A Perspective on Miniature Soft Robotics: Actuation, Fabrication, Control, and Applications

Soft robotics enriches the robotic functionalities by engineering soft materials and electronics toward enhanced compliance, adaptivity, and friendly human machine. This decade has witnessed extraordinary progresses and benefits in scaling down soft robotics to small scale for a wide range of potential and promising applications, including medical and surgical soft robots, wearable and rehabilitation robots, and unconstructed environments exploration. This perspective highlights recent research efforts in miniature soft robotics in a brief and comprehensive way in terms of actuation, powering, designs, fabrication, control, and applications in four sections. Section 2 discusses the key aspects of materials selection and structural designs for small‐scale tethered and untethered actuation and powering, including fluidic actuation, stimuli‐responsive actuation, and soft living biohybrid materials, as well as structural forms from 1D to 3D. Section 3 discusses the advanced manufacturing techniques at small scales for fabricating miniature soft robots, including lithography, mechanical self‐assembly, additive manufacturing, tissue engineering, and other fabrication methods. Section 4 discusses the control systems used in miniature robots, including off‐board/onboard controls and artificial intelligence‐based controls. Section 5 discusses their potential broad applications in healthcare, small‐scale objects manipulating and processing, and environmental monitoring. Finally, outlooks on the challenges and opportunities are discussed.

as cells with soft materials embody powering, sensing, and actuation into soft miniature robots for autonomous motion and manipulation. [10]Combining such embodied materials intelligence and structural intelligence from novel designs of robotic structures [11] could largely reduce the onboard control burden for achieving autonomous and intelligent miniature soft robotics in full functionalities. [12]lthough there are several reviews on small-scale soft robotics, they focus on either soft actuating materials, [9] or bioinspired soft micro-/nanoswimmers that mimic simple single cell-based microorganisms such as bacteria and sperm cells, [5] or diverse locomotion modes, [4] or clinic applications. [6]Miniature soft robotics require comprehensive and multidisciplinary efforts in integrating materials intelligence for powering and actuation, structural intelligence of novel robotic structural designs, smallscale advanced manufacturing, and control systems for achieving diverse robotic functions in confined spaces.In this perspective, we briefly overview state-of-the-art miniature soft robotic systems in a more comprehensive way ranging from materials, designs, actuation, powering, fabrication, to control and applications.We first discuss materials and structural selection by revisiting the materials and structures library often used in conventional large-scale soft robotics, as well as emerging materials such as soft living biomaterials for powering and actuation in Section 2.Then, we will discuss the fabrication methodologies, especially in small-scale manufacturing techniques in Section 3, control systems for onboard/off-board and machine learning-based controls in Section 4, as well as their potential applications in Section 5.In the end, outlooks toward highly integrated and intelligent miniature soft robotics are discussed in Section 6.

Soft Materials and Structural Designs for Small-Scale Actuation and Powering
In this section, we will first revisit the materials library often used in conventional mesoscale soft robotics and reexamine their scalability for potential actuation and powering at small scale.Generally, the soft materials library for soft robotics include two major categories in terms of tethered and untethered powering for actuation: one is mechanical-based actuation under mechanical force and pressure such as the widely used fluidic elastomer actuators (FEAs).Their deformation is driven by pneumatic or hydraulic pressurization of patterned fluidic channels.The other is stimuli-responsive-based actuation in response to one or multiple external stimuli such as heat, light, solvent, magnetic field, and electric fields.The materials library includes electro-active soft materials under electric fields such as dielectric elastomer actuators (DEAs) and piezoelectric actuators (PEAs), thermal-or photo-or photo-thermal responsive materials such as liquid crystal polymers (LCPs), hydrogels, and shape memory alloys (SMAs) or polymers, solvent-responsive materials such as gels and hydrogels, and magneto-responsive elastomers (MREs).Normally, each material is not limited to one stimulus, which could be potentially designed and synthesized through combining two or more stimuli to allow for multistimuli actuation.For example, a hybrid magneto-responsive and LCP-based actuator could be actuated by both magnetic field and temperature or light.In the following, we will discuss and examine their materials scalability and structural forms for actuation and powering at small scales, as well as emerging soft active materials such as soft living biomaterials for small-scale actuation.

Tethered Actuation and Powering
FEAs are often utilized in macroscale soft robots due to their easy and robust fabrication and controls.The benefits of small-scale FEAs reside in lightweight, low energy consumption, and potential handling of small-scale sized objects.Despite the benefits, to implement FEAs at small scales, it needs to address several challenges in terms of fabrication, actuation, and powering.For miniaturized FEAs, the size of fluidic channels dramatically shrinks from millimeter to several or tens of micrometers.The high resolution makes the traditional molding and demolding or 3D printing approaches used in macroscale FEAs become very challenging.Alternative ways include lithography and microfluidic fabrication developed in microfluidics, [13] which make it promising to downscale to small sizes.Meanwhile, it still requires tethered actuation such as small-size air or fluidic tubes to powered micropumps for pressurized deformation.Such tethered tubes and valves could severely interfere and impact their actuation behavior, especially during locomotion due to the relatively larger size and heavy weight of tethered tubes, etc.One solution is to integrate the tubes, valves, and powering systems onboard in an electronics-free manner.The powering system can be an onboard small gas container that supplies constant pressure to actuate the FEA through oscillation response of the connected bistable valves.Another solution to pressurize FEAs without the tethered tubes and pump is to utilize the liquid-to-gas phase transition, as seen in a solar-driven pneumatic soft robot. [14]An energy efficient film was employed to absorb energy from sunlight.When coupled with low-boiling point fluid (34 °C), the photothermal energy can be used to pressurize the pneumatic artificial muscle for dedicated grasping.
[17][18] Nonetheless, considering the development of fabrication technologies in microelectronics, the DEAs may have higher potential than FEAs at small-scale (e.g., submillimeter or micron) soft robots. [19,20]A simplest DEA can be a sheet of elastomers with thin electrodes on both sides.When external voltage is applied, the thicknesses of DEAs are reduced through Maxwell stress, while the elastomers have an in-plane expansion. [15,18]To magnify the deformations, the DEAs are often fabricated into different structures, such as laminated structures and tubular structures.The advantages of DEAs include easy control and fast response (in the range of tens to hundreds Hz [15,19,20] ).However, DEAs usually need very high voltages (e.g., hundreds [15,19] to thousands [17,20] of volts) for actuation provided by high voltage power supplies, which hinders its applications in untethered scenarios, e.g., flying soft microrobot.The spacing between two electrodes is preferred to be small for lower actuation voltage.Figure 1a shows the prototype of a tethered miniature soft aerial robot actuated by cylindrical shell-like DEAs [19] .It employs a multiple layer fabrication process for low-voltage DEAs (500 V) made of elastomer and carbon nanotube (CNT) as electrodes.
PEAs are another type of electro-active actuators which needs high electric voltages.The materials with piezoelectric effect can generate a strain under applied external electric field.PEAs have been widely used in soft robotics due to their superior performances, such as high precision positioning, fast response, large output forces, and easy but stable control, which makes it promising for small scale applications that require high precision.As the generated strains are usually small, PEAs are often made into unimorph and bimorph structures, which contain one or two actuation layers (or active layers) and one passive layer, for larger deformation and better control.Piezoelectric effects have been found in different categories of materials.The most used piezoelectric materials in soft robots include lead zirconium titanate (PZT) [21] and polyvinylidene difluoride (PVDF). [22]Figure 1b shows one example of PEA-based insect-scale fast moving soft robot. [23]It consists of a curved body with a piezoelectric PVDF unimorph structure and a leg.The periodic extension and contraction in the PVDF layer driven by an alternating current (AC) voltage with certain frequency change the shape of the robot for fast locomotion (20 body length/s) near the resonant frequency of the structure.
Despite the promise and benefits of DEAs and PEAs, they still require wired actuation tethered to power supplies because it will be very challenging to integrate the power and controls onboard of miniaturized or submillimeter scale soft robots.Similar to FEAs, the tethered wires will be expected to largely hinder their motion and deformation at small scales.Thus, untethered and remote actuation and controls (e.g., heat or light) can be an alternative way for small-scale actuations as discussed below.

Untethered Actuations and Self-Powering
Untethered actuation can be readily achieved in external stimuliresponsive soft materials.These soft intelligent materials can harvest energy directly from the environmental stimuli such as heat, light, magnetic field, moisture, and chemical solvents for self-powering.By converting thermal, photothermal, electromagnetic, and chemical energy into mechanical energy, they can spontaneously deform their shapes in response to such external stimuli.Thus, such embodied materials intelligence can intrinsically and simultaneously encode onboard sensing, powering, and actuation into their materials, which makes them promising for implementation in untethered and even autonomous and intelligent miniature soft robotics.Their motions can be controlled by external stimuli.For example, light can provide both temporal and spatial resolution for selectively controlling the Figure 1.Tethered and untethered miniature soft robots actuated and powered by various stimuli-responsive materials.a) A tethered microaerial robot actuated by a cylindrical shell-like multilayered DEA.Adapted with permission. [19]Copyright 2022, Wiley.b) A tethered insect-scale fast moving soft robot actuated by a PEA-based curved body.Reproduced with permission. [23]Copyright 2019, AAAS.c) A light-powered untethered microswimmer actuated by an electrospun liquid crystal elastomer (LCE) microfiber.Reproduced with permission. [27]Copyright 2021, AAAS.d) An untethered steering photoresponsive U-shaped hydrogel mirocrawler powered by green light.Adapted with permission. [32]Copyright 2021, Mary Ann Liebert, Inc. e) An untethered submillimeter-scale crab-like terrestrial robot actuated by SMA-SiO 2 shell composite as reversible dynamic joint actuators under laser heating.Reproduced with permission. [33]Copyright 2022, AAAS.f ) An untethered ferromagnetic ribbon-based multimodal soft robot navigating a synthetic stomach phantom actuated by external magnetic fields.Reproduced with permission. [36]Copyright 2018, Springer Nature.g) An untethered ferromagnetic wire-based steering soft robot navigating a complex vasculature with an aneurysm.Reproduced with permission. [37]Copyright 2019, AAAS.h) A light-powered untethered millimeter-scale soft robotic ray actuated by biohybrid heart muscle actuators.Reproduced with permission. [41]opyright 2016, AAAS.i) A light-powered untethered centimeter-scale soft walking robot actuated by biohybrid optogenetic skeletal muscle tissues.Adapted with permission. [42]Copyright 2023, AAAS.
actuation and deformation in different parts of a small-scale soft robot.
LCPs are a category of stimuli-responsive materials with twoway (reversible) shape memory effect.When exposed to heat or light, the materials will change from nematic state (N) to the isotropic state (I) due to the orientation change of the liquid crystal director field and shape morphing occurs.The morphing shapes can be predefined with mesogen alignment by mechanical deformation [24] or molds. [25]LCPs can take different 1D, 2D, and 3D structures at small scales. [25,26]The simplest structure is an electrospun LCE microfiber [27] (Figure 1c).Benefiting from its microscale fiber diameter, it can act as an artificial muscle with fast response and high power density powered by environmental heating or near-infrared (NIR) light.Such microscale artificial muscles can be further integrated with different robotic structures to design various miniature soft robots such as the lightpowered untethered microswimmer shown in Figure 1c.Other simple LCP structures such as rods, rings, helix, and twisted ribbons can be fabricated through soft molding methods alongside mechanical ways of stretching, bending, and twisting to align the mesogens. [12,28]More complex 2D or 3D shapes can be fabricated through additive manufacturing such as direct ink writing (DIW) printing, where the mesogens can be aligned by the shearing during extrusion from the nozzles. [29]ydrogels are another kind of active materials that can respond to different external stimuli, such as temperature, pH, light, and ionic strength. [30]Hydrogels can expand or shrink in volume isotopically when absorbing or losing water in the networks, making it more beneficial for aquatic locomotion.Like LCPs, the actuation process is also reversible.The actuation time can be as long as minutes or even hours because of the slow diffusion processes.The actuators can be fabricated at a small scale because the actuation is size independent. [31]Figure 1d demonstrates a photoresponsive U-shaped hydrogel mirocrawler powered by green light. [32]Aiming the laser beam at either arm can steer the microcrawler for left or right turning by contracting or expanding the irradiated area.Au nanoparticles are embedded for enhancing the green light absorption for raising the localized temperature that triggers the shrinkage in thermoresponsive hydrogels.Its speed can be tuned by the frequency of irradiation.However, limit actuation bandwidth and slow response to external stimuli may limit its usage toward practical applications.For example, the U-shaped microcrawler in Figure 1d shows a slow moving speed of 18 μm s À1 (about 0.1 body length/s). [32]MAs or shape memory polymers (SMPs) are another type of thermo-responsive active materials that can recover to the preprogramed shapes by phase transition.The advantages of SMAs or SMPs include high power density and various actuation motions.However, they are difficult to be manufactured into complex shapes in very small scales.Moreover, most SMAs or SMPs are limited to one-way shape memory effect.Therefore, composites with SMA wires or sheets embedded are more often utilized for small scale actuators. [33,34]Figure 1e shows a submillimeter-scale crab-like terrestrial robot using SMA-SiO 2 shell composite as dynamic joint actuators under laser heating. [33]To overcome the challenge in small-scale fabrication of complex SMA architecture, a mechanical self-assembly method is employed to create the small-scale 3D crab-like architecture through compressive buckling of a prescribed 2D SMA pattern under prestrain release.Upon laser heating to the phase transition temperature, the SMA drives the flattening of the 3D architecture.To address the one-way shape memory limitation in SMA, a thin layer of SiO 2 shell is deposited on either side of SMA to provide the elastic restoring force during cooling.Thus, such a double layer composite structure allows for a reversible actuation using a one-way shape memory material.The power consumption and extreme working temperature of SMA need to be considered in small-scale soft robotics applications, especially those involving biological tissues.
MREs are a type of active materials that respond to magnetic field. [35]The MREs usually contain magnetic responsive particles (e.g., NdFeB, iron oxide, etc.) and elastomer matrix (e.g., silicone, polyurethane, etc.).Compared to other active materials and actuators, MREs can be easily made into different shapes and programmably magnetized in very small scale, such as ribbons [36] (Figure 1f ), wires [37] (Figure 1g), and even complex 3D shapes. [37,38]The fast-responsive deformation and shape changing of MRE-based actuators such as bending and rotation can be remotely controlled by changing the magnitudes or directions of external electromagnetic fields (Figure 1f ).Such shape-changing and steering mechanisms enable multimodal locomotion of jumping, swimming, rolling, and climbing in unstructured aqueous environments, e.g., inside human body (Figure 1f ).By coating the steering MRE wire with a lubricating hydrogel skin, it can largely reduce the friction of a submillimeter-scale soft continuum robot during navigation [37] (Figure 1g).Its active omnidirectional steering and navigation capabilities enable navigation through complex and constrained environments, e.g., hard-to-access complex vasculature with an aneurysm [37] (Figure 1g).Apart from directly deforming MREs by magnetic fields, varying magnetic actuation can induce liquid-to-gas phase transition of a liquid through inductively heating dissipated ferromagnetic particles by a portable induction heater.The phase transition from liquid-to-gas expands its volume by a factor of 1600 and develops sufficient pressure for pneumatic artificial muscle without a pump. [39]The remotely controlled, small-scale MRE-based steering and navigation soft robots hold great promise in medical applications.To approach the versatile application in biomedical engineering, the limits of the working space and precise multidegrees for control need to be addressed in the future.

Soft Living Materials
Soft living materials are hybrid biomaterials that consist of living cells or organisms and artificial soft materials.Comparing to the man-made materials, living biomaterials have the natural advantages for replicating life-like movements and locomotions to develop the new-generation miniature soft robots.To date, soft actuators made of biomaterials vary from bacteria, muscle cells and tissues, to insect tissues. [10]Biomaterials are usually combined with other materials (e.g., polydimethylsiloxane (PDMS), CNTs, etc. [40] ) to form biohybrid actuators for better functioning.Such emerging biohybrid soft robots in recent years have shown large output forces, fast locomotion speeds, and large travel distances.They have also demonstrated the capabilities of onboard adaptive actuation, sensing, and control for multitask manipulation and locomotion at small scales.
Figure 1h,i shows two examples of muscle-driven biohybrid miniature soft swimming and walking robots guided by light, respectively.The bioinspired millimeter-scale swimming robot mimics the undulating movement of a ray fish for propelling. [41]t is composed of a microfabricated gold skeleton and soft elastomer body powered by patterned rat heart muscle cells (Figure 1h).Inspired by the muscle-tendon-bone architecture, the centimeter-scale walking robot is composed of 3D-engineered skeletal muscle tissue as biological actuators forming around an asymmetric hydrogel scaffold, and an onboard battery-free wireless optogenetic device [42] (Figure 1i).Stimulation of the optogenetic muscles using light leads to cyclic contractions in the muscles, which deform the bonded artificial materials and structures to propel both biohybrid robots.Localized light stimulation at different body parts of the robots leads to maneuverable motions in a remotely controllable manner.Despite the challenges such as short lifetime of the living cells, soft living biomaterial-based actuators are still very promising for small-scale soft robots. [10]

Structural Designs
Due to the constraints in small-scale manufacturing, the structural complexity is rather limited in miniature soft robots. [33]As seen in the showcase examples actuated by different stimuliresponsive materials in Figure 1, most miniature soft robots take simple geometrical and structural forms in 1D, 2D, and 3D. [4,5]D forms include filaments, fibers, rods, and wires with millimeter or submillimeter diameters with or without coatings of other functional thin shells (e.g., Figure 1c,g).2D forms include different shaped ribbons, rings, arcs, curved thin plates, lattices, and multilayer stacks of planar structures (e.g., Figure 1b,d,f,h,i).3D forms include cylinders, spheres, domes, helical or twisted ribbons or filaments, and origami structures via active folding [43] (e.g., Figure 1a).More complex and functional 3D shapes and architectures can be created through advanced small-scale manufacturing, buckling-driven 2D-to-3D mechanical selfassembly (e.g., Figure 1e), 4D printing, [44] and pop-up kirigami approach via cutting. [45]o enable reversible deformation and motions, the structure forms are integrated with soft intelligent materials or soft composites for stimuli-responsive actuation.The deformation that drives the terrestrial or aqueous motions or manipulation often includes reversible contraction/expansion, bilayer bending, wave-like undulation, and twisting through the control of applied external stimuli.Among them, reversible contraction and expansion is the most fundamental deformation mode often observed in artificial and biological muscles.It can lead to other bending, undulation, and twisting deformation modes through rationally designed layered or inhomogeneous actuated structures for different terrestrial, aqueous, and aerial motion, and manipulation.
For terrestrial motion, it is mainly propelled by asymmetric frictions generated at the body-substrate or leg-substrate interface in 2D or 3D structural forms. [46,47]The asymmetric frictions are generated and driven by actuated bending or contraction deformation in either the soft body or the integrated active or passive legs or combined under tethered electricity or remote light and magnetic field (e.g., crawling, Figure 1b,d-f,i).Rolling motion is often employed in 1D structural forms due to their round cross sections driven by contraction or bending-induced torque under steady-state heat sources (e.g., hot surfaces) [12] or by directional light. [48]For aqueous motion, it often employs flapping or oscillation motions of soft body or wings or fins for propulsion. [49]The miniature soft robots often take 2D or 3D structural forms at millimeter and submillimeter scales under tethered electricity or remote light and magnetic field (e.g., Figure 1c,h).At the micro-or even nanoscale, aqueous motion in small-scale soft robots is achieved by oscillatory bending or rotating tails in the structural forms of wavy, helix, or twisted ribbons for propulsion actuated by a magnetic rotating field.Their propulsion mimics the swimming of sperm cells and bacteria.Similar flapping wing motion is often used in miniature or microaerial robots powered by tethered electricity (e.g., Figure 1a).For manipulation and cargo delivery, it requires both manipulation and navigation functionalities.Open or closedform origami structures are often used to grasp and encapsulate cargos or drugs inside or onboard for navigation and delivery powered by remote magnetic field. [50]

Manufacturing and Fabrication Methodologies
Besides addressing some grand challenges through structure design and material selection for miniature soft robotics, the processing and fabrication methods are crucial to scale down the dimension of soft machine.Miniature soft robotics has potential to complementarily promote the blooming of advanced manufacturing at small scale.On the one hand, maturing of the micro-/nanoengineering makes it practically integrate logical circuits, sensors, and actuators; thus, it enables the potential downscaling of soft robotics.On the other hand, embodying miniature soft robotics with advanced functionalities such as adaptive shape morphing ability and compatible interaction with human and environment can benefit many aspects, including ultraprecise manufacturing for dedicate manipulation, better understanding of the design principle, and working mechanism of biological systems.Envision of different advanced fabrication techniques for more complex architectures utilized for soft or compliant miniature systems will be discussed in this section.

Lithography
Miniature soft robotics can benefit from the development of semiconductor industry.Mature manufacturing techniques include deposition, photoresist, lithography, etching, and ionization with ultraaccuracy.The techniques enable dexterity of shape morphing ability conjugating to compliant and continuous mechanism and integration of the robotics system.As shown in Figure 2a, Miskin et al. [51] fabricated a graphene-based biomorphs utilizing photolithography techniques (atomic layer deposition and chemical vapor deposition) and assembled to an origami cell-sized machine in response to light or pH.Recently, Reynolds et al. [52] developed microscopic robots (100-250 μm in size) with onboard control benefited from semiconductor industry.The robots were controlled by an application-specific IC (ASIC), which consists of %1000 transistors in 100 μm and can achieve complex functions, including quadruped locomotion, microsurgery, and sense and respond to their environments with high level of computational intelligence.
Soft lithography is a collection of techniques for prototyping and replicating the device at micro-or nanoscale based on stamping, molding, and embossing with the elastomeric materials. [53]oft lithography can be applied to a variety of materials to generate promising and controllable 3D patterns and stacked devices that have been widely applied to surface chemistry, microelectromechanical systems (MEMS), and optical devices.Due to simplicity and cost-effectiveness, soft lithography was introduced to generate the microfluidic logic to regulate fluid flow for onboard control of artificial tentacles of the octobot. [54]Coupling with molding and 3D printing, the octobot can be integrated with chemical reservoir for untethered, solely soft and autonomous robotics function in centimeter scale as shown in Figure 2b.Moreover, researchers can establish multilayer soft lithography, precise interfacial bonding, The bimorph bends when the glass is strained relative to the graphene.The structure was patterned in thick pads of photoresist.The device can then fold and unfold in response to environmental changes.Each layer must be of comparable rigidity for the device to bend efficiently.The glass layers are fabricated to 2 nm thicknesses using atomic-layer deposition.Reproduced with permission. [51]Copyright 2018, PNAS.b) Schematics of fabrication of an autonomous soft octobot.A microfluidic soft controller is prefabricated and loaded into a mold.Matrix materials are poured into the mold and fugitive and catalytic inks are EMB3D printed.Adapted with permission. [54]Copyright 2016, Springer Nature.c) Schematic illustrations of the self-built DLP-based multimaterial 3D printing system that allows rapid material exchange between puddles and cleaning of printed parts with air jet.Reproduced with permission. [57]Copyright 2019, Wiley.d) Schematics of mechanical self-assembly fabrication of a submillimeter-scale crab-like terrestrial robot through compressive buckling.Reproduced with permission. [33]Copyright 2022, AAAS.e) Design of the milliDelta is based on origami-inspired engineering and made using PC-MEMS manufacturing techniques.Reproduced with permission. [62]Copyright 2018, AAAS.f ) Tissue-engineered ray with four layers of body architectures.Reproduced with permission. [41]Copyright 2016, AAAS.and easy folding for soft microactuators. [55]Precise stacking and bonding of 12 individual layers tolerate a complex shapes and 3D microfluidic circuit to independently control the bending of each leg.As such, soft lithography techniques can simplify the design of the computational circuit replaced by microfluidic channel, which leave the space for the integration of sensing and feedback control system to allow the miniature soft robotics exploring in the unconstructed environment.

Additive Manufacturing
Additive manufacturing or 3D printing utilizes the computer-aiddesign (CAD) to guide deposition of material and construction of a 3D object.Additive manufacturing was introduced to fabricate the precise and sophisticated molds for fluid-driven soft actuators. [47]Such a rapid mold fabrication method [56] leaves an open space to integrate power sources, actuators, control board, and sensing systems due to its feasibility in segmentation and assembly by bonding agent.
Recent advances in 3D printing provide a powerful platform for directly printing out the miniature soft robotics, benefiting from exploration of 3D printability of versatile materials such as nanomaterials, stimuli-responsive materials, and even biomaterials, as well as high printing resolution and accuracy.As shown in Figure 2c, Zhang et al. [57] developed a customized digital light processing (DLP) 3D printing platform for multimaterial systems.The submillimeter-scale pneumatic actuators can be printed to bestow different deformation modes for camouflage and manipulation in narrow spaces.Moreover, Cui et al. [58] reported the a 3D printed miniature robotic metastructure closely integrated with sensing, actuation, and control.The charge of the additive manufacturing machine can be programmed for the complex microarchitecture of multipiezo-active and conducting materials.Such a proprioceptive microrobot can navigate in obstacle courses integrated with feedback controller and selfsensing elements.

Mechanical Self-Assembly
To address the limitations in additive manufacturing, an alternative solution for creating sophisticated 3D small-scale structures is the mechanical self-assembly pop-up approach [59] (Figure 2d), which is inherently quick, efficient, and straightforward by engineering strain and buckling instability of multilayer systems for complex architectures at small scale.
To facilitate the fabrication methods for complex 3D structures and overcome planar limitation inherent to MEMS [60] , researchers have developed pop-up book-like MEMS methods to self-assemble the highly articulated 3D compliant structures.This technique is inspired by the printed circuit board (PCB) fabrication [60] for miniature invasive surgical manipulator, [61] microaerial vehicle powered by soft artificial muscles, [17] and high-precision and high-speed miniature Delta robot as shown in Figure 2e. [62]ne example is the aforementioned submillimeter-scale walking robots shown in Figure 1e. Figure 2d shows its corresponding schematic fabrication process. [33]In principle, the 2D precursors incorporate programmable bonding sites attached to prestretched silicone elastomer.Releasing the engineered prestrain will compress the 2D precursors to bend or fold to different 3D pop-up architected configurations at small scales.Conceivably, coupling stimuli-responsive materials such as LCE and hydrogels, more complex robotic architectures can be fabricated and actuated by external stimuli such as light, temperature, pH, and humidity.
An important feature of the mechanical self-assembly manufacturing technique is that it can fabricate a large number of samples simultaneously operating in a parallel mode; [33] thus, it holds great potential for mass production.However, challenges also remain to create free-standing small-scale architectures, [63] especially for structures made of soft materials that could not provide sufficient structural stiffness for support.Multilayered designs that combining soft active materials and layered stiff polymers (e.g., polyimide) can be used to enhance the structural stiffness.To generate freestanding architectures, additional thin layer of stiff coatings on the fabricated architectures can be deposited to fix the 3D shapes. [33]

Tissue Engineering
Tissue engineering can be referred to integrating biology with engineering for reproducing cells or cellular products to maintain and improve biological tissue functions.Cells are often utilized as building blocks to "seed" into some biocompatible scaffolds to realize tissue formation and reconstructions.Extensive advances in genetic and tissue engineering enable the construction of bio-or biohybrid soft robots at small scale. [64]esearchers developed a tissue-engineered biorobot inspired by the batoid fish [41] as shown in Figure 1h and 2f.The biorobot can be triggered by light stimuli to control its locomotion through combining soft materials and tissue engineering.The muscle layer of tissue-engineered ray was patterned with living rat cardiac cells, which were genetically engineered to contract upon light stimuli.The locomotion speed and direction can be manipulated through different frequency of light pulses.Assembled with bioelectronic and flexible control system, the onboard control of muscle driven miniature biohybrid robots can be achieved as well. [42]

Other Fabrication Techniques
In addition to the aforementioned fabrication methodologies, other straightforward methods are developed coupling with novel structure designs and stimuli-responsive materials to scale down the size of soft robots.Wang et al. [65] developed a large arrays of hybrid magnetic micropillars directly fabricated utilizing traditional molding replica methodology.The micro arrays can achieve multiple soft robotics manipulation of small-scale objects by magnetic field.Such precise molding replication methods can be developed to fabricate core-shell magnetic micropillars [66] for reprogrammable magnetic actuation.Hu et al. [36] programmed MREs in submillimeter scale for multimodal locomotion through simple laser cutting.Wu et al. presented insect-inspired soft robots made of a curved unimorph to achieve a record high speed and dexterity enabled by thin film engineering. [23]Yi et al. [67] proposed a facile fabrication strategy to transform 2D magnetic thin sheets to 3D shapes based on automated roll-toroll processing, which allows the high-throughput fabrication for the miniature soft machines.Sun et al. [68] harnessed the wettability and zero stiffness of ferrofluids for high reconfigurability to realize multimodal motions and dedicated manipulation of small droplets and boosts drug delivery in future.

Control Systems for Miniature Soft Actuators and Soft Robots
Recent studies have uncovered various control systems for miniature soft actuators/robots based on onboard and off-board ways. [2]Appropriate selection of control systems relies either on the unique material and structural design features (i.e., physical intelligence [11] ) of soft robots or on the implementation requirements. [69]Different from the biological neuron-based computational intelligence, physical intelligence [11] utilizes both mechanical intelligence of structures and materials intelligence of smart materials to reduce the control burden via self-sensing and self-actuation capabilities without human-like brain.
Onboard control system generally integrates mechanical/electrical circuits alongside actuation systems onto/within the body of soft actuators/robots to wirelessly perform unique functionalities or dynamics.Off-board control system usually uncouples the physical components from the main body of soft robots in either tethered or untethered ways, e.g., the tethered electricaldriven (Figure 1a,b) or pressure-driven (Figure 2c) soft actuators, and untethered soft actuators by remote external stimuli of light and magnetic field, etc. (Figure 1c-1i).Benefitting from the development of the aforementioned fabrication techniques, miniatures soft robots can achieve rather complicated functionalities such as dexterous object manipulations and intelligent locomotion with autonomous external environmental information sensing, detection, interaction, and decision-making capabilities.In the following, we briefly discuss some representative onboard and off-board control systems of miniature soft robots/actuators with unique physical intelligence characteristics.

Tethered Control Systems
The majority of off-board control systems need tethered wires (electrically conductive and pneumatic/fluidic tubes).Figure 3a(i) displays the details of pneumatic control system [70] for the miniature soft robots with air chambers (Figure 3a(iii) [57] ).With unique air channel patterns [57] and/or extra structural components, the end effectors can deform to perform grasping, catching, and holding functionalities by means of quasistatic deformation, such as bending (Figure 3a(iii)), elongation, twisting, or combined deformation.They can also be dynamically actuated for achieving periodic locomotion, such as walking, running, and jumping. [1,2]As shown in Figure 3a(ii), given the success of wireless pneumatic control system integrated with signal processing MEMS control panel and smaller sized valve controller at macroscales (centimeter to meter), future miniature pneumatic soft robots are also expected to be designed with wireless control systems.
Another representative off-board control system is based on electric energy source; the control system and the end effectors relate to conductive wires.Utilizing material intelligence, versatile miniature soft robots with electrical control systems have been invented, including the electromechanical, electrothermal, electromagnetic, and electrochemical materials. [4]Electricbased miniature structures can be constructed with both small sizes and complicated control architectures to achieve highly complex quasistatistic and dynamic functionalities/locomotion.As shown in Figure 3b(ii), by mimicking the flying mechanism of insect wings with flexible joints, Ma et al. [71] constructed a dielectric material actuated microflyer with multilayered control systems (Figure 3b(i)) to efficiently tune the flying directions, magnitudes, and speeds.Similarly, the end effector in the created miniature Delta robot (milliDelta) shown in Figure 2e can be digitally programmed and controlled to have parallel order paths, [62] enabling dexterous manipulations such as complex trajectories tracking and stable object holding.

Untethered Control Systems
The untethered off-board control system depends highly on the physical intelligence of the miniature soft robots, including special structural topologies and the intrinsic responses of the smart materials. [11]For example, as shown in Figure 3c, Ren et al. [72] reported an autonomous microswimmer with opened hollow channel that can trap air bubbles.Its off-board control system includes acoustic transducer and permanent magnet.Simultaneously steered by the acoustic transducer and tuned by magnetic field, this microswimmer can maneuver to manipulate (transport and collect) single particles such as cells, virus, and bacteria in confined environments.Another example of untethered off-board system is based on its intrinsic materials property.As shown in Figure 3d, by merging nanoscale magnetic particle into liquids, Fan et al. [73] reported an ultraflexible ferro-liquid miniature soft robots that can pass through in-plane narrow channels.The control systems of this type of magnetic robot are intrinsically determined by the specific manipulations and/or arrangements of external permanent/ electromagnets.With the help of four symmetrically placed electromagnets (Figure 3d, left), the rolling and squeezing deformations of the ferro-liquid soft robot can be uniquely programmed by a posture and distance tunable ball-shaped permanent magnet.

Onboard Control Systems
Onboard control systems can integrate both control panel and even the power source onto the main body of soft robots.Onboard controlled miniature soft robots are expected to equip with capabilities of performing complex functions, processing information, and responding to and communicating with external environments.As shown in Figure 4a, Reynolds et al. [52] created an autonomous miniature soft robot with onboard MEMS control system with specific circuits (see Figure 4b).Silicon PVs (photovoltaics materials) are also combined into the soft body to supply power for the control system and the locomotion structural components (legs for crawling) after exposure to light.The MEMS control system includes about 10 3 transistors, diodes, resistors, and capacitors to engender a series of phasechangeable wave orders to operate the locomotion gaits and speeds of the robots.The PVs based light-to-electric energy system can be a very promising strategy for onboard powering systems in miniature soft robots because other methods such as mechanical circuits based electronic-free control/actuation strategy [74] or chemomechanical phase changes [75] are almost impossible to downscale the sizes, as well as to integrate with the robot body at microscales to achieve complex functionalities.

Artificial Intelligence
Truthfully, next-generation miniature soft robots should be with onboard control systems, and thus equip with information processing capability to autonomously tackle with and adapt to various application scenarios and environments without human interventions.However, unlike living animals with biological self-sensing, nervous information processing, and self-decisionmaking capability, miniature soft robots could perform autonomous behaviors with the help of artificial intelligence.In the last decade, machine learning has been intensively studied and utilized in soft robot designs. [76,77]avored by machine learning with prestored or ongoing corrected dataset, soft robot can be facilitated to achieve high-level planning, control, state estimation capabilities, and address the hysteresis and nonstationary issues.For the control system, as shown in Figure 4c, there are mainly three common routes to intelligently control robotic behaviors by collecting data through a) learning the inverse kinematics/statics of the end effectors, b) learning the forward dynamics to predict informative control, and c) direct learning from human controllers. [76]Applications of the data-driven machining learning technique have benefitted traditional soft robots with promising results to conduct precise manipulations and avoid any dynamically induced failures.Given the generality, machine learning techniques are expected to benefit the future designs of miniature soft robots with more intelligent control systems for complex tasks.Reproduced with permission. [57,70]Copyright 2019 and 2018, Wiley and AAAS.b) Dielectric material based.Reproduced with permission. [62,71]Copyright 2018 and 2013, AAAS.c) Acoustic and magnetic based.Reproduced with permission. [72]Copyright 2019, AAAS.d) Magnetic based.Reproduced with permission. [73]Copyright 2022, AAAS.

Applications
5.1.Applications in Healthcare, Object Processing, and Environmental Monitoring Practically, soft miniature robots have engaged in applications in real life, including healthcare, small object processing and manipulation, and environmental monitoring.One of the most promising applications of soft miniature robots is in the healthcare industry.These robots have the potential to revolutionize medical procedures by enabling minimally invasive surgeries [78] (Figure 5a(i)), targeted drug delivery [79] (Figure 5a(ii)), and precise medical imaging [80] (Figure 5a(iii)).Among different external stimuli actuations, magnetic actuated ferromagnetic miniature soft robots hold great promise in medical applications due to their superior steering and navigation capabilities in a remotely controllable manner, especially in complex unstructured and confined in vivo environments.Soft miniature robots can be used in microsurgery, allowing for precise manipulation of small structures and tissues within the body, e.g., ultragentle grasping of cells and tissues by self-folding origami grippers made of biocompatible elastomers and thermo-responsive polymeric materials. [81]As a critical aspect of most microsurgery, conventional suturing requires physical manipulation of suturing devices and tools.Li et al. [78] reported wireless small-scale suture and clamp devices based on coiled ferromagnetic artificial muscles actuated by radio frequency (RF)-magnetic heating (Figure 5a(i)).To overcome the small-force limitation in miniature soft actuators, they utilized snap-through structural instabilities to generate large force to close the wound.Soft miniature ferromagnetic robots can also be used to deliver drugs to specific areas of the body, reducing the risk of aftereffects and improving the efficacy of the treatment.The drugs can be carried on top of the planar multilegged miniature robots [79] (Figure 5a(ii)) or loaded inside a 3D container such as a millimeter-scale ferromagnetic Kresing origami robot. [50]The structure-enabled multimodal morphing capability makes it specialized across various tasks, including moving in different terrains and cargo delivery with the expanded biomedical functionality. [50]The multimodal morphologies can further be coupled with multifunctional modules, such as circuit, temperature sensing, UV sensing, and pH sensing. [82]Soft miniature robots can also find potential Optically powered onboard control system.Reproduced with permission. [4]Copyright 2021, Wiley.c) Data-driven machine learning for control system.Reproduced with permission. [76]Copyright 2020, Wiley.
applications in endoscopy, allowing physicians to visualize the inside of the body without invasive surgeries (Figure 5a(iii)).
Another practical application of soft miniature robots is smallscale miscellaneous objects handling as shown in Figure 5b.Miniature soft robots are ideal for handling delicate and complex small-scale objects, such as those used in electronics manufacturing or in the food industry. [14]Soft robots can be designed for gentle and precise manipulation, allowing them to handle fragile materials without causing damage. [83]Miniature soft robots can also be used in precision assembly, enabling the precise placement of small components.
Swarm soft miniature robots can also be used in environmental monitoring, where they are particularly useful for monitoring water quality and air pollution as shown in Figure 5c(i). [84]These miniature robot swarms can be designed to be highly sensitive to environmental parameters such as temperature, pH, and dissolved oxygen, allowing them to provide real-time data on environmental conditions (Figure 5c(ii)). [85]They can also be utilized in exploration of hazardous environments, such as oil pipelines or nuclear reactors, where they can be designed to withstand extreme temperatures and pressures.

Integrating Artificial Intelligent with Soft Robotics
The unique compliant nature of soft robotics results in challenges related to nonlinear behavior and hysteresis, which stem from the inherent viscos or hyper elasticity and infinite degree of freedom.By leveraging machine learning techniques, researchers have been able to address the challenges associated with the design, control, and interaction of soft robots, thereby expanding their range of applications and potential impact.For instance, Du et al. [86] developed a simulation model for Starfish soft robot and implemented it in a differentiable simulator.By introducing gradients from a differentiable simulator, they can narrow down the simulation-to-reality gap and improve its open-loop control.In addition, Sundaram et al. [87] developed a soft robotics glove with Reproduced with permission. [78]Copyright 2022, AAAS.ii) Demonstration of drug transport in a stomach model under wet environment.Reproduced with permission. [79]Copyright 2018, Springer Nature.iii) Conceptual drawing of the proposed soft robotic system for intratympanic injections within a patient's ear canal.Reproduced with permission. [80]Copyright 2021, IEEE Xplore.b) Miniature soft robots for delicate manipulations.i) A light-driven soft robotic gripper grasping on a 60 g boiled egg excited with direct sunlight.Reproduced with permission. [14]Copyright 2021, Wiley.ii) Demonstration of the hybrid micropillars applied as a microscale pick-transfer-place (P-T-P) device (magnetic microtweezers).Reproduced with permission. [65]Copyright 2020, Wiley.iii) Grasping a human hair by a kirigami soft gripper.Reproduced with permission. [83]Copyright 2022, Springer Nature.c) Miniature soft robots in environmental monitoring.i) Color change of domains containing porous silicone substrates after traveling through water polluted with colored oil for DraBot.Reproduced with permission. [84]Copyright 2021, Wiley.ii) The schematic diagram of MFMLS robot with functional modules (circuit, positioning, oil detection, temperature sensing, UV sensing, and pH sensing) and their response under various stimuli.N, North; S, South.Reproduced with permission. [85]Copyright 2022, AAAS.
uniformly distributed sensors; they utilized deep conventional neural networks to identify the grasping objects and estimate their weight.Furthermore, inspired by the human perceptive system, Thuruthel et al. [88] embedded a redundant and unconstructed architecture of sensors into a soft actuator.They modeled the kinematics of such actuator in a nonlinear manner and applied a general machine learning approach to estimate the reaction force when the soft actuator has interaction with objects to achieve a synthetic analog.Besides, we envision that machine learning can help identify and design with the desired properties, such as flexibility, strength, and durability for the soft robotics materials.Integrated with advanced fabrication methods such as 4D printing, machine learning will improve the robustness and efficiency with optimized fabrication parameters.

Outlooks
Besides the potentials of the soft active materials, there are several challenges needed to be addressed for small-scale soft robots.Specifically, the electric actuators usually have the advantages such as higher actuation accuracy, fast response, and large output force; however, one of the major drawbacks that may limit the application is the extremely large actuation voltages.The photo-responsive active materials can be generally easier to be synthesized and fabricated; however, they can be only used in open space without light being blocked.In terms of the MREs, the actuator and materials themselves are easy to be fabricated; nonetheless, they usually need large actuation magnetic field and can only be actuated in a very small space due to the decay of the field.In addition, the recyclability and sustainability of miniature soft robots are crucial to minimizing their environmental impact.The assembly strategy can be employed to modularize and recycle the magnetic soft microrobots. [89]However, there are also challenges that need to be addressed, including the development of suitable recycling processes and standardized protocols through material design toward environmentally sustainable and economically viable.To date, the small-scale soft robots are mainly prepared and controlled in laboratory conditions, and it is usually difficult to integrate small-scale sensors, batteries, CPUs, and even other robotic parts on the small-scale robots, which limits the control difficulty and functionality of the robots.
Despite numerous efforts to shrink down the size of soft robotics, several practical concerns in respect of fabrication methods need to be addressed toward the comparable rigid robotics functions for potential clinic and medicine applications. [6]The resolution of additive manufacturing and soft lithography platform needs to be leveled from submillimeter scale to microscale and multiple printable materials need to be explored for integration with soft robotics system.More importantly, mass production of miniature soft robotics is highly demanded considering their small size scale for creating swarm miniature soft robotics.Such swarm miniature soft robots could find great potential applications in healthcare and environmental monitoring.Multidisciplinary efforts including microbiology, structural genetics, ethology, and nanomaterials engineering are expected to synchronize for low-cost and mass-productive miniature swarm robotics for high physical intelligence and humanmachine intelligence.
Current designs of control system for soft robot are extremely undeveloped compared with conventional mescoscale robots, especially in terms of autonomous information processing and decision-making capabilities.Moreover, constrained by minimal structural size, commonly used actuation equipment and devices for macroscale soft robots are generally hard to directly shrink down to the micro/nanoscale.However, existing researches about miniature soft robots also give some indications for future designs of the control and actuation systems.For example, novel physical intelligences via constructing special structural forms and fabricating with active materials (such as electric, magnetic, thermal, chemical, optical, acoustic, and their combinations based actuated materials) can be uniquely used or integrated to design highly autonomous robots.Another strategy will be the introduction of data-driven machine learning with miniature soft robots to conduct multiple and complex functionalities.Most importantly, future miniature soft robots designs should also borrow inspirations from living bacteria, virus, etc. and construct totally soft body through living cells while integrating with nervous cell-based control system under the human control.
Overall, the physical intelligence enabled by energetically programmed morphologies advances soft robots' capability of operation, especially under complex and unstructured environments, where external control is unnecessary. [11,12]Considering that the physical intelligence is at a preliminary and simplistic level, we can emulate the autonomous operation in biological organisms to encode the intelligence in the human-made soft machines.We envision that the functional materials coupled with the morphological and energetic program facilitate this emulation, which could yield a host of applications under real-world unstructured environments in biomedical engineering, conservation biology, and soft robotics.

Figure 2 .
Figure 2. Representative fabrication methods for miniature soft robotics.a)The bimorph bends when the glass is strained relative to the graphene.The structure was patterned in thick pads of photoresist.The device can then fold and unfold in response to environmental changes.Each layer must be of comparable rigidity for the device to bend efficiently.The glass layers are fabricated to 2 nm thicknesses using atomic-layer deposition.Reproduced with permission.[51]Copyright 2018, PNAS.b) Schematics of fabrication of an autonomous soft octobot.A microfluidic soft controller is prefabricated and loaded into a mold.Matrix materials are poured into the mold and fugitive and catalytic inks are EMB3D printed.Adapted with permission.[54]Copyright 2016, Springer Nature.c) Schematic illustrations of the self-built DLP-based multimaterial 3D printing system that allows rapid material exchange between puddles and cleaning of printed parts with air jet.Reproduced with permission.[57]Copyright 2019, Wiley.d) Schematics of mechanical self-assembly fabrication of a submillimeter-scale crab-like terrestrial robot through compressive buckling.Reproduced with permission.[33]Copyright 2022, AAAS.e) Design of the milliDelta is based on origami-inspired engineering and made using PC-MEMS manufacturing techniques.Reproduced with permission.[62]Copyright 2018, AAAS.f ) Tissue-engineered ray with four layers of body architectures.Reproduced with permission.[41]Copyright 2016, AAAS.

Figure 4 .
Figure 4. Representative onboard control systems for miniature soft actuators/robots.a,b)Optically powered onboard control system.Reproduced with permission.[4]Copyright 2021, Wiley.c) Data-driven machine learning for control system.Reproduced with permission.[76]Copyright 2020, Wiley.

Figure 5 .
Figure 5. Practical applications miniature soft robots.a) Miniature soft robots in healthcare industry.i) A wireless bistable clamper demonstration.Reproduced with permission.[78]Copyright 2022, AAAS.ii) Demonstration of drug transport in a stomach model under wet environment.Reproduced with permission.[79]Copyright 2018, Springer Nature.iii) Conceptual drawing of the proposed soft robotic system for intratympanic injections within a patient's ear canal.Reproduced with permission.[80]Copyright 2021, IEEE Xplore.b) Miniature soft robots for delicate manipulations.i) A light-driven soft robotic gripper grasping on a 60 g boiled egg excited with direct sunlight.Reproduced with permission.[14]Copyright 2021, Wiley.ii) Demonstration of the hybrid micropillars applied as a microscale pick-transfer-place (P-T-P) device (magnetic microtweezers).Reproduced with permission.[65]Copyright 2020, Wiley.iii) Grasping a human hair by a kirigami soft gripper.Reproduced with permission.[83]Copyright 2022, Springer Nature.c) Miniature soft robots in environmental monitoring.i) Color change of domains containing porous silicone substrates after traveling through water polluted with colored oil for DraBot.Reproduced with permission.[84]Copyright 2021, Wiley.ii) The schematic diagram of MFMLS robot with functional modules (circuit, positioning, oil detection, temperature sensing, UV sensing, and pH sensing) and their response under various stimuli.N, North; S, South.Reproduced with permission.[85]Copyright 2022, AAAS.