Thin-Film-Shaped Flexible Actuators

Human‐like and creature‐like systems are one of the most representative imaginary blueprints of future robots. To fulfill this blueprint, the development of high‐performance actuators across different length scales is indispensable. Owing to their mechanical compliance and conformability to curvy surfaces of living organisms, flexible actuators have emerged as an essential direction of next‐generation actuators. This review focuses on thin‐film‐shaped flexible actuators (TFFAs), a rising family of flexible actuators, aiming to provide an overview of the state‐of‐art status in this exciting direction. The designs, manufacturing, and mechanisms of various TFFAs are summarized, according to their key composing materials/mechanisms, including, for example, nanomaterials, liquid crystal elastomers, shape memory polymers/alloys, hydrogels, biohybrids, and other mechanisms/materials. The representative applications of TFFAs are introduced, ranging from biomedical uses, robots for environment explorations, to haptic interfaces and reconfigurable electronics. Finally, the grand challenges and open opportunities are discussed in detail.


2D TFFAs Based on Nanomaterials
Various nanomaterials, such as carbon nanomaterials and metallic nanomaterials, have been used for the fabrication of TFFAs in both 2D and 3D forms, thanks to their high performances and versatile synthesis approaches.
Given their excellent opto-/thermal-electrical properties and highly porous nature, carbon-based nanomaterials (e.g., CNTs, graphene, graphite, and so on) are superior at conversion of various stimuli [74,75,93] (i.e., input energy in different forms, such as light and electricity) into different usable forms of driving powers (e.g., local heating and redox reactions), which makes them promising candidates for uses in TFFAs. In terms of the fabrication of 2D TFFAs, 3D printing, laser processing, and lithography were normally used. For CNT-based TFFAs, their actuations usually relied on optothermal, [84][85][86] electrothermal, [79][80][81][82][83] and electrochemical mechanisms. [77,78] The commonly used structures of actuation layers in CNT-based TFFAs include CNTs randomly embedded in polymeric matrix [122] as well as aligned continuous CNT thin films. [123] The former one usually exploited bilayer or multilayer constructions, due to the requirements of strain mismatches between different layers to enable desired bending deformations. For example, Hu et al. [124] reported a CNT/PDMS bilayer TFFA capable of light-driven bending deformations (i.e., light intensity of 250 W cm À2 ). The latter (i.e., aligned continuous CNT films) could render anisotropic mechanical or electrical properties for TFFAs. For instance, Li et al. [83] reported a TFFA ( Figure 2a) composed of aligned CNT sheets laminated with a PDMS layer. Complex patterns can be defined by laser cutting to enable mechanical (i.e., modulus) and electrical (i.e., conductivity) anisotropies, as well as directional bending capabilities (e.g., T-shape with different cutting angles). Shi et al. [125] developed a triple-layered NIR-driven TFFA (i.e., CNTs/PDMS/ fluorescent gel) to mimic the function of skin chromatophores in cephalopods. Graphene and its derivatives form another huge family of carbon materials, which exhibit excellent electrical/ thermal properties, high flexibility, and diverse capabilities in energy conversion (e.g., photo-to-thermal, electrical-to-thermal, and chemical-to-thermal), also stand as excellent options for uses in TFFAs. Figure 2b presents a 2D graphene-based light-driven TFFA for swimming robots. [126] In particular, the laser-induced graphene layer enables the light-to-thermal conversion, allowing generation of driving forces through the Marangoni effect. In Figure 2c, Dong et al. [96] demonstrated a multistimuli-responsive TFFA consisting of a graphite layer, which realized programmable bending deformations via humidity and temperature changes.
Metallic nanomaterials, such as silver nanowires (AgNWs) and magnetic particles, are commonly used materials for TFFAs. Due to their high electrical conductivity and excellent optical properties, AgNWs have been widely adopted for TFFAs. For instance, Amjadi and Sitti [100] demonstrated a multistimuli-responsive TFFA (Figure 2d) consisting of a thermoexpansive polypropylene (PP) layer (coefficient of thermal expansion (CTE): %137.5 Â 10 À6 K À1 ) and a hygroexpansive paper layer coated with AnNW/poly (3,4-ethylenedioxythiophene) polystyrenesulfonate (PEDOT:PPS). Different from electrically controlled actuators, TFFAs based on magnetic particles (e.g., iron and nickel-or silicon-based alloys of iron, barium hexaferrite, neodymium-iron-boron) could enable fast-responsive and untethered deformations under magnetic fields. For example, Cui et al. [102] presented magnetically controlled microscopic TFFAs by programming the arrays of single-domain nanomagnets on connected panels, as shown in Figure 2e.

2D TFFAs Based on LCEs/LCNs
2D TFFAs could enable both 2D-to-2D and 2D-to-3D shape changes. For instance, by harnessing both the large deformation of LCE and rapid actuation of dielectric elastomer (DE), Fowler et al. [17] achieved enhanced directional actuation of LCE-based TFFAs with a rapid response (strain rate: 18% s À1 ) (Figure 3a). Adapted with permission. [83] Copyright 2015, American Chemical Society. b) Patterned pentagram TFFAs is driven by light-induced Marangoni effect. Adapted with permission. [126] Copyright 2021, Wiley-VCH. c) Schematic illustration and optical images of graphene-oxide-based TFFAs capable of bending/rolling in response to humidity/light/temperature change. Adapted under the terms of the CC-BY license. [96] Copyright 2019, The Authors, published by Springer Nature. d) Illustration of the actuation mechanism and scanning electron microscope (SEM) images of AgNWs-based TFFAs capable of electrically controlled bending/folding. Adapted with permission. [100] Copyright 2016, American Chemical Society. e) Schematic illustration and SEM images of magnetic-nanoarrays-based TFFAs capable of self-folding at desired local regions. Adapted with permission. [102] Copyright 2019, Springer Nature. f ) Optical images of the light-driven helical TFFAs, capable of grasping objects. Adapted with permission. [84] Copyright 2016, American Chemical Society. g) Schematics of the 3D printing technique and various 3D-to-3D shape morphing processes of magnetic-particle-based TFFAs. Adapted with permission. [71] Copyright 2018, Springer Nature.
www.advancedsciencenews.com www.advintellsyst.com By mechanics-guided design and fabrication strategy, Wu et al. [43] prepared TFFAs based on LCE metamaterials with an unprecedented biaxial actuation strain of À53% and a negative thermal expansion of À33 125 ppm K À1 . To achieve 2D-to-3D shape changes, Li et al. [139,140] developed network-shaped micro-TFFAs capable of topological transformations, through molding and magnetic field-induced alignment. Using surface effect-induced approach to spatially align local molecules, Ware et al. [138] realize shape changing of LCE-based TFFA from 2D configurations into complex 3D ones (Figure 3b). Xia et al. [135] realized high-resolution (e.g., 100 nm) alignment of LCE-based TFFA using microchannels (Figure 3c). 3D/4D printing was also adopted to fabricate LCE-based TFFA, owing to their possible alignment of mesogens by shear stresses applied during ink extrusion. [31,50,142] For instance, Kotikian et al. [50] reported the design and additive manufacturing of LCE-based TFFAs with spatially programmed nematic order. Wang et al. [31] printed TFFAs consisting of LCE filaments with tunable actuation behavior and mechanical property, achieving versatile shape changes ( Figure 3d). Using photolithography, Martella et al. [134] demonstrated microhands that could grab microscopic objects by autonomous operation of a light-responsive LCE actuator ( Figure 3e). By means of stretch-induced alignment, Yang et al. [147] reported photo-crosslinked LCE-based TFFAs capable of a diversity of shape changes ( Figure 3f ). Recently, Liu et al. [143] developed a LCE-based TFFA robotic surface with biomimetic structures (e.g., LCE networks were used as artificial muscle, skeleton, and skin), allowing for large and reprogrammable 3D shape morphing.

3D TFFAs Based on LCEs/LCNs
LCE-based TFFAs with 3D as-fabricated configurations were often prepared through residual stress-induced assembly [19,136,148,149] and mechanical deformation-induced assembly. [144,150,151] For instance, by controlling the residual stress, Zeng et al. [19] prepared 3D iris-shaped TFFA capable of adjusting its aperture upon varied incident light intensity. Later, they fabricated a miniaturized "Ω"-like robot, which is capable of mimicking caterpillar locomotion on different substrates under spatially uniform visible light ( Figure 3g). [136] Ahn et al. [150] demonstrated a LCE-CNT-based 3D soft robot that offers three different motion modes, including crawling, squeezing, and jumping ( Figure 3h). As shown in Figure 3i, Xiao et al. [141] reported an electrothermally controlled conveyor soft robot that could push an object forward by alternating activation of its two "leg"-like TFFAs. In addition, Barnes et al. [144] programmed LCEs directly using mechanical deformations, resulting in TFFAs with complex 3D-to-3D shape morphing capabilities. By combining the light-active LCE film with a passive frame, Gelebart et al. [151] developed a 3D oscillatory robot based on the self-shadowing mechanism ( Figure 3j). Recently, Li et al. [44] reported reconfigurable 3D TFFAs through the buckling-guided assembly. To further equip such LCE-based TFFAs with complex 3D-to-3D shape morphing capabilities, Pang et al. [152] combined controlled strain distribution of LCEs with the buckling-guided assembly, yielding a set of millimeter-scale soft actuators (from 1 to 10 mm) with various 3D as-fabricated configurations (Figure 3k).

TFFAs Based on Shape Memory Polymers
Owing to their shape memory effect and excellent activation performances, SMPs processed by advanced fabrication methods (e.g., 3D printing, [47,153,154] photolithography, [40] and cutting [45,155,156] ) were also widely used to enrich the morphing capabilities of TFFAs.

2D TFFAs Based on SMPs
SMP-based TFFAs could be triggered by various external stimuli, including temperature, [45,157] light, [158][159][160][161] electrical current, [46,156,162] and magnetic field. [155] As shown in Figure 4a, Liu et al. [160] achieved programmed folding sequence of SMPbased TFFAs under optical actuation (i.e., blue light results in the out-of-plane configuration, while red light yields the in-plane configuration). By combining SMPs, paper, and resistive circuits, Robert Woods' group [162] fabricated electrothermally controlled foldable TFFAs. Later on, this group [46] utilized such SMP-based TFFAs in the form of origami designs to develop crawling robots capable of 2D-to-3D shape transformation ( Figure 4b). To enable TFFAs with multifunctional shape manipulations, Ze et al. [155] introduced magnetically active SMP (i.e., polymeric matrix embedded with two different magnetic particles) in the preparation of TFFAs with untethered, fast, and reversible shape transformation/locking (Figure 4c). In this design, the low-coercivity particles contribute to the softening of the matrix by magnetically induced heating (applying B h ), while the high-remanence particles with reprogrammable magnetization profiles allow rapid and reversible shape changes under applied magnetic fields (B a ).

TFFAs Based on SMAs
Among various shape memory materials used in soft actuators, SMAs are particularly attractive due to their high actuation stress and convenient actuation by Joule heating. [165] SMAs are a group of alloys that exhibit a reversible change in both shape and rigidity, due to their transformation between monoclinic martensite and cubic austenite phases during heating and cooling. [166] Depending on whether a reversible actuation could be accessed, the shape memory effect of SMAs can be categorized as either one-way or two-way.
In order to obtain 2D-to-3D shape morphing, SMA wires were usually embedded in elastomers [32,[170][171][172] or integrated with fabrics. [173,174] For example, by encapsulating SMA wires in PDMS matrices, a TFFA capable of rapid bending/twisting actuations was demonstrated. [32] Lee et al. [170] developed a miniaturized SMA-based TFFA (length: %5 mm) that could generate a 390 μN force with the bending angle up to 80° (Figure 5b). Using the typical textile manufacturing technique, Buckner et al. [174] fabricated robotic fabrics consisting of SMA and SMP fibers, as well as printable in-fabric sensors, allowing simultaneous actuation and sensing.

3D TFFAs Based on SMAs
The reported SMA-based 3D TFFAs were manufactured mainly using the residual stress-induced assembly [175][176][177][178][179] and bucklingguided assembly. [33] For instance, Huang et al. [175,177] reported the use of the stress-induced assembly in the development of a 3D TFFA composed of U-shape SMA wires and layers of stretched and unstretched elastomers with high thermal conductivities. This actuator could bend with a curvature change of %60 m À1 in 0.15 s, and could be activated with a frequency of above 0.3 Hz, using just a pair of miniature 3.7 V lithiumpolymer battery. Furthermore, using MEMS processing methods, microscale 3D TFFAs could also be manufactured. [176,178] For example, as shown in Figure 5c, a microscale wrappershaped SMA-based 3D TFFA was prepared via photolithography and etching process. [178] By harnessing the force balance between . TFFAs based on SMPs. a) Optical images of sequential shape morphing process for SMP-based TFFAs by using selective light absorption. Adapted with permission. [160] Copyright 2017, AAAS. b) Design illustration and optical images of self-assembled robots equipped with SMP-based TFFAs at strategic locations. Adapted with permission. [46] Copyright 2014, AAAS. c) Illustration of SMP-based TFFAs actuated by two different magnetic fields, and sequential optical images of the grippers lifting an object in both soft and stiff states. Adapted with permission. [155] Copyright 2019, Wiley-VCH. d) 3D-to-3D shape morphing processes of bird-like and elephant-like TFFAs by using crystalline shape memory polymer network with thermo-and photoreversible bonds. Adapted with permission. [164] Copyright 2018, AAAS.

TFFAs Based on Hydrogels
Hydrogels, consisting of crosslinked polymeric network infiltrated with water (e.g., hydrogen bonding, covalent/non-covalent bonding, among others), are able to offer considerable actuation strains (e.g., >10 times in volume) during hydration and dehydration, under different external stimuli, such as temperature, light, pH, and others, [180,181] which makes them promising candidates for TFFAs. Various processing approaches (e.g., cutting, molding, lithography, 3D assembly, and the rest) have been employed to fabricate hydrogel-based TFFAs with complex geometries and multiple deformation modes across different length scales, showing great potentials in applications like drug delivery, [52,182] tissue engineering, [183] and biomedical devices. [184,185]

2D TFFAs Based on Hydrogels
Additive manufacturing (e.g., 3D/4D printing) and subtraction technologies (e.g., laser cutting, photolithography) are most commonly used approaches for hydrogel-based 2D TFFAs. In terms of the additive manufacturing, by controlling the alignment of cellulose fibrils during 3D printing, Gladman et al. [49] fabricated hydrogel-based TFFAs that could form complex 3D morphologies in aqueous environment (Figure 6a). Using a multinozzle 3D printing system, Zheng et al. [186] prepared hydrogel composites consisting of component (i.e., poly(N-isopropyl acrylamide)) with considerable volume contraction and nonresponsive component (i.e., P(AAc-co-AAm)) to further improve the response speed and output force of hydrogel-based TFFAs. Through molding techniques, Li et al. [187] fabricated photoactive TFFAs capable of programmable 3D origami-like deformations upon irradiation. In addition, using PDMS molds, Zhu et al. [188] prepared hydrogelbased self-powered TFFAs made of hydrophobic and hydrophilic groups, allowing movements on water surfaces.
www.advancedsciencenews.com www.advintellsyst.com have also been widely exploited. As shown in Figure 6b, a fourfooted robot (i.e., equipped with hydrogel-based TFFAs) fabricated by mechanical cutting can realize unidirectional crawling through light activation. [190] By photopatterning, Wu et al. [193] prepared single-layer hydrogel-based TFFAs exhibiting differential shrinkage and elastic moduli, which can be utilized to yield stress mismatch for complex shape morphing. By the use of periodically photopatterned hydrogel sheets, Wang et al. [53] succeeded in Figure 6. TFFAs based on hydrogels. a) Optical images of deformation sequences for 3D-printed hydrogel-based TFFAs during the water absorption. Adapted with permission. [49] Copyright 2016, Springer Nature. b) Deformation of a light-driven four-footed TFFA. Adapted with permission. [190] Copyright 2020, Springer Nature. c) Illustration of the photolithographic patterning process of the hydrogel and optical images for the cooperative deformation of TFFAs induced by swelling. Adapted with permission. [53] Copyright 2017, AAAS. d) DNA sequence-directed photopatterned hydrogel TFFAs. Adapted with permission. [54] Copyright 2017, AAAS. e) Electrically controlled snapping of domal hydrgel-based TFFAs with a mechanical bistability. Adapted with permission. [28] Copyright 2022, AAAS. showing a synergistic deformation mode, in which neighboring domains could mutually interact and cooperatively deform ( Figure 6c). Furthermore, via photolithography, Cangialosi et al. [54] reported centimeter-scale TFFAs containing multiple domains that could undergo different deformations upon receiving various DNA sequences (Figure 6d).

3D TFFAs Based on Hydrogels
Hydrogel-based TFFAs capable of 3D-to-3D shape morphing have received increasing attention, due to their unique capabilities in drug delivery, clinical, and other applications. For example, Lee et al. [72] printed a doubly curved hydrogel-based TFFA with embedded microfluidic channels, demonstrating a swelling-induced fast actuation through the snap-through mechanism. Breger et al. [182] prepared thermally responsive microgrippers consisting of a swellable soft hydrogel (i.e., pNIPAM-AAc) and a nonswellable stiff segmented polymer (i.e., polypropylene fumarate), allowing the formation of stable as-fabricated 3D configurations by the stress-induced assembly. Li et al. [194] developed hydrogel-based magnetic TFFAs with complex 3D magnetization, resulting in programmable 3D-to-3D actuation behaviors.
Recently, through a two-step photolithographic polymerization followed by swelling-induced stress mismatch, Li et al. [28] fabricated a bistable domal TFFA. Particularly, the applied electric field could redistribute its mobile ions that direct the migration of water molecules, therefore triggering the actuation (Figure 6e). Lu et al. [195] prepared a 3D TFFA using thermo/ion dual-responsive hydrogel, showing great capabilities of shape fixing and morphing.
Pressure-driven actuators were also studied extensively in recent years. The relatively large volumes and bulky forms of traditional pressure-driven actuators set limitations of their widespread applications. To address this issue, many pressure-driven TFFAs were developed. [238][239][240][241][242] For example, by welding laser-cut sheets of thermoplastic polyurethane, Moghadam et al. [239] reported a TFFA made of five different pneumatically soft actuators (e.g., bending, rotating, contracting) capable of both planar and spatial motions.

Surgery
Minimum-invasive or noninvasive approaches, such as biopsy [182,255,256,301] and catheter, [254] are crucial for early diagnosis, interventions, and treatments of various diseases. The unique advantages of TFFAs (e.g., flexibility and high degree of integration) make them excellent candidates for less/noninvasive clinical surgeries.
For instance, Leong et al. [301] developed untethered thermobiochemically actuated microgrippers remotely controlled by external magnetic fields to assist in vitro biopsies. To demonstrate the feasibility of microgrippers for in vivo uses, Gultepe et al. [255] prepared grippers with sizes ranging from 300 μm to 1.5 mm (Figure 9a), which can extract tissue samples from real organs (i.e., a porcine liver) and hard-to-reach places within a live animal (i.e., a porcine biliary tree). Additionally, Breger et al. [182] designed magnetically controlled thermoresponsive hydrogel grippers that enable enhanced biocompatibility and possibly endow biodegradability in such surgical applications. Later on, using a novel molding approach, Jin et al. [256] fabricated multifingered grippers to allow manipulation of few cells or an individual cell.

Rehabilitation
Wearable robotic systems often require a high flexibility to provide an enhanced comfortableness, and therefore, soft materials such as biocompatible elastomers and fabric networks are broadly adopted. To improve their rehabilitation effect, TFFAs with a reliable actuation are demanded. For instance, Zhu et al. [267] reported a TFFA (called fluidic fabric muscle sheets) composed of fabric network with integrated fluidic transmissions, enabling wearable, and biomedical devices for rehabilitation. Inspired by natural auditory systems, Yan et al. [266] developed a piezoelectric fiberbased TFFA featuring bidirectional conversion of mechanical vibrations (i.e., sound generation and reception) and electrical signals. As shown in Figure 9b, a shirt using such woven TFFA could precisely measure the direction of an incoming acoustic impulse as well as auscultate cardiac sound signals. In addition to the above wearable rehabilitation devices, in vivo rehabilitation/organ assisting devices have also been developed using TFFAs. For example, Roche et al. [265] reported a TFFA-based soft robotic sleeve that could act as a cardiac ventricular assisting device (Figure 9c). In addition, Yang et al. [268] prepared an implantable soft robot using magnetically responsive TFFA to assist urination, by improving the contractility of the detrusor muscle in underactive bladders (Figure 9d). Adapted under the terms of the CC-BY 4.0 license. [220] Copyright 2020, The Authors, published by Wiley-VCH. b) Optical images of the locomotion process of 3D TFFAs made of piezoelectric composites in a driving cycle. Adapted with permission. [36] Copyright 2019, AAAS. c) Fabrication process and optical images of electromagnetically controlled TFFAs for crawling soft robots. Adapted under the terms of the CC-BY license. [23] Copyright 2020, The Authors, published by Springer Nature. d) Elastically and plastically foldable electrothermal micro-origami structures. Adapted with permission. [244] Copyright 2020, The Authors, published by Wiley-VCH. e) Schematic illustration of the fabrication process and performance of microscale helical TFFAs activated by the phase transition. Adapted with permission. [247] Copyright 2013, Wiley-VCH.

Reconfigurable Bioelectronics
Thanks to the advancements in flexible electronics, reconfigurable bioelectronics, featuring shape-changing capability, long-term stability, and biocompatibility are playing increasingly important roles in in vivo tissue engineering. The use of TFFAs in such devices could further broaden their applications in biomedical research. Figure 9. Applications of TFFAs for biomedical research. a) Ex vivo tissue excision using untethered TFFA-based grippers. Adapted with permission. [255] Copyright 2013, Wiley-VCH. b) Woven-fabric-integrated TFFAs for sound-direction detection and stethoscope applications. Adapted with permission. [266] Copyright 2022, Springer Nature. c) A soft robotic sleeve made of 3D TFFA for cardiac surgeries. Adapted with permission. [265] Copyright 2017, AAAS. d) An artificial bladder by using magnetic TFFAs for assisted urination. Adapted with permission. [268] Copyright 2022, AAAS. e) A temperature-driven TFFA integrated with electrodes, twining on peripheral nerve for stimulation and recording. Adapted under the terms of the CC-BY 4.0 license. [298] Copyright 2019, The Authors, published by AAAS. f ) A TFFA with microelectrode arrays for mapping electrical activity in brain organoids. Adapted under the terms of the CC-BY 4.0 license. [299] Copyright 2022, The Authors, published by AAAS. g) Temperature-controlled TFFAs as self-latching drug delivery devices. Adapted under the terms of the CC-BY 4.0 license. [259] Copyright 2020, The Authors, published by AAAS. h) Schematics of wireless soft millirobots based on magnetically controlled TFFAs, for on-demand drug delivery. Adapted under the terms of the CC-BY 4.0 license. [257] Copyright 2022, The Authors, published by AAAS. i) Transfer of the therapy patch by small-scale magnetic soft robot in stomach. Adapted with permission. [300] Copyright 2022, AAAS. j) A breathable, shrinkable, hemostatic patch made of LCE-based TFFAs for enhanced skin regeneration. Adapted with permission. [43] Copyright 2021, Wiley-VCH. k) A strain-programmed patch with TFFA for diabetic wounds repair. Adapted with permission. [183] Copyright 2022, Springer Nature. l) An autonomously swimming biohybrid fish integrated with TFFAs, designed with human cardiac biophysics. Adapted with permission. [212] Copyright 2022, AAAS.
www.advancedsciencenews.com www.advintellsyst.com For example, using SMP-based TFFA, Zhang et al. [298] prepared a self-climbing twining electrode driven by body temperature (Figure 9e), enabling the formation of conformal neural interfaces to reduce heart rate and record the action potential of the sciatic nerve. Hao et al. [184] reported a hydrogel-based device that can deform into sophisticated geometries to perform in situ monitoring of mammal organs. Recently, using TFFA consisting of self-folding polymer layer and conductive metal electrodes, Huang et al. [299] prepared wafer-integrated shell microelectrode arrays (MEAs) caps for high signal-to-noise ratio sensing and 3D spatiotemporal brain organoid recording (Figure 9f ).

Drug Delivery
Given the reversible and controllable shape morphing capability, TFFAs hold great potentials for drug delivery. For instance, Chang et al. [260] reported a wireless, radio-frequency controlled ionic polymer-metal composites actuator that can deliver targeting drugs within aqueous environment. Li et al. [302] fabricated a soft microrobot for targeted drug delivery using pH-responsive hydrogel-based TFFAs under the remote control of magnetic fields. Inspired by the above work, Ghosh et al. [259] demonstrated untethered thermally actuated microdevices with sharp microtips that can be stably attached on the mucosal tissue for 24 h, prior to the drug release (Figure 9g). To reach the diverse 3D surfaces of disease sites inside the human body and perform minimally invasive therapy (e.g., drug delivery), Wu et al. [257] prepared an untethered magnetically controlled soft millirobot with microstructured footpads, featuring strong adhesion on wet 3D tissue surfaces and on-demand drug delivery upon the varied pH values (Figure 9h).

Tissue Engineering
TFFAs could be designed to offer a conformable integration with various tissues and, therefore, have been explored for tissue engineering. For example, Dong et al. [300] demonstrated a therapy process operated by a TFFA-based magnetic robot (Figure 9i). To accelerate skin regeneration while avoiding scars and keloid, Wu et al. [43] fabricated a shrinkable and breathable hemostatic patch (Figure 9j) using a LCE-based metamaterial with an actuation temperature (46°C) close to the body temperature. Later, relying on a hydration-based shape-memory mechanism, Theocharidis et al. [183] prepared a strain-programmed healing patch for diabetic wounds using TFFA consisting of a dry adhesive layer of hydrogel bound to a prestretched hydrophilic elastomer layer, achieving an enhanced wound recovery (Figure 9k). In addition to the wound repair, biohybrid robots based on TFFAs have a direct and immediate impact on tissue engineering and regenerative medicine. [208,211,212] Very recently, Lee et al. [212] reported an autonomously swimming biohybrid fish (Figure 9l) using TFFA consisting of cardiac cells embedded in polymer frames, by replicating human cardiac biophysics. This study shows great potentials for a more granular analysis of structure-function relationships in cardiovascular physiology.

Grippers
TFFA-based soft robotic grippers, benefiting from their intrinsic compliance and continuous deformations, present high degrees of freedom in terms of their shape changes, thereby facilitating the grasp of arbitrarily shaped 3D objects. [272] For example, Shintake et al. [223] reported soft grippers using DE-based TFFAs with electroadhesion, allowing the grasp of deformable/fragile objects and flat objects like paper (Figure 10a). Wang et al. [171,172] reported TFFA-based soft grippers composed of three identical fingers of tunable stiffness for adaptive grasping. Roh et al. [156] reported a millimeter-scale soft gripper using SMP-based TFFAs, which was able to perform manipulation of heavy objects and grasping of organisms at micro-to-millimeter scale. This gripper could achieve simultaneous temperature/ pressure sensing and stimulation of the grabbed objects.

Locomotive Robots
Robots with various locomotion modes can implement the exploration of unstructured environments. In terms of terrestrial environments, the reported TFFAs-based robots could perform the following locomotion modes: crawling, [18,36,231,304] walking, [33,303] rolling, [142] jumping, [229,305] and climbing. [152,216,269] For instance, Liang et al. [231] developed an agile insect-scale crawling robot with the use of a piezoelectric-based TFFA and electrostatic footpads, enabling fast moving and swift rotational motions. Using 3D TFFAs consisting of soft elastomer with embedded curved SMA wires, Li et al. [309] demonstrated a crawling robot, allowing periodic bending and recovery. Using electrohydrostatic TFFAs, Chen et al. [229] demonstrated robots capable of rapid, consecutive, and steered jumping (Figure 10b). Wu et al. [310] fabricated a laser-controlled crawling soft robot using 2D TFFAs based on thermally responsive hydrogels, which could move over complex terrains. Recently, Pang et al. [152] developed a set of climbing microrobots using LCE-based TFFAs, allowing simultaneous climbing and transitioning between different curvy surfaces.
In terms of aqueous environments, TFFA-based robots could propel themselves on [188,191,271] or in [101,306,311,312] the liquid media, realizing complicated gaits. Ren et al. [101] fabricated a jellyfish-like robot (i.e., 5.6 mm in diameter) based on magnetic TFFAs that can be used to transport small objects underwater. Inspired by the deep-sea snailfish, Li et al. [312] developed a robot using DE-based TFFAs for deep-sea exploration. Remarkably, this robot can swim freely at a depth of 3,224 m under ocean, and flap its fins at a depth of 10 900 m under the Mariana Trench (Figure 10d).
Flying robots (e.g., microfliers or vehicles) represent a rising robotic family. [41,307,313] In this context, TFFAs driven by electrostatic forces with high frequency, fast response, and low power consumption are promising for relevant applications. [217,308] Helps et al. [217] developed a flying robot using swingable TFFAs controlled by time-varying electrostatic fields.

Haptic Interfaces
Haptic interfaces are playing crucial roles in virtual reality (VR) and augmented reality (AR). A challenge in this field involves developments of miniaturized, micro-to-nano scale actuators that could offer similar or better actuation performances, when comparing with the existing macroscale counterparts. Owing to their lightweight, compact design and large actuation strain, TFFAs are promising to fulfill the requirements of nextgeneration haptic devices. Given the very broad family of haptic interfaces, this section mainly focuses on the following two aspects: touch screen [275][276][277]280,282] and wearable haptic interfaces. [273,274,314]

Touch Screen
Because of their flexible characteristics and excellent actuation performances, the use of TFFAs in touch screen could enhance user experiences. For example, Jun et al. [279] developed a DE-based TFFA that could generate four different surface textures reflecting touch feeling from relatively smooth to rough (Figure 11a). Using bistable TFFAs with patterned CNT electrodes, Qiu et al. [276] fabricated a 4 Â 4 pneumatic tactile display with Braille standard resolution (Figure 11b). To enhance the user interface of tactile electronic devices, multimodal feedback is necessary. Van et al. [282] demonstrated an electrically controlled button with piezoelectric TFFAs, enabling simultaneous audio and haptic feedback (Figure 11c).

Wearable Haptic Interfaces
Soft wearable interfaces are the most common forms of VR/AR haptic feedback systems. Figure 11d shows a kinesthetic feedback glove using textile electrostatic clutches for VR applications. [278] In detail, the applied capacitive TFFA-based clutches are able to convert different applied voltages to output frictional forces. As shown in Figure 11e, Leroy et al. [283] fabricated a 5 Â 5 array of wearable hydraulically amplified TFFA, allowing both out-of-plane and in-plane motions to generate normal and shear forces. Ji et al. [281] reported a feel-through DE-based TFFA (Figure 11f ), providing a broad vibrotactile feedback spectrum (from 1-500 Hz).  [223] Copyright 2015, Wiley-VCH. b) Sequential images of jumping motion for soft robots with HASEL-based TFFAs. Adapted under the terms of the CC-BY license. [229] Copyright 2021, The Authors, published by Springer Nature. c) An untethered jellyfish-inspired soft millirobot actuated by magnetically responsive TFFAs. Adapted under the terms of the CC-BY license. [101] Copyright 2019, The Authors, published by Springer Nature. d) A snailfishinspired, self-powered soft robot with DE-based TFFAs for deep sea exploration. Adapted with permission. [312] Copyright 2021, Springer Nature.

Other Reconfigurable Devices
In addition to the above, TFFAs also show great potentials in reconfigurable surfaces [137,253,287,290,291] and reconfigurable electronics. [288,293,315,316]

Reconfigurable Surfaces
Attributing to their flexible and morphable nature, TFFAs based on smart materials (e.g., LCE, DE, and SMP) could be reshaped by harnessing their shape memory effects, electrophysical/ thermal properties, and the rest. [137,287,290,295] Remarkably, through inverse design strategies, programmable, and even reconfigurable surfaces could be fabricated using such TFFAs. For instance, Hajiesmaili et al. [295] reported that tunable Gaussian curvatures could be realized via DE-based TFFAs.
Recently, Bai et al. [253] presented a film-like mechanical metasurface driven by Lorentz forces (Figure 12a). In particular, by implementing an in situ stereo-imaging feedback system combined with a digitally controlled actuation scheme, such TFFA-based metasurface could follow a self-evolving route to morph into a wide range of 3D shapes. Inspired by cilia-like structures, many artificial robots have been developed for microfluidics, biomedical research, and so on. [15,289,291,317] For example, through the use of independently controlled polypyrrole (PP)-based TFFAs, Ren et al. [317] proposed an electrically driven soft-robotic ciliated epidermis, enabling fluid manipulation inside tubular structures and enhanced fluid transportation near dynamically bending/expanding surfaces. By employing cilia-like unit cells that could generate nonreciprocal motions, Wang et al. [15] developed a TFFA-based metasurface allowing almost arbitrary control and switch of desired flow patterns (Figure 12b).

Conclusion and Outlook
To create the future devices/systems with countless humanmachine interactions, the rising trend of miniaturization and flexibility in electronics and robotics has always been pushing to its limit. As a core component of these devices/systems, actuators with high flexibility, very tiny body size, fast response, large actuation strain, and driving force are in unprecedentedly high demand. TFFAs, standing as a promising solution, have been systematically reviewed in this article. In particular, this review summarized various types of existing TFFAs, according to their key composing materials, including nanomaterials, LCEs, SMPs, SMAs, hydrogels, biohybrids, and other stimuli-responsive materials. The broad-ranging applications of TFFAs have been discussed, spanning biomedical uses (e.g., surgery, rehabilitation, reconfigurable bioelectronics, drug delivery, and tissue engineering), robots for environment explorations (e.g., grippers and locomotive robots), to haptic interfaces (e.g, touch screen and wearable interfaces), and reconfigurable devices (e.g, reconfigurable surfaces and reconfigurable electronics). Notably, key merits of representative TFFAs, covering their composing materials, driving mechanisms, deformation types, moduli, geometric features, fabrication methods, and actuation performance, are summarized in Table 1, featuring a macroscopic vision of the diverse TFFA family.
The advantages, drawbacks, and potential usages of the current fabrication methods for 2D and 3D TFFAs are summarized and discussed as followings. In terms of 2D TFFAs, planar fabrication techniques including laser cutting, molding, printing, and photolithography are commonly used. The laser cutting shows great material compatibility and high manufacturing speed, but the unavoidable heating effect could significantly reduce its manufacturing resolution especially for some soft materials (e.g., LCEs, hydrogels, and so on). This method would be suitable for fast and relative large-scale fabrications. The soft lithographybased molding can well address the above heating issue to prepare 2D TFFAs at microscales. However, the well-controlled surface chemistry (i.e., possible demolding) and unvaried volume during solidification of molding limit its applicable range of materials. This approach would fit for the fabrication of some polymerbased 2D TFFAs with feature dimensions down to a few tens of microscales. Printing methods, such as 3D or 4D printing, allow direct fabrication of 2D TFFAs with desired geometries, but they suffer from slow fabrication speed and restricted applicable materials. Therefore, this method might fit the direct fabrication of soft polymer-based 2D TFFAs with moderate resolutions. Photolithography shows high manufacturing resolution (up to nanoscale) with a relatively large material library, while it might not be suitable for certain soft materials (e.g., LCEs, silicone). This route would be good at preparing micro-to-nanoscale 2D TFFAs with SMA (e.g., Ni-Ti) and semiconductor material (e.g., silicon).  [72] www.advancedsciencenews.com www.advintellsyst.com In terms of 3D TFFAs, many direct and indirect approaches, including 3D/4D printing, residual-induced rolling/folding, conformal curving, and buckling-guided assembly, are usually employed. 3D/4D printing can directly form 3D TFFAs with complex geometries using some soft materials, while remaining challenges lie in the expanding the accessible range of functional materials (e.g., high-performing semiconductors, 2D materials, and so on). Therefore, such methods are good at direct fabrication of 3D TFFAs with complex geometries at millimeter-tocentimeter scales. Rolling/folding methods are compatible with a variety of functional materials, but their formed 3D configurations (e.g., simple arc shapes or tubular shapes for rolling) are relatively simple. These methods would be good to form 3D TFFAs at small scales (e.g., microscale) using functional thinfilm materials. Conformal curving methods are facile to implement with a good material compatibility. However, the formed structures are directly determined by the surface configuration of the 3D substrate, which leads to their limited structural complexities and relatively large dimensions. As such, conformal curving methods might be suitable for fast preparation of TFFAs with simple curved 3D configurations at large scales. Buckling-guided 3D assembly allows the fabrication of 3D TFFAs with complex geometries and a broad range of applicable materials, from microscale to centimeter-scale. This method could be used for fabrication of 3D TFFAs with heterogeneous structures, multimode deformations, and small dimensions.
Despite these impressive achievements, several grand challenges still stand in the vibrant research field of TFFAs.
There are still some grand challenges standing in the manufacturing of 2D/3D TFFAs using soft materials (e.g., hydrogels and LCEs). In terms of 2D TFFAs, the intrinsic chemistries of soft materials have led to limited synthetic approaches (e.g., polymerization and gelation), which might not be compatible with current high-precision planar microfabrication processes (e.g., cutting, molding, and photolithography). Therefore, the reported sizes of 2D TFFAs based on soft materials could barely reach several microns or smaller. In terms of 3D TFFAs, except the above issues on precision manufacturing, the low rigidity and complex constitutive behaviors of soft materials have set great difficulties in maintaining their as-prepared 3D configurations as well as the accurate prediction of their deformation modes. These would hinder the broad applications of soft materials in 3D TFFAs. Transformative technologies that allow reliable heterogeneous integration of different functional materials (in particular, hard, and soft materials) in high-performance TFFAs are still lacking. Additionally, it remains a challenge to well solve the performance tradeoff between the actuation deformation and the driving force of TFFAs, in view of the thin film configuration.
Future opportunities of TFFAs might lie in the following directions: 1) developing novel synthetic approaches highly compatible with existing planar microfabrication techniques (e.g., lithographic methods and molding) that could take advantage of their high manufacturing resolution to create high-performance TFFAs at microscale and even sub-microscale; 2) broadening the applicable material sources of additive manufacturing approaches (e.g., 3D and 4D printing) and inventing new