FaBrick: An actuated fabric building platform for constructing scalable and versatile soft robots

Abstract

Soft robots have shown promise across applications such as wearable assistance, gripping, manipulation, and locomotion, where adaptability, safety, and compliance are essential. These functions are largely enabled by a small set of fundamental actuation modes: contraction, bending, twisting, rotation, and stiffening; but existing soft actuators that realize these motions are constructed using distinct materials and fabrication methods. This lack of standardization fragments the design process, limits scalability, and restricts accessibility for nonspecialized users. To address these challenges, this paper introduces FaBrick, a modular actuated fabric material made through a scalable manufacturing process that uses heated, patterned rolling to produce it at arbitrary lengths. Through folding, cutting, sealing, and stapling, FaBrick material can be configured to replicate the contractile, twisting, rotational, and bending motions of other soft pneumatic actuators. Each actuator type is fabricated, modeled, and experimentally validated, and the actuators are integrated into (i) wearable robotic limbs, (ii) a one-piece adaptive gripper, (iii) an elephant trunk–inspired manipulator, and (iv) a quadruped-like locomotion robot. Together, these results establish FaBrick as a scalable, versatile, and user-friendly foundation for building integrated soft robotic systems from a single material platform.

INTRODUCTION

Soft robots have emerged as a new class of machines capable of interacting safely and adaptively with people and complex environments. Their inherent compliance can passively compensate for disturbances introduced by unobservable features, therefore enabling functions that are challenging for traditional rigid systems, such as conformal grasping, body-safe physical assistance, and locomotion over unstructured terrain. These characteristics make soft robots particularly suited for four major application domains: wearable assistance, gripping, manipulation, and locomotion (1–5). In wearable robotics, soft actuators can augment or restore human motion with lightweight, body-conforming interfaces (6–24). In gripping and manipulation, their adaptable surfaces can securely grasp fragile or irregular objects without complex sensing or control (25–27). In locomotion, compliant structures enable crawling, climbing, or rolling without rigid links and joints, which can better adapt to the environment (28–30).

By examining these core applications, we recognize that most soft robotic systems can be described using five principal actuation motions: contraction, bending, twisting, rotation, and stiffening (31–37). Researchers have explored a variety of actuation strategies to achieve these motions, including thermal, electrical, hydraulic, and pneumatic methods (38–40). Pneumatic actuators, in particular, are attractive because they are lightweight and inexpensive, yet different motions are typically built as separate designs, with distinct geometries, materials, and fabrication steps, which slows prototyping and makes it difficult to combine or scale them within a single system (23, 35, 41–47).

We sought to develop a single material platform that could address three qualities we view as important for flexible and accessible soft robotic systems: scalability, versatility, and user-friendliness. Scalability describes how easily actuators can be scaled up to accommodate the force and size requirements of various applications. Versatility describes the extent to which a single material can be configured to produce different types of actuation. User-friendliness refers to how easy it is to fabricate actuators and integrate them into working systems. Niiyama et al. (48) pioneered printable pouch motors that demonstrated both versatility and user-friendly fabrication, but their reliance on heated two-dimensional (2D) routing processes limited scalability due to fixed build plate size. Lee et al. (49) developed a modular pouch-motor fabrication method that could be tiled into large systems, addressing scalability but still requiring substantial fabrication steps such as sewing, cutting, and laminating that made it more difficult for users to adopt them. At the system level, Omni-Skin introduced a reconfigurable soft skin capable of wrapping around diverse objects, offering partial versatility. However, the skin is constructed by assembling many McKibben actuators in parallel, such that actuation remains fundamentally contraction based, which limits some potential tasks it can achieve. Scaling the system therefore requires fabricating and integrating an increasing number of individual McKibben actuators, a process that remains labor intensive and difficult to extend to larger formats. As a result, while Omni-Skin enables reconfigurable coverage, its fabrication and scaling are constrained by the discrete nature of its actuator construction (50). Collectively, these approaches advanced important aspects of versatility, scalability, and user-friendliness, but none provided a unified material platform that inherently integrates all three.

To realize such a platform, we developed FaBrick, a fabric-based, pneumatically actuated material, whose name is inspired by modular building blocks such as bricks. Using only a few basic hand tools to cut, seal, and staple, the same sheet can be configured to achieve all of the major actuation modes (contraction, bending, twisting, rotation, and stiffening) commonly used in soft robots (Fig. 1). We validate this platform by analytically and experimentally characterizing the behavior of each actuator type and then combining them into higher-level robotic systems representing the four core soft robotic applications: a wearable robotic sleeve that replicates muscle-like assistance, a single-sheet gripper capable of adaptive grasping, an elephant trunk–inspired arm for manipulating the gripper, and a quadruped-like locomotion robot that moves using coordinated bending and contraction. Being able to generate this full spectrum of behaviors from one material highlights the platform’s versatility.

Fig. 1. FaBrick can be made into several types of actuators and then combined into integrated systems.

The FaBrick is an actuated material platform that allows users to rapidly create contraction, bending, twisting, rotation, and stiffening motions, which can then be combined to build higher-level devices across the core application domains of soft robotics.

The platform enables users to easily construct soft robotic components and assemble them into integrated systems using only scissors, staplers, and soldering irons, reflecting its high user-friendliness. To ensure scalability, we developed a continuous, patterned, heated rolling process for manufacturing FaBrick. This process produces the material continuously at arbitrary length and is a second core contribution of this work alongside the FaBrick material itself.

Together, these demonstrations highlight how FaBrick can serve as a material platform for building soft robotic systems of increasing complexity. The ability to fabricate multiple actuator types and structures from a single roll of material is especially valuable in resource-limited or rapidly changing environments. These scenarios include field robotics, field medicine, space missions, and educational settings, where access to molds, specialized equipment, or large fabrication tools may be limited. By lowering the barrier to entry to designing and fabricating soft actuators, FaBrick enables a broader range of users, including those without specialized equipment or fabrication expertise, to explore, prototype, and iterate on soft robotic ideas. In doing so, FaBrick could help lower the threshold for participation in the soft robotics field and foster innovation across research, education, and application domains.

RESULTS

Design of FaBrick

Inspired by the advantages that modular building blocks, such as bricks, can provide, we aimed to bring the same design philosophy to soft robotics. Building blocks demonstrate three key strengths:

1) Scalable, as the same set of parts can be used to build systems of vastly different sizes

2) Versatile, as the same standardized pieces can be reconfigured into countless creations

3) User-friendly, as individual pieces can be easily made and put together to form complete assemblies

In an analogous way, we developed FaBrick, a fabric-based actuatable material that serves as a set of modular building blocks for soft robots (Fig. 2A). Unlike traditional building blocks, which are static structural elements, each unit within FaBrick is itself an active, pneumatically driven pouch motor. A pouch motor is a thin, airtight pouch that inflates into a bulged shape when pressurized, shortening its projected length and generating force in the process (24, 48, 51–55). The material consists of arrays of pouch motor cells in both row-wise and column-wise directions, with each cell having open channels connected to their neighboring cells for air flow, allowing a single air inlet to inflate the entire sheet of material (Fig. 2, B and C). By selectively sealing off these air channels, the sheet can be divided into functional regions. Then, through simple fabrication operations (cutting, heat-sealing, and stapling), distinct actuation modes can be produced including contraction, twisting, bending, rotation, and stiffening. Furthermore, this architecture allows actuators created from the same sheet to be directly combined into higher-level systems, enabling rapid construction of wearable devices, grippers, manipulators, and locomotion platforms from a single roll of material.

Fig. 2. Design and manufacturing of the FaBrick.

(A) Raw FaBrick material. (B) Close up view of FaBrick when inflated. (C) Inflated region of FaBrick. (D) Close up view of the metal 3D printed patterned roller. (E) Heated, patterned rolling machine for large-scale manufacturing. (F) Schematic of the working principle of the heated, patterned roller. (G) Manufacturing of FaBrick in action.

FaBrick is mass producible (scalable)

Scalability is a crucial but often unmet requirement in soft robotics. While many soft actuators perform well at the laboratory scale, they are difficult to reproduce in large quantities or areas. Existing fabrication methods, such as heat pressing, 3D printing, or casting, are typically batch based, constrained by tooling size, and dependent on manual alignment or curing steps, preventing continuous production and large-scale deployment (24, 48–51, 53, 54).

To address these limitations, we developed a patterned, heated rolling manufacturing method for FaBrick (Fig. 2, E to G). The roller-based system can continuously laminate two TPU-coated fabric layers using a patterned metal roller with raised sealing features (Fig. 2D). As the fabric passes between the heated patterned roller (top) and a compliant rubber roller (bottom), heat and pressure fuse the material only at the raised regions, forming sealed pouches while leaving connecting channels open. This continuous process enables rapid production of extended sheets of the actuated material directly from standard textile rolls, demonstrating FaBrick’s potential for mass-producible soft robotic systems. This manufacturing capability makes it easy to scale actuators to meet the force and size requirements of an application, thus fulfilling the “scalability” goal defined in Introduction. Because the material can be produced continuously on a roll-to-roll basis, scaling up an actuator to make it larger or more powerful is simply a matter of cutting a longer or wider strip from the same roll, with additional pouch columns in parallel increasing force and additional rows in series increasing stroke, relationships that are modeled and experimentally validated in the sections that follow. Detailed manufacturing parameters are provided in Materials and Methods.

FaBrick can accomplish major actuation types (versatile)

Versatility is a defining strength of a building platform for soft robotics. It enables a single platform to be made into different actuators to perform a broad range of motions such as contraction, bending, twisting, rotation, and stiffening. However, most existing soft actuators are designed for one specific function, with each new motion requiring separate materials, geometry, or fabrication methods (31–6). This specialization makes it difficult to rapidly prototype or integrate different actuator types within the same device.

FaBrick overcomes these limitations at the material level. Its grid architecture allows multiple actuation modes to be created from the same sheet through simple cut, seal, and staple operations (Fig. 3). By following the fabrication steps shown in Fig. 4, a single roll of material can be transformed into the different types of actuators without changing the underlying actuation principle or material composition. Detailed fabrication techniques are provided in Materials and Methods.

Fig. 3. Different actuation motions can be achieved using the FaBrick.

(A) Contraction: Selective inflation of left, right, or both columns produces asymmetric or symmetric linear contraction. (B) Twisting: Cutting and sealing the grid into a parallelogram allow the actuator to twist a cylindrical core. (C) Rotation: Arranging pouch cells in a stacked configuration produces torque about a central axis with tunable rotation angles. (D) Bending: Stapling the inflatable layer to a strain-limiting layer yields bending with curvature controlled by staple spacing. (E) Stiffening: Isolating and inflating selected regions rigidify the structure for load support or shape locking.

Fig. 4. FaBrick enables fabrication of all major soft actuator types using simple patterning steps.

(A) Contraction actuator: Sealing selected regions of the sheet (yellow and orange) and assigning separate inlets allows independently controlled contraction. (B) Stiffening structure: Tightly rolling the sheet along its width and adding an external collar that constrains radial expansion produces a pneumatic stiffening element. (C) Twisting actuator: Cutting the sheet into a parallelogram, then sealing and disconnecting the columns, and wrapping it around a cylinder converts linear contraction of each column into torsional motion. (D) Rotation actuator: Folding a rectangular sheet in an accordion shape and stapling at the pivot forms a bellow-like hinge that produces rotational output upon inflation. (E) Bending actuator: Creating alternating fold segments and stapling the inflatable layer to a strain-limiting layer at a prescribed spacing generates controlled bending behavior.

FaBrick reduces fabrication complexity (user-friendly)

Fabrication complexity is one of the central barriers to adopting soft robotic actuators. Many established actuator types require specialized molds, laser-cutting templates, precise alignment fixtures, or multistep curing and assembly processes. This makes them difficult to manufacture outside well-equipped laboratory settings. In contrast, FaBrick is designed so that the most difficult and precision-dependent step of sealing the pouches is completed during the material production stage through the patterned, heated rolling process. As a result, users of the material do not need to interact with any of the underlying lamination, sealing, or alignment steps.

Once produced, the FaBrick sheet behaves as a ready-to-use actuated material. Actuators of different types can be created using only simple, accessible tools:

1) Scissors to cut the sheet to a desired outline or length

2) A soldering iron to selectively close or reroute air channels through heat-sealing

3) A stapler to join separate layers of fabric

With these three low-tech operations, cut, seal, and staple, a user can rapidly form contraction actuators, twisting bands, bending modules, rotational bellows, or hybrid combinations without molds, fixtures, CAD templates, or postprocessing. This reduces actuator creation time to a matter of minutes and makes the platform accessible to users ranging from researchers and educators to clinicians and field operators. By transforming complex pneumatic actuation into lightweight, intuitive fabrication steps, FaBrick lowers the practical barrier to entry and enables soft robotic systems to be prototyped, repaired, and reconfigured entirely with simple hand tools.

Predicting and validating actuator behavior

Predicting how a FaBrick sheet behaves under pressure requires models that are both accurate and interpretable. Since structures of different types and dimensions can be constructed from FaBrick depending on the number and arrangement of pouches, we developed single-cell models that can be extended to multipouch arrangements. Our framework therefore begins by modeling the deformation and volume change of a single inflated pouch and then uses geometry-driven transformations to derive force, deformation, and pressure relationships for contraction, twisting, and rotation actuators and to predict bending direction for bending actuators (Fig. 5). The modeling pipeline is grounded in the principle of virtual work. Changes in enclosed volume under pressure generate mechanical work, allowing us to compute forces and torques directly from geometry. This approach lets users predict actuator behavior before fabrication: How long a strip should be cut for a desired stroke, how many columns to add for a target force, or how many folds to include for a desired rotation. These models therefore provide insights for designing FaBrick structures analytically rather than by trial and error, making it a programmable material.

Fig. 5. Modeling of actuators.

(A) Modeling of contraction actuator as a pillow, with lightly shaded red as the cross section formed with major arc (L) in length direction and lightly shaded blue as the cross section formed by minor arc (W) in width direction, sharing the same bulge height h. (B) Modeling of twisting actuator is derived from the linear contraction model in part (A), assumed to be a 2D planar actuator wrapped around a cylinder at an angle. Force versus displacement can be then converted into torque versus angle relationship. (C) Modeling of rotation actuator from using F = PA(h) relationship, and by modeling change of A with respect to h, the force versus height relationship can then be obtained, which can then be converted into torque versus angle. (D) Modeling of bending actuator is done geometrically, the angle θ correlates to the stapling distance, and the total bending radius is determined through the number of segments and the stapling distance.

Contraction actuator

Pouch motors generate contractile forces when pressurized. Assuming that the pouch material is inextensible, as a pouch bulges outward, its length decreases until a maximum enclosed volume is reached. A prediction of the force generated by a pouch motor can be obtained using a virtual-work formula (Eq. 1), where P is the internal pressure, V is the enclosed volume, F is the force and q_L is the actuator length, and δq_L and δV correspond to the change in actuator length and volume, respectively. For a virtual displacement of the system from static equilibrium, the principle of conservation of energy requires that the mechanical work performed by the actuator force balances the change in fluid energy. This relationship can be expressed as

$$-F \mathrm{\delta }{q}_{\mathrm{L}}=P \mathrm{\delta }\mathrm{V}$$ (1)

Dividing both sides by δq_L and taking the limit as the virtual displacements go to zero then yields the following force-displacement relationship (48, 56)

$$F=-P \frac{\mathit{dV}}{{\mathit{dq}}_{\mathrm{L}}}$$ (2)

This means that the predicted force depends directly on the accuracy of $\mathit{dV}/{\mathit{dq}}_{\mathrm{L}}$. Any geometric approximation that incorrectly estimates how the pouch volume changes as it contracts will produce a proportional error in the force-displacement curve. In the below sections, we show how we derive an expression for the enclosed volume V in terms of q_L and then take the derivative of that expression to get the $\mathit{dV}/{\mathit{dq}}_{\mathrm{L}}$ relationship.

Different from conventional pouch-motor models, which assume the geometry of a fully inflated pouch to be cylindrical, we assume it to be a pillow shape to more accurately capture the actual behavior. For wide pouches with large length-to-width aspect ratios, ignoring the geometry of the pouch near its ends and treating the inflated shape as a cylinder produce acceptable estimates of $V\left(\mathrm{\theta }\right)$ and its derivative (38, 48, 57).

However, the FaBrick sheet uses a smaller 2:1 (width:length) aspect ratio because narrower pouches conform more naturally to curved and cylindrical surfaces, enabling the sheet to function beyond planar applications. The reduced width also provides finer spatial resolution when cutting the sheet, allowing users to define actuation zones with greater precision without wasting material or sacrificing functionality. This geometric flexibility is essential for treating the sheet as an actuated material platform rather than a single-purpose device. This same compact geometry, however, also makes volume predictions much more sensitive to the assumed pouch dimensions: When the aspect ratio decreases, the end-cap regions constitute a much larger fraction of the inflated shape. As a result, the cylindrical approximation substantially distorts both the total volume and, critically, the rate at which volume changes with displacement. This end-region geometry has received sustained attention in prior work. Predicting the bulged shape of an inflated flat pouch is itself a classical mathematical problem, known as the paper-bag problem (58), and later analyses have refined the predicted shapes of inflated thin sheets (59, 60). Do et al. (61) modeled the end caps as elliptical arcs but used this geometry to predict the contact area of a pouch compressed between two parallel plates, rather than the enclosed volume needed to predict force. Other works avoided the end-cap problem in hardware altogether: Wang et al. (folded pneumatic artificial muscle) (62) and Greer et al. (series pneumatic artificial muscles) (63) folded the pouch ends or constrained them with O-rings so that the inflated pouch stays closer to its cylindrical idealization. However, a closed-form expression for the enclosed volume of the actual pouch shape, from completely flat to fully bulged, is still missing. Our pillow model fills this gap, and because it captures the geometry that actually develops during inflation, it enables more informed design choices, allowing actuator behavior to be explored analytically before fabrication.

To capture the true deformation behavior of these compact pouches, we model the inflated geometry as the intersection of two orthogonal extrusions of circular arcs, a closer representation of the actual “pillow-like” bulging shape (Fig. 5A). The pouch is a rectangle with length L and width W, and the two arcs are named the major arc and the minor arc. The major arc describes the bulging behavior of the pouch in the length direction, shaded in red in Fig. 5A, with arc radius r_L. The minor arc describes the bulging behavior in the width direction, shaded in blue in Fig. 5A, with radius r_W. The contracted length and width are represented by the chord length q_L in the length direction and q_W in the width direction.

We also found that the stretch of the fabric can result in an overestimation of force near $q_{\mathrm{L}}=L$ (when the pouch is deflated and flat) compared with experimental data, yielding a higher root mean square (RMS) error (Table 1, conventional cylinder model). This overestimation increases as pressure rises, a trend also reported by Niiyama et al. (48). We posit that this can likely be attributed to some pressure-induced material stretching that creates a prebulge even when the displacement is zero upon initial inflation. To capture this effect, we introduce a pressure-dependent stretch term k, a correction factor for the realistic arc length $L_{\text{stretched}}$ and width $W_{\text{stretched}}$, represented in Eq. 3. Meaning, when the pouch dimensions are designed to be L and W, upon inflation, it will result in slightly larger values of $L_{\text{stretched}}$ and $W_{\text{stretched}}$ and produce a higher volume

$$L_{\text{stretched}}=L+\mathit{kP}, W_{\text{stretched}}=W+\mathit{kP}$$ (3)

Table 1. NRMSE for contraction, twisting, and rotation models under inflation and deflation strokes.

The contraction part compares three models: the conventional cylinder model, which does not take fabric stretch into consideration; cylinder with stretch, a model that still uses cylinder as baseline geometry but takes fabric stretch into consideration; pillow with stretch, which uses pillow shape as baseline geometry and takes stretch into consideration.

Stroke	Pressure	NRMSE
		Contraction			Twisting	Rotation
		Conventional cylinder	Cylinder with stretch	Pillow with stretch (our model)
Inflation	69 kPa (10 psi)	3.02 × 100	1.17 × 10−1	2.78 × 10−2	6.97 × 10−2	3.83 × 10−2
	138 kPa (20 psi)	4.17 × 100	1.22 × 10−1	1.59 × 10−2	9.32 × 10−2	3.21 × 10−2
	207 kPa (30 psi)	5.05 × 100	2.07 × 10−1	1.95 × 10−2	1.13 × 10−1	3.55 × 10−2
Deflation	69 kPa (10 psi)	3.02 × 100	9.25 × 10−2	2.07 × 10−2	1.29 × 10−1	2.44 × 10−2
	138 kPa (20 psi)	4.17 × 100	9.34 × 10−2	2.93 × 10−2	1.58 × 10−1	4.26 × 10−2
	207 kPa (30 psi)	5.05 × 100	8.62 × 10−2	4.65 × 10−2	1.87 × 10−1	4.16 × 10−2

While the major and minor arc lengths of the pouch are now assumed to be $L_{\text{stretched}}$ and $W_{\text{stretched}}$, at their maximum extension, the pouches have chord length q_L and chord width q_W equal to L and W, respectively. The bulging behavior of the pouch is therefore governed by the contracted length q_L, which decreases from the designed length L in the flat state to $L/\mathrm{\pi }$ in the fully inflated state. This relationship holds under several underlying assumptions about the inflation and deflation process:

1) The stretch in warp and weft direction of the fabric are the same, thus k can be used for both length and width direction of the pouch.

2) The major and minor arc profiles for a given pressure are inextensible: The major arc length is $L_{\text{stretched}}$ and the minor arc length is $W_{\text{stretched}}$.

3) The major arc and minor arc share the same peak height $h_{\text{peak}}$.

The first assumption provides an initial offset to the pouch arc lengths; the second assumption gives a geometric constraint for the pouch throughout the inflation and deflation process; the third assumption links the two arcs through the common height h and makes the minor arc derivable from the major arc, shown in Eq. 4.

Here, we use the Sagitta relation (64) with a constant offset so that the coordinate system is centered at the midpoint of the chord. This formulation relates the arc radii r_L and r_W to the local height above the chord as a function of the chord lengths q_L and q_W. The chord coordinates x and y are defined such that $x\in [-\frac{q_{\mathrm{L}}}{2},\frac{q_{\mathrm{L}}}{2}]$ and $y\in [-\frac{q_{\mathrm{W}}}{2},\frac{q_{\mathrm{W}}}{2}]$, with the midpoint of the chord located at x = 0 and y = 0. The resulting height profiles $h_{\mathrm{L}}\left(x\right)$ and $h_{\mathrm{W}}\left(y\right)$ describe the arc height above the chord in the length and width directions, respectively, with the maximum height occurring at the midpoint. Subtracting the constant term $\sqrt{r_{\mathrm{L}}^2-{\left(\frac{q_{\mathrm{L}}}{2}\right)}^2}$ for the x direction and $\sqrt{r_{\mathrm{W}}^2-{\left(\frac{q_{\mathrm{W}}}{2}\right)}^2}$ for the y direction shifts the reference from the circle center to the chord line, ensuring that the height is zero at the chord endpoints and equal to the peak height at the midpoint

$$h_{\mathrm{L}}\left(x\right)=\sqrt{r_{\mathrm{L}}^2-x^2}-\sqrt{r_{\mathrm{L}}^2-{\left(\frac{q_{\mathrm{L}}}{2}\right)}^2}, h_{\mathrm{W}}\left(y\right)=\sqrt{r_{\mathrm{W}}^2-y^2}-\sqrt{r_{\mathrm{W}}^2-{\left(\frac{q_{\mathrm{W}}}{2}\right)}^2}$$ (4)

Equation 5 is a special solution of Eq. 4, where the peak height of the major arc $h_{\mathrm{L}}\left(0\right)$ and minor arc are equal, where $x=0$ and $y=0$

$$h_{\text{peak}}=h_{\mathrm{L}}\left(0\right)=r_{\mathrm{L}}-\sqrt{r_{\mathrm{L}}^2-{\left(\frac{q_{\mathrm{L}}}{2}\right)}^2}=h_{\mathrm{W}}\left(0\right)=r_{\mathrm{W}}-\sqrt{r_{\mathrm{W}}^2-{\left(\frac{q_{\mathrm{W}}}{2}\right)}^2}$$ (5)

Since variable r_L in Eq. 5 is still unknown, we use the arc-length formula for a circle (Eq. 6) that relates the arc length $L_{\text{stretched}}$, the arc radius r_L, and the chord length q_L

$$L_{\text{stretched}}=2r_{\mathrm{L}}\text{arcsin}\left(\frac{q_{\mathrm{L}}}{r_{\mathrm{L}}}\right), W_{\text{stretched}}=2r_{\mathrm{W}}\text{arcsin}\left(\frac{q_{\mathrm{W}}}{r_{\mathrm{W}}}\right)$$ (6)

With the above equation and $L_{\text{stretched}}$ and $q_{\mathrm{L}}$ known, we can find r_L numerically and substitute back into Eq. 5 to find $h_{\mathrm{L}}\left(x\right)$. Similarly, the same relationship for $W_{\text{stretched}}$, r_W, and q_W also stands, as shown in Eq. 6. Since $h_{\text{peak}}=h_{\mathrm{L}}\left(0\right)=h_{\mathrm{W}}\left(0\right)$, we can use Eqs. 5 and 6 to work out q_W and r_W, which will then give us $h_W\left(y\right)$ at every point along the minor arc. Now, we have $h_{\mathrm{L}}\left(x\right)$ and $h_{\mathrm{W}}\left(y\right)$ fully defined. In Eq. 7, $H(x,y)$ is the height of the pouch, and the $\text{min}(h_{\mathrm{L}},h_{\mathrm{W}})$ term captures their geometric intersection, forming the pillow-like shape. The outer $\text{max}(0,\cdot )$ ensures that only positive heights are integrated

$$H(x,y)=\text{max}\left\{0,\text{min}\left[h_{\mathrm{L}}\left(x\right),h_{\mathrm{W}}\left(y\right)\right]\right\}$$ (7)

The volume of the pillow-shaped pouch as a function of pouch length q_L is obtained by integrating the intersection of two orthogonal circular arc profiles along the length and width directions. The double integral computes the enclosed volume of one-half of the pouch, which is then multiplied by two to form a full pouch

$$V(x,y)=2{\displaystyle {\int }_{-q_{\mathrm{L}}/2}^{q_{\mathrm{L}}/2}}{\displaystyle {\int }_{-q_{\mathrm{W}}/2}^{q_{\mathrm{W}}/2}}H(x,y)\mathit{dx} \mathit{dy}$$ (8)

Then, from Eq. 2, the force-displacement relationship at various pressures in L and W directions becomes

$$F_{\mathrm{L}}(P,q_{\mathrm{L}})=-P \frac{\mathit{dV}}{{\mathit{dq}}_{\mathrm{L}}}, F_{\mathrm{W}}(P,q_{\mathrm{W}})=-P \frac{\mathit{dV}}{{\mathit{dq}}_{\mathrm{W}}}$$ (9)

The length change of one pouch can be obtained through subtracting the contracted pouch length (or the chord length) q_L by the nominal pouch length L

$$\mathrm{\Delta }L=L-q_{\mathrm{L}}$$ (10)

Multiple cells can be combined in series and parallel arrangements to generate larger contractions and forces. $\mathrm{\Delta }{L}_{\text{array}}$, the total length change of a chain of pouches connected in series, is the length change of each pouch multiplied by the number of pouches $n_{\text{series}}$; connecting pouches in series does not affect the force output

$$\mathrm{\Delta }{L}_{\text{array}}=n_{\text{series}}\mathrm{\Delta }L$$ (11)

To increase force, $n_{\text{parallel}}$ number of pouches can be combined in parallel, where the net force $F_{\text{array}}$ is the sum of the forces from each pouch (or chain of pouches)

$$F_{\text{array}}(P,q_{\mathrm{L}})=n_{\text{parallel}}{F}_{\mathrm{L}}(P,q_{\mathrm{L}})$$ (12)

To validate the virtual-work model, we ran force versus displacement experiments on an 8 × 1 (row × column) contraction actuator with 14 mm–by–28 mm (length-by-width) cells and a 3-mm gap between each cell. We fitted the constant $k=2.3\times {10}^{-9}$ to our model, yielding the lowest RMS error between model and experiment. The pillow-shaped model captured both the contraction ratio and the force-displacement trend more accurately than the conventional cylindrical approximation. While the cylindrical model predicts an idealized contraction of ∼30.6% (40.7 mm for our setup), the measured contraction was only 27.1% (36 mm), and our pillow model predicted nearly the same value of 27.4% (36.5 mm), confirming that the finite-length geometry must be considered for pouch actuators operating at aspect ratios near 2:1. Notably, the agreement improved at higher pressures, where the pouch shape adheres more closely to the assumed pillow geometry. We further evaluated the model across pressure levels of 69 kPa (10 psi), 138 kPa (20 psi), and 207 kPa (30 psi).

Across all conditions, the simulated force-displacement curves of the pillow model closely tracked the experimental data (Fig. 6Aa) with lower normalized RMS errors (NRMSEs) than the conventional cylinder model and the cylinder-with-stretch model across all tested pressure ranges, highlighting that a single parameter sufficiently captures fabric compliance and energy losses (Table 1). To test whether force scales linearly with the number of columns, we compared single- and double-column actuators. As expected, the two-column array generated nearly twice the force of the single column (Fig. 6Ab), confirming that parallel pouch-cell architectures sum forces predictably and can be designed using the modular approach without additional model tuning. Together, these results demonstrate that the proposed pillow model with stretch not only predicts contraction and force with high fidelity but also generalizes across pressure conditions and multipouch configurations, forming a reliable foundation for actuator and device-level design within the FaBrick platform.

Fig. 6. Data collection.

(A) (a) Comparison between conventional model, our proposed model, and experiment data. (b) Force versus displacement for single versus double columns in parallel at 20 psi. (B) (a) Modeled force and experiment force versus angular displacement. (b) Torque versus angular displacement of four- versus two-column actuators at 20 psi. (C) (a) Model versus experiment data for a single pouch. (b) Comparison of rotation actuators made with single pouch versus double pouches in parallel at 20 psi.

Twisting actuator

The twisting actuator converts the linear contraction of columns of pouches into rotational motion by wrapping them around any cylindrical structure, provided the cylinder is split into top and bottom parts that can freely rotate relative to one another through a joint. To create a twisting actuator, we cut a parallelogram-shaped piece of FaBrick and wrap it around the cylinder at an initial skew angle (Fig. 4C). The top edge of the wrapped sheet is fixed to the top plate, and the bottom edge is fixed to the bottom plate, ensuring that the sheet cannot slip or change length along its boundaries.

When the actuator is pressurized, each column of pouches attempts to shorten. However, because the ends of the parallelogram sheet are rigidly attached to the top and bottom of the cylinder, where the axial contraction is completely restricted, the only admissible motion is tangential displacement around the cylindrical surface. As a result, the contraction force produced by each pouch column will only generate a torque that rotates the top plate relative to the bottom. In effect, the wrapped sheet behaves similar to a helical band that twists the cylinder when tensioned.

The twisting actuator model builds directly upon the contraction actuator model (Fig. 5B). Wrapping the linear actuator around a cylindrical surface at a defined helix or wrap angle transforms linear contraction of the actuator into rotational motion of the cylindrical structure, thereby converting the force-displacement relationship into a torque-angle relationship. Geometrically, the twist angle θ of the actuator is determined by the change in length of each pouch column as it shortens. To simplify the analysis, we assume that no out-of-plane bulging occurs as the actuator contracts; that is, pouch bulging does not increase the helix diameter.

Given a cylinder height H and initial or unpressurized actuator length $L_0$, a wrap length C can be obtained, defined as the maximum arc length the linear actuator can travel

$$C=\sqrt{L_0^2-H^2}$$ (13)

$L_{\text{Actuator}}$ is the instantaneous actuator length during contraction, while S represents the circumferential arc length generated by the actuator as it induces rotation of the cylinder

$$S\left(L_{\text{Actuator}}\right)=C-\sqrt{{L_{\text{Actuator}}}^2-H^2}$$ (14)

and the total twist angle of the actuator ${\mathrm{\theta }}_{\text{Twist}}$ can then be defined as

$${\mathrm{\theta }}_{\text{Twist}}\left(L_{\text{Actuator}}\right)=\frac{S\left(L_{\text{Actuator}}\right)}{r}$$ (15)

Substituting Eq. 14 into Eq. 15 gives us

$${\mathrm{\theta }}_{\text{Twist}}\left(L_{\text{Actuator}}\right)=\frac{C-\sqrt{{L_{\text{Actuator}}}^2-H^2}}{r}$$ (16)

After converting the actuator linear displacement into rotation angle, we then convert the actuator force, which is the contraction force predicted from the linear actuator model, into torque ${\mathrm{\tau }}_{\text{rotation}}$. ϕ is the wrap angle between the actuator and the top surface of the cylinder, which increases as the actuator contracts or as the cylinder rotates. ϕ can be obtained through

$$\mathrm{\varphi }=\text{arcsin}\left(\frac{H}{L_{\text{Actuator}}}\right)$$ (17)

The circumferential component of the actuator force, $F_{\text{circ}}$, from linear force $F_{\text{Actuator}}$ can then be calculated using Eq. 18

$$F_{\text{circ}}(P,{\mathrm{\theta }}_{\text{Twist}})=F_{\text{array}}(P,{\mathrm{\theta }}_{\text{Twist}})\text{cos}\left(\mathrm{\varphi }\right)$$ (18)

$F_{\text{circ}}$ can then be multiplied by the cylinder radius r to get the torque. The angular displacement versus torque relationship can thus be expressed as

$$\mathrm{\tau }(P,{\mathrm{\theta }}_{\text{Twist}})=F_{\text{circ}}(P,{\mathrm{\theta }}_{\text{Twist}})r$$ (19)

To validate the performance of the twisting actuator against our model, we ran a set of experiments measuring the torque output of an actuator with 4 columns × 8 rows, where each cell had the same length and width (14 mm by 28 mm) as in the contraction model, wrapped around a cylinder of height H = 92.6 mm and radius r = 25 mm. We compared two curves: the simulated torque-angle curve and the experimental torque-angle data (solid line in Fig. 6Ba; torque for two- versus four-column actuators in Fig. 6Bb). The NRMSE results are reported in Table 1 for all pressures tested: 69 kPa (10 psi), 138 kPa (20 psi), and 207 kPa (30 psi).

Rotation actuator

The rotation actuator leverages the pouch’s out-of-plane bulging to generate torque. It is created by folding a strip of pouch cells in an accordion pattern so the pouches stack directly on top of each other (Fig. 4D). When inflated, each pouch bulges outward, and this bulging causes adjacent layers to press against one another. Because all pouches share the same pivot point, their bulging produces forces that rotate about a common center, resulting in a net rotational motion and torque when multiple layers are stacked.

The pillow/virtual-work treatment ($-F \mathit{dH}=P \mathit{dV}$) used for the contraction actuator does not carry over directly to pouches stacked together. We therefore adopt a contact-area description, as Do et al. (61) also do, since the force transmitted across each pouch interface is well captured by pressure times the instantaneous contact area. However, our accordion-folded rotation actuator has a pivot that converts the linear bulging into rotational motion, requiring an additional step to map bulge height to rotation angle and force to torque through the lever arm.

The force generated by the bulging of each pouch can be approximated as the product of the internal pressure and the contact area between adjacent layers, which is the area over which the inflated pouch presses against its neighbors. This relationship can be expressed as

$$F_{\text{bulge}}=\mathit{PA}$$ (20)

where A is the instantaneous contact area between the inflated pouch and the parallel plates confining it. w_L is the width in the length direction, shaded in red, and w_W is in the width direction, shaded in blue (Fig. 5C). As the distance h between the plates increases, this contact area decreases. We approximate the contact area as

$$A\left(h\right)=w_{\mathrm{L}}\left(h\right)w_{\mathrm{W}}\left(h\right)$$ (21)

To compute $w_{\mathrm{L}}\left(h\right)$ and $w_{\mathrm{W}}\left(h\right)$ in terms of h, we assume that the cross sections of each pouch are rectangles with semicircular arcs appended to the short sides (Fig. 5C) and that the total perimeters C_L and C_W of the cross sections remain constant during inflation, as written in Eq. 22

$$C_{\mathrm{L}}\left(h\right)=2L=\mathrm{\pi }\mathrm{h}+2w_{\mathrm{L}}, C_{\mathrm{W}}\left(h\right)=2W=\mathrm{\pi }\mathrm{h}+2w_{\mathrm{W}}$$ (22)

Simplifying Eq. 22 gives us Eq. 23

$$w_{\mathrm{L}}\left(h\right)=L-\frac{\mathrm{\pi }h}{2}, w_{\mathrm{W}}\left(h\right)=W-\frac{\mathrm{\pi }h}{2}$$ (23)

This geometric assumption allows us to express w_L and w_W as functions of h, the distance between the parallel plates, and thus derive A(h) and the corresponding force-height relationship below

$$F_{\text{bulge}}(P,h)=\mathit{PA}\left(h\right)=P w_{\mathrm{W}}\left(h\right) w_{\mathrm{L}}\left(h\right)$$ (24)

After obtaining the height-force curve, we can then assume that the height of the pouch is the arc length of the circle that the pouch is rotating about to convert the vertical motion into rotational motion. The distance between the pivot point and the pouch center defines the lever arm, allowing the height displacement to be mapped into angular displacement

$${\mathrm{\theta }}_{\text{Rotate}}\left(h\right)=\frac{h}{R}$$ (25)

and the force to be mapped into torque by multiplying lever arm R by force $F_{\text{bulge}}$

$$T\left[P,{\mathrm{\theta }}_{\text{Rotate}}\left(h\right)\right]=F_{\text{bulge}}(P,h) R$$ (26)

Similar to the contraction actuator, we can then multiply the torque by adding more columns of pouches in parallel with the current helix and increase the number of rows for increasing the total amount of angular rotation. The final equation is shown below

$$T(P,{\mathrm{\theta }}_{\text{Rotate}})=P(L-\frac{\mathrm{\pi }\mathrm{R}{\mathrm{\theta }}_{\text{Rotate}}}{2})(W-\frac{\mathrm{\pi }\mathrm{R}{\mathrm{\theta }}_{\text{Rotate}}}{2})R$$ (27)

We ran an experiment to validate our torque versus angular displacement model, in which we measured the force reading at each angular displacement and multiplied by the lever arm to obtain a torque reading (Fig. 6Ca). The resulting RMS errors are shown in Table 1, with R = 36 mm, L = 14 mm, and W = 28 mm, at 69 kPa (10 psi), 138 kPa (20 psi), and 207 kPa (30 psi).

Beyond validating the single-fold behavior, we tested whether the actuator follows predictable scaling laws when arranged in series or parallel. Adding more folds in series increased the total angular displacement with little change in torque, while adding more pouch stacks in parallel increased torque proportionally with minimal change in displacement. These trends are consistent with classical pneumatic actuator scaling and were confirmed experimentally (Fig. 6Cb). Together, these results show that both the model and physical hardware behave in a modular manner, which is an essential property for building larger rotational mechanisms within the FaBrick.

Bending actuator

The bending actuator is fabricated by stapling an inflatable layer of FaBrick on top of an un-inflated strain limiting layer of FaBrick using normal staples, such that when the inflatable layer is inflated, the bulging and contraction of the pouches is constrained by the strain-limiting layer, producing bending rather than pure shortening. The staples are placed on the dead space of the fabrics instead of directly on the pouches so that they do not cause leaking. We refer to the region between adjacent staples as a “segment”; the more segments an actuator has, the longer it will be, making it suitable for larger applications. For each actuator, the user can choose the number of segments, the number of pouches in each segment, and the stapling distance between each segment to achieve the desired bending behavior (Fig. 4E).

We did not use the virtual-work or area-pressure-force models here because bending in this actuator is governed primarily by geometric interactions. Accurately, modeling force would require capturing nonlinear behavior, spatially varying deformation, and fabric stretch, which makes the analytical formulation intractable and not especially meaningful for this actuator’s intended use. Unlike the contraction or twisting actuators, the bending actuator is not designed as a force-output device; its value lies in its ability to adapt, wrap, and achieve controlled curvature. For these reasons, a geometric model that relates curvature to staple spacing provides the most relevant and interpretable description of its behavior.

The model relates the bending curvature and direction to the distance between staples. When the staple spacing is small, the pouches adjacent to each staple dominate the deformation: The increased bulge height between neighboring pouches pushes them against one another, causing the actuator to bend toward the strain-limiting layer and form an angle ${\mathrm{\theta }}_{\text{Bend}}$ (Fig. 5D). However, as the staple distance increases beyond a threshold, approximately when

$$\begin{array}{cc}\hfill \text{staple distance}=& (\text{contracted distance} \text{per} \text{pouch})\times \hfill \\ & (\text{number of pouches} \text{per} \text{segment})\hfill \end{array}$$

the deformation mechanism shifts; now, pouch contraction causes bending in the other direction. For our case, with pouch length L = 14 mm and an experimental contraction rate of 27.1%, this gives $q_{\mathrm{L}}=14 \text{mm}\times (1-0.271)=10.2 \text{mm}$, the contracted distance per pouch. Having two pouches per segment for our design, the threshold of staple distance was ∼1.46L for the actuator to start bending in the other direction. Beyond the threshold, the pouches between the staple points become the dominant contributors, pulling the staple lines closer together as the fabric curls around the inflated pouches. The resulting bending is therefore governed by the curvature of the inflated pouches rather than their local interaction, producing a reversal in bending direction. This transition defines the critical stapling distance separating the forward- and reverse-bending regimes. Although deriving a force versus displacement model remains challenging, this geometric framework provides useful design guidance for controlling bending direction and curvature range in future FaBrick configurations.

Soft robotic systems can be made from FaBrick

User-friendliness in soft robotics goes beyond the fabrication of individual actuators; it determines how easily those actuators can be assembled into complete, functional systems. In many existing soft robotic designs, different actuation modes, such as contraction, bending, and twisting, are realized using distinct actuator geometries, materials, or fabrication workflows, making it cumbersome to combine multiple functions within a single device. For example, a wearable system that requires both linear contraction and joint rotation typically relies on distinct types of separately fabricated actuators for each of those motions, which must be mechanically and pneumatically interfaced post hoc, adding complexity and limiting rapid iteration.

By contrast, FaBrick enables a fundamentally different design workflow. Because all actuator types are derived from the same continuous actuated fabric and share a common pneumatic and mechanical interface, diverse actuation modes can be combined directly within a single structure. The same cut, seal, and staple operations used for fabrication also enable simpler attachment between actuators, allowing designers to build larger assemblies that combine contraction, twisting, rotation, and bending functions within a single integrated system. This material-level compatibility makes it straightforward to construct multifunctional systems. Figure 7 demonstrates several such integrated systems built from FaBrick, highlighting how user-friendliness at the material level opens new possibilities for rapid system-level design and reconfiguration.

Fig. 7. FaBrick actuators can be combined into integrated systems for a wide range of applications.

(A) Wearable assistive and rehabilitation devices: Contraction and rotation actuators are integrated to support biceps and shoulder lifting motions, while paired twisting actuators provide clockwise and counterclockwise wrist rotation. (B) Adaptive soft gripper: A single FaBrick sheet is patterned into contraction and bending actuators; the contraction layer opens the fingers and the bending layer closes them, enabling stable grasping of objects with diverse shapes and weights. (C) Elephant trunk–inspired manipulator: A stiffening structure forms a backbone, while contraction actuators generate bending and extension; an added twisting actuator enables in-plane gripper rotation for improved alignment before grasping and manipulation. (D) Quadruped locomotion robot: Stiffening structures serve as weight-bearing legs, and contraction actuators generate forward and backward strokes; time-lapse images show continuous walking gait enabled by alternating actuation of diagonal leg pairs.

Wearable robots

One of the core benefits of the FaBrick platform is its lightweight and conformable nature, making it well-suited for wearable applications. By combining contraction actuators with rotation actuators, FaBrick can reproduce assistive motions such as bicep-like motion for lifting the forearm using the contraction actuator, and shoulder-like motion for lifting the upper arm using the contraction and rotation actuators (Fig. 7A). It can also recreate wrist pronation and supination motion using the twisting actuator. We laid two twisting actuators on top of each other in opposite directions to achieve rotation in both clockwise and counterclockwise directions. In future devices, these motions could support rehabilitation and daily assistance for conditions such as stroke recovery, Parkinson’s disease, and postsurgical or post-fracture therapy.

Grasping

We fabricated a gripper entirely from a single FaBrick sheet by patterning its upper half into contraction actuators and its lower half into bending actuators (Fig. 7B). Then, the contraction half is folded back onto the bending half and stapled at the end. When this planar sheet is wrapped around a cylindrical support structure, it forms a multifingered gripper without requiring any additional assembly. Upon inflation, the contraction layer shortens, pulling the fingers open, while the bending layer inflates to curl the fingers inward, enabling a firm grasp around the target object. By selectively controlling the inflation sequence, the gripper can transition smoothly between open and closed states, adapting passively to objects of varying shapes. We demonstrated stable grasping of items across a range of shapes and weights, illustrating the versatility and compliance of the FaBrick material. This single-piece, pattern-defined approach eliminates the need for discrete joints or rigid connectors, highlighting FaBrick’s potential for scalable and easily manufacturable soft end-effectors.

Manipulation

We further demonstrated a multi–degree-of-freedom manipulator arm inspired by an elephant trunk (Fig. 7C). The arm consists of three parallel FaBrick actuation columns arranged side by side, each independently pressurized for selective inflation and deflation. By coordinating the pressure in each column, the arm can generate swinging and bending motions, enabling directional control. To showcase integrated functionality, we attached the FaBrick -based gripper to the tip of the arm, allowing grasping and release of lightweight objects such as a snack bag. A twisting actuator was also incorporated at the wrist section to introduce rotational alignment of the gripper, enabling in-plane orientation adjustments before grasping. This modular combination highlights FaBrick’s capacity to achieve complex manipulation behaviors through simple control of a unified material system.

Locomotion

We also explored a simple form of locomotion using a FaBrick strip with two independently controlled actuation columns running along its length. Each foot is powered by both columns, so pressurization produces bending either forward or backward. Critically, each foot integrates an inflatable stiffening structure that can be pressurized to support the weight of the entire system, enabling the feet to act as load-bearing elements rather than purely compliant appendages (Fig. 7D). By coordinating these two columns, we implement a four-foot gait where diagonal pairs of feet move together. In a typical cycle, feet 1 and 3 first bend forward, while feet 2 and 4 remain flat, anchoring the body. Next, feet 2 and 4 bend forward, pulling the body ahead. The pressures are then inverted so that feet 1 and 3 bend backward, while feet 2 and 4 anchor, followed by feet 2 and 4 bending backward to reset. Repeating this sequence generates net forward motion, demonstrating that FaBrick can realize basic locomotive behaviors through simple, pattern-based actuation without additional rigid joints.

DISCUSSION

The results presented here demonstrate a new type of functional material platform that can be used as base units for constructing soft robots. Rather than building each actuator as a separate device, FaBrick platform establishes a single textile substrate from which multiple actuation modes can emerge through simple geometric patterning. This shift from discrete, task-specific actuators to reconfigurable, scalable material platforms suggests a new design paradigm for soft robotics, where functionality is encoded directly in the material structure.

Manufacturing

The heated roller-based fabrication process demonstrated consistent and repeatable sealing of pouch geometries but can still benefit from further tuning of process parameters. Minor dimensional variation arises from uneven heat transfer and roller surface irregularities, underscoring the need for precise control of temperature, pressure, and rolling speed. While the current system is constrained by the laminator’s width, the rolling design inherently supports continuous fabrication of arbitrary length, offering a clear path toward scalable production. Material selection also provides flexibility; the same heated, patterned rolling process accommodates fabrics across a range of thicknesses and stiffnesses, which can be selected to tune the response-speed and blocked-force trade-off (see the “Material selection” section and figs. S2 and S4). Adjusting pouch geometry, such as aspect ratio, spacing, or size, further tailors actuator force and compliance without altering the underlying process, highlighting the scalability of the manufacturing framework.

Pattern geometry must balance various effects that influence performance. For example, our models assume that the enclosed volume in channels is negligible compared to that of the pouches. This restricts how large they can be relative to the pouches, but narrower channels slow inflation and lead to uneven pressure distributions. To balance these considerations, we chose an air-channel width of 3 mm for pouches with 14 mm–by–28 mm dimensions. We also specified the dead spaces between pouches to be 3 mm because narrower spaces leave too little seal area for the layers to remain bonded when the pouches are inflated. Pouch size also affects performance. Reducing pouch size will result in reduced achievable force and contraction ratio. As the pouch size gets smaller, the dead-space widths do not scale down with the pouch so the overall contraction ratio falls, per-pouch volume decreases so the force drops, and effects coming from fabric properties such as thickness and stiffness become a larger fraction as the pouch force drops. Smaller pouches are therefore most useful when fine spatial resolution is required within a tightly constrained footprint, rather than when high force or stroke is the priority.

The spatial resolution and maximum patternable area are limited by manufacturing constraints. Currently, the maximum material width is limited by the roller width, which is 335 mm. This can be easily resolved with custom made rollers of desire. A longer and larger-diameter roller increases the area patterned per revolution and can further increase single-pouch length and width.

Actuator performance

At the actuator level, FaBrick demonstrates that diverse motions such as contraction, twisting, bending, rotation, and stiffening can be derived from a single material platform using only common, readily available tools. This material-level versatility establishes a foundation for rapid and standardized actuator design. Despite these advances, several challenges and trade-offs are worth noting when building actuators with the FaBrick platform.

First, flow resistance is a potential issue across all actuator types and is most acute in the bending actuators due to narrow channels and localized kinking, which can slow inflation and venting. The rotation (bellows) actuator experiences similar challenges when multiple folds press flat: Depending on where the air inlets are, pouches farther from the inlet can see delays in inflation compared with pouches closer to it. These effects can be mitigated through geometric refinement and material selection. Preliminary tests show that using a softer fabric enhances airflow (fig. S2), and threading thick woven strings through the channels helps maintain open flow paths. Increasing the number of air inlets could further improve uniformity and response. These findings point to clear design guidelines for future iterations, emphasizing the importance of channel geometry, flow routing, and fabric compliance in achieving faster, more uniform actuation.

Second, beyond flow resistance during inflation, the shared-channel architecture produces a coupled effect during external loading: Mechanical contact on a subset of pouches can transiently redistribute air to surrounding neighbors. This is an inherent consequence of having all pouches in the sheet interconnected. Here, we monitored supply pressure at the inlet during all experiments, but resolving local pressure inside individual pouches of a connected array is not possible with the current sensing setup, and we do not have a closed-form model for the fluid redistribution dynamics. In the demonstrations of Fig. 7, the actuators completed their intended motions under contact loading, but the effect of pressure variability within each actuator was not measured. Modeling and quantifying this phenomenon are left for future work.

Third, one actuator-level caveat is that the twisting mode, unlike contraction, rotation, and bending, requires an external rotatable structure (typically an axle-mounted cylinder) to convert column-wise contraction into tangential rotation. This contrasts with actuator families that embed the motion structure directly in the actuator architecture itself; fiber-reinforced elastomeric enclosures (FREEs) (65), for instance, twist because of their fiber-wind angle alone, without external scaffolding. However, for applications such as wearable devices, this dependency on external structures often becomes an asset. For instance, the wrist-rotation element of the wearable sleeve in Fig. 7A uses the existing twisting degree of freedom of the wrist. The thin, low-profile fabric form factor wraps cleanly around these existing structures without adding bulk. We therefore treat the external structure requirement as a trade-off of the FaBrick approach for the twisting mode, rather than a fundamental limitation.

Fourth, the performance of any specific actuator type may not match that of a custom-made design tuned for a single application. Purpose-built bending actuators such as PneuNets (30) and twisting actuators such as FREEs can be tuned more aggressively against a single force, torque, or stroke target than FaBrick can, because their geometries can be optimized end to end rather than constrained to a common fabric substrate. We view this as an acceptable trade-off because the FaBrick platform offers two advantages that are more fundamental than peak performance. First, motion type is not encoded into the actuator architecture: A PneuNet bends because of its bellows pattern, and a FREE twists because of its fiber-wind angle, so each device is committed to a single motion at the moment of fabrication. FaBrick derives its motion type from how the same fabric is cut, sealed, and stapled, so the same roll of material and the same fabrication process produces all five motion modes, and producing a different motion is a matter of cutting differently rather than retooling. Second, scaling up is process level rather than tool level. A larger- or higher-output PneuNet generally requires a new mold of the corresponding size, and a larger FREE requires a new fiber-winding setup, whereas a larger or more powerful FaBrick-based actuator is obtained by cutting a longer or wider strip from the same roll, with no new tooling. FaBrick’s value lies in providing one material and one process that span a broad motion vocabulary and a broad size range from a single substrate.

Data collection

The analytical model developed for FaBrick is not only valuable for matching experimental behavior but also serves as a practical design tool for end users. Because the model predicts contraction force, torque, angular displacement, and bending curvature directly from internal pressure and pouch geometry, users can determine actuator dimensions before fabrication. For linear actuators, the force-displacement curve indicates exactly how long a strip must be cut to achieve a desired stroke and how many columns to arrange in parallel to reach a target load. For twisting and rotational actuators, the torque-angle prediction similarly informs the quantity of folds or columns required to attain a specified rotation or moment. This predictive capability allows FaBrick to function as a programmable material. Designers can reason directly from model outputs to actuator performance, enabling rapid, geometry-driven customization without iterative prototyping.

Model-experiment comparisons confirm this utility. For the contraction actuator, the “pillow” pouch model captures contraction behavior more accurately than the often-used cylindrical approximation, owing to its coupling of the major and minor arcs through a shared bulge height that more closely captures the true pillow-like pouch deformation. At higher pressures, the contraction ratio converges toward model predictions, as the pouch geometry stabilizes and fabric stretch becomes more uniform. Across all pressure levels, hysteresis was observed in the force-displacement response. Specifically, the measured force was consistently lower during inflation stroke (increasing pouch volume) and higher during deflation stroke (decreasing pouch volume). We hypothesize that this hysteresis arises from flow-induced pressure nonuniformity within the pouch. During inflation, internal pressure may require time to equilibrate along the length of the actuator, resulting in a reduced effective force. Conversely, during deflation, compression of the enclosed volume may transiently elevate internal pressure above the regulated inlet value, leading to increased force output. Although pressure was monitored at the inlet and a 1-s dwell time was introduced between displacement steps, spatial pressure gradients within the pouch may still persist. For the twisting actuator, the measured torque-angle response follows the overall trend predicted by the model but exhibits deviations at mid-range angles. We hypothesize these deviations are attributed to 3D bulging of the inflated pouch, which lifts the actuator surface out of plane. This out-of-plane deformation increases both the effective arc length and moment arm, resulting in slightly higher torque and larger total twist than predicted by the 2D model. Together, these observations suggest that future analytical models would benefit from explicitly accounting for 3D deformation and internal pressure distribution dynamics to further improve predictive accuracy.

However, there was still hysteresis observed in all tests at different pressure levels, with the resulting force being lower for the inflation stroke (pouch volume increases) and force being higher for the deflation stroke (pouch volume decreases). We hypothesize that the hysteresis may be due to restricted flow causing pressure to be nonuniform distributed inside the pouch. During the inflation stroke, the pressure could take time to equalize, resulting in a slightly lower force, and for the deflation stroke (or downstroke), while volume is being compressed, the internal pressure may be higher than what we set it to be, resulting in higher force.

Usability and integration

At the system level, unlike custom-built or premolded actuator systems, FaBrick can be directly reconfigured into different actuator types and combined into complete robotic assemblies. This simplicity demonstrates the platform’s user-friendliness: Functional actuators and robotic systems can be produced without molds, 3D printers, or proprietary materials. The wearable sleeve, single-sheet gripper, manipulator, and quadruped prototypes illustrate how diverse actuator behaviors can be combined into integrated, multifunctional systems primarily using one material. This design paradigm reduces fabrication complexity and enables rapid reconfiguration, making FaBrick a promising approach for deployable and field-ready soft robots.

The models introduced in this work were not used to size the demonstrations in Fig. 7; the wearable sleeve, gripper, manipulator, and quadruped were each built and tuned empirically. The demonstrations and the model validation experiments therefore illustrate distinct claims. The demonstrations show that cut, seal, and staple operations alone can produce recognizable systems across all four core application domains. The model validation experiments show that pouch actuator force and torque can be predicted in closed form within the geometric range we tested. Going forward, the primary value of the model is as a forward design tool, allowing a user to reason from a target force, torque, or stroke back to a pouch geometry and pressure before fabrication; retrospective application to the present demonstrations is left for future work.

Outlook

By establishing generalizable models and scalable manufacturing methods for a modular actuated fabric material, this work offers a promising new design and fabrication paradigm for soft robotics. The FaBrick platform elevates the fabric beyond a passive structural material, enabling it to function as an actively actuating medium that can be easily configured to generate different motion. The same sheet can therefore serve as a passive structural material or be patterned to actively generate motion. By merging fabrication scalability with versatility and system-level user-friendliness, it establishes a foundation for building soft robots the way we build with bulk materials, such as cutting, joining, and shaping rather than assembling discrete components. Because FaBrick can be patterned and reconfigured using only simple tools, it lowers the barrier to entry for creating custom soft robotic systems in laboratories, clinics, classrooms, or even remote environments where specialized infrastructure is unavailable. Its continuous, roll-to-roll manufacturing approach also suggests a path toward large-area, deployable morphing structures and wearable assistive textiles that could be produced at industrial scales. Beyond pneumatics, the same architecture could integrate distributed sensing, electrical actuation, or morphing control networks, enabling a new generation of programmable fabrics that blur the boundary between material and machine. In this way, FaBrick points toward a future in which actuation becomes a universal material capability, democratizing the design and construction of adaptive robotic systems across disciplines.

MATERIALS AND METHODS

Material selection

Candidate textiles including Stretchlon nylon laminates and TPU-coated ripstop composites were evaluated for seal strength, stiffness, and ease of processing. The selected material, black TPU-coated nylon taffeta (Seattle Fabrics), provided consistent heat sealing, moderate stiffness, and commercial availability. Preliminary mechanical testing revealed anisotropy between warp and weft directions, with the weft exhibiting higher elasticity. Aligning the warp direction with the primary contraction axis increased force output and reduced unwanted stretch. Each pouch cell measured 14 mm by 28 mm, connected by 3-mm air channels with 3-mm dead space between adjacent rows and columns.

To characterize the effect of fabric choice on actuator response speed and force output, two TPU-coated nylon fabrics were compared: a thinner red TPU-coated rip-stop nylon (DIY Packraft) and the standard thicker black TPU-coated nylon taffeta (Seattle Fabrics) used elsewhere in this work. Identical 2 × 8 contraction actuator arrays were fabricated from each fabric using the same heated, patterned rolling manufacturing process. Force-displacement measurements were performed at 138 kPa using the linear actuator test jig described in the “Experimental setup for evaluating model predictions” section (fig. S1A). The experiment demonstrated that the black fabric has higher force output compared to the red fabric (fig. S4). Inflation and deflation timing were recorded across 10 consecutive cycles per condition on the same 2 × 8 array geometry at 138-kPa supply pressure, with inflation time defined as the interval from inlet pressurization to steady output force, showing that the red fabric has faster inflation times compared to the black fabric (fig. S2B). To achieve a balance between force output and response speed, the black fabric was used for all actuators in this work.

Manufacturing of FaBrick sheets

FaBrick sheets were fabricated using a patterned, heated rolling laminator (Fig. 2) modified from a commercial heated lamination system. The setup included two rolls of heat-sealable fabric positioned above and below the roller assembly, with their TPU-coated sides facing inward toward each other. During lamination, the two fabric layers were fed simultaneously into the heated roller, where the raised sealing pattern defined the pouch and channel geometry.

The stock rubber rollers of the laminator were replaced with custom 3D-printed stainless-steel sleeves featuring raised sealing features mounted on a stainless-steel core. The roller temperature was maintained between 150° and 160°C, and the rotational speed was set to ∼10 rpm, balancing seal quality and fabrication throughput. A tensioning mechanism kept both fabric rolls stretched tightly during feeding, preventing slack or wrinkles that could disrupt uniform sealing. Downstream of the heated roller, a pair of unheated rubber rollers continuously pulled the laminated fabric forward. This separation of pulling and sealing functions ensured that the heated rollers were responsible solely for forming airtight seals, resulting in a consistent and repeatable seal pattern across the entire sheet.

The process produces a continuous grid structure that can be fabricated at arbitrary lengths and is limited only by the roller width, which is 335 mm due to machine constraints rather than a fundamental limitation. The modular roller design further enhances versatility. The metal roller sleeves can be quickly swapped to create different pouch geometries or aspect ratios without altering the underlying system. Together, these features make the fabrication process suitable for mass production of FaBrick sheets and large-scale actuator arrays, supporting applications ranging from small scale actuators to large-scale wearable devices.

Actuator fabrication

FaBrick’s grid architecture allows multiple actuator types to be fabricated from the same sheet through simple cut, seal, and staple operations (Fig. 4). Once the actuators are fabricated, pneumatic connectors can be attached directly to any open channel to form an inlet port, allowing standard air tubing to interface with the actuator without additional fixtures or hardware. We apply a small amount of epoxy to securely attach the connector and the fabric to ensure an air-tight seal at the inlet, although a properly matched interface between the air-channel opening and the pneumatic connector can create a sufficiently airtight seal without adhesive for up to 207 kPa (30 psi).

Contraction actuator

A contraction actuator is created by cutting out a rectangular section of FaBrick and sealing along its perimeter with one or more inlets left open for pressurization (Fig. 4A). When inflated, each pouch bulges out of plane, shortening its length and generating contraction. When a sheet is wrapped into a cylinder, inflating the sheet will inflate all pouches together and result in axial contraction. The sheet can also be divided into independently pressurizable regions by selectively sealing the channels between pouches. In Fig. 4A, a central seal separates the sheet into two regions (orange and yellow), each consisting of an 2-column × 8-row array. When this patterned sheet is wrapped into a cylinder, pressurizing different regions produces directional contraction. Inflating the orange region deflects the actuator to the right, inflating the yellow region deflects it to the left, and inflating both regions simultaneously yields pure axial contraction (this deflection results solely from asymmetric contraction and is distinct from the dedicated bending actuator introduced later, which uses a strain-limiting layer to generate curvature). Actuation force and stroke can be tuned by modifying the number of active rows or columns or by adjusting the array aspect ratio. This example illustrates how simple planar patterning on FaBrick can create either linear contraction or directional bending depending on which regions are pressurized.

Twisting actuator

A twisting actuator is created by cutting a parallelogram-shaped piece of FaBrick and cutting the connections between adjacent columns, while keeping the top and bottom edges intact so that the sheet remains a single piece (Fig. 4C). Cutting opening the intercolumn connections isolates each column so that they can contract independently rather than being forced into uniform shortening. During twisting, neighboring columns will contract independently to achieve twisting. If the columns remained joined along their full height, then this differential motion would be mechanically incompatible, causing the sheet to behave similar to a linear actuator and buckle instead of rotating.

To generate twisting, the processed parallelogram sheet is wrapped around a cylindrical support that provides a rotational degree of freedom, typically realized by a bearing- or axle-mounted cylinder. The top and bottom edges of the parallelogram are attached rigidly to this rotatable cylinder. Upon inflation, each column’s independent contraction redirects into tangential force around the circumference, producing controlled rotation. The twist angle and torque can be tuned by adjusting several design parameters. A smaller initial skew angle and a larger number of columns increase torque, while the number of rows increases angular displacement. Fabrication involves sealing along diagonal paths while preserving airflow within each column and providing open inlets for pressure. The resulting actuator produces twisting motion, demonstrating how complex rotational behaviors can emerge from planar patterning on the FaBrick sheet.

Rotation actuator

A rotation actuator is created by cutting a rectangular section from the FaBrick sheet and sealing the internal channels so that only the top and bottom rows remain inflatable (Fig. 4C). The sheet is then folded into an accordion-like geometry, and the top and bottom rows are aligned and stapled together at a chosen pivot location to form the rotation joint. When pressurized, the pouch cells in each fold expand outward and generate a moment about this pivot, with each fold acting as an individual torque-producing unit.

The actuator’s performance can be tuned by adjusting several design parameters. The rectangle height (which sets the lever arm) and the number of cells in each fold both contribute to torque generation, while the number of inflatable rows determines angular displacement. This enables users to tailor the output torque and angular displacement to the needs of a specific application. By modifying only the folding pattern and mounting geometry, without changing the material or fabrication method, the same FaBrick sheet can provide either high-torque, small-angle motion or larger-angle, lower-torque motion. This demonstrates how rotary actuation emerges directly from planar patterning and simple folding operations, enabling seamless integration of FaBrick into a wide range of torque-driven mechanisms.

Bending actuator

A bending actuator is created by cutting two pieces of material, a strain-limiting layer and an inflatable layer, from the FaBrick sheet according to the geometry shown in Fig. 4D. The inflatable layer is then stapled on top of the strain-limiting layer. A bending segment is defined as the section between the adjacent staples, which contains two pouch cells for our design. The distance between staple points determines how much each segment can expand and therefore governs the resulting bending behavior.

When the pouches inflate, the inflatable layer expands, while the strain-limiting layer constrains extension, causing the actuator to curve toward the constrained side. Closely spaced staples restrict local expansion and produce tighter curvature, whereas wider spacing allows more gradual bending. Beyond a critical spacing, as demonstrated in the 23-mm staple spacing case in Fig. 4D, the inflated pouches will contract to a length smaller than the length of the strain-limiting layer between the staple spacing and thus pull the structure in the opposite direction. This enables reverse bending using the same material and layout. By tuning the number of cells per segment and the staple-to-staple distance, FaBrick can achieve tunable bending with only simple planar patterning and minimal assembly.

Stiffening structure

A stiffening element is fabricated by rolling a rectangular FaBrick sheet tightly along its width direction and securing the rolled cylinder inside an external rigid collar that constrains radial expansion (Fig. 4B). With all internal channels left open, pressurizing the rolled sheet causes each pouch layer to inflate against its neighbors and against the collar, producing a column whose flexural stiffness scales with internal pressure. We characterized this behavior with a three-point bending test on rolled stiffening structures of six diameters (26, 28, 30, 32, 34, and 36 mm). At 207 kPa (30 psi), the peak resisting force at 20-mm cross-head deflection ranged from 25.7 N (36 mm in diameter) to 45.0 N (26 mm in diameter), compared with 3 to 4 N for the same elements at atmospheric pressure, a 7 to 15× stiffness increase from pressurization (fig. S3D and see Supplementary Text for experimental setup details). This element is what supports the legs of the quadruped demonstration (Fig. 7D) and the internal structure of the elephant-trunk manipulator (Fig. 7C). It is functionally distinct from the other classes of actuators discussed in this paper because it is operated only at full inflation and acts as a switchable structural member rather than a continuously varying actuator. We do not provide a closed-form model for it in this work, but modeling and characterizing the stiffness of FaBrick structures could be explored in future work.

Experimental setup for evaluating model predictions

Force-displacement data for the linear actuators were measured using a load cell (CALT DYLY-103) mounted on a precision linear guide stage to ensure axial alignment (fig. S1A). Torque for the twisting actuators was recorded using a torque load cell (ATO Static Torque Sensor) coupled to a geared NEMA 23 stepper motor that incrementally rotated the actuator under controlled angular steps (fig. S1B). Torque for the rotation actuators was tested on a separate custom made jig using the same load cell setup as the linear tests. The jig has a rotatable platform that can be fixed at a range of angular displacements. At each angle, the load cell measures the force, and the torque is calculated as the product of the measured force and the corresponding lever-arm distance between the load cell and the pivot point (fig. S1C).

The pressure was held constant at 69 kPa (10 psi), 138 kPa (20 psi), and 207 kPa (30 psi) while sweeping over a range of displacements. For the linear actuator, data were collected at 0.2-mm displacement intervals, and for the twisting and rotation actuators, data were collected at 1° angular intervals. A built-in 0.5-s pause was applied at each step to allow internal pressure to equalize before recording. The control program automatically reversed travel motion once zero force or torque was detected, ensuring that the full linear and angular displacement range was captured. Data were collected for three trials. For each trial, the inflation stroke (volume increasing stroke) and deflation stroke (volume decreasing stroke) were collected. The control code is designed such that, during the inflation stroke, the motion reverses and initiates the deflation stroke as soon as the sensor detects zero force or torque. Baseline calibration was performed before every test to zero both force and pressure sensors.

The NRMSE is obtained using the below equation, where N is the number of data points and y is the force or torque data, normalized by range of force or torque data $y_{\text{max}}-y_{\text{min}}$

$$\text{NRMSE}=\frac{\sqrt{\frac{1}{N}{\displaystyle \sum _{i=1}^N}{(y_i-{\hat{y}}_i)}^2}}{y_{\text{max}}-y_{\text{min}}}$$ (28)

Supplementary Materials

The PDF file includes:

Figs. S1 to S4

Supplementary Text

Legends for movies S1 to S6

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S6

Data S1