Actuation Mechanisms and Applications for Soft Robots: A Comprehensive Review

Abstract

Soft robots, which exhibit distinguishing features in terms of compliance, adaptability, and safety, have been expansively adopted in various niche applications. For soft robots, innovative actuators have been designed based on smart materials enabling the robots to perform flexible and versatile functions, whereas extra spaces and accessories to accommodate motors and power devices have been eliminated to achieve structural optimisation. Herein, different types of actuation mechanisms for soft robots are summarised to reflect the state-of-the-art research and applications. Major characteristics of the actuation mechanisms are updated. Design methodologies of the actuation mechanisms are discussed in detail. Furthermore, their advantages, disadvantages, and application potential are compared and summarised. In the end, based on our knowledge and understanding, new thoughts and recommendations to further develop the actuation mechanisms are put forward. This review is useful to support the conclusion that, through incorporating actuation mechanisms and advanced intelligent technologies, soft robots tend to create disruptive innovations in applications.

1. Introduction

Smart material-enabled soft robots exhibit greater degrees of freedom and produce muscle-like behaviours [1]. Compared with a rigid robot, a soft robot demonstrates the advantages of reducing the complexity of mechatronic design, minimising the requirement for algorithmic programming, and interacting with unknown environments in a flexible means. Soft robots have been extensively applied in broad prospects, such as industrial production, agriculture, biomedical science, disaster resilience, etc. [2,3,4]. For soft robots, some critical parts are made of smart materials, like polymers, rubbers, silicones, or other soft materials [5,6,7,8,9,10]. The materials are used to design actuation mechanisms to drive soft robots to fulfill complex functions and adapt to changes in complex environments intelligently and autonomously [11,12,13,14]. Some typical soft robots and their actuation mechanisms are illustrated in Figure 1, including shape memory alloy (SMA) (Figure 1a,b), fluid (Figure 1c–f), electro-active polymer (EAP) (Figure 1g–i), magnetic (Figure 1j,k) and origami (Figure 1l).

Almost all soft robots are based on the design principles of bionics [14]. By imitating the behaviours of various creatures in nature, such as reptiles, annelid, molluscs, and humans (bionic hands, arms, etc.), different types of soft robots have been designed, like snake-shaped robots [27,28], inchworm robots [29,30], earthworm robots [31,32], jellyfish robots [33,34,35], robotic hands [36,37,38,39,40], auxiliary gloves [41,42,43], etc. Such a bionic soft robot can achieve a single function, such as multi-gait crawling [44,45,46,47], frictional crawling [48,49,50], jumping [51,52], and rolling [53,54], or it can achieve multiple functions, such as two modes of scrolling and crawling [55] and three modes of straight crawling, steering and rolling [56]. The combination of different actuation mechanisms can make a soft robot more adaptable in different terrain environments, such that it could be used to be underwater, on land, or in the low sky.

Some previous reviews for soft robots (e.g., [2,57]) have focused on the motion modes or the applications of soft robots. In some reviews [58,59,60,61,62], soft robots and their actuation mechanisms have been summarised. For instance, in [61], a survey of magnetically actuated bio/soft robots was conducted, and the future prospects of actuation mechanisms used to improve the performance of different types of soft robots have been discussed. In [62], the actuators of soft robots have been divided into three types, i.e., variable length tendon, fluidic actuation, and electro-active polymer (EAP). In the research, the sensing mechanisms, structures, and applications of soft robots have been introduced. Nevertheless, more comprehensive evaluations and discussions on the latest development of soft robots and their actuation mechanisms are highly imperative. For instance, some newly emerging types of actuations, like light [63,64] and chemical [65], have not been discussed yet.

To bridge the aforementioned knowledge gap, this paper is aimed to update the latest progress of actuation mechanisms and applications for soft robots. The objectives include:

(i)

The actuation materials and mechanisms of soft robots are summarised according to seven categories, i.e., SMA, fluid, EAP, electric and magnetic, light, sound, and chemical;

(ii)

Furthermore, according to the progress of soft robots in recent years, extra categories, including SMA-based wire, spring design, origami-based pneumatic, and dielectric elastomer actuator (part of EAP), are also summed up;

(iii)

The actuations of soft robotics under the above categories are reviewed in terms of characteristics, mechanisms, and typical structures. The advantages and disadvantages of the categories are analysed quantitatively and qualitatively;

(iv)

In addition, challenges for the future development of soft robots are put forward to bring more insights to practitioners and researchers.

2. Actuation Materials and Mechanisms for Soft Robots

In the following, actuation mechanisms are summarised according to the aforementioned seven categories. Simultaneously, smart materials, driving mechanisms, applications, and technical difficulties for the actuation mechanisms are also presented.

2.1. Shape Memory Alloy (SMA)

SMA is a specific alloy that, after being deformed, can return to its initial shape when heated [66,67]. As shown in Figure 2, the effects of the shape changes can be categorised into three types: one-way shape memory effect (OWSME), two-way shape memory effect (TWSME), and pseudo-elasticity [68,69]. Under a certain temperature, stress, and strain, SMA can exhibit two different phases, three different crystal structures (detwinned martensite, austenite, and twinned martensite), and six possible transformations [70]. These phenomena create OWSME, TWSME, and pseudo-elasticity, respectively.

Typical SMAs include Ni-Ti-based (nitinol), Fe-based, Cu-based, etc. Ni-Ti-based SMA is the most commonly used shape memory material, owing to its high thermal stability, strain recovery, and corrosion resistance [71,72]. That is, in terms of heat resistance, transformation, and recovery stress, Fe-based and Cu-based SMAs are not as good as Ni-Ti-based SMAs, so they are less used to design actuations of soft robots [73,74,75]. Ni-Ti-based functional components with shape memory effects can be flexibly produced using additive manufacturing processes [58], which supports the design of small, exquisite, highly automated, and reliable soft robots. Therefore, Ni-Ti-based SMA has attracted increasing attention and has been widely applied in biomedical, aerospace, robotics, mechanical, and other fields [76]. SMA-based actuators can be mainly divided into the spring form and the wire form. More technical details are given below.

2.1.1. SMA Spring

The SMA spring actuator employs a spring as the main body of a soft robot to connect its front and rear moving components. When the spring is deformed and expanded, the movable components operate. As shown in Figure 3a, an earthworm-like micro-robot achieves a locomotion action based on the SMA spring actuation mechanism [77]. As shown in Figure 3b, when the robot is energised, the front foot of this robot produces an electro-adhesive force to ensure the robot is fixed on the ground. Then, upon the electric current is applied, the spring is heated to shrink, making the rear foot move forward. Finally, when the rear foot is anchored, the spring stretches and returns to its original shape, causing the front foot forward. In Figure 3c, the high state of the front foot or the rear foot means the robot is moving forward, and the low state means that it is in an anchored state. That is, the high state of the SMA spring means deformation and contraction, and the low state means elongation. In the first stage (from the 2nd to 4th time interval), the SMA spring contracts, causing the rear foot to move towards, whereas in the second stage (from the 4th to 6th time interval), the front foot is pulled by the elongated SMA spring.

Moreover, the robot with three locomotion actions has eight basic modules and a separation [56]. In this research, SMA springs connect each module, and the stiffness of each SMA spring is 2.92 N/mm [56]. When the temperature rises, the springs will deform to make the robot take a locomotion action. A small crawling robot was developed to imitate the movement of Caenorhabditis elegans [78]. Adjacent motion modules were connected by SMA springs, which could contract up to 50% of their extended lengths. The SMA spring actuator can produce a maximum force of approximately 0.3 N to make the robot move like a real worm. Inspired by the muscular organs and modularisation of sea stars and octopuses, Pan et al. designed a bionic spherical robot [79]. Driven by SMA springs, the robotic feet bend up and down to help the robot crawl on the ground and roll down slopes. With the excitation of 3.8 V, the SMA spring generated a driving force of 0.1 N, and the soft robot was driven at a speed of 0.133–0.147 mm/s with a maximum load of 850 g.

2.1.2. SMA Wire

The general design concept of SMA wire driving is to install an SMA wire into the main body of a soft robot. By applying the electric current, the SMA wire deforms, bending the main body and generating sliding frictions. Kim et al. designed a bionic inchworm robot (shown in Figure 4a) [80]. As shown in Figure 4b, when the temperature of the SMA wire exceeds its phase transition temperature, the SMA wire returns to its initial shape. When the temperature is reduced, the wire is under relaxation. In Figure 4c, during the 2nd time interval, the SMA wire starts to deform and arch up (at the high state), directly driving the rear foot to move forwards. At the same time, the front foot moves back. Then, in the 4th time interval, the SMA wire relaxes and recovers to its original shape, the front and rear legs spread out, and the robot advances to a certain distance.

Wang et al. developed a crawling soft robot to mimic the worm movement [81]. The main body of the robot is a thin tetragonal composite plate with two pairs of SMA wires placed on the four sides of the robot. Periodic currents are applied to deform the SMA wire, enabling the robot to turn and crawl. In a single motion cycle, the moving distance of the robot is 54 mm, which is about one-third of its own body length, and the maximum linear speed can reach 3.6 mm/s. Inspired by the movement of caterpillars, Lin et al. designed a rolling soft robot using silicone materials and SMA wires [82]. The minimum weight of the robot is only 5 g, and the SMA wires installed on both sides of the robot can provide a driving force of 10 mN. Under excitation, the SMA wire deforms and makes the robot curl into a wheel shape, and the rolling speed can reach 0.5 m/s.

From the manufacturing perspective, the spring form is more complicated than the wire form. From the structural perspective, the wire form can be flexibly bent to connect two kinematic joints, whereas the spring form needs to be fixed in a specific position and can only be extended. From the perspective of the bearing capacity, the spring form can produce a greater output force, and the spring form-enabled soft robot can bear a higher load. Therefore, in comparison with the spring form, the wire form can be employed to design a versatile soft robot to play the role of a hand, a body, and a leg to grasp, stretch and move objects [15,83,84].

In addition to spring and wire forms, SMA can also be made into other forms. The research in [85] is a bell-shaped jellyfish robot, which body is composed of eight SMA compound actuators. The deformation of SMA is controlled by voltage changes to simulate the swimming motion of a jellyfish. During 14 cyclic movements, the average thrust and power consumption of the jellyfish robot are 0.0039 N and 16.74 W, respectively, and the maximum speed reaches 5.42 cm/s. RoBeetle, designed by Yang et al., is powered by the catalytic combustion of methanol fuel with high specific energy [86]. Its four legs are made of SMA actuators. Methanol vapor flows out of the container and reaches the surface of the SMA legs. The SMA shrinks when the catalytic surface is exposed to fuel and oxygen, and expands when the fuel flow stops. The robot exhibits a maximum average stride of 2.83 mm, and a maximum average speed of 0.76 mm/s. It can carry a maximum payload of 230 mg (about 2.61 times the weight of the prototype).

As a flexible material that can be deformed and used in widespread applications, SMA still has some technical limitations, no matter whether the form is spring, wire, or other. Firstly, in the repeated phase-changing process, SMA exhibits functional fatigue, which will greatly shorten the service life of the SMA actuator and reduce the driving efficiency [87]. Secondly, most SMA actuators have hysteresis. At present, ferromagnetic shape memory alloys (FSMAs) can be rapidly deformed in a magnetic field owing to the magnetic particles contained [88]. However, the required magnetic field is high, which limits its applications.

2.2. Fluid

Fluid can be designed as an actuation mechanism that converts the pressurised energy of fluid into mechanical energy to realise the motion of a soft robot. The basic principle is to change the pressure in the fluid chamber in a soft robot, expand the channel in the robot, and therefore deform the robotic structure controllably. There are three types of fluid-based actuation mechanisms for soft robots, i.e., pneumatic, hydraulic, and pneumatic–hydraulic [89]. Fluid is a commonly used actuator having the advantages of convenient operation and control, large output power, and easy automation.

2.2.1. Pneumatic

A pneumatic actuator is a practical and versatile pneumatic device. It can operate in an antagonistic way like human muscles, and only generate traction when powered by compressed air [90]. Pneumatic utilises air pressure to generate a driving force to actuate a soft robot. It demonstrates the characteristics of quick response speed, good safety, and strong flexibility [91]. In pneumatic robots, gases are pressurised into embedded channels and elastic cavities [60]. This method generates stress at the gas–solid interface, thereby causing the structure of the soft robot to deform. It is usually assumed that different pressure levels of the gases in the internal cavity of the robot propagate fast enough so that all the gases in the structure have a uniform pressure level [92].

The control and movement mechanism of the pneumatic is shown in Figure 5. The soft robot in Figure 5a consists of an air pump, an air chamber, and two feet [93]. Figure 5b shows that the air pump inflates or deflates the air chamber for elongation or contraction. At the same time, the front and rear feet move to realise the movement of the robot alternately. Figure 5c describes that, during the above entire cycle, the air chamber completes elongation and contraction, and the two feet are anchored or moved collaboratively to realise the movement of the robot. In the 2nd time interval, the pump inflates the chamber to elongate, driving the front foot moving forwards. In the 4th time interval, the front foot is anchored, and then the pump deflates, making the chamber contract and pull the rear foot.

In applications, Ning et al. designed pneumatic devices to imitate worms to achieve a Ω-shaped crawling motion [94]. The robot is composed of eight air chambers, which are in a sawtooth shape. When inflated, the main body of the robot swells and arches upward, and then returns to its original shape after deflation. Then, a crawling process is completed. The robot can shrink the length of its body down to 40 mm (the initial length is 98 mm), the maximum moving distance is 2 mm in one cycle (0.4 s), and the average speed reaches 2.2 mm/s. In terms of materials, Terryn et al. utilised the self-healing ability of Diels–Alder (DA) polymers to design soft-handling robots, soft grippers, and artificial muscles [95]. Under the air pressure of 25 kPa, each actuator produces a gripping force of 0.25–0.32 N and the robotic hand can grab objects up to 92.8 g. If the robot hand is damaged when grasping sharp objects, it can be healed automatically at a high temperature (80 °C) through the DA reaction of the material. This characteristic expands the application field of pneumatic robots. In [96,97], some pneumatic actuation-based soft robots for pipeline-related applications were developed. By integrating a pneumatic device that has telescopic characteristics, a soft robot was designed to monitor and dredge pipelines. When it is possible to make pneumatic actuation-based soft robots smaller, the robots can be applied to medical treatments, such as treating vascular diseases. In addition, researchers leveraged the elongation characteristics of pneumatic actuators to mimic the process of biological navigation [98] and the movement of octopus tentacles [99].

2.2.2. Origami-Based Pneumatic

Conventional pneumatic actuation-based soft robots are made of soft materials (e.g., silicone), which expand and elongate themselves after being inflated. If the action of intaking air is continued, such a soft robot will deform excessively, which will affect the robot’s movement. Origami is made of folding papers or polymers. Owing to bidirectional stability, origami is a structural innovation of pneumatic soft robots. Origami itself is not flexible and can not deform, but after being folded, it will contract like a spring to avoid unconstrained deformation.

Hong et al. devised a pneumatic origami muscle actuator (POMA) for a soft robotic hand [39]. The chamber of the robot is an origami structure using a 0.2 mm polyethylene film, which can withstand air pressure of 100 kPa. The chamber’s outer layer is a spandex fabric that can be stretched to 200% of its original length, which helps the inner chamber bend. The overall weight of the soft glove is about 200 g, and it can generate a constant driving force and transmit it in various grasping states. Yu et al. presented a pneumatic crawling soft robot based on Miura-Ori [100]. It consists of two pneumatic foldable actuators that act as legs. Each actuator is made of soft materials and a paper skeleton, the total weight is about 2.52 g, and the load can reach 10 g. The working pressure of the PFA is from −10 kPa to 6 kPa, and the maximum strain under positive and negative pressure is 0.19 and 0.43. By controlling the unfolding and folding sequences of the two PFAs, the linear and turning motions of the crawling robot are realized, and the speeds are about 5 mm/s and 15°/s.

Since the origami robot is actuated by using the crease memory of a paper, the movement of the robot can only depend on the folding mode and cannot be changed. Thus, it has low utilisation. Moreover, repeated folding can also make origami vulnerable.

2.2.3. Hydraulic

A hydraulic actuation makes use of liquids to flow directionally with the excitation of the electric currents. Even though this actuator can generate a greater driving force, it has not been sufficiently developed due to the restrictions of the volume and weight of the hydraulic device. In the robotic field, Ueno et al. developed a hydraulic robot that utilises electrically conjugated fluid for actuator design [101]. It consists of a rubber bellow, a pair of needle-ring electrodes, a shrunk water tank, and a grounding foot. The functional fluid undergoes a directional flow with a high voltage of more than 4 kV and an average power consumption of about 77.5 mWs, causing the expansion of the bellow and the movement of the tilting for the soft robot. Its weight is only 1.9 g, but it exhibits a superior migration speed capability of 5.2 mm/s on a horizontal acrylic surface. In [20], a hydraulic robotic arm was developed with the inspiration of an octopus. The robotic arm can simulate the elongating, shortening, bending, and stretching movement of an octopus. The arm has good stretch properties and can be stretched from 100 mm to 160 mm, a 60% increase in length. The diameter is also reduced by 20% (from 20 mm to 16 mm). Additionally, Lu et al. devised a soft strain and pressure sensor based on a liquid conductive metal, which can be easily integrated into a typical soft pneumatic actuator to realise a sensing function [102].

2.2.4. Pneumatic–Hydraulic

The combined pneumatic–hydraulic actuation mechanism has been mostly used for underwater robots. By taking the changes of the internal and external air pressures, a soft robot can absorb or discharge water to make the robot dive or float up. Utilising the elasticity of the dielectric elastomer and pneumatic–hydraulic combination, Godaba et al. designed a gas–liquid joint-driven soft jellyfish robot [103]. The overall weight of the robot is 261 g, and the load capacity is 241 g, so it is equivalent to its own weight. The robot has two chambers. The internal chamber is made of a dielectric elastomer membrane (actuation pressure is 38.97 kPa). When the air pressure in the internal chamber decreases, the water enters the external chamber and the robot descends. When the air pressure in the internal chamber increases, the water is discharged and the robot rises. Through the descending–rising actions of the robot, specific functions are fulfilled. In 20 s, the robot can generate a maximum buoyancy force of about 0.15 N and rise to 18 cm with an average speed of 9 mm/s.

In comparison with other types of actuators, fluid actuators can generally output larger forces and respond to external factors much faster. Nevertheless, fluid actuators require external devices for control and execution. The external devices are large in size and heavy in weight, so the fluid-driven soft robots are difficult for miniaturisation. At present, fluidic actuators are mainly used in semi-automated soft robots like grippers, crawlers, or underwater robots [104,105,106,107], and most of them are pneumatically driven. A trend is to make fluidic actuators to achieve a centimetre level, with which soft robots can be developed to be fully automated.

2.3. Electroactive Polymer (EAP)

EAP is a specific polymer having the ability to change its sizes and shapes by stimulating an electric field. By converting the elastic potential energy into mechanical energy, EAP can enable soft robots to move. EAP exhibits the advantages of large deformation, high energy density, and a lightweight and compact structure. It is therefore a promising driving material to actuate soft robots [57,59,108]. There are different actuation mechanisms of EAP, including polarisation, molecular phase transition, and mass/ion transport. The polarisation mechanism is driven by dielectric elastomers and piezoelectric polymers. The molecular phase transition mechanism is activated by shape memory polymers and liquid crystal elastomers. The mass/ion transport mechanism is mainly used to activate some specific materials, such as gels and conductive polymers [109]. From the material perspective, EAP can be generally divided into two categories according to their actuation mechanisms, that is field-activated [110] and ionic [111]. Dielectric elastomers, liquid crystal elastomers, electro-strictive polymers, polymer electrets, and ferroelectric polymers belong to the field-activated category. Ionic polymer–metal composites, ionic gels, carbon nanotubes, and conductive polymers belong to the ionic category [112,113]. Both of the two categories show remarkable performance in their respective applications. Below are some EAPs that are commonly used in the development of soft robots.

2.3.1. Dielectric Elastomer

Owing to inherent flexibility, large strain, high efficiency, high energy density, and fast response, dielectric elastomer actuators (DEA) have been widely used in designing soft robots [114,115]. DEA is a kind of motion-generating material, which is similar to human muscle in terms of force, strain (displacement per unit length or area), and driving pressure/density [116,117]. This material mainly includes silicone and acrylic. The former has a faster response speed, whereas the latter can produce a greater deformation [118]. In addition to the material preparation produced by the planar method, DEA can also be realised by 3D printing, and its mechanical properties are adjustable [119].

The development of worm-like soft robots is taken as an example for illustrating DEA. Gu et al. leveraged the characteristics of DEA to develop a wall-climbing soft robot (Figure 6a) [120]. In Figure 6b, the elastomeric body of the robot can be expanded and bent at a high voltage (6 kV). When the voltage is lost, the robotic body will contract to achieve a forward motion. The robot can obtain a fast climbing speed owing to the rapid periodic deformation of the elastomeric body. It can be seen from Figure 6c that, when DEA is at a high state, the rear foot is anchored but the front foot moves forward. The two feet exchange the motion state when DEA is at a low state. In a movement cycle, at the first stage (from 2nd to 4th time interval), the front foot is driven by DEA deformation, and at the second stage (from the 4th to 6th time interval), the rear foot is pulled by the restored DEA.

There are various applications of EDA to actuate soft robots. Examples as illustrated in [121,122] are bionic swimming robots that can conduct underwater exploration. At actuation voltages of above 3 kV, their swimming speed can reach more than 1.5 mm/s. The research in [123] shows a micro robot that mimics the annelid. The robot has a small size (20 mm in diameter, and 45 mm in length) and a light weight of 4.7 g. The maximum moving speed of the robot is 2.5 mm/s, and the load capacity is more than twice its own weight. Multi-legged crawling robots in [124,125] are attractive because of their high load capacity (more than 900 g) and fast speed (about 9 mm/s). The rolling robot presented in [126] possesses the advantages of not only a higher rolling speed (36.27 mm/s on average) but also a larger speed–mass ratio (about 41.22 mm/s·g). Therefore, it has good environmental adaptability. Additionally, the bionic caterpillar robot in [127] was designed with an integrated artificial nervous system.

2.3.2. Liquid Crystal Polymer

Liquid crystal polymers are materials that combine the properties of polymer elastomers and liquid crystals. They exhibit large, rapid, and reversible changes in shape after applying certain external stimuli (an example is illustrated in Figure 7 [128]). Meanwhile, they are between ordered crystalline solids and isotropic liquids. While maintaining a certain degree of molecular order, liquid crystals have similar fluidity as liquids [128]. These external stimuli include temperature, light irradiation, or electronic devices (heating by resistance or direct piezoelectric response) [129,130].

The most important application of liquid crystal elastomers for soft robot design is light-driven robots. Rogoz et al. presented a millimetre-scale snail robot with liquid crystal elastomer materials [131]. The position where the elastomer material deforms and bends changes with the position of laser irradiation. After one cycle of laser irradiation, the elastic body moves to a certain distance at a speed of 0.3 mm/s, and then it will proceed to the next cycle. Therefore, the snail can continue to wriggle. A wave-shaped robot, which was reported in [132], curls and crawls like a caterpillar under the light. After the robot absorbs the green light, thermal actuation is triggered and the length shrinks to 20% of the original length, resulting in a total tensile stress of 2.5 kPa. The uniqueness of the robot is its small size (14.8 mm in length) and it can pass through a gap of 0.9 mm. In addition, with a CW power of 2.5 W and 0.4 Hz scanning beam frequency, the robot walks for a distance of 14.2 mm at an average speed of 0.24 mm/s within 59 s. A transporter robot presented in [133] fuelled by light consists of four light-driven liquid crystalline polymer films that play the roles of legs. The robot has an overall length of 2 cm and a weight of 20 mg. In one motion cycle, it can move forward by 4 mm at a speed of 0.5 mm/s. It is a fully light-responsive transporter robot with controlled multi-freedom locomotion and the capability to deliver loads (about 50% of its weight). Wei et al. designed a bionic hand by assembling a photonic film in a liquid crystal elastomer material. It can trigger finger bending and deformation under visible light [134]. It can be imagined that a soft robot that can move quickly and continuously through light-triggered photonic actuators and liquid crystal elastomer photonic films.

2.3.3. Ionic Polymer–Metal Composite (IPMC)

The IPMC material [135] consists of a fluoropolymer ion exchange membrane, which is sandwiched between two conductive precious metal layers (such as platinum, gold, etc.). The mechanism of the IMPC material is illustrated in Figure 8a. The two main characteristics of IPMC are low activation voltage and large bending strain caused by the movement of cations in the polymer matrix. IPMC demonstrates strong performance in a variety of sensing and actuating applications [136]. IPMC actuators have various advantages, such as low driving voltage, relatively fast response, large displacement, the ability to be activated in water or wet conditions, and easy miniaturisation [137].

Carrico et al. printed a crawling robot with IPMC [138]. By applying voltage (<5 V), an IPMC-based actuator stretches and contracts, driving the robot to move along the pipe like a caterpillar or an inchworm (see Figure 8b). IPMC-based actuators have potential applications in the field of underwater robots. The research in [139] presented a fish-shaped robot that can imitate fish swimming in an S shape, and control the speed by the voltage frequency. When the input voltage frequency is 4.0 Hz, the speed is the fastest to reach 7 mm/s. The research in [140] put forward a bionic jellyfish robot, and the curve shape of the IPMC-based actuator can be successfully applied to generate smooth curve motions through heat treatment. When the excitation frequency is 1 Hz and 2 Hz, the vertical floating displacement of the robot is about 0.15 mm and 0.3 mm, and the floating speed is 0.057 mm/s and 0.045 mm/s. Li et al. designed a small capsule underwater robot [141]. The pectoral fin and caudal fin are made of IPMC. Under the condition of a symmetrical square wave with a driving voltage of 3 V and a frequency of 1.5 Hz, various swing combinations of the fins can be realised to drive the movement of the robot. The robot can swim forward at a speed of 4.83 mm/s and turn around at a speed of 4.5–5°/s.

2.3.4. Ionic Gel

Ionic gel is a kind of reticular polymer that can expand in water and is affected by the pH and electric field after being put into aqueous solutions [142] (examples are illustrated in Figure 9). Through the interaction between the polymer and water, the shape and volume of ionic gel are able to change, thereby obtaining the functions of actuators or sensors. According to the formed gel network, gels are divided into two types: chemical gels and physical gels. While in chemical gels, the cross-linked gel network is formed by irreversible covalent bonds, resulting in gels with high elastic modulus and low strain tolerance. In physical gels, cross-linked networks are reversibly formed through non-covalent interactions, such as hydrogen bonding, π stacking, and electrostatic interactions. Since the non-covalent bonds of physical gels can be destroyed by the application of external heat and reformed upon cooling, the gels exhibit sol–gel transitions. At each temperature, the gels experience transition from a gel state to a sol state. However, the sol state is temporary and will return to the gel state after cooling [143]. A soft tactile sensor composed of ionic gel and elastomer was presented and the relationship between its impedance with current frequency was illustrated by Hara et al. [144]. In future work, they will integrate the pressure sensor for soft robot design.

In addition, there are some other polymers applied to make soft robotics. In reference [145], a tank-shape climbing robot (Tankbot) was designed using non-patterned flat and elastomer adhesive treads that operate as tracks. These palm-sized robots are lightweight (60–150 g), fast (up to 12 cm/s), and relatively energy efficient (efficiency up to 0.18). Driven by the motor in the robot, the elastomer tracks enable the Tankbot to climb over obstacles at 16 mm height. Sun et al. demonstrated a kind of artificial muscle polymer called a twisted coil actuator (TCA) [146]. The actuator can provide opportunities for developing untethered soft robots or robotic hands due to its excellent bending performance and programmable motions. A single TCA weighs 0.3 g and has a maximum tensile force of 0.78 N (stress is 8.59 MPa). It can overcome a peak friction force of 0.4 N and achieve a free stroke of 48%. In [147], a rolling robot (210 g total weight) composed of 6 curved fluidic elastomer actuators was put forward. With the catalysed decomposition of hydrogen peroxide, the internal air pressure increases to 24.1 kPa, causing the elastomer to expand and deform. Under the alternate action of six fluidic elastomer actuators, the robot rolls forward. At each step, the robot rolls forward for approximately 40.2 mm.

In general, an EAP-based soft robot exhibits the advantages of large deformation, high energy density, lightweight and compact structure, etc. However, the robot also has the following technical difficulties:

(1)

It requires a high voltage above kilovolts with high power consumption;

(2)

It has a slow response time, which needs much time to deform;

(3)

Because of the softness of these polymers, their load-bearing capacities are less than other actuations.

2.4. Electric and Magnetic

Electric can convert electrical energy into other forms, such as mechanical and magnetic, leading electrical soft robots to move. Electrical robots possess the ability to move and achieve functions under wireless control. An electric motor can be used as a direct actuator, as determined in [148,149]. After receiving motion signals, the motor drives the grippers or limbs of robots directly, and, ultimately, the robots will climb trees or swim under water.

Moreira et al. developed an inchworm robot capable of two-anchor motion that can crawl on flat and inclined surfaces [150]. It is a three-segment body connected with 3D-printed lever arms and two servo motors. This mobile robot is able to carry supplies for search and rescue missions. The speed of the robot can reach 2.54 cm/s, and it can crawl on the horizontal plane and the slope of 19° with a load of 0.45 kg. Another form is a wall-climbing robot designed based on electrostatic adhesion mechanisms [151]. When a high voltage of 3 kV is applied to the pad electrodes of the robot, each pad can generate a holding force of about 3.2 N on the glass surface, supporting the robot to complete straight-line crawling and turning. The maximum speed is 2 cm/s, and the maximum single turning angle is 30°. By controlling the adhesion force of each foot, the robot can climb similarly to a real gecko. In [152,153], electric motors were used to design electric-motor-driven soft grippers.

A magnetic actuator can penetrate most materials and control the direction/scale of the magnetic field efficiently and accurately. In the magnetic field, the magnetic particles on the robot will deform as the magnetic field changes, such as elongation, contraction, bending, or other forms, driving the robot to move. This actuation has the advantage of being wirelessly controlled by a remote untethered magnetic field without physical contact.

In [154], a modular and untethered robot named Limpet II (weight: 450 g; size: 250 × 250 × 140 mm) was developed. It utilises the hybrid electromagnetic module (EMM) as its core module and a 450 mAh 11.1 V Li-Po battery as power, allowing the capabilities of adhesion and locomotion. The maximum force achieved by the EMM is 1.8 N. With an actuation current of 1500 mA, a single module can lift objects with a mass below 184 g. As a sensor, it also can be used for inspecting and monitoring offshore structures and nearby conditions. Niu et al. introduced a 54 mm long magnet worm-like robot actuated by permanent magnetic patches [155]. The magnets are embedded in the robotic feet, and each adjacent magnet has opposite magnetisation directions. When the speed of the driving magnet reaches 150 mm/s or more, the robot has a stride of about 6 mm and an average speed of 38.8 mm/s. If the robot prepares to move, another permanent magnet will generate a moving magnetic field. Joyee and Pan demonstrated an unconstrained full 3D-printed soft robot with a multi-modal motion capability involving linear crawling and turning locomotions [25]. It is very light (only 0.23 g), but has a carrying capacity ten times greater than its own weight. Affected by the external magnetic field, the robot changes its motion state and delivers medicines to a designated location at a speed of 3.1 mm/s.

Electrical and magnetic actuators have the advantages of quick and accurate reactions, low cost, remote control, and easy maintenance. Nevertheless, the technical difficulty of the actuators is the same as their drivers, motors, or coils. Meanwhile, the actuators are not scalable when compared with SMA and EAP. Furthermore, electrical actuators exhibit limited flexibility and high rigidity while magnetic actuators need an external strong magnetic field.

2.5. Light

Light-driven soft robots are designed based on the light-responsive effect of polymers, such as liquid crystal elastomers, liquid crystal gels, and some ionic polymers, to deform and move their bodies [156,157,158]. This technology, which is to convert light into mechanical energy, is found in possession of some distinct functions, from cleanliness, directional motility, and sensing, to wirelessly controlled interactions between humans and environments [159]. It is foreseeable in the future that the technology will provide novel opportunities for the development of soft robots in space.

Huang et al. fabricated a series of transparent actuators using local surface plasmon resonance semiconductor nanocrystals that respond to infrared light [160]. The actuators have the advantages of wireless and remote control on camouflage soft robots, and significant deformation (curvature up to 0.66 cm⁻¹) under near-infrared light irradiation (200 mW·cm⁻²). It is realised by preparing three semiconductor nanocrystals as photothermal conversion agents to construct a photo-actuator. Wang et al. demonstrated a smart thin-film composite with perception and motility in the field of light-driven robots [161]. The robot is capable of coordinated actuation, proprio–exteroception, and communication. Under visible light with the irradiation of 100 mW·cm⁻², the maximum driving deformations of the light-driven film on tempered glass, printing paper, and sandpaper are 0.49, 0.472, and 0.443. In addition, the temperature and light are controllable by electric signals, driving the robot uninterruptedly. Using the light-driven film, kirigami soft robots with multitasking and information feedback functions, such as walkers, robotic hands, and centipedes, can be designed.

As a new actuation mechanism being developed in recent years, light-driven robots have some aspects for further optimisation, including: (1) this kind of actuator requires a high-intensity laser to make the photosensitive part of the robot heat up and move around, which consumes a lot of electric energy during this period, so the energy conversion efficiency needs to be improved; and (2) due to the light nature and soft material, light-driven robots can be miniaturised, but they can also result in a small output force, small load bearing, and limited application scenarios.

2.6. Sound

A sound-actuated robot senses external sound through the internal voice coil to generate vibrations and performs modular communications for monitoring. This actuator consists of coils, permanent magnets, and circuits. Its working principle is that the coil collects external sound waves and vibrates, cuts the permanent magnet to generate current, and the coil carries the current to drive the corresponding sensor.

Nemitz et al. presented a 160 mm long bionic robot driven by 3D printed voice coil actuators, relying on electromagnetic coils [162]. Powered by a 7.4 V battery, the robot can move 25 body lengths in 12 min. A series of electromagnetic coils are embedded in the robot (just like annelids) to transmit vibration and realise surface movement. It cannot only be used for sound input and output but also to achieve the purpose of communications within functional modules of the robot. This low-cost robot shows great potential in applications, such as pipeline monitoring and search and rescue scenarios.

The issues facing sound actuations are that it is susceptible to external noise, and the stability is low. Moreover, its output driving force is still not high enough, which limits its mobility.

2.7. Chemical

A chemical reaction actuator is to use the mechanisms of chemical combustion or explosion to convert chemical energy into mechanical energy for actuation purposes. These materials are mostly gases with a high-volume energy density, such as alkanes and other hydrocarbons. During the process of reaction, a large amount of gas is generated in such a soft robot. After the expansion of gas, the air pressure rises and ejects the robot to a designated position.

Bartlett et al. designed a soft robot that can be activated by butane combustion to complete unfettered jumping movements [163]. This robot autonomously jumps up to 0.76 m high (six body height) and demonstrates directional jumping of up to 0.15 m (0.5 body length, 20% of jump height) laterally per jump. In addition, Shepherd et al. presented a three-legged robot that burns methane and explodes to jump [164]. It uses electric sparks to ignite methane gas to achieve rapid combustion, releases a large amount of gas (71 kPa), and causes the internal gas network to make a jump action. In each explosion, the robot jumps to a height of 30 cm, and the jumping speed reaches 3.6 m/s. A chemical catalytic decomposition of a pneumatic robot jointly developed by Wehner et al. is controlled by microfluidic logic [165]. It automatically adjusts the fluid flow to catalytically decompose the vehicle’s single propellant fuel supply. The gas (approximately 50 kPa) produced by the decomposition of the fuel expands the fluidic network downstream of the reaction site (expected to expand to 160 times its original volume), thereby causing movements.

There are also other innovative chemical reaction actuation mechanisms applied to soft robots, such as electrolysed water actuators. These types of robots use electrolysed water to decompose hydrogen and oxygen, push the robots to swim forward when exhausting backwards, and change the speed by adjusting their voltage. Nevertheless, the chemical reaction actuation mechanisms exhibit the following limitations:

(1)

Chemical soft robots include reaction devices, which are difficult to miniaturise;

(2)

Some robots driven by explosions are designed for jumping purposes, but the positioning accuracy of the robots is not high.

3. Applications

Compared with traditional rigid robots, soft robots have less complexity of mechanical design or algorithms and more freedom of movement, and thereby they can better adapt to external environments [5,62]. Therefore, soft robots demonstrate more application potential. The early applications of soft robots are mainly on bionics and motion [14]. With more research on deformable materials and increasing uses in soft robot designs, soft robots are increasingly applied to wider applications, such as medicine and rehabilitation, control and exploration, industry and production, and interaction and contact [4].

3.1. Bionic and Motion

Bionic and motion are the basic applications of soft robots. Inspired by observations of animals’ body structures, muscles, and bones, researchers designed bionic soft robots to mimic crawling, jumping, and rolling [14]. Biomimetic objects are generally creatures with simple movement processes that are easy to imitate, such as inchworms, caterpillars, jellyfish, and octopuses. In this way, soft robots can use flexible structures to move in natural environments and quickly adapt or interact with the environments [62].

A type of soft robot can be designed according to the working principle of an inchworm [81]. The inchworm utilises the stretching of abdominal muscles to drive the front and rear legs to cyclically grasp and move to achieve an Ω-shaped continuous crawling motion. The used materials are required to be soft and deformable, and the driving method generally adopts SMA and EAP. When an electric current is applied, the actuating material deforms, mimicking the alternating movement of the front and back legs of the inchworm. SMA wires are longitudinally embedded in a soft polymer to mimic the longitudinal muscle fibers that control the abdominal contraction of the inchworm during movement. Another example is a DEA-driven bionic inchworm robot, which uses reinforcing fibers to suppress the deformation of the dielectric elastomer in other directions [166]. The deformation and bending of the dielectric elastomer can be effectively converted into forward motions. This robot’s average speed can reach 1.1 mm/s, and each motion cycle can advance 3.8 mm.

A type of soft robot can be designed according to the working principle of a caterpillar. This kind of bionic robot simulates the crawling action of caterpillars by driving friction between the moving body and the ground. The materials used are generally polymers, and the actuation methods include fluid, light, and electromagnetic methods. Sheng et al. designed a robot driven by pneumatic actuators. The biological structure and morphological movement of caterpillars are simulated by the synergistic movement of the deformation of the bellows body and the friction feet. The robot can move at a speed of 1.05 cm/s (0.16 body length per second) [167]. A light-driven bionic caterpillar robot realises rubbing motion by light-induced deformation of heated liquid crystal elastomers [23]. This robot can adapt to various unstructured environments, and its movement speed can reach 0.25 mm/s.

A type of soft robot can be designed according to the working principle of a jellyfish. A jellyfish swims by contracting and expanding epidermal muscle fibers as well as the thrust of water currents. Researchers used elastic polymers to mimic the muscle fibers of jellyfish. By applying an electric current or changing the internal pressure, the polymer is deformed to simulate the swimming of the jellyfish. Christianson et al. used DEA materials to imitate the epidermal muscle fibers of jellyfish, and employed fluid electrodes to apply voltage and deform them, so the average swimming speed of such a designed robot could reach 3.2 mm/s, and it could work underwater silently [35].

A type of soft robot can be designed according to the working principle of an octopus. The arms of octopuses can achieve some fine and complex movements by changing the shape of the arms by shortening, elongating, bending, twisting, and deforming the bodies of octopuses [14]. It can change the shape of the bionic octopus robot and grasp objects by using elastic polymers to simulate the muscles of octopus brachiopods and drive them with fluids or electric currents. For instance, a robotic arm was designed to mimic octopus brachiopods with embedded cables that replicate the function of octopus brachiopod muscles for propulsion and object grasping [168].

3.2. Medical and Rehabilitation

The materials used in medical soft robots can achieve good biocompatibility with human bodies and tissues. In medical and rehabilitation, soft robots can be used in minimally invasive surgery, drug delivery, rehabilitation training, and wearable devices [2]. According to application scenarios, medical robots can be divided into two categories: implantable and wearable [13]. Nevertheless, there exist guarantee conditions, such that the human body will not produce allergic, immune, or rejection reactions to the soft robot, and the robot will not cause secondary damage to human tissues while performing its functions. Implantable soft robots can be used in minimally invasive surgery and drug delivery, and the driving method used is often magnetically driven. For instance, a minimally invasive surgical robot, which is with magnetic-driven steering and navigation capabilities, can be reduced to less than a few hundred microns in diameter and can travel through blood vessels [169]. Another example is a drug delivery robot, which can move and turn in the intestinal model under the action of a magnetic field to demonstrate the virtual drug delivery process. Its average speed can reach 3.1 mm/s [25]. Wearable robots can help our limbs perform rehabilitation training or directly act as prosthetic limbs, which require greater output forces and are suitable for fluid-driven methods. In oral rehabilitation, a pneumatically driven robot can achieve an elongation rate of 800–1000% after driving, which is suitable for sensitive areas of the human body and assists in the treatment of patients with mandibular movement disorders [170]. In hand rehabilitation, a low-cost pneumatic hand rehabilitation robot helps fingers perform flexion/extension training with a maximum output force of 110 kPa [38]. In ankle–foot rehabilitation, a wearable robot driven by pneumatic artificial muscle actuators helps humans with orthopedic ankle and foot or assist walking [171].

3.3. Control and Exploration

On the basis of motion, more soft robots will be applied to the motion control of robots and the exploration of unknown environments to perform tasks that rigid robots are unable to accomplish [8]. By changing the direction, speed, and shape of their motion, soft robots can quickly adapt to and explore their environments. They will either be applied to flexible production lines to participate in industrial production or be arranged in unreachable environments as detectors for scientific research. For instance, a pneumatic robotic hand can change the opening and closing of the gripper and the intensity of the gripping force to grip objects [39]. A small bionic robot can disguise itself as a caterpillar, an inchworm, and so on, through the change of shape to pass through a narrow space [172]. A crawling robot with four legs changes the posture of the four legs to realise dynamic turning [173]. Underwater robots change the pressure inside and outside the body to dive into the bottom of the water to explore [107]. Jumping robots use explosion and combustion reactions to advance over obstacles [164].

3.4. Industry and Production

In industrial production, rigid robots are prone to be damaged when colliding with objects due to their high hardness. Nevertheless, the material of a soft robot can be in flexible contact with objects, eliminating stress concentration and reducing potential damage [174]. In addition, soft robots exhibit good expandability and can completely wrap and cover contacted objects. The requirement for the output forces of soft robots by industrial applications is greater, and the actuators to meet the requirement are fluid and electric, and the shape is also mainly based on robotic carriers and robotic hands [175]. For instance, a quadruped soft robot, which uses deep reinforcement learning to train the walking gait of the robot, was designed as a transportation tool in the industry [176]. Another example is a flexible soft robotic gripper, which establishes an extreme learning machine by identifying and grasping shapes [177]. It can help improve the control accuracy in industrial production.

3.5. Human–Robot Collaboration

Soft robots can interact with human operators for collaboration by changing their shapes. A soft-growing pneumatic navigation robot designed by Greer et al. is capable of interacting and turning during human–robot collaboration when encountering obstacles, and finally planning a navigation path [178]. In terms of human interactions, due to the soft and easily deformable materials used, soft robots can safely interact with humans, so humans suffer less damage and load [179]. Soft robots can cooperate with humans as assistive robots (also known as cobots), such as a ten-fingered robotic hand [83]. The robot integrated the embedded SMA and a piezoelectric transducer (PZT) flexure sensor, enabling real-time interaction with human operators for precise control of working objects.

4. Further Discussions

In this paper, the movement mechanisms, characteristics, applications, and technical limitations of actuators for soft robots are discussed. In this session, based on some indicators, these actuators are further discussed to summarise their advantages, disadvantages, and application potentials (to facilitate readability, the advantages, disadvantages, and application potentials of each actuation mechanism are enumerated in Table 1, and key parameters of typical soft robots are shown in Table 2). Meanwhile, each actuation mechanism is evaluated quantitatively in terms of the output force/carrying capacity, miniaturisation, lightweight, efficiency, response speed, and technology maturity (see Figure 10).

In Figure 10, output force/carrying capacity indicates how much driving force a soft robot can generate and how much load it can bear; miniaturisation means how to reduce the size of a soft robot as much as possible; lightweight focuses on whether the robotic weight can be decreased; efficiency reflects the energy conversion rate of a soft robot; response speed shows the maneuverability and sensitivity of a soft robot; and technology maturity describes the current application status of the actuation and whether the technology is mature. For the above indicators, the larger the values, the better the performance of a soft robot. These six indicators interfere with each other, and jointly affect the soft robot’s movement mechanism and function realisation.

SMA, fluid, and EAP are the most widely used actuators in various applications (see Table 3). Compared with SMA and EAP, Fluid has the advantages of a larger output force, stronger load, higher efficiency, and faster response speed. Nevertheless, the above actuators are difficult to realise miniaturisation and lightweight design.

Other actuators are recently developed, but they are making significant progress. Part of them, electric and magnetic, have obvious advantages in miniaturisation and wireless control [61]. In recent years, many soft robots actuated by magnetic are designed, like C-balls, connected and driven via the magnet force [180], thermo-magnetically actuated hybrid soft millirobots based on 3D printing technology [181]. The advantage of miniaturisation and wireless control shows the promising application prospects of magnetic actuation.

For light actuation, the biggest advantage is that its mass and volume can be made smaller. Light-driven robots are within 20 mm in length and weigh around 20 mg. Nevertheless, they need a high-intensity laser as a heat source, which consumes high energy, and the material is soft. The output force, carrying capacity, and deformability are also small. Sound actuation is at a medium average level, and the advantages are not obvious. Nevertheless, in aspects of miniaturisation and lightweight, it has a bit of an advantage. Chemical reaction actuation can provide a great instantaneous force by explosion or combustion, and it also has a good performance in response speed and efficiency.

From the perspective of the output force and carrying capacity, the actuations of fluids and chemicals have obvious advantages. They are able to carry loads more than ten times their own weights and have a fast move speed (see Table 2). In terms of miniaturisation, soft robots actuated by EAP and light like in [23,126,133] have a relatively small size (the length can reach 20 mm).

Response speed and efficiency are the most important factors of soft robots that researchers are concerned about. Because hydrocarbons react violently and release heat quickly during the combustion process, chemical actuation can obtain energy efficiently. For instance, the response speed for such a robot is able to reach 3.6 m/s [164], faster than other actuations. Moreover, the efficiency is also adjustable so that it can also arrive at a middle-to-high level. SMA, fluids, and magnetic require less energy, but can also reach about 5 mm/s in terms of response speed [53,82,95,182]. For EAPs, the voltage required of electrical polymers is relatively high, especially for dielectric elastomers, which could require a high voltage level of more than 3 kV [125,126]. Ionic polymers do not require high-voltage excitation, but the response speed is much lower than that of electrical polymers.

For efficiency, fluids-actuated robots can make full use of air pressure to bear a load ten times greater than their own weights. Simultaneously, the robots have relatively fast response speeds, so the efficiency is higher than other types of actuations. Robots actuated by SMA, electric and magnetic, and sound and chemical, have low energy consumption, so they can be put in the second echelon in the rank of efficiency. Finally, the actuation of EAP needs a high excitation voltage consuming much energy and light-driven has low output power, so the efficiencies of EAP and light are lower than that of others.

Soft robots have more flexible degrees of freedom and continuous deformation capabilities, which enable them to reach special locations untouchable by humans. At the same time, soft robots also demonstrate good environmental adaptability. They can operate in extreme temperatures, high pressure, thin air, heavy pollution, strong radiation, etc. The continuous discovery and application of new flexible materials [183,184], as well as the combination of various sensors [185,186] and actuations, have greatly enriched the types of soft robots. Materials with a softer texture, lighter weight, and faster deformation play an increasingly important role in the shape, volume, function, speed, and aesthetics of the robot. Low-latency sensing technology and sensitive driving speed are beneficial for improving the rapid response of soft robots.

Table 1. Comparison of various actuation mechanisms.

Actuations	Advantages	Disadvantages	Applications
Shape Memory Alloys (SMAs) [187,188,189,190]	High thermal stability. Recoverable strain. Strong corrosion resistance. Easy to automate. Able to be additively manufactured.	Slow response. High energy consumption and low efficiency. Affected by temperature.	Aerospace, automotive, biomedicine, and robotics.
Fluids			Grabbing devices such as robotic hands and robotic arms. Swimming robot.
Pneumatic [191,192,193,194]	Fast response. High security. Strong flexibility.	Need external air pumps. Difficult to miniaturize.
Origami-based Pneumatic [100]	Low cost. Light weight. Quick response. Able to be programmable.	Easy to be vulnerable. Low reuse rate.
Hydraulic [107,195]	Good driving force.	Large weight. Complex structural elements. Inconvenient to maintain.
Pneumatic-hydraulic [103]	Both pneumatic and hydraulic advantages.	Difficult to miniaturize.
Electroactive polymers (EAPs)			Robot gripper. Creeping, underwater, aerial robots. Soft actuators such as artificial muscles.
Electronic [196,197,198]	High energy density. Quick response. Long operating time. Able to perceive itself.	High drive voltage. Prone to aging and failure.
Ionic [10,199]	Low driving voltage. Large bending displacement.	Response time is longer than electronic. Small output force.
Electric & Magnetic [200,201,202]	Automatic control. Fast response. Adapt to enclosed areas.	Large external coil that generates the magnetic field. High power consumption. Small controllable area.	Drug delivery and surgery. Robot crawling device, swimming device, and micro-pump.
Light [128,133]	Able to be controlled wirelessly and remotely. Easy to miniaturize. Have access to human–computer interaction.	Need high-intensity light. Low light-to-heat conversion efficiency of actuation material.	Biomedical applications, multi-modal motion of underwater robots
Sound [203]	Transfer energy through vibration. In-module communication.	Vulnerable to external unrequired noise. Limited mobility.	Sound monitoring, communication, and search and rescue.
Chemical [164,204]	High energy density. Quick response. Untethered. Greater obstacle avoidance capability.	Need to be refueled regularly. High requirements for impact resistance of materials.	Robotic movement and directional jumping.

Table 1. Comparison of various actuation mechanisms.

Actuations	Advantages	Disadvantages	Applications
Shape Memory Alloys (SMAs) [187,188,189,190]	High thermal stability. Recoverable strain. Strong corrosion resistance. Easy to automate. Able to be additively manufactured.	Slow response. High energy consumption and low efficiency. Affected by temperature.	Aerospace, automotive, biomedicine, and robotics.
Fluids			Grabbing devices such as robotic hands and robotic arms. Swimming robot.
Pneumatic [191,192,193,194]	Fast response. High security. Strong flexibility.	Need external air pumps. Difficult to miniaturize.
Origami-based Pneumatic [100]	Low cost. Light weight. Quick response. Able to be programmable.	Easy to be vulnerable. Low reuse rate.
Hydraulic [107,195]	Good driving force.	Large weight. Complex structural elements. Inconvenient to maintain.
Pneumatic-hydraulic [103]	Both pneumatic and hydraulic advantages.	Difficult to miniaturize.
Electroactive polymers (EAPs)			Robot gripper. Creeping, underwater, aerial robots. Soft actuators such as artificial muscles.
Electronic [196,197,198]	High energy density. Quick response. Long operating time. Able to perceive itself.	High drive voltage. Prone to aging and failure.
Ionic [10,199]	Low driving voltage. Large bending displacement.	Response time is longer than electronic. Small output force.
Electric & Magnetic [200,201,202]	Automatic control. Fast response. Adapt to enclosed areas.	Large external coil that generates the magnetic field. High power consumption. Small controllable area.	Drug delivery and surgery. Robot crawling device, swimming device, and micro-pump.
Light [128,133]	Able to be controlled wirelessly and remotely. Easy to miniaturize. Have access to human–computer interaction.	Need high-intensity light. Low light-to-heat conversion efficiency of actuation material.	Biomedical applications, multi-modal motion of underwater robots
Sound [203]	Transfer energy through vibration. In-module communication.	Vulnerable to external unrequired noise. Limited mobility.	Sound monitoring, communication, and search and rescue.
Chemical [164,204]	High energy density. Quick response. Untethered. Greater obstacle avoidance capability.	Need to be refueled regularly. High requirements for impact resistance of materials.	Robotic movement and directional jumping.

Table 2. Summary of the characteristics of the different soft robots.

	Actuation	Speed (m/s)	Source of Energy	Weight (g)	Carrying Capacity (g)	Output Force	Max Displacement per Step	Dimension	Refs.
SMARoll robot	SMA	0.13	<10 V DC	191.2	-	-	50 mm	100 × 50 × 8 mm3	[53]
GoQBot	SMA	0.5	<10 V DC	5–6.2	-	0.05 N	25 cm	100 mm (Length)	[82]
Inchworm robot	Fluid	0.0022	Air pressure	-	50	0.5 N	2 mm	98 × 25 × 17 mm3	[94]
Tube climber	Fluid	0.006	Air pressure	98	1381	13.8 N	2.4 cm	70–90 mm (Length)	[96]
Jellyfish robot	Fluid and EAP	0.009	Gas–liquid pressure difference	261	241	0.15 N	18 cm	140 mm (Height) 59 mm (Radius)	[103]
Annelid robot	EAP	0.0053	3 kV DC	10.3	-	2 MPa	9 mm	170 mm (Length)	[125]
RSR	EAP	0.03627	3.2 kV DC	0.88	-	-	145.09 mm	24.83 mm (Radius)	[126]
Multi-material robot	Magnetic	0.0031	Magnetic field	0.23	3	-	5.5 mm	40 × 5 × 2 mm3	[25]
C-Balls	Magnetic	0.046	Magnetic field	30	-	-	-	-	[180]
Helical robot	Magnetic	0.01	Magnetic field	0.04	-	-	19.45 cm	30 mm (Length) 4.5 mm (Diameter)	[182]
Inching robot	Light	0.00025	200 mW·cm−2 (Light intensity)	-	-	-	2.7 mm	17 mm (Length)	[23]
Transporter robot	Light	0.0005	150–250 mW·cm−2 (Light intensity)	0.02	0.005	0.05 N	4 mm	20 mm (Span)	[133]
Wormbot	Sound	0.056	300 mAh 7.4 V (Battery)	-	-	-	6 mm	160 mm (Length) 27 mm (Radius)	[162]
Tripedal robot	Chemical	3.6	Methane	-	-	71 kPa	30 cm	130 mm (Length) 10 mm (Height)	[164]
Jumping robot	Chemical	0.43 (Horizontal) 0.86 (Vertical)	Butane	510	-	138 kPa	0.6 m	150 mm (Radius) 80 mm (Height)	[52]

Table 2. Summary of the characteristics of the different soft robots.

	Actuation	Speed (m/s)	Source of Energy	Weight (g)	Carrying Capacity (g)	Output Force	Max Displacement per Step	Dimension	Refs.
SMARoll robot	SMA	0.13	<10 V DC	191.2	-	-	50 mm	100 × 50 × 8 mm3	[53]
GoQBot	SMA	0.5	<10 V DC	5–6.2	-	0.05 N	25 cm	100 mm (Length)	[82]
Inchworm robot	Fluid	0.0022	Air pressure	-	50	0.5 N	2 mm	98 × 25 × 17 mm3	[94]
Tube climber	Fluid	0.006	Air pressure	98	1381	13.8 N	2.4 cm	70–90 mm (Length)	[96]
Jellyfish robot	Fluid and EAP	0.009	Gas–liquid pressure difference	261	241	0.15 N	18 cm	140 mm (Height) 59 mm (Radius)	[103]
Annelid robot	EAP	0.0053	3 kV DC	10.3	-	2 MPa	9 mm	170 mm (Length)	[125]
RSR	EAP	0.03627	3.2 kV DC	0.88	-	-	145.09 mm	24.83 mm (Radius)	[126]
Multi-material robot	Magnetic	0.0031	Magnetic field	0.23	3	-	5.5 mm	40 × 5 × 2 mm3	[25]
C-Balls	Magnetic	0.046	Magnetic field	30	-	-	-	-	[180]
Helical robot	Magnetic	0.01	Magnetic field	0.04	-	-	19.45 cm	30 mm (Length) 4.5 mm (Diameter)	[182]
Inching robot	Light	0.00025	200 mW·cm−2 (Light intensity)	-	-	-	2.7 mm	17 mm (Length)	[23]
Transporter robot	Light	0.0005	150–250 mW·cm−2 (Light intensity)	0.02	0.005	0.05 N	4 mm	20 mm (Span)	[133]
Wormbot	Sound	0.056	300 mAh 7.4 V (Battery)	-	-	-	6 mm	160 mm (Length) 27 mm (Radius)	[162]
Tripedal robot	Chemical	3.6	Methane	-	-	71 kPa	30 cm	130 mm (Length) 10 mm (Height)	[164]
Jumping robot	Chemical	0.43 (Horizontal) 0.86 (Vertical)	Butane	510	-	138 kPa	0.6 m	150 mm (Radius) 80 mm (Height)	[52]

Table 3. Chronology of SMAs, fluids, and EAP actuation.

Years		SMAs	Fluids	EAPs
2006		Earthworm robot [77]		Swimming robot [139]
2007		Jellyfish microrobot [205]		Annelid robot [123]
2008		Six-legged robot [206]	Robotic eye [207]
2009		Omegabot [29]	Miniature soft hand [208]	Jellyfish robot [140]
2010		GoQBot [82]		Tankbot [145]
2011		∩-shaped robot [80]	Multigait soft robot [209]
2012			Grasping hand [210]
2013		Meshworm [211]	Snake robot [27]	Folding robot [212]
2014		Starfish robot [213]	Micro inchworm robot [101]	Quadruped robot [124]
2015		Robotic hand [83]	Oral robot [170]	Aligned fibers robot [166]
2016		Soft morphing hand [214]	Tube-climbing robot [96]	Versatile soft grippers [196]
2017		Curved gripper [215]	Self-healing robot [95]	Electronic fish [197]
2018		Gecko-inspired gripper [216]	Caterpillar robot [217]	Feedback control robot [218]
2019		Spherical robot [79]	Soft robotic manipulator [219]	Millimetre-scale snail robot [131]
2020		RoBeetle [86]	Balloon-type robot [99]	Capsule underwater robot [141]

Table 3. Chronology of SMAs, fluids, and EAP actuation.

Years		SMAs	Fluids	EAPs
2006		Earthworm robot [77]		Swimming robot [139]
2007		Jellyfish microrobot [205]		Annelid robot [123]
2008		Six-legged robot [206]	Robotic eye [207]
2009		Omegabot [29]	Miniature soft hand [208]	Jellyfish robot [140]
2010		GoQBot [82]		Tankbot [145]
2011		∩-shaped robot [80]	Multigait soft robot [209]
2012			Grasping hand [210]
2013		Meshworm [211]	Snake robot [27]	Folding robot [212]
2014		Starfish robot [213]	Micro inchworm robot [101]	Quadruped robot [124]
2015		Robotic hand [83]	Oral robot [170]	Aligned fibers robot [166]
2016		Soft morphing hand [214]	Tube-climbing robot [96]	Versatile soft grippers [196]
2017		Curved gripper [215]	Self-healing robot [95]	Electronic fish [197]
2018		Gecko-inspired gripper [216]	Caterpillar robot [217]	Feedback control robot [218]
2019		Spherical robot [79]	Soft robotic manipulator [219]	Millimetre-scale snail robot [131]
2020		RoBeetle [86]	Balloon-type robot [99]	Capsule underwater robot [141]

5. Conclusions and Outlook

Actuators play crucial roles in the structure design, motion control, and function execution of soft robots in applications. In this paper, the characteristics, mechanisms, and typical structures of soft robots with various actuators are discussed. Furthermore, the advantages and disadvantages of the actuators are investigated in detail.

In the future, more research on soft robots can be carried out in the fields of miniaturisation, intelligence, automation control, and multi-function design. Miniaturisation means designing soft robots with lower density and more flexible materials, reducing the weight and volume of soft robots as much as possible. Research is undergoing to design centimetre-scale, millimetre-scale, or even micrometre-scale soft robots that allow them to fit into cramped environments and reach places out of reach of humans. Intelligence means that, with the applications of integrated circuits and artificial intelligence technologies, soft robots can have exquisite structures, precise positioning, and advanced functions. Automation involves programming, sensor applications, and data processing. When instructed, soft robots can be autonomously controlled to accomplish complex operations, so that some risks and uncertainties caused by human participation can be effectively avoided. At present, the functions of soft robots are relatively single, so they can only realise relatively simple walking, grasping, and communication. In future, multi-functions of soft robots are imperative to be applied in practical situations.

In particular, the design of actuation mechanisms for soft robots has some challenges to be addressed in future works. The first challenge is the response speed and efficiency. After a soft robot is stimulated, there will be a delay time, so that the response lags are unable to provide timely feedback. The second challenge is material choice and design. Soft robots need more flexible materials to adapt to more deformation requirements and faster response speed. At the same time, they need to have a certain load capacity to conduct complex tasks and operations. Moreover, material properties constrain the design of complex modelling and control strategies, making it difficult to control precise positions and execute tasks according to geometrically exact models. Therefore, the development of soft robots is still in the laboratory stage.

Taking these factors into consideration, it is unable to design soft robots to perform some complex tasks in small areas. Nevertheless, as new materials and technologies are under continuous development and applications, it is expected to break through these difficulties by designing more advanced soft robots that are smaller in structure, more precise in movement, more flexible in control, and eco-friendly in operation. Together with the aforementioned features, new actuation mechanisms will be essential for soft robots to support complex application requirements, such as medical treatment, manufacturing, surveying, maintenance, information collection, human–computer interaction, electromechanical control, and wider fields.