, we first mounted the bottom plate of the DEA onto a bracket with a hole for the cable to pass through freely with a ruler positioned in-plane with the cable as a scale reference. To perform force-displacement characterization, we hung successively heavier weights from the end of the cable while the actuator was initially unactuated, and then ramped the voltage up to 1250 VDC (Fig. 4B and Supplementary Fig. 6). We used image analysis of the test videos to measure the maximum displacement reached for each weight applied.

A Force-displacement curve for two single DEAs and two double DEAs. The plot shows the mean and standard deviation across five trials for each test condition. Additionally, the blocked force of the DEA (not transmitted through the cable) at 1250V (Fig. 3A) is plotted on the horizontal axis for comparison. B Schematic of the experimental setup. A weight is hung on the cable to provide the force, the DEA was actuated, and the displacement, d, of a dot on the cable was measured. Supplementary Fig. 6 shows a photo of the setup. C Force at the fingertip of a rolled DEA-actuated finger. Comparison between the measured force and the prediction obtained using a quasi-static model. D Force distribution in the finger-DEA system.

Figure 4 A shows the force-displacement results for the cable-DEA devices. For each cable-DEA actuator, we found an approximately linear relationship between displacement attained as a function of applied load. In the free displacement condition (when the force is 0N, on the vertical axis), the results for cable actuation were comparable to the results shown previously for DEA extension in Fig. 3C. As heavier weights were applied, the behavior diverged, with the double DEAs being able to lift equivalent masses to larger displacements. All actuators exhibited a relatively linear force-displacement curve (R2= [0.99,0.96,0.87,0.99] with respect to the legend shown in the plot).

The abscissa of the linear fit (representing the expected blocked force where the displacement is 0mm) for the cable-DEA configuration differs significantly from the measured blocked force, from 1.2 to 2.3 × the blocked force values shown in Fig. 3A, and also plotted on the horizontal axis of Fig. 4A. We surmise that this discrepancy is due to the impingement of the top end-plate on the Teflon tube, causing it to bend, as weights are attached to the cable. To avoid excessively deforming the Teflon tube, we limited the maximum weight to ≈ 0.5N. Force-displacement curves of the unactuated DEAs show a gradual then sharp increase in force, indicating the initial compression of the elastomer, followed by bending of the tube (Supplementary Fig. 7). The competing effects of the reaction force of the compressed tube and the decreased blocked force output from the DEA due to compressive prestrain (due to increased dielectric thickness, assuming incompressibility of the silicone elastomer) is shown in Supplementary Fig. 8. In the weight-lifting experiments shown in Fig. 4A, the weights are in the regime where the reaction force contribution of the bent Teflon tube dominates and artificially increases the apparent force output of the DEA (Supplementary Fig. 8, preload 0N - ≈ 0.5N). This result illustrates the tradeoff in including tubing to prevent adhesion and friction between the cable and the silicone of the DEA, and points to future device designs that can mitigate the effect of the tubing, or alternately, leverage the mechanics of the tubing for additional functionality.

In order to demonstrate the capabilities of the cable-DEA system, we used it to actuate a compliant robotic gripper, a multi-degree-of-freedom Sarrus linkage, and a soft end effector.

The DEA-driven gripper was composed of two independently-driven fingers held together by a 3D printed frame as shown in Fig. 1. The fingers were created using laminate manufacturing techniques (see Materials and Methods for fabrication details, and Supplementary Fig. 9 for dimensions) and were attached to the flat faces of the frame, and the double rolled DEAs were slid into place from the side of the frame which constrains their bottom end-plates. The Kevlar cable was attached to the moving link of the finger, as shown in Fig. 4D. As the rolled DEAs are actuated, they expand vertically, which causes their top end-plate to move upwards. Since the top of the Kevlar cable is attached to the top end-plate, actuation of the DEA creates tension in the cable, resulting in motion of the finger. The input signal we used to actuate our device was a ramp and hold at 1250V for a predefined time of five seconds. Both parts of the gripper were actuated simultaneously by the same signal, creating the gripping motion around the target object.

We created a model of quasi-static force transmission for our finger-DEA system, shown in Fig. 4C, and estimated the expected force at the fingertip as a function of the actuator force using the following expression:

where β is the angle that the distal linkage bends relative to the proximal linkage, θ is the angle between the central axis of the DEA and the cable between the attachment point of the finger to the exit point of the rolled DEA, and k = Ykaptonwt3/12l = 0.26mN/m is the stiffness of the hinge64. In order to find a good balance between a high output force and sufficient displacement of the fingertip, we designed our fingers with a ratio of lengths OA to OB of approximately 0.73. We compared this model with what we manufactured using an MDB-2.5 lbs load cell and the single-strip, single-rolled DEA. We placed the finger and load cell horizontally and measured the output force at different angles when the DEA was actuated at its operating voltage of 1250 V (Supplementary Fig. 10). Figure 4C shows the results of this experiment. We notice that the trend set by the model is followed by our measurements, particularly at low finger joint angles.

The DEA-powered gripper was capable of grasping a wide variety of small objects as depicted in Fig. 5 and Supplemental Video 2. Using a pair of 1.42 g actuators, our gripper was able to grasp and lift objects of various sizes and shapes up to 5.45 g, a payload-to-actuator mass ratio of 1.88. We believe that heavier objects could be lifted by creating a system that improves the transfer of forces from the DEA to the fingertip. The gripper prototype here performs a pinch grasp with a low surface contact area and with the fingers not in perfect opposition with each other (as in a parallel gripper), resulting in a less stable grasp. We expect to improve grasp stability by generating a more enveloping grasp with multiple linkages comprising each finger and appropriate cable routing, and, additionally, increase the number of actuators driving each finger to increase grasping forces at a single degree-of-freedom or to actuate multiple degrees-of-freedom.

Objects include (A) a wireless earbud (5.45 g); (B) a binder clip (2.80 g); (C) a push pin (0.46 g); (D) a pencil (3.60g); (E) a piece of candy (5.14 g); and (F) a penny (2.53 g). The total weight of the gripper (structure, two actuators, two fingers) is 4.42 g; the weight of a single actuator is 1.42 g. The DEAs were actuated simultaneously by applying a ramp to 1250 V and holding.

In addition to simultaneously driving the DEAs for 1-DOF motions, we could drive each DEA within a matrix individually to enable multi-degree-of-freedom motions. In Fig. 6 and Supplemental Video 3, two cable-DEAs were used to actuate a multi-degree-of-freedom linkage, a modified Sarrus linkage (Supplementary Fig. 11) Two single-rolled DEAs were mounted to the top of the linkage, and their cables were secured to the bottom plate or side links of the mechanism. By actuating both DEAs, the bottom plate was lifted vertically (Fig. 6A, B, D, E); actuating one of the DEAs caused rotation and/or translation of the bottom plate (Fig. 6C, F). Moving the cable connection point to the side links offered mechanical advantage in lifting the bottom plate to a higher position (Fig. 6A, B vs. D, E). This difference in displacement was more pronounced when driving the DEAs at higher frequencies (Fig. 6A, D vs. B, E and Supplemental Video 3). This demonstration illustrates the utility of transferring the DEA actuation through a cable to increase control authority of a higher degree-of-freedom mechanism. Without the cable, the DEAs would need to be mounted inside of the Sarrus linkage, which would limit the minimum enclosed volume in the linkage and would also inhibit the full range of deformation that the Sarrus could undergo (e.g., packing to a fully flat configuration).

The two DEAs are mounted to the top plate of the linkage, and the cables are fixed to the bottom plate using a silicone cord lock. A screw is used to weigh down the bottom plate, keeping the linkage open. A–C In a direct-drive configuration, the cables are secured to the bottom plate. D–F The cables can also be attached to the side links to increase mechanical advantage, resulting in a higher displacement of the bottom plate. A, B, D, E Actuation of both DEAs to 1 kV lifts the screw in quasi-static (A, B) and at 15 Hz (D, E). The bottom half of each photo is split, showing the unactuated and actuated state on the left and right. C, F Actuating one of the DEAs induces rotation towards the actuated DEA in direct drive (C), and causes a diagonal translation when the cable is attached to the side link (F).

A cable DEA was also used to actuate a soft silicone “finger”, with the cable being routed through a bent L-shaped channel (Fig. 7). Actuating the DEA caused the tip to bend upwards from the starting configuration (Fig. 7A). We actuated the DEA with a sinusoidal signal with a 1 kV amplitude at frequencies from 1 Hz to 25 Hz (Fig. 7B, C). However, due to the current limit of the high voltage amplifier (Trek 610E), the actual applied voltage decreased as frequency increased (e.g., at 25 Hz, the maximum voltage reached was ≈ 500 V). This reduction in voltage, combined with the dynamic properties of the silicone tip and DEA, altered the maximum tip displacement at different frequencies (Supplemental Video 4). A future iteration of this demonstration can include a longer flexible tube, similar to a Bowden cable, allowing for truly distant actuation.

A The cable is routed inside an L-shaped pipe from the DEA, through the T-connector, and affixed to the tip of a silicone “finger''. B, C Actuating the DEA at different frequencies results in different tip displacements.

In this work, we demonstrate a method for scaling up the size of rolled DEA artificial muscles and an architecture that transforms the extensional actuation of the DEA to cable-driven actuation. Increasing the volume, and thus work output, of rolled DEAs can be achieved by sequentially rolling multilayer DEA strips (Fig. 2). While the in-plane area of a DEA multilayer strip (prior to rolling) could ideally be expanded to arbitrary dimensions, there are practical limits due to constraints on the working area of fabrication equipment and environmental cleanliness, such as the presence of airborne dust particles. The advantage of using multiple, shorter strips primarily lies in compatibility with DEA multilayering equipment, particularly spin coating which has a limited platter size, to obtain consistent and thin dielectric layers (sub-50μm), and stamp-transferring of electrode materials. Additionally, using multiple strips allows for pre-assembly inspection and quality control to avoid incorporating dust-contaminated segments into a larger rolled DEA.

In comparison to previous rolled actuators, where a single strip, typically with a single active layer, is rolled to form the actuator, splitting the flat strip into segments introduces unactuated zones at the margins of each strip and the possibility of misaligning strips. The effect of the unactuated regions is most apparent in free displacement measurements, where actuators with multiple strips underperformed the single-strip rolled DEA (Fig. 3C, single vs. double-half), but is negligible in the blocked force condition (Fig. 3A). As the strips get longer, deleterious effects appear in both test cases, with the double DEA specimens underperforming the theoretical predictions more severely compared to the single DEA. We believe that this can be attributed to misalignment errors during rolling, the effects of which become more apparent as the strips lengthen. For example, a 1° rotational misalignment results in a 12.3% reduction (2.1 mm) in active height (17 mm) over the course of a 120 mm long active region (Supplementary Fig. 1E). This analysis points toward the necessity of improving the rolling procedure through automated positioning and handling of strips.

Efforts to realize the full active volume should also include reduction of defects in the electrode and dielectric layers during manufacture that result in regions that are inactivated by the pre-clearing process, as well as refining the pre-clearing process to minimize the amount of eliminated electrode material, and thus inactivated volume. Further investigation in defining the limiting factors of the roll-on-roll scale-up method, such as critical aspect ratios and dimensions to prevent buckling during operation, tradeoffs in fabrication efficiency and reliability between layer stacking and rolling longer strips, and electrical requirements for driving such large capacitors, particularly for high speed or resonance which require high voltages and currents, will be critical for engineers to design linkages and other mechanisms that can be efficiently driven by large rolled DEAs. Additionally, the overall mechanical integrity of the connection between the silicone-based DEA and the carbon fiber end-plates could be improved. The durability of DEAs formed with self-clearing CNT electrodes and silicone dielectric has been well-characterized into the millions of cycles62. Here, though the combination of carbon conductive adhesive and cyanoacrylate we used here was sufficient for the hundreds or thousands of actuation cycles the devices underwent, this interface is critical and warrants further investigation. The integrity of this interface is dependent on the geometry of the rolled DEA and, thus the alignment of the edges during the rolling process. Improvements in the rolling process to produce flatter ends on the DEAs will improve both mechanical and electrical interfacing, since the rolled DEA must mate with the flat carbon fiber endcap.

For robotics applications, remote actuation, including cable-driven mechanisms, allows for more freedom in placement of bulky motors and actuators and is particularly useful for preserving the low-inertia and lightweight properties of smaller, more intricate end-effectors and mechanisms. This approach is analogous to the architecture of vertebrate musculoskeletal systems, and has been used for cable-driven snake robots65,66, continuum manipulators67, and robot hands3,68,69. Here, we have converted the extensional actuation of the rolled DEA architecture to cable actuation by threading a cable through tubing inserted in the hollow core of the rolled DEA and anchoring the cable end to one of the end caps on the roll. This architecture allows us to take advantage of DEA’s combination of compliance (backdrivability), light weight, and direct linear actuation, compared to the electromechanical motors that have typically been used for cable-actuation. We evaluated this architecture through weightlifting experiments and demonstrated its utility in driving single-degree-of-freedom hinges (Fig. 4). The weightlifting experiments yielded a linear force-displacement relationship that aligned with the free displacement experiments. However, as increasing weights were lifted, the tubing used to sheath the cable began to play a role in the compressive deformation of the unactuated DEA, sharing the load from the hanging weight. This interaction resulted in a shifted force-displacement curve, resulting from a combination of the tube bending (Supplementary Fig. 7), potential delamination between rolled layers, buckling of the walls, and the changes in thickness of the dielectric layer which reduces the field, and thus the output force (Supplementary Fig. 8) as a result of the precompression prior to actuation. We note here that, assuming the materials are incompressible and isotropic, Eq. (2) fails to predict the change in blocked force as a result of compression, and furthermore, neglects any boundary effects. We hypothesize that as the rolled DEA is compressed, the walls of the DEA (i.e., the dielectric layers) thicken non-uniformly, with greater increase in thickness at the mid-plane of the rolled DEA, and least change in thickness at the end-plates, where the DEA strip is rigidly adhered. In combination with potential delamination and buckling of the walls, actuation is no longer acting purely in the axial direction, with some components of the generated force directed radially, thus reducing the force measured by the load cell. This interaction emphasizes the need for incorporating strain-limiting elements in the actuator to limit unwanted deformation for more effective actuation. These radial stiffeners would act similarly to a spring where radial deformation is limited, but axial deformation and bending would still be allowed to occur. In this work, the tubing was a passive element in the actuator structure, however, future work could explore intentionally functional cable sheathing. For example, it could be designed to serve its cable protection role while minimizing any mechanical response under compression; akin to the spring roll actuators38,39,40, the sheathing could serve to pre-tension the DEA, or act as a bistable element that could enhance the actuation force or displacement.

To illustrate the utility of this architecture, we used a cable-DEA (double DEA, 1.42g each) to actuate the distal linkage of a lightweight hinged finger (765 mg) (Figs. 1, 5). A pair of these fingers was driven simultaneously to pinch grasp objects with widths ranging from 1.4mm to 1cm, and weighing up to 1.23 × the mass of the assembled gripper. The flexibility of the cable and the compliance of the actuator allow for backdrivability of the finger in both flexion and extension, increasing its robustness when operating in dynamic or uncertain environments. Here, the extension of the finger back to its resting position is due to the restoring force of the hinge material. Future work can exploit the small footprint of the cable attachment and use an antagonist actuator to actively extend the finger joint; moreover, such an actuator pairing for a single degree-of-freedom enables control of the joint compliance, allowing for both accommodation of hand positioning error in a higher compliance mode, as well as precision in positioning the fingers or manipulating an object in a higher stiffness mode. Particularly for robotic applications, it would be advantageous to incorporate self-sensing techniques to determine the actuator state without the need for external sensors70,71,72. With these improvements, a coupled pair of cable-DEAs can simultaneously serve as an actuator, stiffness controller, and sensor, enabling closed-loop control of an end effector’s position and compliance.

In addition to the single-degree-of-freedom joints in the gripper fingers, we also used cable-DEAs (single-rolled DEAs) to change the configuration of a multiple-degree-of-freedom Sarrus linkage (Fig. 6) and to actuate a continuously flexible (infinite degrees-of-freedom) silicone end-effector (Fig. 7). In both cases, the DEAs were too large to be directly integrated into the actuated structure, necessitating a cable-driven architecture. This concept becomes even more vital when considering independent control of closely located degrees of freedom (as in the Sarrus linkage), or when the structure being actuated is too soft to support the direct integration of the actuator (as with the silicone end-effector). As more actuators are integrated into a system, one important design consideration is the choice between fewer double-rolled DEAs or twice as many single-rolled DEAs. While the single-rolled DEAs have a higher energy density, this metric only takes into account the mass of the actuator and does not capture the overhead volume required for each actuator (end-plates, hollow core). Additional considerations include the time and cost of manufacturing overhead for each device and physical interface, and overall system reliability through redundancy of components. It will be crucial to consider these tradeoffs when designing systems that use many DEAs to actuate more complex structures. By using a cable-driven scheme, the high energy density of DEAs can be leveraged without strict requirements on co-locating actuators and passive structures. The work presented here provides a necessary building block for realizing our vision of using collectives of cabled DEAs to drive complex multi-degree-of-freedom linkages.

The fabrication process of the rolled DEAs begins with the creation of compliant electrodes (Fig. 2A). We used a spray deposition process developed by Cohen, et al.60 and an electrode formulation presented by Cohen, et al.61. This process, described in Fig. 2A, consists of spraying a CNT ink mixed with isopropanol (IPA) (1:2 by volume CNT ink/IPA) using an atomizer and nozzle. The CNT ink consists of CNTs dispersed in DI water at a ratio that if was diluted 16 ×, would yield 17% optical transmittance (~100 mg CNT in 100 g water). The mixture is sprayed onto a polytetrafluoroethylene (PTFE) filter paper with 0.45 μm pores through a 125 μm Mylar mask, both held down using pins, vacuum, and weights, and drenched with a 1:1 mixture of IPA and a deionized (DI) water ahead of the spray. The nozzle and atomizer hover above the target zone to uniformly cover the desired surface with the ink. Each spray creates two batches of three rectangular electrodes of dimension 120 × 20 mm. Once the spray is complete, the mask is removed and the filter paper with the CNT electrodes is placed in a 60 °C oven for 10 min to dry and cure.

The second manufacturing step consists of layering the electrodes and thin silicone elastomer (Wacker Elastosil P7670) films (Fig. 2B). The DEAs we manufactured in this study were made of three pairs of electrodes “A” and “B”. These electrodes are uniformly aligned in an alternating pattern, where each electrode of type B is vertically offset by 3 mm from the preceding electrode A, resulting in a 17 mm overlapping active region. The layering process is detailed in Fig. 2B, and starts by spin coating a layer of elastomer onto an acrylic substrate. We spin the substrate at 2500 rpm for the first layer and 1500 rpm for the subsequent layers in order to create roughly 32 μm layers. This difference in speed is due to the fact that the first layer is the only one that is layered directly onto the acrylic substrate rather than on another elastomer film.

The substrate is then placed in a vacuum chamber to degas for three minutes before curing in a 70° oven for 30 min. We then let it cool to room temperature for ~10 min and prepare our electrode filters for the stamping phase.

Ahead of stamping, debris on the filter paper is removed using an air gun and an anti-static gun successively. We then place the substrate on an alignment board using two pins at the bottom of the substrate. We carefully align the bottom of our electrode sheet face down using these same two pins and gently unroll the rest of the filter paper onto our freshly cured elastomer. Once in place, we gradually press the paper down using a roller to properly transfer the electrodes. Stamping motions going from the center of the substrate outwards, as well as along and across the electrodes, were among the most effective methods to transfer a large amount of CNT particles. The last step of this layering process consists of removing the filter paper to reveal the transferred electrodes. We found this step particularly prone to elastomer delamination compared to previous iterations of smaller DEAs. Care was taken to reduce roller stress during stamping (<60kPa) and to avoid rolling along the edge of the filter paper, and to peel the filter paper along the longest edge of the electrode. The next layer of elastomer is immediately cast and spin coated on top of the stamped layer, and the process repeats.

The entire layering process is repeated a total of six times, and the first three steps of the process are repeated one last time to encapsulate the outermost electrode. At the end of the process, three six-electrode DEA strips are produced per substrate.

The strips are then cut from the substrate using a vinyl cutter. The cut has to be made in a way that it exposes the electrodes on the long side of the rectangular strips while leaving 2mm of spare elastomer on the short edge to help with rolling and preventing shorting. The flat strip in Fig. 2C and Supplementary Fig. 1C shows how the DEA should be cut, with dimensions of 124 × 20 mm. Once cut, the DEA strip is placed under the microscope to be rolled as shown in Fig. 2C using tweezers. Once the first strip is fully rolled, the second strip is aligned and rolled around the first roll, resulting in a thicker roll as shown in Fig. 2D. No additional adhesive or silicone prepolymer is used to bond layers together; the inherent tackiness of cured silicone is sufficient to hold the layers together in the roll.

The final assembly step of the rolled DEAs consists of capping, wiring, and tubing the DEAs in order to electrically connect the roll and convert its expansion into cable tension. This process, detailed in Fig. 2E, begins with laser cutting end-plates from a three-ply carbon fiber sheet. These plates serve both as a means of conducting the voltage to the DEAs through the exposed electrodes, as well as structural elements to hold the inner tubing and to attach the wire. A first end-plate is attached to the bottom of the roll using carbon conductive adhesive (Electron Microscopy Sciences, Carbon Conductive Adhesive 502) and reinforced by applying cyanoacrylate adhesive (Loctite 495) around the base of the DEA. We then insert a 1.4 mm diameter PTFE tube inside the DEA through the hole in the end-plate, and push it until it is almost level with the top of the DEA roll. This tube is meant to minimize the friction of the wire inside the DEA to avoid any negative impact on the actuation of the DEA, such as an electrical short. After making sure it is properly centered and that it does not touch the inner walls of the DEA, we secure the tube by gluing it to the bottom end-plate. We then attach a non-conductive 0.36 mm diameter Kevlar wire to the other end-plate using cyanoacrylate adhesive. Finally, we feed the free end of this Kevlar wire through the tube and attach the top end-plate to the DEA using carbon adhesive followed by cyanoacrylate adhesive on the contours of the roll.

Insulated copper wires were also attached to the end-plates of the roll to simplify the electrical connection to the amplifier. To do this, we start by removing the protective layer around the two ends of the wires using a scalpel, and soldering a pin to one end of each wire. We then attach the other end of the wire to the end-plate of the DEAs using silver conductive paint, and reinforce the junction with a small layer of cyanoacrylate adhesive once cured.

Potential defects caught inside the DEA structure during the manufacturing and rolling processes need to be cleared out before the DEA can be used. Given the high operating voltages—up to several kV—some structural defects in the elastomer can cause dielectric breakdown which can severely damage the device and lead to premature failure. To solve this issue, we use a pre-clearing setup and script developed by Cohen et al.61. The protocol consists of applying a gradually increasing voltage to the DEA while monitoring the output current to spot potential resistive behaviors which are characteristic of defects. If, at a certain voltage, the measured current exceeds a predefined limit, the ramp stops, and we switch to an alternating current (AC) burst mode, which involves sending the highest voltage signal before the breakdown at a high frequency. This AC burst causes the controlled destruction of the CNT particles located around the defect, which isolates it and prevents further breakdown in the vicinity. Once the AC burst has cleared away the defective region, the ramp input continues until another fault is detected at a higher voltage. This process stops when the DEA can withstand the target voltage for three seconds without any breakdown.

When choosing the operating voltage, we had to find a compromise between reaching high actuation voltages and retaining a high capacitance. Higher voltage tends to damage the DEA more when experiencing a defect, which reduces the capacitance, decreasing the force and displacement that can be generated. To find the optimal voltage that maximizes force and displacement, we gradually pre-cleared two of our DEAs at different voltages until we noticed a significant drop in the capacitance, and measured the evolution of their capacitance with an LCR meter at each step. Using this information and the characteristics of our DEA, we characterized the evolution of the capacitance retention Cre and the predicted Maxwell stress σ61,62. We found that the operational limit for our system, defined by the point where Cre and σ drop, corresponds to 39.1 V/μm, which for our 32 μm layers translates to a voltage of 1250 V.

The fingers were designed using the laminate manufacturing technique presented by Ramirez Serrano, et al.73. We chose to use a 50 μm thick flexible polyimide (Kapton) sheet as the flexible core of the finger, and 125 μm thick glass-reinforced epoxy resin laminate (FR4) sheets to stiffen the base and fingertip, bonded together with adhesive layers. These sheets were laser cut into the desired shapes and stacked together using hot-melt adhesive (GBC Octiva Hot Mount adhesive (GBC-9300712)) and a laminator (Fellows Voyager 125) as shown in Supplementary Fig. 9A. A long and thin rectangular hole was cut across the width of the FR4 and adhesive sheets to create the flexure joint of the finger. The stack is then laser cut to outline the fingers, which are shown in Supplementary Fig. 9B. We cut a hole in the distal segment of the finger for attaching the cable, and added silicone (Smooth-On EcoFlex 00-30) fingertips to increase the friction coefficient of the finger to enhance grasping for heavier and more geometrically complex objects, as shown in Supplementary Fig. 9C.

To assemble the gripper, we designed and 3D printed the base structure displayed in Supplementary Fig. 9D. The rolled DEAs’ bottom end-plates are slotted into the structure’s sides in a position where they can expand upwards and where their cables are not constrained. The fingers are first attached to the structure and then connected to the cables hanging from the DEA rolls. The cable was passed through a hole in the distal linkage of the finger, then knotted to prevent the cable from pulling back through.

The Sarrus linkage was constructed using the same laminate manufacturing method as for the fingers above, substituting cardstock for the FR4 material. The designs for each layer are shown in Supplementary Fig. 11. To mount the actuators, we designed and 3D printed a structure that could be bolted to a rigid frame and had slots for the rolled DEAs’ bottom end-plates. This part was taped (using double-sided tape) to the top plate of the Sarrus linkage. The cables were then threaded through the 3D printed part and top plate of the Sarrus mechanism while the DEAs were slotted in place. The cables were passed through holes in the bottom plate of the Sarrus linkage and threaded (with the help of a needle) through two small pieces of silicone, which locked the cable in place via friction (similar to a cord lock), in place of a knot.

A silicone tip was actuated by a single-rolled DEA via a cable passed through an elbow in a T-connector with barbed tube fittings. The silicone tip was fabricated by casting silicone elastomer (EcoFlex 00-20, Smooth-on Inc.) inside of a transfer pipette. After curing, the silicone was removed and trimmed to length. The silicone tip was adhered to a 3D printed adapter (Sil-Poxy, Smooth-On Inc.) to mount it on the barbed fitting. A notch was cut out of the silicone near to base to allow for bending of the tip. To assemble the structure, the DEA and cable were threaded and slotted into the 3D printed adapter. The cable was snaked through the elbow of the T-connector and then through the adapter for the silicone tip. Both adapters were fitted onto the T-connector, and the cable was secured tautly into the silicone tip by sewing it through the material.

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

The code for DEA preclearing and operation is available from the corresponding author upon reasonable request.

Li, M., Pal, A., Aghakhani, A., Pena-Francesch, A. & Sitti, M. Soft actuators for real-world applications. Nat. Rev. Mater. 7, 235–249 (2022).

Article Google Scholar

El-Atab, N. et al. Soft actuators for soft robotic applications: a review. Adv. Intell. Syst. 2, 2000128 (2020).

Article Google Scholar

Odhner, L. U. et al. A compliant, underactuated hand for robust manipulation. Int. J. Robot. Res. 33, 736–752 (2014).

Article Google Scholar

Mosadegh, B. et al. Pneumatic networks for soft robotics that actuate rapidly. Adv. Funct. Mater. 24, 2163–2170 (2014).

Article Google Scholar

Kurumaya, S., Nabae, H., Endo, G. & Suzumori, K. Design of thin McKibben muscle and multifilament structure. Sens. Actuators A Phys. 261, 66–74 (2017).

Article Google Scholar

Seok, S. et al. Meshworm: a peristaltic soft robot with antagonistic nickel titanium coil actuators. IEEE/ASME Trans. Mechatron. 18, 1485–1497 (2013).

Article Google Scholar

Lee, Y. et al. Magnetically actuated fiber-based soft robots. Adv. Mater. 35, 2301916 (2023).

Article Google Scholar

Hu, W., Lum, G. Z., Mastrangeli, M. & Sitti, M. Small-scale soft-bodied robot with multimodal locomotion. Nature 554, 81–85 (2018).

Article Google Scholar

Li, S. et al. Digital light processing of liquid crystal elastomers for self-sensing artificial muscles. Sci. Adv. 7, eabg3677 (2021).

Article Google Scholar

Rothemund, P., Kellaris, N., Mitchell, S. K., Acome, E. & Keplinger, C. HASEL artificial muscles for a new generation of lifelike robots—recent progress and future opportunities. Adv. Mater. 33, 2003375 (2021).

Article Google Scholar

Taghavi, M., Helps, T. & Rossiter, J. Electro-ribbon actuators and electro-origami robots. Sci. Robot. 3, eaau9795 (2018).

Article Google Scholar

Helps, T., Romero, C., Taghavi, M., Conn, A. T. & Rossiter, J. Liquid-amplified zipping actuators for micro-air vehicles with transmission-free flapping. Sci. Robot. 7, eabi8189 (2022).

Article Google Scholar

Gu, G.-Y., Zhu, J., Zhu, L.-M. & Zhu, X. A survey on dielectric elastomer actuators for soft robots. Bioinspiration Biomim. 12, 011003 (2017).

Article Google Scholar

Mirvakili, S. M. & Hunter, I. W. Artificial muscles: mechanisms, applications, and challenges. Adv. Mater. 30, 1704407 (2018).

Article Google Scholar

Guo, Y., Liu, L., Liu, Y. & Leng, J. Review of dielectric elastomer actuators and their applications in soft robots. Adv. Intell. Syst. 3, 2000282 (2021).

Article Google Scholar

Hajiesmaili, E. & Clarke, D. R. Dielectric elastomer actuators. J. Appl. Phys. 129, 151102 (2021).

Article Google Scholar

Pelrine, R., Kornbluh, R., Pei, Q. & Joseph, J. High-speed electrically actuated elastomers with strain greater than 100%. Science 287, 836–839 (2000).

Article Google Scholar

Tang, C., Li, B., Fang, H., Li, Z. & Chen, H. A speedy, amphibian, robotic cube: resonance actuation by a dielectric elastomer. Sens. Actuators A: Phys. 270, 1–7 (2018).

Article Google Scholar

Qin, L., Tang, Y., Gupta, U. & Zhu, J. A soft robot capable of 2D mobility and self-sensing for obstacle detection and avoidance. Smart Mater. Struct. 27, 045017 (2018).

Article Google Scholar

Ji, X. et al. An autonomous untethered fast soft robotic insect driven by low-voltage dielectric elastomer actuators, Sci. Robot. 4, 37 (2019).

Kofod, G., Paajanen, M. & Bauer, S. Self-organized minimum-energy structures for dielectric elastomer actuators. Appl. Phys. A 85, 141–143 (2006).

Article Google Scholar

Araromi, O. A. et al. Rollable multisegment dielectric elastomer minimum energy structures for a deployable microsatellite gripper. IEEE-ASME Trans. Mechatron. 20, 438–446 (2015).

Article Google Scholar

Chen, F.-Y., Ren, Z.-X. & Lau, G.-K. Maximal strengths of dielectric elastomer fingers for a passive grip. Smart Mater. Struct. 31, 045014 (2022).

Article Google Scholar

Franke, M. et al. Dielectric elastomer actuator driven soft robotic structures with bioinspired skeletal and muscular reinforcement. Front. Robot. AI, 7, (2020).

Hau, S., Rizzello, G. & Seelecke, S. A novel dielectric elastomer membrane actuator concept for high-force applications. Extrem. Mech. Lett. 23, 24–28 (2018).

Article Google Scholar

Lee, D.-Y. et al. A wearable textile-embedded dielectric elastomer actuator haptic display. Soft Robot. 9, 1186–1197 (2022).

Article Google Scholar

Kovacs, G., Düring, L., Michel, S. & Terrasi, G. Stacked dielectric elastomer actuator for tensile force transmission. Sens. Actuators A Phys. 155, 299–307 (2009).

Article Google Scholar

Duduta, M., Hajiesmaili, E., Zhao, H., Wood, R. J. & Clarke, D. R. Realizing the potential of dielectric elastomer artificial muscles. Proc. Natl Acad. Sci. 116, 2476–2481 (2019).

Article Google Scholar

Giousouf, M. & Kovacs, G. Dielectric elastomer actuators used for pneumatic valve technology. Smart Mater. Struct. 22, 104010 (2013).

Article Google Scholar

Maas, J., Tepel, D. & Hoffstadt, T. Actuator design and automated manufacturing process for DEAP-based multilayer stack-actuators. Meccanica 50, 2839–2854 (2015).

Article Google Scholar

Kunze, J. et al. Design, manufacturing, and characterization of thin, core-free, rolled dielectric elastomer actuators. Actuators 10, 69 (2021).

Article Google Scholar

Baltes, M., Kunze, J., Prechtl, J., Seelecke, S. & Rizzello, G. A bi-stable soft robotic bendable module driven by silicone dielectric elastomer actuators: design, characterization, and parameter study. Smart Mater. Struct. 31, 114002 (2022).

Article Google Scholar

Zhao, H. et al. Compact dielectric elastomer linear actuators. Adv. Funct. Mater. 28, 1804328 (2018).

Article Google Scholar

Xu, S., Nunez, C. M., Souri, M. & Wood, R. J. A compact DEA-based soft peristaltic pump for power and control of fluidic robots. Sci. Robot. 8, eadd4649 (2023).

Article Google Scholar

Xu, S., Chen, Y., Hyun, N.-S. P., Becker, K. P. & Wood, R. J. A dynamic electrically driven soft valve for control of soft hydraulic actuators. Proc. Natl Acad. Sci. USA 118, e2103198118 (2021).

Article Google Scholar

Jiang, S., Tang, C., Dong, X., Liu, X.-J. & Zhao, H. Soft pocket pump for multi-medium transportation via an active tubular diaphragm. Adv. Funct. Mater. 33, 2305289 (2023).

Article Google Scholar

Pei, Q., Rosenthal, M., Stanford, S., Prahlad, H. & Pelrine, R. Multiple-degrees-of-freedom electroelastomer roll actuators. Smart Mater. Struct. 13, N86 (2004).

Article Google Scholar

Kovacs, G., Lochmatter, P. & Wissler, M. An arm wrestling robot driven by dielectric elastomer actuators. Smart Mater. Struct. 16, S306–S317 (2007).

Article Google Scholar

Pei, Q., Pelrine, R., Stanford, S., Kornbluh, R. & Rosenthal, M. Electroelastomer rolls and their application for biomimetic walking robots. Synth. Met. 135, 129–131 (2003).

Article Google Scholar

Li, J., Liu, L., Liu, Y. & Leng, J. Dielectric elastomer spring-roll bending actuators: applications in soft robotics and design. Soft Robot. 6, 69–81 (2019).

Article Google Scholar

Gupta, U., Qin, L., Wang, Y., Godaba, H. & Zhu, J. Soft robots based on dielectric elastomer actuators: a review. Smart Mater. Struct. 28, 103002 (2019).

Article Google Scholar

Rosset, S. & Shea, H. Small, fast, and tough: Shrinking down integrated elastomer transducers. Appl. Phys. Rev. 3, 031105 (2016).

Article Google Scholar

Guo, Y. et al. Dielectric elastomer artificial muscle materials advancement and soft robotic applications. SmartMat 4, e1203 (2023).

Article Google Scholar

O’Halloran, A., O’Malley, F. & McHugh, P. A review on dielectric elastomer actuators, technology, applications, and challenges. J. Appl. Phys. 071101, 2008 (2015).

Shintake, J., Rosset, S., Schubert, B., Floreano, D. & Shea, H. Versatile soft grippers with intrinsic electroadhesion based on multifunctional polymer actuators. Adv. Mater. 28, 231–238 (2016).

Article Google Scholar

Shintake, J., Cacucciolo, V., Shea, H. & Floreano, D. Soft biomimetic fish robot made of dielectric elastomer actuators. Soft Robot. 5, 466–474 (2018).

Article Google Scholar

Ouyang, F. et al. An optimization design method of rigid-flexible soft fingers based on dielectric elastomer actuators. Micromachines 13, 2030 (2022).

Article Google Scholar

Tang, C. et al. A pipeline inspection robot for navigating tubular environments in the sub-centimeter scale. Sci. Robot. 7, eabm8597 (2022).

Article Google Scholar

Lau, G.-K., Lim, H.-T., Teo, J.-Y. & Chin, Y.-W. Lightweight mechanical amplifiers for rolled dielectric elastomer actuators and their integration with bio-inspired wing flappers. Smart Mater. Struct. 23, 025021 (2014).

Article Google Scholar

Chen, Y. et al. Controlled flight of a microrobot powered by soft artificial muscles. Nature 575, 324–329 (2019).

Article Google Scholar

Kim, S. et al. Laser-assisted failure recovery for dielectric elastomer actuators in aerial robots. Sci. Robot. 8, eadf4278 (2023).

Article Google Scholar

Shi, Y. et al. A processable, high-performance dielectric elastomer and multilayering process. Science 377, 228–232 (2022).

Article Google Scholar

Ozsecen, M. Y., Sivak, M. & Mavroidis, C. Haptic interfaces using dielectric electroactive polymers. In Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2010, 7647, p. 960–966, (SPIE, 2010).

Zhao, H. et al. A wearable soft haptic communicator based on dielectric elastomer actuators. Soft Robot. 7, 451–461 (2020).

Article Google Scholar

Nagai, T. & Shintake, J. Rolled dielectric elastomer antagonistic actuators for biomimetic underwater robots. Polymers 14, 4549 (2022).

Article Google Scholar

Amin, H., Assal, S. F. M. & Iwata, H. A new hand rehabilitation system based on the cable-driven mechanism and dielectric elastomer actuator. Mech. Sci. 11, 357–369 (2020).

Article Google Scholar

Sarban, R., Jones, R. W., MacE, B. R. & Rustighi, E. A tubular dielectric elastomer actuator: Fabrication, characterization and active vibration isolation. Mech. Syst. Signal Process. 25, 2879–2891 (2011).

Article Google Scholar

Madsen, F. B., Daugaard, A. E., Hvilsted, S. & Skov, A. L. The current state of silicone-based dielectric elastomer transducers. Macromol. Rapid Commun. 37, 378–413 (2016).

Article Google Scholar

Yuan, W. et al. Fault-tolerant dielectric elastomer actuators using single-walled carbon nanotube electrodes. Adv. Mater. 20, 621–625 (2008).

Article Google Scholar

Cohen, A. J. et al. Batch-sprayed and stamp-transferred electrodes: a new paradigm for scalable fabrication of multilayer dielectric elastomer actuators. Adv. Funct. Mater. 32, 2205394 (2022).

Article Google Scholar

Cohen, A. J., Yuen, M. C., Corrigan, M. N., Xu, S. & Wood, R. J. Facilitating dielectric elastomer actuator multilayer fabrication and performance through low-contact-resistance hybrid electrodes, scalable vacuum filtration, and adaptive pre-clearing, Adv. Mater. Technol. 10, 2400787, (2024).

Jiang, S., Tang, C., Liu, X.-J. & Zhao, H. Long-life-cycle and damage-recovery artificial muscles via controllable and observable self-clearing process. Adv. Eng. Mater. 24, 2101017 (2022).

Article Google Scholar

Peng, J. et al. Dielectric elastomer actuators with low driving voltages and high mechanical outputs enabled by a scalable ultra-thin film multilayering process. Adv. Funct. Mater. 34, 2411801 (2024).

Article Google Scholar

McClintock, H. D. et al. A fabrication strategy for reconfigurable millimeter-scale metamaterials. Adv. Funct. Mater. 31, 2103428 (2021).

Article Google Scholar

Racioppo, P. & Ben-Tzvi, P. Design and control of a cable-driven articulated modular snake robot. IEEE/ASME Trans. Mechatron. 24, 893–901 (2019).

Article Google Scholar

Wang, T. et al. Mechanical intelligence simplifies control in terrestrial limbless locomotion. Sci. Robot. 8, eadi2243 (2023).

Article Google Scholar

Hwang, M. & Kwon, D.-S. Strong continuum manipulator for flexible endoscopic surgery. IEEE/ASME Trans. Mechatron. 24, 2193–2203 (2019).

Article Google Scholar

Aukes, D. M. et al. Design and testing of a selectively compliant underactuated hand. Int. J. Robot. Res. 33, 721–735 (2014).

Article Google Scholar

Wang, L., DelPreto, J., Bhattacharyya, S., Weisz, J. & Allen, P. K. A highly-underactuated robotic hand with force and joint angle sensors. In 2011 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 1380–1385 (2011).

Rosset, S. et al. Self-sensing dielectric elastomer actuators in closed-loop operation. Smart Mater. Struct. 22, 104018 (2013).

Article Google Scholar

Prechtl, J., Baltes, M., Kunze, J., Seelecke, S. & Rizzello, G. Towards Sensorless Configuration Estimation in Multi-DoF Soft Robotic Structures Driven by Rolled Dielectric Elastomer Actuators. In 2022 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), pp. 1152–1158, (2022).

Rizzello, G., Serafino, P., Naso, D. & Seelecke, S. Towards sensorless soft robotics: self-sensing stiffness control of dielectric elastomer actuators. IEEE Trans. Robot. 36, 174–188 (2020).

Article Google Scholar

Ramirez Serrano, F., Steinhardt, E. & Wood, R. J. Considerations for the design and rapid manufacturing of pop-up MEMS devices. Adv. Mater. Technol. 9, 2301940 (2024).

Article Google Scholar

Download references

The authors would like to thank Mady Corrigan, Dr. Alyssa Hernandez, and Dr. Andy Cohen for sharing their expertise and providing helpful comments. This work was supported by the National Science Foundation under the Materials Research Science and Engineering Center program (award number DMR-2011754) and the Army Research Office (award number W911NF2210219). Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

These authors contributed equally: Michelle C. Yuen, Théo Keroullé.

John A. Paulson School of Engineering and Applied Sciences, Harvard University, Allston, MA, USA

Michelle C. Yuen, Théo Keroullé, Siyi Xu & Robert J. Wood

ETH Zurich, Zurich, Switzerland

Théo Keroullé

Department of Mechanical Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, USA

Siyi Xu

You can also search for this author inPubMed Google Scholar

M.C.Y. and T.K. contributed equally to this work. M.C.Y. and T.K. built the devices, designed and performed the experiments, and prepared all figures. T.K. drafted the manuscript. M.C.Y. wrote the main manuscript text and supplementary information. S.X. provided guidance on fabrication and experimental design. R.J.W. provided guidance and funding for the project. All authors reviewed and edited the manuscript.

Correspondence to Robert J. Wood.

The authors declare no competing interests.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Yuen, M.C., Keroullé, T., Xu, S. et al. Electrostatic artificial muscles for cable-driven actuation of compliant mechanisms. npj Robot 3, 12 (2025). https://doi.org/10.1038/s44182-025-00030-7

Download citation

Received: 23 September 2024

Accepted: 26 April 2025

Published: 02 June 2025

DOI: https://doi.org/10.1038/s44182-025-00030-7

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Electrostatic artificial muscles for cable-driven actuation of compliant mechanisms | npj Robotics