Light-driven small-scale soft robots: material, design and control

Small robots for drug transportation, environmental detection and military reconnaissance have been a popular research topic in the field of robotics. Recently, people have proposed using light-driven actuators to make flexible and remote-controllable small robots. Herein, we reviewed the research on light-driven soft robots in recent years. First, we summarized and compared the performance and fabrication method of light-driven actuators. Then, we classified and summarized the structures of robots according to their move mode. After that, we described how to control the robot. Finally, the challenges of light-driven robots are discussed.


Introduction
Small robots, ranging in size from a few centimetres to hundreds of microns, are now perceived as promising devices for detecting confined environments, drug transportation, and military reconnaissance.In order to be effective in these applications, small robots need to meet specific requirements, including simple and low-cost production methods, good motion performance, and easy controllability.Traditional motors, with their large size, monotone output motion, and high cost, do not satisfy these requirements.Therefore, researchers proposed using soft actuators to make small robots [1].These soft actuators are capable of deforming and driving the robot to move after being physically or chemically stimulated.This innovation has the potential to address the limitations of traditional motors and enhance the efficiency of small robots in various applications.* Author to whom any correspondence should be addressed.
Researchers have developed small soft robots that can be driven by various stimuli, such as light [2][3][4], temperature [5,6], electricity [7][8][9][10][11][12], magnetism [13][14][15] and humidity [16,17].Among these stimuli, light-driven robots have garnered considerable attention due to their potential to operate without relying on batteries or wires and to enable further size reduction.The encoding of information, such as the wavelength, switching frequency, and irradiation location of light, allows these robots to perform a variety of actions.Moreover, sunlight is a ubiquitous and clean energy source, sunlight-driven robots are also attracting growing interest from researchers [18][19][20].
The present review conducts an overview of actuators, structures and control of light-driven small-scale robots.In section 2, various light-driven actuators are introduced, including those based on biological materials and artificial materials, along with their corresponding manufacturing methods.In section 3, we classify the structures of light-driven robots according to their motion modes, including crawling, rolling, jumping, swimming, manipulating and self-locomotive.In section 4, we present research results on controlling  [23].CC BY 4.0.Swimming robot.Reprinted with permission from [24].Copyright (2018) American Chemical Society.Manipulating robot.[25].John Wiley & Sons.[© 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim].Self-locomotive robot.Reprinted with permission from [26].Copyright (2020) American Chemical Society.light-driven soft robots.In section 5, the challenges of lightdriven robots and probable solution are discussed.The structure of article is shown as figure 1.

Light-driven actuator of robots
As a key part of a light-driven robot, light-driven actuators influence the movement capability and structural design of robots [27].According to materials, light-driven actuators include two categories: based on biological materials and based on artificial materials.This section begins by introducing these materials, followed by methods to fabricate the actuators.The comparison of different photo-responsive materials and actuators is shown as table 1.

Biological materials
Biological materials have incomparable advantages over artificial materials in terms of energy density and humanfriendliness [28][29][30].These makes biological materials an ideal material for making actuators.Microorganisms, muscle, and nerves-muscle are commonly employed as biological materials.Microorganisms are suitable for making micronsized actuators due to their size.Muscle tissue can produce a large contractile force and contractile displacement upon light irradiation.In addition, muscle tissue has self-repair and self-assembly capabilities that improve the ability of actuators to adapt to the environment.Nerves-muscle combines the advantages of muscle tissue with improved controllability performance.
Numerous microorganisms are phototropic, making them excellent for creating light-driven actuators.For example, Zhang et al proposed a micromanipulator fabricated through two types of heliotropic microbes, Chlamydomonas reinhardtii (CR) and Pandorina morum (PM) [31].The micromanipulator were controlled by two modes of light: spotlight and geometric shape light.The method avoids the need for excessive processing of biological materials, however, it is not suitable for organisms that lack phototaxis.
To make non-photosensitive biological materials such as muscle tissue respond to light, researchers utilize optogenetics to modify the biological materials.This involves the transfer of light-responsive proteins into living cells using viral or microbial agents.Consequently, the modified biological materials are subject to external application of light for processing [32].For example, Asano et al fabricated photosensitive myotubes by inducing C2C12 myoblasts (Yaffe and Saxel, 1977) to differentiate [33].These myoblasts experienced optogenetic processing and expressed channelrhodopsin-2 (ChR2), enabling their contraction to be controlled by light.Suzumura et al firstly transferred ChR2 into insect heart muscle (Drosophila melanogaster DV tissue) [34].Yamatsuta et al transferred ChR2 into Drosophila larvae and fabricated a type of light-driven peristaltic pump [35].
To overcome the reliance of biological materials on chemical nutrients, researchers have developed methods to enable cells to convert light energy into chemical energy directly.Walter et al expressed proteorhodopsin (PR), a light-powered proton pump, in Escherichia coli cells.These cells could convert light energy into chemical energy when respiration was inhibited [36].In addition, the study also improved the controllability of the processed cells.This is because researchers can regulate the energy obtained by the cells according to their needs.
Although biological material actuators have been greatly developed, there are still many flaws.Biological materials generally have a short service life and spontaneous motion of biological materials may disrupts the controllability of actuators.Furthermore, existing biological materials must be immersed in a liquid environment to remain viable.As a result, there are few reports of robots based on biological material that can move in a dry environment thus far.

Artificial materials
Actuators based on artificial materials consist of photoresponsive agents and main body material [37].Photoresponsive agents have the ability to convert light energy into heat energy or alter the shape of molecules upon different wavelengths of light, such as near-infrared (NIR), ultra-violet (UV) and visible light.Based on the principle of photoresponse reaction, these agents can be categorized into photothermal agents and photochemical agents [38].As the main     component of actuators, the main body materials need to be soft and have good elasticity.Common main body materials include liquid crystal elastomers (LCEs) and networks (LCNs) [39][40][41][42], polydopamine (PDA) [43], polydimethylsiloxane (PDMS) [44][45][46], polydimethylsiloxane (PDMS) [44][45][46] and hydrogels [2,47].LCEs and LCNs molecules line up along a certain direction at room temperature, As the temperature increases, the order of molecules is lost, leading to the macroscopic shape of the material swells.Hydrogels have the opposite properties to LCEs.They contract at high temperatures due to water loss and swell by absorbing water from the environment when the temperature decreases.
Familiar photothermal agents include carbon-based materials [43,[48][49][50], MXene [51][52][53][54][55][56], metallic nanoparticles [25,44,[57][58][59], semiconductor nanostructures [60], and NIR dyes [61].These agents can absorb light and convert it into thermal energy.Then, heat diffuses throughout the main body materials and induces actuator deformation.Hu et al integrated carbon nanotubes (CNTs) and PDMS material to fabricate an actuator [46].Upon light irradiation, the heating of CNTs expands PDMS and actuator exhibits a fast response speed and ultralarge deformation from tubular to flat.MXene have remarkable photothermal absorption, abundant surface functional groups, and layered structures.The surface functional groups on MXene nanosheets can absorb water at room temperature and release water when the photothermal reaction causes the temperature to rise.Liu et al proposed an actuator composed of MXene (Ti 3 C 2 T x ) layer, cellulose film, and polyimide (PI) film [54].The actuator can bend to track the incident light a like sunflower.Inorganic rigid agents may affect the flexibility of the actuator substrate resulting in reduced response speed of actuators.To solve this problem, Li et al proposed the use of metal dichalcogenides as photothermal agents, and a two-dimensional heterojunction assembled of WSe2 and graphene was introduced into PDMS to fabricate actuators with rapid deformation [62].
Common photochemical agents include azobenzene and stilbene dye.The principle of actuators based on photochemical agents can be summarized as follows: photochemical agent molecules are arranged in the main body materials according to certain rules, and the shape of the molecules will change upon light irradiation, resulting in changes in the macroscopic shape of the materials.For example, Cheng et al provided a thin film actuator that was fabricated by introducing azobenzene into the LCNs [22].Xiang et al reported a kind of photochemical agent called photoswitchable hexaarylbiimidazole (HABI) [47].The agent is used to fabricate an underwater actuator driven by visible light.Osaki et al fabricated a transparent actuator by adding stilbene dye (Sti) into hydrogels [63].
There are differences between photothermal and photochemical actuators in many aspects, such as the response speed and wavelength of response.This inspires researchers to make better actuators with both photothermal and photochemical properties.For example, Lu et al doped polymer-grafted gold nanorods into azobenzene liquid-crystalline dynamic networks (AuNR-ALCNs) to fabricate an actuator that exhibits different deformations at different wavelengths of light [25].
Lahikainen et al provided an actuator consisting of LCNs and azobenzene [64].The actuator can exhibit a variety of deformations through the synergistic use of photochemical and photothermal responses.It is worth noting that most photochemical agents also release heat during photochemical reactions.On the one hand, it hinders the accuracy of photochemical actuator deformation.On the other hand, it improves the deformation capacity of actuators.Liu et al provided a photochemical/photothermal actuator, and both kinds of photo-response reactions are induced by azobenzene [65].

Fabrication of actuators 2.3.1 Actuators based on biological materials.
The actuators are composed of both biological and non-biological materials.The non-biological materials create the necessary conditions for the growth and adhesion of the biological materials.Nonbiological materials must exhibit good biocompatibility and be easily fabricated into complex 3D structures.Commonly used nonbiological materials include PDMS and hydrogels.PDMS can be fabricated with complex microstructures, and it can be easily modified to achieve biocompatible conditions.Hydrogels can provide nutrients for materials through their pores.Moreover, hydrogels are materials for 3D bioprinting techniques.
To obtain actuators based on muscle tissue, the muscles are usually seeded on the 3D structure of nonbiological materials allowing the muscle tissue to automatically attach and develops into rings, strips, or other structures.For example, Anand et al fabricated an actuators based on C2C12 cells (figure 2(a)) [66].PDMS structure was set on the Si chips in a petri dish.The PDMS structure was coated with fibronectin while all the other surfaces were coat with pluronic F127.Then cells were plated on all the 2D surfaces.After few hours, the cells start moving towards the 1D structure and eventually cover the whole structure.Aydin et al fabricated an actuator based on nerves-muscle (figure 2(b)) [67].First, muscle cells were assembled on a glass coverslip and muscle strip was formed after two days.Then, neutrals cells were embedding and coculture with muscle tissue.The neural cluster was experienced optogentic processing and can response light stimulation.After that, the research team provided a method enables co-culture of a neuronal cluster with up to four target muscle actuators (figure 2(c)) [68].First, a 3D cell culture platform was designed.Then, the skeletal myoblasts and neurospheres were sequentially seeded on the platform.This study demonstrated that the function of muscles and neural networks was improved through their codevelopment.
The actuators power by microorganisms can generate in two ways.One is seeding microbes into the microstructure.while the other relies on the self-assembly of microorganisms with microstructures.Vizsnyiczai et al designed a type of 3D microstructure.Initially, the structures are immersed in clean motility buffer, and then cells are added (figure 2(d)) [69].The structures could self-assemble with several E. coli strains.As figure 2(d) shows, the body of the bacteria is limited to a microchamber (green), and the flagellum is left outside.The microactuators could rotate upon light irradiation.Steager et al proposed a type of microactator that is obtained by synthetizing bacteria and microstructures [70].The microstructure is a 50 µm equilateral triangle with a thickness of 10 µm, and the bacteria were seeded on the plate.After 8-16 h, the microactuators begin to swim.
Fabricating biological actuators with complex shapes is still a challenge.Although 3D bioprinting have been used to fabricate muscle tissue with complex shapes [71], there are few reports about light-driven actuators fabricated by these methods.In addition, the majority of present actuators cannot change shape after complete fabrication.This reduces the flexibility in fixing actuators on robots.Therefore, it is necessary to make modular actuators that are compatible with robots of different structures in the future [72].

Actuators based on artificial materials.
Integrating the photo-responsive agent with the main body material is a crucial aspect of creating an artificial actuator.Common methods for achieving this integration include adding the agent into the solution of the main body and waiting for solidification, dropping the main body into the solution of the photoresponsive agent to produce a coat, and adhering an agent layer to the main body.The molecular arrangement of the material and the distribution pattern of the agent also impact the performance of the actuators.Several methods have been developed to adjust the molecular alignment and the distribution pattern [73][74][75].For example, Wani et al developed a method that uses a laser projector with a microelectromechanical system to program the molecular orientation of liquid crystal polymer networks [76].A flower-like actuator was fabricated in this way (figure 3(a)).By controlling the orientation of the molecules on the petals, the petals can bend in different directions.
Film actuators that can bend and twist are often made to assemble robots.For example, Yang et al integrated graphene oxide (GO) and poly(N-isopropylacrylamide) hydrogels to fabricate a film actuator with a signal layer [77] (figure 3(b)).By utilizing a direct current electric field, GO is gradient and oriented in the poly hydrogel.These single-layer actuators has excellent material stability.However, they face challenges in achieving programmable deformation and easy fabrication.Therefore, researchers have proposed bimorph actuators with an active layer and a passive layer.The active layer is used to respond to light.The passive layer can perform various functions, such as support, heat dissipation, and respond to other stimuli.By changing the angle between the axes of the passive layer and active layer, the actuators could realize bend and twist.For example, Zuo et al fabricated a bending actuators and a twisting actuators (figure 3(c)) [61].The active layer is stuck on the passive layer with a crossed angle of either 0 • or 45 • .Then, the actuators are heated for 2 d.The actuator with an angle of 0 • could bend, and the actuator with an angle of 45 • could twist.The active can be made into various patterns to realize program deformation.For example, Jiang et al provided a film actuator consisting of an active layer and a passive layer with different shapes to realize more complex deformation [78].Weng et al provided a method called water welding to assemble several actuators together [49].Through this method, researchers can first make several simpleshaped actuators and then assemble them into a complex actuator.Due to their shape being similar to paper, film actuators easily fabricate complex 3D structures through process similar to folding [79,80] and kirigami [22,41,81].For example, Chen et al fabricated Y-shape and H-shape actuators are produced actuator through integrating the concepts of kirigami in a light-driven actuator with specific molecular orientations (figure 3(d)) [41].Yu et al developed a facile and scalable approach to fabricate actuators through kirigami [82].The actuators fabricated by this method could be used to devise a variety of light-driven soft robots, such as artificial flytraps, walkers, and dolphin-like swimmers.
Another common shape of actuators is a line, which can extend and contract [83,84].Compared with film actuators, line actuators can integrate better with soft robots at arbitrary locations.Meng et al provided a light-driven actuator with a line shape [85].First, the mat is fabricated by the electrospinning technique.Then, the mat is twisted into yarns, and yarns are twisted continuously to form fiber coiling.After whole yarns are coiled, thermal treatment is employed to maintain the writhing shape.The line-shaped actuator exhibit great tensile actuation, and 1000 cycles are performed without an obvious decline in actuation performance.Liu et al provided a facile and stable method to fabricated line actuators (figure 3(e)) [86].First, the LCE ink is extruded by a syringe pump and exposed to UV light for a few seconds to lock the LC alignment and the filament shape, Then, the line was collected by a rotating mandrel at the bottom.
To achieve a more complex 3D shape, 3D printing technology has also been used for light-driven actuator fabrication, especially actuators with a size within 1 mm.For example, Wang et al proposed a 3D-printable photoresponsive AuNR/LCE composite ink [87].By fine tuning the ink formulation and printing parameters, the 3D-printed AuNR/LCE filament can reach 27% actuation strain upon NIR light irradiation.For some 3D printing methods, light is crucial.To prevent photoresponsive materials from interfering with the manufacturing process, Hsu et al developed a 3D printing method (figure 3(f)) [88].The method is divided into two steps: (i) Print a 3D microstructure by laser.(ii) Diffuse the photoresponsive agent into the 3D microstructures.Through this method, a kind of microscale graphene actuator is fabricated.Smart responsive materials, including photoresponsive materials and thermoresponsive materials, are used as printable materials for 3D printing.The structures fabricated by this method can change their shape upon external stimulation.This concept is frequently referred to as '4D printing' [89].

Structure of light-driven soft robots
A wide range of light-driven robots with diverse structures have been developed.Typically, these robots are constructed by integrating one or more light-driven actuators with supplemental auxiliary components.These auxiliary materials play a crucial role in enhancing support [93], energy storage [94] and hydrophobicity [57].In the ensuing section, the classification of robot structures based on their movement models and the introduction of their typical structures will be presented.The comparison of robots in different structures is shown as table 2. Based on line-shape actuators -Lift a 0.5 g object - [107] (Continued.)

Crawling robots
The crawling robots are similar to caterpillars and snakes.
They have a slender body and move forward through bending body to generate friction.For example, Yang et al provided a snake robot (figure 4(a)) [95].The robot can crawl in two kinds of modes as snakes, concertina locomotion (the front part stretches forward by head uplift) and serpentine locomotion (the body curves on the ground and stretches forward).
As figure 4(a) shows, two light-driven actuators were installed on the robot body as joints, the actuators have different active layer widths to generate different bending angles.
To give the robot a steering ability, Three-legs and fourlegs robots were also made.Wang et al provided a three-legs robot based on biological materials (figure 4(b)) [96].The muscle strips were installed on robot legs.Light on different legs allowed the robot to move in different directions.Pilz da Cunha et al provided a walk-like robot with four legs (figure 4(c)) [97], the soft robot using four identical bimorph actuators as legs (yellow part) connected with a polymeric hub (blue part), which hosts bimorph actuators acting as soft robotic arms (red part) for carrying cargo.Sung et al proposed a walk robot that consists of three legs with joint structure [93].Every legs can be divided into two separate regions.
In crawling locomotion, different part of the robot obtain friction in different directions.One produces power for locomotion, and the other obstacles hinder locomotion.To enhance the former and reduce the latter, Song et al provided a crawling robot with bristle like earthworm (figure 4(d)) [98].The U-shaped forelegs and S-shaped hindlegs fabricated by elastic awns were installed on the body of robot.This structure can increase forward friction and reduce backward friction.
Rehor et al provided a 100 µm long crawling robot [99].At this scale, the ratio between body length and its thickness is smaller.Therefore, the bending of body became less pronounced.To solve these problems, the robot is designed in the U-shape and can move by contraction and expansion of the body.Miskin et al provided a robot at micronscale consist of a body containing standard silicon electronics and four legs fabricated from actuators [100].The silicon body generate current under laser irradiation, which causes the bending of the leg.

Rolling robots
The rolling robot typically takes the form of a standard cylindrical or spherical structure.When illuminated by light, the robot undergoes local deformation, disrupting its original stable state and causing it to initiate rolling movement.For instance, Wang et al demonstrated the creation of a rolling robot by curling a bilayer actuator into a roll [101].Upon light irradiation, the outermost layer unfolds, causing the robot to roll forward.Cheng et al presented a roll robot with steering capability (figure 5(a)) [22].The structure of the robot can be seen as a ring surrounded by several strip actuators.Through the bending of the actuators in contact with the ground, the robot can roll in a controlled direction.
Compared with crawling, rolling robots have faster speed and higher efficiency.However, rolling robots can only work on flat ground.To solve this problem, Yan et al provided a robot that can transform between different movement models, including crawling, rolling, and oscillating (figure 5(b)) [102].
The robot is fabricated by a film actuator.By adjusting the position of the light, the robot can move in different modes.

Jumping robots
Jump necessitates the ability to store and release energy in a short time.Therefore, the design of energy storage structure is the core problem of jumping robot design.For example, Ahn et al provided a robot that can crawl and jump over a barrier (figure 5(c)) [94].The robot is fabricated by a film actuators, two magnets are attached to the robot head and end.The robot formed a closed loop by magnet attraction.Upon light irradiation, the robot gradually stores elastic energy, and sudden release resulted in a jump.However, the magnets need to be attached manually so that they cannot jump continuously.To solve this problem, Hu et al provided a jump robot inspired through a kind of insect called springtall (figure 5(d)) [103].The robot is designed into a three-leaf panel fold structure by using a film actuator.Upon local light irradiation, the inner folded part stores energy and results in a jump movement.Then, the outer folded side recovers and becomes the inner folded side in the next jump.Through this structure, the robot obtains a continuous jump capability.Li et al provided a global robot composed of binary iron oxide nanoparticles (IONPs) and poly hydrogel composites (figure 5(e)) [23].When the bottom of the ball is exposed to light, the water in the hydrogel is evaporated through the photothermal effect of IONPs, and the local global shape is changed through the vapor, which ultimately leads to the robot jumping.

Swimming robots
Inspired by aquatic life from bacteria to fish, many swimming robots at various scales have been provided.For example, Park et al provided a swim robot based on biological materials, the robot adopt a movement strategy like batoid fish called the median and/or paired fin (MPF) mode [104].The robot consisted of serpentine-patterned muscle tissues for producing power, a gold skeleton for storing elastic potential energy as the muscles contracted and an elastomer body.Xu et al provided a swim robot that can suspended in water (figure 6(a)) [105].The robot consists of a muscle tissue actuator, and a hydrogel wing could respond to light.Upon NIR irradiation, the hydrogel wing bends to reduce the thrust produced by the muscle actuator and then the robot enters a stationary state.Tian et al provided a robot that swim through beat the water like a dolphin (figure 6(b)) [24].The body of the robot is constructed with VHB, while the tail is created using a film actuator.However, current dolphin robots can only float on the water and cannot swim underwater.Yin et al provided a jellyfish soft robot that can swim in water [106].He et al provided a robot that swims by a line actuator pull flexible hinge [107].Upon light irradiation, the actuator pulls the hinge by contraction, generating a push forward.Yang et al provided a threelegged soft robot that can float on water similar to a water spider through combining light-driven actuators with superhydrophobic material [57].The legs made from actuators that respond different wavelengths of light, so the three-legged robot can move in different directions.
Robots that use the surface tension of the fluid to move are also provided [108].The robot uses the photothermal to change the tension of the surrounding liquid for motion.The robot's body is usually a thin sheet and does not undergo large deformation during movement.These robots consist of a hydrophobic material and light-driven actuators.The hydrophobic material makes robots float on the water, and actuators power the robot.For example, Wang et al provided robots that are pushed by the Marangoni effect (the liquid in the high surface tension region will move toward the low surface tension region) [109].The actuators consist of a photothermal layer and a hydrophobic PDMS layer (figure 6(c)).Upon light irradiation, the photothermal induced surface tension changes, and then the robot is propelled by surface flow generated through the Marangoni effect.
Robots hundreds of microns long have been developed.For example, Weibel et al presented a method to control microorganism (Chlamydomonas reinhardtii) to transport objects through light [110].Xin et al used a single cell (green microalgae) to fabricate a microrobot (figure 6(d)) [90].Upon light irradiation, the micromotors could move in biological media such as cell culture medium and saliva.Then, reconfigurable motor arrays consisting of several micromotors are realized.Robots based on artificial materials have also been created.At the microscopic scale, friction and adhesion are typically large, which means that surface effects are the main factor in determining the movement of robots, while the weight of the object and its inertia are unimportant [111,112].To solve these problems, Li et al provided a microrobot that can swim in a blood-mimicking viscous glycerol solution (figure 6(e)) [113].The robot adopted a rocket-like triple tube structure.To track a single robot by an imaging system, a gold layer is coated on the body of the robot to improve reflectivity.Upon NIR irradiation, the robot is pushed by the self-thermophoresis force, and a strong heat flow exits from the tubular structure, as a result of concentrated photothermal energy within the tube.Palagi et al fabricated long cylinders and flat discs using LCE.Both cylinder and flat disc robots realize traveling-wave motions such as peristalsis of annelids [114].Obara et al provided a microrobot attempt three-link swimmer model to break the scallop theorem: the displacement of objects in Newtonian fluid is zero if the deformation of the object is a reciprocal process [115].The robot was divided into three panels with different lengths, and two short panels are propulsive upon light irradiation.
Compared with traditional robot arms and hands fabricated by hard materials, light-driven manipulating robots have unique advantages in grasping brittle objects and irregularly shaped materials.The manipulating robots structure can be divided into two categories: based on film actuators [116,117] and based on line actuators [107].Robots that combine film actuators work by bending and twisting of the actuators.For example, Lan et al designed a robot arm that could grip and lift objects although a film actuator [118].Lu et al provided a light-driven soft robot arm that combines a lifting arm and a gripping hand (figure 7(a)) [25].The lift arm could contract upon NIR light, and the grip hand could bend to grip upon UV light.Lyu et al provided a pick-and-place robot that consists of three actuators, a rotating base, a lifting unit and a suction cup-based gripper (figure 7(b)) [119].The gripper is similar to a suction cup, which is inspired by cephalopods.Upon light irradiation, the gripper deforms to a conical state, generates negative pressure, adheres the object to the cup, and the structural actuators can rotate or lift under electric stimulation.The line actuators contract upon light and pull the body of robot into motion, which mimics the movement pattern of a human arm pulling a bone through muscles.He et al provided a weight-lifting artificial arm fabricated by line light-driven actuators [107].The LCE microfiber bundle contracts upon light irradiation and causes the arm to bend.
In addition, the arm and hand structure can also be carried on other robots.For example, Huang et al provided a robot combines with a gripper, body and flagellum that could transport cargo in the water (figure 7(c)) [120].The flagellum is actuated to swing by UV and white light.The gripper on the robot head is also controlled by light.Wang et al installed a bionic arm on the body of a fish robot that was also assembled by a light-driven actuator (figure 7(d)) [121].The bionic arm is controlled by an NIR laser.It consists of the multidegree-offreedom arm and a grip hand.The bionic arm can grasp and release the object, and with the light-driven swimming of the fish body, the robot realized object transformation.

Self-locomotive robots
The robot introduced above needs to frequently adjust the light parameters to make continuous motion.The structures that can achieve self-locomotive under constant light were also proposed.The spontaneous contract ability of biological materials, including cardiac muscle and insect muscle has been demonstrated, and the self-locomotive robots based on biological materials has been provided.For example, Lee et al provided a self-locomotive swimming robot [122].The robot consisted of two layers of muscle tissue, a gelatin body and a plastic fin.The muscle tissue is mounted on both sides of the gelatin body to make the robot swim continuously.The robot can swim in the body and/or caudal fin (BCF) mode.
The mechanism of self-locomotion in robots based on artificial materials can be outlined as follows: the movement of the actuators causes a change in the light they are exposed to, resulting in a light-mechanical feedback loop [43].For example, Dong et al developed a soft butterfly robot [26].In sunlight, the wings change from a flat to an upward-curved position, a part of the wing is obscured by shadow causing the wing to flatten out and the cycle repeats.The robot could swing at 4.49 Hz (figure 8(a)).Liu et al provided an actuator that can aim the light source and continuously oscillate [54].When exposed to light, the actuator can deform in the direction of the light (aim).Then, the actuator bends downward, shading the light and allowing the actuator to recover and repeat the cycle (oscillate) (figure 8(b)).Hu et al provided an inch-worm-like robot capable of autonomous crawling towards a light source, achieved by removing a part of the robot's head to expose its tail to light (figure 8(c)) [21].When light irradiates the head of the robot, the robot crawl forward, and the illuminated area on the tail is reduced, which leads to continuous crawling.Yang et al provided a bimorph actuator that can produce two kinds of different self-oscillations by changing the light irradiation position (figure 8(d)) [43].When the light illuminated tail, the robot performed sit-up motion.When the robot illuminated the side, the robot performed set-up/twisting mix motion.

Control of light-driven robots
Controlling light-driven robot involves adjusting light parameters such as light intensity and switching frequency according to their relationship with actuator performance.The control strategies include manual closed-loop control, automatic open-loop control and automatic closed-loop control.This section first describes the methods used to adjust the light to target motion.Then, each control strategy is described individually.Finally, the multiple field coupling control robot is introduced.

Adjustment of light
To effectively control the light-driven robot, it is essential to determine the illumination parameters necessary to achieve the desired robot motion.Therefore, a comprehensive study of how various light parameters influence the performance of robots is crucial.In summary, the speed of the robot tends to rise with an increase in light intensity.The effect of lighting switching frequency on the robot is contingent upon the response time of the actuators.Moreover, factors such as the position of laser irradiation and the wavelength of light can also significantly impact the robot's motion.Yin et al tested the swimming speed of the robot under different light frequencies and powers [106].Based on the experimental results, the appropriate light parameters were chosen and robot was controlled to swim through the obstacles.
In order to achieve more precise control, the researchers are trying to establish a theoretical model between the light parameters and the movement of the robot.Shahsavan et al researched the relationship between photothemal actuator deformation underwater and light intensity [128].Several underwater movement models, such as crawling, walking, jumping, and swimming, were demonstrated.Wu et al proposed a relationship between the LCE actuator deformation and its temperature [129].The deformation of the LCE actuator is a result of changes in the molecular alignment orientation.To characterize the order of LCE molecules, a parameter Q was established.Subsequently, the relationship between Q and temperature was determined.Li et al demonstrated that polarization impacts the bending angle and oscillation frequency of the actuators [130].The finite element method is also employed to analyze the relationship between light and robot motion, and it is suitable for the analysis of a variety of actuators [91].However, because of the heavy computational burden of FEM software (e.g.ANSYS, COMSOL), applying FEM to real-time control is difficult [131]  Manual closed-loop control involves researchers manipulating the motion of robot by manually adjusting the light, while observing their movement.For example, Hui et al developed a microscopic crawling robot that can be driven by adjusting a laser manually [132].A microscope is accessible for researchers to observe the movement of the robot.In the future, this robot may have applications in eye surgery.
Manual control strategies offer several advantages, such as simple control systems and high success rates.This method is still considered the safest approach for certain surgical robots.However, it suffers from a significant control delay, and it presents challenges for a single person to adjust multiple lighting parameters simultaneously.

Automatic open-loop control.
In open-loop control systems, it is essential to model and understand the relationship between light and robot motion state, with predetermined light parameters being set to achieve the desired robot motion, with predetermined light parameters set to achieve the desired robot motion.In this type of control system, the output signal (robot motion state) does not influence the input to the system (light illumination).For example, Yongdeok et al developed a robot fitted with LEDs and programmed the LEDs to switch on and off in a specific sequence, enabling the robot to move along a predetermined path [133].
Open-loop control allows rapid adjustment of the lighting parameters, but without feedback the robot cannot correct for deviations caused by model inaccuracies.Consequently, the open-loop control strategy is unsuitable for complex motion.Using this way, the robot arm can transfer objects to a fixed distance.Dong et al created a light-driven microrobot with closed-loop control by genetically and neurally engineering nematode worms [125].The laser excited the worm muscles which were then monitored by a camera.An image of the microrobot was obtained, and its movements were regulated through the DMD chip using a computer.Using an objective approach, the living soft robot successfully navigated a plastic maze via a controlled strategy.Wani et al presented an automated artificial flytrap (figure 9(b)) [125].When an object enters the field of view and causes optical feedback, the actuator bends and clamps, mimicking natural flytraps and demonstrating its effectiveness.
Obtaining feedback is a crucial aspect of achieving closedloop control.Gu ¨ell-Grau et al provided an opto-magnetic actuator that could reflect different colors of light at different bending angles (figure 9(c)) [135].A camera was used to analyze the hue changes of images in real time with a precision of 0.05 • .This work has important implications for lightdriven robots that rely on vision sensors for control.Wang et al employed film actuators can convert deformation into an electrical signal.Then, the actuators were assembled into a hand capable of measuring temperature and the softness of the clamped object (figure 9(d)) [136].Wang et al provided a lightdriven crawl robot can communicate with the receiver by magnetic coupling (figure 9(e)) [137].Researchers can get information about the location of robots through an external coil array.This work promotes cooperation in a group of robots.

Multiple field coupling control
Multiple field coupling control, which can be controlled by different stimuli at the same time, inherit the advantages of different control methods and can perform more complex sports [139].Zhang et al provided a bilayer actuator can be controlled by light and moisture field (figure 10(a)) [140].The legs consisting of nanosized graphite (Nano-G) and polyvinylidene fluoride could respond to moisture simulation.The other legs, consisting of GO, could respond to light simulation.Pilz da Cunha et al provided a soft robot arm that can be controlled by to light and magnet field (figure 10(b)) [141].The robot arm is composed of a grasping hand driven by light and a flexible stem driven by magnetic field.Upon UV light irradiation, the grasping hand closes, and the actuator retains its deformed state after stopping the UV light.Upon blue light irradiation, the grasping hand reversibly opens.As the magnetic field rotates, the stem can rotate and bend.Tang et al provided a dual-responsive actuator that can controlled by light and electricity (figure 10(c)) [142].The actuator consisted of an MXene film, a hydrophobic biaxially oriented polypropylene (BOPP) film and hygroscopic bacterial cellulose (BC) films.When the electricity and optical energies are effectively converted to heat by MXene, the BOPP layer expands, and the BC layer loses water rapidly, therefore, the actuator changes from flat to bent.
On the basis of the multiple field coupling control, the researchers further propose the concept of multiple field synergistic control, which requires several stimulation synergies to complete the movement.The effects of various stimulations are not isolated, and one stimulation will contribute to the effect of another stimulation.For example, Han et al provided a strip actuator that could respond to light and magnetic stimulation (figure 10(d)) [143].The actuator could twist in light and magnetic couple fields.The magnetic field can adjust the axial of twisting and light can control actuator bends.Then, a crab robot (CraBot) that could move toward any direction by synergistic controlled of magnetic and light was fabricated.Tan et al fabricated a self-repairing actuator that worked under synergistic control of light and electricity [144].The actuators can deform and repair themselves under electrical stimulation.Upon exposure to NIR light, the actuator undergoes a photothermal reaction, and the toughness increases, which leads to an increase in the deformation speed of the material and a decrease in the temperature required for self-repair.Ha et al provided a kind of microrobot for treating brain tumors that functions through the synergistic work of electricity and light [145].Electricity field was used to induce actuator activity, and light was used to induce photothermal reactions to eliminate tumor cells.

Challenges and probable solutions
Light-driven small soft robots, as an emerging technology, still encounter numerous challenges.The challenges and probable solutions can be summarized as follows.
(1) Challenges of actuators.Current light-driven actuators still have certain disadvantages that limit robots practicality, including long response time, limited motion patterns and poor repeatability.To solve this problem, better actuators need to be designed in terms of materials and the manufacturing process.In addition, it is necessary to select actuators with different characteristics according to the move mode of robots.For example, swimming robots require actuators with fast response and large deformation amplitudes, while gripping robots necessitate high output force from the actuator.(2) Challenge of structure design.An increase in the number of actuators is often a prerequisite for an increase in the flexibility of the robot's movements.But increasing the number of actuators also leads to an increase in control complexity.To address this contradiction, it is imperative to propose an innovative structural design.In the conventional design, the lack of coordination between multiple actuators hinders efficiency.Therefore, it is important to explore methods for reusing actuators for multiple functions and orchestrating collaboration among different actuators to enhance the robot's motion flexibility.This approach not only enhances the robot's flexibility but also streamlines the control system and reduces its complexity.
(3) Challenge of sense.Unlike traditional robots which can utilize various sensors to sense their motion state and environment, the light-driven robot's sensing capabilities are still at a rudimentary stage.It can only perceive minimal information with low accuracy, greatly limiting its potential applications.To address this limitation, two potential solutions have been identified.The first approach involves modifying existing electronic sensors to make them smaller, lighter, and capable of being powered by light instead of batteries.Alternatively, the second approach involves studying the driver, which integrates the sensor function.These strategies aim to enhance the sensing capabilities of light-driven robots, thereby expanding their potential utility and effectiveness in various applications.(4) Challenge of control.To achieve as accurate control as a rigid robot.More accurate models need to be built, and suitable algorithms are also necessary to determine the input light sources.Machine learning (ML) has been used for the control of soft robots, and using artificial intelligence to control light-driven soft robots may be a potential approach.However, there are few reports about using ML to control light-driven soft robots due to the lack of a dataset and 'black box' between the robot and control.The development of migration learning and unsupervised learning techniques may be able to solve the problem that lacks highquality datasets.(5) Challenges of non-translucent environments.Because of the physical properties of light, it remains difficult to enable robots to move in environments where light cannot penetrate.Current solutions primarily depend on multiple field control.Moreover, we posit that installing a light source inside the robot is a viable approach.

Summary
This paper provides an overview of the most recent studies on light-driven robots, with a focus on their actuators, structures, and control systems, while also introduced the challenges currently encounter.Light-driven robots have a wide range of applications, including drug transportation, environmental detection, and military reconnaissance.The development of light-driven robots required across various research fields, including physics, chemistry, materials science, manufacturing technology, and control science.It is anticipated that in the future, light-driven robots will exhibit better locomotion, sensing ability, and controllability, offering potential benefits across a variety of fields.

4. 2 . 3 .
Automatic closed-loop control.Closed-loop control offers superior stability, adaptability, and robustness compared to manual and open-loop control.Wu et al developed a closed-loop control robot arm (figure 9(a)) [134].The actuators of robot can generate electrical signals based on the degree of actuation bending.When arm lift, the controller receives the electrical signals and adjusts the illumination intensity.

Figure 10 .
Figure 10.Multiple field coupling control robots.(a) Robot controlled by light and moisture simulation.Reprinted from [140], Copyright (2020), with permission from Elsevier.(b) Soft robot arm controlled by light and magnet field.Reproduced with permission from [141].Copyright © 2020 the Author(s).Published by PNASCC BY-NC 4.0.(c) Actuator that can bend upon light or electricity stimulation.Reprinted with permission from [142].Copyright (2022) American Chemical Society.(d) Crab robot (CraBot) that could move in any direction by light and magnet field synergistic control.[143].John Wiley & Sons.[© 2021 Wiley-VCH GmbH].

Table 1 .
Comparison of different photo-responsive materials and actuators.

Table 2 .
Comparison of robots in different structures.