Stiffening in Soft Robotics

A Review of the State of the Art

By Mariangela Manti, Vito Cacucciolo, and Matteo Cianchetti

he need for building ro bots with soft materials emerged recently from considerations of the li mitations of service robots Tin negotiating natural environments, from observation of the role of comp liance in animals and plants [1], and even from the role attributed to the physical body in movement control and intelligence, in the so-called embodied intelligence or morphological computation paradigm [2]–[4]. The wide spread of soft robotics relies on numerous investigations of diverse materials and technologies for image licensed by ingram publishing actuation and sensing, and on research of control tech niques, all of which can serve the purpose of building robots with high deformability and compliance. But the core due to material deformation (which is commonly experi challenge of soft robotics research is, in fact, the variability enced when dealing with soft materials-based actuators) and controllability of such deformability and compliance. and focus on systems whose activation determines a In nature, there are some examples of soft structures change of mechanical properties (in terms of elastic mod with variable stiffness: the octopus is able to selectively ulus) but without motions. stiffen parts of its arms to use them as a modifiable skele There are two main approaches in the literature (Table 1): ton, and the squid with its tentacles and the elephant with the use of active actuators arranged in an antagonistic man its trunk can do the same to transmit very high forces. ner and the use of semiactive actuators that can change their Tongues in general and human lips are other examples. elastic properties (Young’s modulus). In this article, we Stiffness tuning is the way soft structures (natural or artifi review the two approaches, providing further categorization cial) can effectively interact with the environment. Inher and surveying the main works that illustrate the current per ent softness enables dexterity and safe interactions formance of such technologies. (preventing damage to the environment and to the entity), but stiffening is required to increase the forces transferred Active (Antagonistic Arrangement) to the environment when necessary. Thus, this has to be The antagonistic method to vary the stiffness of a structure done on demand and possibly without deformation. or a joint consists in the use of active elements that are This article reviews the approaches for on-demand applied to contrast each other (active–active) or are coupled stiffening obtained through direct activation of soft actu with passive structures (active–passive structure). The ation technologies, which are being explored for soft equilibrium position of the system depends on the com robotics. We exclude from this review the stiffness change bination of the equilibrium positions of the constituent elements, so in some cases it is possible to actuate the indi

Digital Object Identifier 10.1109/MRA.2016.2582718 vidual units while leaving the whole system at rest. This Date of publication: 13 September 2016 feature allows varying the stiffness and the equilibrium

1070-9932/16©2016IEEE September 2016 • IEEE ROBOTICS & AUTOMATION MAGAZINE • 93 position of the system independently. Generally, this effect and nonantagonistic elements, electroactive polymers (EAPs) can be implemented through joint antagonistic actuation are one of the most effective. Being made completely of soft or structure antagonistic actuation. While recent and com materials, dielectric elastomer actuators (DEAs) are theoreti plete reviews of systems that involve rigid joints can be cally the most suitable solutions for electrically activated soft found in [5] and [6], here we will focus on systems with robots, but some limitations on fabrication processes and continuous compliant/soft structures that can vary their material features are still obstacles for large-scale exploitation. stiffness and that are more suitable for soft material- A general setup to use DEAs as isometric variable-stiffness based robots. devices has been proposed in [7] [Figure 1(a)]. Exploiting the ability to vary stiffness and integrating low Active–Active weight and compact dimensions, researchers have applied Virtually all the combinations of the active actuation technol EAP technology to build a hand rehabilitation orthosis [8]. ogies listed in Table 1 are possible implementations of this The stiffness is controlled only by means of the applied volt kind of antagonistic coupling (whether the same or two dif age, thus easily enabling the implementation of a control algo ferent technologies are used), but in fact only some of these rithm to design the stiffness curve. solutions have been attempted so far. Among the solutions Another example where actuators based on the same tech where the same technology has been adopted for antagonistic nology exploit their arrangement to generate a stiffening is

Table1. The classification of the main soft robotics actuation technologies.

Technology Acronym Working Principle Subgroups Active Flexible fluidic actuators FFA A flexible inflatable structure, McKibben, fluidic elasto- (energy actuated by fluid. By patterning the mer actuator, PneuNets, introducing or flexible/elastomeric structure or by PneuFlex dissipating) adding different materials to intro- duce asymmetries, the structure can be designed to deform into specific shapes. Shape memory SMM Materials capable of recover- Alloys, polymers, com- materials ing a predetermined geometric posites shape (memorized) after a plastic deformation by inducing a phase transformation. At the base of this shape memory effect, there are the different mechanical properties associated with different stable phases. Electroactive polymers EAP Polymers able to undergo shape Dry (electric) deformation when electric fields Wet (ionic) are applied.

Tendon-driven actuators - Although not soft actuators per se, - if the tendons are remotely pulled, they present very low flexural stiffness and are often used in continuous soft robots. Semiactive Material jamming - Systems composed of an external Granular, layer (energy membrane and a filler. By apply- dissipating only) ing a vacuum, the membrane collapses on the filling material, increasing the density and the stiffness of the entire system. Electro- and magneto- ERM, Materials embedding magnetic or ERMs: fluids, elastomers rheological materials MRM electric particles. When subjected MRMs: fluids, elastomers to an external magnetic or electric field, the particles orient and build chains in response to the inter- particle interaction. This translates into an increased resistance to deformation. Low melting point LMPM Materials featuring a low melting Alloys, polymers materials temperature, in which a phase change can be rapidly obtained with a thermal input.

94 • IEEE ROBOTICS & AUTOMATION MAGAZINE • September 2016 represented by the use of flexible fluidic actuators (FFAs). contracting McKibben is encircled with elongating actuators, These kinds of actuators generally exploit elastomeric cham thus enabling the possibility of increasing the stiffness of the bers patterned so that when inflated they create asymmetric structure (without deformations) by activating them simulta deformations directed toward specific directions. But if used neously [Figure 1(b)]. in a trilayer design (not shown, but theorized in [9]), a simul One example of variable-stiffness structures using a com taneous activation provides a method of actively controlling bination of different active actuation technologies is based on stiffness without inducing bending (opposite forces cancel out tendons coupled with a pneumatic chamber [Figure 1(c)]. if the actuators are mirror arranged). With less design flexibil The chamber is made of an internal extensible bladder and an ity (in terms of free parameters) and more difficult fabrica external sleeve, which constrains the radial expansion. The tion, the same approach has been exploited using McKibben increase of pressure in the chamber tends to elongate the actuators. These actuators can be considered a specific class of structure, while the cables tend to shorten it, thus producing FFAs, and they consist of a flexible pneumatic chamber and a an antagonistic effect. The physical parameters related to the braided structure wrapped around it. Two different actuation stiffness are the pressure inside the chamber and the tension effects exist in these structures, one in contrast to the other, in the cables. This mechanism has been explored starting regulated by the initial angle θ0 between the fibers of the from the first preliminary examples of continuum robots braided structure and the axis of the actuator. The conse [11]–[12] and is still used in more recent robots [13]–[15]. quence is that by varying the initial angle θ0, elongation or These systems focus on varying the stiffness of the bending shortening can be obtained. To produce a stiffening behavior modes, while the cable lengths limit the extension. Passive without inducing deformations, the elongating and contract elongation is not allowed (the cables do not have tensile com ing McKibben actuators have to be coupled so they act antag pliance), but compression is possible instead, and its stiffness onistically. This is the case reported in [10], where a central depends on the internal pressure.

Electrodes (1) Tendons Inner Bladder External Sleeve Dielectric (2)

(3)

X (4) (i)

–X (5) (ii) X (a)

Contracting Actuator

Extending (iv) Actuator (iii)

(b) (c)

Figure 1. (a) A scheme of two DEAs used to form a variable-stiffness actuator. In (1), the two DEAs are unconnected and in unloaded equilibrium positions. When connected (2), the structures are prestressed in tension. Thus, the activation of one of them (3 and 4) produces a release of the prestress and a consequent displacement x of the central reference point. When the actuators are both activated (5), they both decrease the lateral prestress, which in turn decreases the stiffness of the system (not of the actuators themselves) without changing the equilibrium position (isometric stiffness variation). (b) The simultaneous activation of contracting and expanding McKibben actuators proposed in [10] to obtain a variable-stiffness device. (c) (i) A description of the main components of tendon-fluidic mechanisms; their implementation in (ii) and (iii) the AirOctor [12] and (iv) a surgical manipulator able to shrink almost completely [15].

September 2016 • IEEE ROBOTICS & AUTOMATION MAGAZINE • 95 Active–Passive Structure trunks), researchers created a robot inspired by the octopus When one of the actuation technologies listed in Table 1 is [18]. Muscular hydrostats [Figure 2(b)] feature perpendicu coupled with a passive structure that prevents deformations, lar muscle bundles, whose simultaneous activation allows an a stiffening occurs. McKibben actuators have already been isometric increase of the stiffness of the structure. The same introduced as elongating or contracting elements that can be principles have been proposed in [19] and [20], where per combined to form a variable-stiffness structure; but, on the pendicular longitudinal and transverse actuators can inde other hand, when θ0 is exactly 54.7°, the braided structure is pendently control arm position and stiffness. A combination not able to be deformed by inflation, and the structure of a braided structure, showing the same features already dis increases its stiffness without producing any movement. cussed in the case of the McKibben actuators, with SMA coils Figure 2(a) shows a finite element modeling (FEM) simula used as actuators forms the continuous octopus-inspired arm. tion of the McKibben actuator in such a situation. In [16] While the single actuation units produce local deformations and [17], a robotic system able to independently vary its (bending, elongation, and shortening), the simultaneous stiffness and its equilibrium position has been produced by activation of transverse and longitudinal bundles results in coupling a McKibben actuator with θ0 = 54.7° with other the stiffening of the whole structure. actuators able to produce motion (e.g., tendons [17] or other In this work, the SMA actuators are connected to a braid McKibben actuators with θ0 ≠ 54.7° [16]). ed sleeve to couple their antagonistic action and to allow Despite being very versatile, McKibben actuators have a deformation propagation. But a stiffness variation can some miniaturization limits (mainly connected to the size of also be obtained using simpler structures. In [21], a multi the pipes and to the manufacturing process). Thus, when the layered variable-stiffness system based on SMA transverse scale starts to be prohibitive for fluidic actuators, smart mate wires has been proposed. When not activated, the SMA rials like shape memory alloys (SMAs) are preferable. This is wires let the layers slide on each other, but activation of the the case, for example, of some biomimetic robots. To exploit wires increases the friction force among the layers, increas the high dexterity featured by muscular hydrostats in ani ing the flexural stiffness of the entire structure. The same mals (e.g., octopus arms, squid tentacles, and elephant group in [22] also explored the possibility of using EAPs

(i) (ii) Displacement 50 mN Top Element

(iii) (iv) Bottom Element

Initial State 0 MPa (b) 50 mN 50 mN F x y (i) F x y (ii) F

x 0.1 MPa 0.2 MPa y Form Closure No Form Closure (iii) (a) (c)

Figure 2. (a) An FEM simulation of a McKibben actuator with i0 = 54.7° from [16]. (b) The scheme of the muscle arrangement in the octopus muscular hydrostats from [20]: in (i), four main longitudinal muscles (L) are arranged along the arm length; transverse muscles (T) are arranged in a radial configuration along four arcs connecting the external connective tissue with the central channel, which contains nerve fibers (N); and oblique muscles (O) wrap the whole structure [67]. (ii) A histological transverse section of the same muscular system [67]. Longitudinal (iii) and transverse (iv) views of the robotic actuator proposed in [19], with SMA coils replacing the longitudinal and transverse muscle bundles of the muscular hydrostats. (c) The function of a multilayer variable-stiffness device: (i) compliant, (ii) partly stiffened, and (iii) stiff (courtesy of M. Henke).

96 • IEEE ROBOTICS & AUTOMATION MAGAZINE • September 2016 (namely DEAs) instead of SMAs to enable the stiffness In a similar approach, Cheng et al. [32] developed a robust, change. The mechanism is based on the same principle, and modular, and highly articulated manipulator that utilizes jam the results show the possibility of increasing the stiffness by ming of granular media for achieving local stiffness control, up to two orders of magnitude. while actuation cables along the robotic manipulator allow the control of its shape and position [Figure 3(b)]. Jamming mech Semiactive (Intrinsic Rigidity Tuning) anisms provided the ground for the development of a new par Other than antagonistic arrangement, variable-stiffness adigm for soft robots, as referenced in [33]. The novelty is also mechanisms can be obtained by using semiactive actuators related to the capability of introducing selective anisotropies in whose working principle relies on the modulation of the the behavior of the material. A soft mobile robot [Figure 3(d)] intrinsic passive mechanical properties of the material itself. has been developed by using jamming-based cells arranged on This translates into the capability of varying the energy dissi the external surface of a sphere-shaped robot. By jamming spe pation of the system. cific cells with respect to others that remain in a fluid-like state, it is possible to steer the deformation along preferred directions Jamming-Based Systems that can be used to initiate a rolling locomotion pattern. With a Among the suitable strategies for changing the stiffness of a similar strategy, Kaufhold et al. [34] implemented a variable- soft robot, the jamming-based systems are emerging with a stiffness mechanism based on granular jamming to locally new set of possibilities. Despite already being investigated for interrupt the symmetrical behavior of their amoeba-like robot a long time (granular jamming is a well-known issue in agri [Figure 3(e)]. The rotating magnetic system used to induce a culture and the food industry), it recently gained the attention vibration in the robot implements a random circular locomo of roboticists because of its capability of enabling a reversible tion until the stiffness change brakes the symmetry, causing a transition between a fluid-like and a solid-like material with modification in the locomotion direction. out (or with very limited) volume variation [23], [24]. The simplicity, feasibility, and reliability of the jamming-based There exist two systematic approaches that exploit the jam technologies are enabling their use in completely new and inno ming phenomenon: membranes filled with granular matter or vative scenarios. In recent years, new application areas have with thin sheets. Both refer to the same activation principle: emerged, such as the stiffness modulation used as feed vacuum triggers the “phase change,” increasing the relative back for haptic or tactile interfaces [35]–[37] [Figure 3(f)]. shear stress experienced by the particles or layers in the elastic But the most straightforward and promising exploitation sce membrane. Without attempting to catch the phenomenon nario for this technology is undoubtedly the medical field. In from a physical point of view, some research works have tried minimally invasive surgery (MIS) procedures, clinicians need to identify the key factors affecting the jamming mechanism at instruments that are flexible enough to enable insertion the macroscale. Among the most accredited are analyses of the through body cavities without damaging tissues but that are number, shape, and dimension of particles or of overlapping also able to stiffen enough for applying forces on the target surfaces (contact surfaces); the mechanical properties of the site. Up to now, different endoscopes with these requirements elastic membrane; and the vacuum level and the shear stress have been developed by exploiting the jamming mechanism experienced by the system [25]–[28]. for actively varying stiffness [38]. The STIFF-FLOP surgical One of the first and most effective efforts to exploit such a manipulator is one of these. Three FFAs arranged at 120° are phenomenon is represented by the universal gripper devel used to generate omnidirectional bending and elongation and oped by Brown and colleagues [29]—a robotic end effector combined with a central hollow channel that hosts the granu able to pick up unfamiliar objects of widely varying shape and lar jamming mechanism to enable a selective stiffening of up surface properties [Figure 3(a)]. When pressed onto a target to 46% [39]–[41] [Figure 3(c)]. object, the gripper flows around it and conforms to its shape. Particle jamming has interesting features, such as high Upon application of a vacuum, the granular material con deformability in the fluid state and a drastic stiffness increase tracts and hardens quickly to pinch and hold the object. This in the solid state without a significant change in volume. basic yet effective functionality has been increased in a more However, it requires a substantial volume of granular material recent work by Amend et al. [30], where they introduced the to achieve a significant stiffness variation. From this point of possibility of adding a positive pressure in the elastomeric view, layer jamming technology could represent an even bet membrane to make the recovery phase faster and to increase ter alternative. Here, overlapping surfaces present a large con the reliability and the error tolerance of the gripper, while tact area that translates into an increased friction force that decreasing the force needed on target objects. can be generated through vacuum application. A snake-like Used in an anthropomorphic gripper in combination with manipulator based on this principle has been developed by PneuFlex actuators (a particular kind of FFA), a granular jam Kim and colleagues [42], [43]. The cylindrical shape of the ming-based system can allow a selective stiffening or a shape manipulator develops by overlapping layers into a helical locking of any bending state by retaining the motion of the shape for maximizing the friction effects between layers elastic top side of the fingers. This effect, in combination with [Figure 3(h)]. The scalability of the technology is not particu the flexible but nonstretchable material placed in the bottom larly remarkable, but it allows obtaining a manipulator with side, increases the flexural stiffness [31]. features compatible with various MIS applications.

September 2016 • IEEE ROBOTICS & AUTOMATION MAGAZINE • 97 Tube Fittings Outer Latex on End Caps Membranes Tension Cables × 4

1 23 45 Abdomen

Inner Springs Entire Cavity Filled Trocar with Ground Coffee Spooler Motors × 4 Three Connected Gripper Gripper Air is Object Can be (b) STIFF-FLOP Approaches Deforms Evacuated Manipulated Modules in Soft State Around Object from Gripper (a) FA FB F F A B External Sheet Silicone Unjammed Cells F1 Granular Jammed Cells Jamming F C F Fluidic Actuators Expanded Brushless dc Motor C Actuator Rotor with Permanent Magnets Silicone Inelastic Frame Fluidic Connection Actuators Element External 50 mm 30 mm Sheet Fluid Granular Jamming Segment with Variable Compliance Silicone Compliant Structure with Pipes Permanent Magnets (d) (e) (c)

240 Inner Flaps Outer Flaps Middle Strap

(i) Unjammed, Flexible Compartment dpitch

cone (ii) Jammed, Stiff Compartment (i)(Center Line ii) (iii) After Evacuation of Air

(f) (g) (h)

Figure 3. (a) A schematic drawing of the jamming-based gripper for picking up a wide range of objects without the need for active feedback [29]. (b) A CAD drawing showing a cross-sectional view of the manipulator developed by Cheng et al. [32]. (c) A CAD drawing of the overall architecture of the STIFF-FLOP manipulator, with the design of the single module showing all the structural elements (soft-bodied module with fluidic and the stiffening chambers), adapted from [40]. (d) A side view of the proposed jamming-based soft robot with three unjammed cells and the internal actuator partially inflated [33]. (e) The basic configuration of the locomotion system of the amoeba robot [34]. (f) The jamming technology for haptic feedback [35]. (Photo courtesy of MIT Media Lab, Tangible Media Group.) (g) The working principle of three different jamming-based approaches tested on top of a PneuFlex actuator. From top: granular jamming, layer jamming with overlapping fish-scale-like layers, and layer jamming with two stacks of three interleaved layers. The PneuFlex actuator is shown on the bottom, and the jamming chamber is indicated by dashed lines [31]. (h) In (i), an assembled one-side flap pattern, (ii) a section view of a double-side flap pattern, and (iii) a real assembled double-side flap pattern [43].

Other more classical configurations rely on multilayer They are usually well suited to automotive applications for structures that can be “jammed” to realize planar thin adaptive bumper and shock absorber systems and for damp interfaces with tunable stiffness [44]. In this case, the phe ing control in adaptive orthotic devices. If coupled with the nomenon is maximized due to the allowed dimensions, use of soft materials, the same principle can be exploited to which imply a large contact surface between layers, but at increase the stiffness of soft structures. Majidi et al. [45] pro the same time the geometrical pattern of the system usu posed a self-contained system with tunable stiffness capability ally limits the allowed workspace. based on microconfined MR domains [Figure 4(a)]. The application of a magnetic field causes an alignment of the Magneto- and Electrorheological Materials magnetic particles, which opposes to the stretching of the Magnetorheological (MR) and electrorheological (ER) mate elastomer, translating into an increased longitudinal stiffness. rials are mostly known as fluids able to change their rheologi But the use of MR fluids (MRFs) in soft robotics is currently cal properties when a magnetic or electric field is applied. limited by a series of drawbacks related to such issues as particle

98 • IEEE ROBOTICS & AUTOMATION MAGAZINE • September 2016 Shake Table Motion Micropatterned Ribbons

Power Load Cell MRE Base Isolator Supply Soft Polyurethane Shake Table (i) 250 µm 500 µm (b)

A +Q F L mL H 10 mm n ∆ 500 µm Ribbon Surface (ii) –Q (i)(ii) Electric 50 Displacement SSN S N 40 50 m = 1 18.00 ) 30

15.00 ( n = 100 12.00 f 20 S N S N S N n 10 9.00 10 6.00 3.00 0 10–35 mT 0.00 0.0 0.5 1.0 1.5 (iii) Zero Magnetic Field Magnetic Microdomains Simulation Theory (iii) (iv) (iv)

(a) (c)

Figure 4. (a) In (i), an ultrasoft polyurethane elastomer is embedded with rigid micropatterned ribbons that slide past each other. The tabs are enclosed in a chamber filled with MR fluid. In (ii), the surface of each ribbon is patterned with an array of aligned microchannels. In (iii), in the absence of a magnetic field, the MR microparticles are randomly dispersed. In (iv), under an external field of 10–35 mT, the microparticles form magnetic domains that are confined in the microchannels [45]. (b) The experimental setup for evaluating and characterizing the performance of the MRE base isolator prototype proposed in [48] (courtesy of Y. Li). (c) In (i), an ERE with layered mesostructures in an undeformed state and (ii) undergoing simple share deformation by an applied force and voltage. In (iii), the distribution of electric displacement in the layered mesostructures calculated by a finite-element model. In (iv), the normalized effective permittivity f as a function of the layer orientation i calculated by the numerical model and theory [53]. settling, sealing issues, and environmental contamination, As with MRFs, ER fluids have their elastomeric ver which makes another member of the MR material family more sions, but the use of ER elastomers (EREs) is much less preferable to enable tunable stiffness: MR elastomers (MREs) widespread. This is probably due to the shortcomings of and foams. While MRFs can be coupled with elastomers the use of ferroelectric particles, as the maximum-yield through fluidic channels, a sensibility to external magnetic stress they generate is, on average, two orders of magni fields can be directly embedded in the elastomer material by tude lower. A few studies have focused on the experimen including the MR particles during the fabrication/curing pro tal evaluation of the stiffness change, and an interesting cess. This is the case with MREs, which consist of a polymeric area is represented by the theoretical study of the distribu matrix with embedded micro- or nanosized ferromagnetic par tion of the ferroelectric particles in the elastomers. The ticles, such as carbonyl iron [46]. Such materials have been role of anisotropy has been deepened from a theoretical already investigated and applied in various engineering fields, point of view by Liu et al. [52] in the microscopic structure especially as vibration reducers or isolation systems. A recent and by Cao and Zhao [53] in the mesoscopic structure and comprehensive survey of current applications, manufactur [Figure 4(c)]. Even if not validated, the theoretical values ing processes, and modeling approaches can be found in [47], suggest stiffness variations of 300 kPa if the layered meso but so far studies have focused on MREs’ static properties of structures are arranged in parallel configuration. stiffness change, aiming at developing adaptive layered struc tures. In [48], an adaptive MRE base isolator shaped as a lami Low Melting Point Materials and Glass nated structure demonstrated the capability of remarkably Transition-Based Softening changing the lateral stiffness of the isolator by up to 1,630% A very recent and impressive example of the exploitation of under a medium level of magnetic field [Figure 4(b)]. MRE- low melting point materials (LMPMs) as variable-stiffness based sandwich beams with adjustable rigidity have been theo systems has been provided by Cheng et al. [54]. They found retically evaluated in [49] and [50], while experimental that an extraordinary functionality is enabled by an verifications are reported in [51]. extremely low-cost and commercially available material:

September 2016 • IEEE ROBOTICS & AUTOMATION MAGAZINE • 99 (i) (ii) (iii)

Strut Coating

1 mm 6 mm (v) (iv) 1 mm y Epoxy Wax Coating (Bounded x Strut by Dotted Curve) Cross Section No One Wax Two Wax Wax Layer Layers (After Compression) 500 µm 100 µm

(a)

(i) Prestretched DEA (i) (iv) Silicone Matrix

LMPA Channel 4 mm w LMPA Substrate p (ii) Rigid (iii) Soft Cross Section h hm pb wm X (ii) Electrode V hpt 5 mm x w y a Elastomer (c) Electrode Soft State Liquid Metal Heating Element

(1) LMPA: ON (2) LMPA: ON Field’s Metal DEA: OFF DEA: OFFON Copper–Lead

(d) (5) Time L ~ ther L >> ther l ~ ther << L LMPA: OFF l DEA: OFF (4) (3) LMPA: OFF LL Rigid State DEA: ON

(iii) VDEA > 0 VDEA = 0

VDEA = 0 VDEA > 0

(b) (e)

Figure 5. (a) In (i), a schematic of a cross section of a wax-coated printed lattice; (ii) microscope images of a (left) uncoated and (right) coated polyurethane foam; (iii) pictures of an uncoated (left) and coated (right) 3-D printed lattice; (iv) microscope images of vertices in the uncoated, single-coated, and double-coated 3-D printed lattices; and (v) microscope images of cross-sectional views of a wax-coated printed lattice strut embedded in epoxy (courtesy of N. Cheng) [54]. (b) Mechanism of the variable-stiffness DEA proposed in [60]: in (i), the actuator consists of a prestretched DEA attached onto an LMPA substrate. In (ii), the activation of the LMPA makes the structure soft, allowing the DEA to shape the structure, while the deactivation of the LMPA keeps the desired shape. In (iii), bidirectional actuation can be obtained by placing a DEA on each side of the LMPA-based substrate. (c) In (i), a variable-stiffness device showing low-melting-point alloy (LMPA) tracks embedded in PDMS. In (ii) and (iii), a side view of a device supporting a nut when the LMPA is (ii) in a solid state and (iii) in a liquid state. In (iv), a drawing of a device showing the cross section [59]. (d) An illustration of the composite composed of an elastomer sealing layer (top), a liquid-phase Joule heating element (middle), and a thermally activated layer of Field’s metal or SMP (bottom) (courtesy of C. Majidi) [61]. (e) A schematic of the general strategy used to accelerate stiffness transitions in thermoplastic structures utilizing a vascularized material design in bulk thermoplastic, which decreases the characteristic length scale of the material [57].

100 • IEEE ROBOTICS & AUTOMATION MAGAZINE • September 2016 wax. Flexible open cells coated in wax [Figure 5(a)] demon temperature become gradually softer. McEvoy and Correll strated a very large stiffness-changing range (three orders used polycaprolactone (PCL) in their robotic material [56] of magnitude), paving the way for their exploitation as and introduced the possibility of locally modifying the locking joint systems or shape-shifting structures. A ther mechanical properties of the material defining its stiffness mal input is also necessary for the nanocomposite based on profile. This is obtained through the installation of polyvinyl alcohol developed by Capadona et al. [55], where nichrome wires wrapped around the PCL bars. An alterna instead of melting, glass transition is responsible for a dra tive solution for a local selective heating has been proposed matic decrease of its elastic modulus (slow but very effec by Balasubramanian and colleagues [57]: vascularized mate tive). In both of these cases, however, heating elements are rials. In this approach, microfluidic networks are embedded not embedded and have to be provided separately, thus in thermoplastic polymers, enabling a rapid control (2.4 ± complicating the structure in an unpredictable manner. 0.5 s) of mechanical properties through accelerated glass– Actually, even if the exploitation of such materials is in its rubber phase transitions [Figure 5(e)]. The exploitation of earliest infancy, other researchers have already demonstrat the glass transition of this kind of material has not been ed that particular materials possess the intrinsic capability of explored very much so far, but potentially all plastics that are softening because of their state change. That is, thermoplas thermolabile can be used. Some other common materials tic materials approaching and going beyond their melting present this capability, such as hot-melt adhesives. Some

Polymer Stiff Layers (Shear Layer) (i) Axial Cover Layer

Base Beam Bending

Deformation Mode Depends Coupled on Polymer Shear Stiffness

(ii) (iii)

High Polymer Low Polymer Shear Stiffness Shear Stiffness Axial

Bending Decoupled

Coupled Decoupled

(a) (b)

1 in

(c)

Figure 6. (a) A schematic representation of a multilayered beam and deformation modes corresponding to high and low polymer shear moduli (adapted from [62]). (b) A generic concept for variable-stiffness materials using segmented, overlapping reinforcement materials combined with a variable-stiffness SMP matrix [63]. (Reprinted with permission from SAGE Publications, Ltd.) (c) Variable-stiffness fibers sewn onto fabric [64].

September 2016 • IEEE ROBOTICS & AUTOMATION MAGAZINE • 101 research works already rely on the use of these materials for to fully exploit this advantage, encapsulation is needed to con fast manufacturing, self-healing, or reconfigurability [58], tain possible liquid pouring. A smart solution is the use of but not for the ability to change its mechanical properties. microfabricated elastomeric structures [59] [Figure 5(c)]. Temperature-sensitive plastics and wax have demonstrated This is based on Cerrolow 117, a low melting point alloy their potential, but higher stiffness variation can be obtained (LMPA), embedded in soft polydimethylsiloxane (PDMS). if metals are used instead. Moreover, in principle the metal The devices demonstrate a relative stiffness change higher itself can be used as a source of heat from electric energy than 25 times (theoretically > 1,000 times) and a fast transi (resolving the issue of providing an external heat source). But tion from rigid to soft states (< 1 s) at low power (< 500 mW).

Table 2. A direct comparison of stiffening mechanisms (qualitative evaluation scale: ++, +, o, -, – from the highest to the lowest values).

Independent Number of Actuation Speed of Speed of Stiffness Stiffness Control and Passive Modes of Allowed Technology Stiffening Depending on Destiffening Depending on Scalability Variation Equilibrium Position Deformation Stiffening Workspace References

Active–Active EAP + EAP ++ Voltage amplifier ++ Discharge + 2 x Yes Large 1, 2 Medium [7], [8]

Fluidics + fluidics + Fluidic inflow o Fluidic outflow - 10 x Yes Large 1, 2 Large [9], [10]

Tendons + fluidics o Motor’s o Motor’s velocity o 103 x Yes Large 1 Large [11 ]–[15] velocity and and fluidic outflow fluidic inflow

ACTIVE Active– Fluidics + braided sleeves + Fluidic inflow o Fluidic outflow - 2.4 x No Large 1, 2 Large [16], [17] Passive Structure SMA + braided sleeve + Electric power - Thermal conditions - 1.5 x Yes Large 1, 2 Large [18 ]–[20] SMA + flexible layers - Electric power - Thermal conditions o 14.6 x No Small 1, 3 Small [21] (antagonistic arrangement) EAP + flexible layers + Voltage amplifier ++ Discharge o 102 x No Medium 1, 3 Medium [22]

Jamming-based Granular jamming + Vacuum pump o Vacuum pump + 40 x Yes Large 1, 2, 3 Large [23 ]–[41] outflow inflow

Layer Jamming + Vacuum pump o Vacuum pump + 10 x Yes Medium 1, 3 Medium [42 ]–[44] outflow inflow

ER and MR materials MR fluids ++ Magnetic field ++ Magnetic field + 37 x Yes Large 2 Large [45]

MREs ++ Magnetic field ++ Magnetic field + 5– 16.3 x No Large 2 Large [46]–[51]

EREs ++ Electric field ++ Discharge + 10 x Yes Large 2 Large [52], [53]

Low melting point Wax – Thermal - Heat source ++ 103 x Yes Small 1, 2, 3 Large [54] materials conditions

Polymers – Thermal o Heat source ++ 86 x Yes Medium 1, 2, 3 Medium [56] conditions

SEMI-ACTIVE Alloys – Electric power - Thermal conditions ++ 25– 104 x Yes Small 1, 2, 3 Large [59]–[61]

2 3

(intrinsic rigidity tuning) Glass transition-based – Thermal o Heat source ++ 10 – 10 x Yes Medium 1, 2, 3 Medium [55], [57] conditions

SMMs Polymers – Heat source - Thermal conditions ++ 2– 100 x No Small 1, 2, 3 Medium [61]–[63]

Alloys – Electric power - Thermal conditions ++ 8– 10 x No Small 1, 2, 3 Medium [64]

Conductive poly- – Thermal - Electric popup ++ 25 x No Medium 1, 2, 3 Medium [65] mers conditions

Chemical-based – Drying - Hydration speed ++ 40– 102 x Yes Small 1, 2, 3 Large [55], [66] conditions

Independent control stiffness and equilibrium position: the ability to vary the stiffness independently from the equilibrium position Passive deformation: the maximum elastic deformation of the structure Number of modes of stiffening: the modes considered are 1) bending, 2) elongation/compression, and 3) torsion. Allowed workspace: the workspace of the robotic device in which the change of stiffness is allowed

102 • IEEE ROBOTICS & AUTOMATION MAGAZINE • September 2016 In [60], the same group also proposed an innovative combi phase-changing metal alloy as the active element able to nation of DEAs and LMPAs embedded in a silicone substrate. change the stiffness of the elastomer composite into which it The device enables functional soft robots with a simplified is embedded. The variable-stiffness composite has a multilay structure, where the DEA generates a bending actuation and er structure that allows stacking the LMPA and the heater the LMPA provides controllable stiffness between soft and close to each other [Figure 5(d)]. At room temperature, the rigid states [Figure 5(b)]. embedded Field’s metal is solid, and the composite remains The use of microfabricated elastomeric channels has also elastically rigid. Joule heating causes the metal to melt and been reported in [61], where Shan and colleagues used a allows the surrounding elastomer to freely stretch and bend.