A Novel Design of Soft Robotic Hand with a Human-inspired Soft Palm for Dexterous Grasping
Haihang Wang, Fares J. Abu-Dakka, Tran Nguyen Le, Ville Kyrki, He Xu
11 A Novel Design of Soft Robotic Hand with aHuman-inspired Soft Palm for Dexterous Grasping
Haihang Wang, Fares J. Abu-Dakka ∗ , Tran Nguyen Le, Ville Kyrki, and He Xu Abstract —Soft robotic hands and grippers are increasingly at-tracting attention as a robotic end-effector. Compared with rigidcounterparts, they are safer for human-robot and environment-robot interactions, easier to control, lower cost and weight, andmore compliant. Current soft robotic hands have mostly focusedon the soft fingers and bending actuators. However, the palm isalso essential part for grasping. In this work, we propose a noveldesign of soft humanoid hand with pneumatic soft fingers andsoft palm. The hand is inexpensive to fabricate. The configurationof the soft palm is based on modular design which can beeasily applied into actuating all kinds of soft fingers before. Thesplaying of the fingers, bending of the whole palm, abductionand adduction of the thumb are implemented by the soft palm.Moreover, we present a new design of soft finger, called hybridbending soft finger (HBSF). It can both bend in the grasping axisand deflect in the side-to-side axis as human-like motion. Thefunctions of the HBSF and soft palm were simulated by SOFAframework. And their performance was tested in experiments.The 6 fingers with 1 to 11 segments were tested and analyzed.The versatility of the soft hand is evaluated and testified bythe grasping experiments in real scenario according to Feixtaxonomy. And the results present the diversity of grasps andshow promise for grasping a variety of objects with differentshapes and weights.
Index Terms —Soft robotic hand, Human-inspired Soft Palm,Grasping
I. O
VERVIEW AND M OTIVATION H ANDS or grippers are essential component for roboticmanipulation as they handle objects with certain posi-tions, orientations and contact forces. They serve as the end-effector interfacing between target objects and robots. Dexter-ous grasping is a prerequisite for task-dependent manipulation,which requires the consideration of important factors suchas the interaction-force, stiffiness/compliance, dexterity andnumber of degrees of freedom [1].Conventional rigid robotic hands for industrial applicationsare generally able to provide high accuracy in position thanksto their sophisticated actuation and sensing mechanisms. How-ever, it is hard to control the contact force between the rigidhand and objects as the rigid structure driven by electricalmotors commonly generates large contact forces. In real-world scenarios, usually, we require grippers to manipulate
H. Wang is with College of Mechanical and Electrical Engineering, HarbinEngineering University, China and with Intelligent Robotics Group at theDepartment of Electrical Engineering and Automation, Aalto University,Finland (e-mail: [email protected]).F.J. Abu-Dakka, T. Nguyen Le and V. Kyrki are with Intelligent RoboticsGroup at the Department of Electrical Engineering and Automation, AaltoUniversity, Finland (e-mail: { firstname.lastname } @aalto.fi).H. Xu is with Electrical Engineering, Harbin Engineering University, China(e-mail: railway [email protected]). Fig. 1: The soft humanoid hand and the grasping objects used in this work. objects with uncertain shapes, sizes and poses in uncertainenvironments [2]. Moreover, when the targeted objects arefragile or delicate, large contact forces can deform or evendamage the objects. Another drawbacks of these rigid handslie in their heavy weight and high cost. Thus, the applicationsof soft robotic hands with passive compliance have attractedattention for an inherently safe and adaptive contact. Softrobotic hands not only can easily adapt to objects of variousshapes and sizes, but also can perform a self-adaptive contactwithout the need of sophisticated control as rigid hands.Furthermore, their soft nature helps to minimize the damageto the manipulated objects.Soft bending actuators, used as fingers, are the main com-ponent of soft robotic hands/grippers. They can be sorted asdifferent types, such as fluidic elastomer actuators (FEAs),cable-driven actuators, shape memory alloys (SMAs), electro-magnetic/magnetic actuator [3]. Amongst these, FEAs haveachieved particular sizable push toward the utilization ofcompliant hands. FEAs are mainly made of silicone rubber a r X i v : . [ c s . R O ] S e p Fig. 2: The proposed pneumatic soft humanoid hand with dexterous palm. (a) Illustrates the 3D model of our hand. (b) Shows air chambers: the red partsshow the inner distribution of the air chambers, which will be pressurized based on SoftRobots plugin. (c) The soft robotic hand prototype. and driven by pneumatic or hydraulic. The pneumatic typeof FEAs are known as soft pneumatic actuators (SPAs). Themost popular SPAs are the pneumatic networks (Pneu-nets)bending actuators designed by Whitesides et al. [4] and fibre-reinforced actuators designed by Galloway et al. [5]. Pneu-netsis bonded by 2 layers: the silicone-based top layer containingnumerous chambers inside (like networks) and the inextensiblebottom layer. When the actuator is inflated, the top layerwill extend, and the actuator will achieve a bending motion.Fibre-reinforced actuator comprises an extensible chamber, aninextensible layer and fibres. Its bending mechanism is similarto the Pneu-nets actuator. The fibre-reinforcement is used tolimit the chamber in axial extension instead of useless radialexpansion. Both types are simple in design, effective and easyto fabricate.In literature, there are different design and application forsoft robotic hands based on Pneu-nets bending actuator [6],[7], [8], [9] and fibre-reinforced actuator prototypes [1], [2],[10], [11], [12], [13], [14]. However, compared with eachother, Pneu-nets actuator has a lower capacity of input pressuredue to its total soft top layer under the same wall thickness,which limits its maximal grasping force. In addition, fibre-reinforced actuator has a lower bending efficiency, whichlimit its bending angle under the same pressure. In order toovercome the shortcomings of them, in this paper, we designa novel hybrid bending soft finger (HBSF) by integrating theinner chambers network structure inspired by Pneu-nets withfibre-reinforcement method.Recently, there have been rapid developments on softrobotic hands and grippers. However, most of them focuson the study of soft fingers and overlook the importance ofthe palm. The fingers were usually assembled together andfixed in a rigid palm or basement. However, the palm playsa considerable role in the grasping functioning. The fixedposition of fingers will greatly limits the grasping scope andpose of the hands.In order to achieve a dexterous manipulation, we refer to thepostural variability of the hand: the higher this variability, themore dexterous we consider a hand (for example of graspingpostures referring to the grasp taxonomies purposed by Feix et al. [15]). Robotic hand with a changeable palm can adjustthe position and orientation of fingers, which can significantlyimprove the postural flexibility of a robotic hand in terms ofsizes and shapes of objects which can be grasped. Sun et al.[8] presented a flexible robotic gripper with a rigid changeablepalm. The distance of the fingers can adjust using sliderand beam mechanism. An opposable thumb is important anduseful to achieve dexterity in robotic hand. However, the rigidchangeable palm can only change the position of the fingers.RBO hand 2 [1] has a soft palmar actuator for enabling thumbabduction. They proposed a new PneuFlex actuator with fiber-reinforcement as the soft fingers. In addition, they also usetwo connected PneuFlex actuators as the base of the thumb toachieve the dexterity of the thumb. The other four fingers werefixed on a 3D-printed scaffold. The assembly angle betweenthumb and the other four fingers is about 120 ◦ , instead of aplane. The soft biomimetic prosthetic hand developed by Fraset al. [11] presents a similar design of the thumb abduction.They used the PneuFlex actuators as in [1]. They applied onefor the thumb and two for the palm. The exoskeleton for fixingthe soft actuators is deformable and based on a 3D scan ofa real human hand. Both hands have soft actuators for thumbabduction. However, the palm motion for the other four fingerswere ignored. The four fingers of their hands can only bendto a certain direction. In contrast of [1] and [11], in order toaugment the human-like palmar function of the soft hand, wepresent a pneumatic soft humanoid palm that can help thumbabduction, the four other fingers splay and palm bend.In this paper, we propose a compact hybrid solution of softhumanoid hand, as shown in Figure 1. For the sake of clarity,the main contributions of this paper are:– A novel design of hybrid bending soft finger (HBSF)by integrating the inner chambers network structure withfibre-reinforcement method.– A novel design of pneumatic soft humanoid palm.The soft robotic hand is made of soft materials only. TheHBSF can robustly grasp a variety of objects with differentweights, sizes, shapes and stiffness. The soft humanoid handconsists of 5 soft pneumatic fingers and 2 parts of the soft palm, which are all independent and assembled together usingsilicone connections.II. D ESIGN AND S IMULATION
A. Soft hand
The design of soft hands/grippers can be divided into twomain morphological types: the anthropomorphic hands and thegrippers with several spatially evenly distributed fingers. Forour hand, we choose an anthropomorphic design in shape witha new dexterous soft robotic palm. Figure 2(c) presents theprototype of the soft humanoid hand, which has a weight ofonly 300 g and is about 1.2 times the size of a typical humanhand.As shown in Figure 2(a), the soft humanoid hand in thispaper has three functional components: finger, palm part A andpalm part B. All these components are actuated pneumatically.The fingers are designed to grasp, grip and manipulate targetedobjects through bending in the grasping axis and deflecting inthe side-to-side axis. The five fingers of the hand are sorted asthumb and four planar fingers (index, middle, ring and littlefinger). Palm part A is used to splay the four planar fingers,which extend the distance between fingers and enlarge thegrasping scope. Palm part B achieves two functions: palmbending and thumb abduction and adduction.Simulation was conducted to guide the design and to ensurethe proposed design can work as intended in real-time. Weused finite element modeling (FEM) to simulate and analyzethe whole soft hand by following the analysis in [17]. Thesimulation in real-time was implemented in SOFA framework[16] with SoftRobots plugin [18]. The mesh file of the softhand consists of 40316 tetrahedra and 11894 nodes.
B. Soft finger1) Function:
In this work, we aim to develop a humanoidpneumatic soft hand with dexterous soft fingers. Among thefive fingers, the functions of the thumb, index and middlefingers are the main ones for grasp manipulation, while ringand little finger are usually play an assistant role. Index andmiddle fingers have two functional goals: bending in theprimary grasping axis (toward the soft palm) and deflectingfrom side to side (perpendicular to the primary graspingdirection). The function of the ring and little fingers in thiswork is only bending motion.Bending motion is the basic and main function for grasping.The bending angles are set to about 180 ◦ like human fingers.Side-to-side deflection motion is also useful and necessary topreform splaying and gripping between two adjacent normalplanar fingers and rotatory movement of the object amongthree fingers’ grasping.
2) Design:
The soft hand fingers, which are able to bothbend and deflect, are made of HBSF based on a modularapproach. HBSFs are used for the thumb, index and middlefingers. The structure of the 3 fingers is identical in shape witha length L of 90 mm, while the thumb is 20 mm shorter thanthe index and middle fingers. The structure of ring and littlefingers is a simplified version of HBSF, which can only bend to the main direction. The two side by side chambers in onesegment of HBSF was omitted into one chamber.In this work, the HBSF combines the advantages ofPneu-nets bending actuators [4] and fibre-reinforced actua-tors [5]. About the configuration, we refer to the humanoidmorphology of the soft fingers in Deimel et al. [1]. Asshown in Figure 3(a), the main body of HBSF consists ofsilicone materials, a strain- limiting layer in inextensible butflexible materials (silk screen) in its bottom, and the fibre-reinforcement thread. The strain- limiting layer determines thebending motion of the finger. And the fibre-reinforcements canprotect the silicone-based chambers from excessive expansion.The tubes are used to pressurize the air cavities insidefingers. A flex sensor (Spectra Symbol FS-L-0095-103-ST)is glued to the bottom flat face. The structure of the HBSF islike a bellow. From the sectional drawing in the bottom planein Figure 3(b), the most recent version used in index/middlefingers has 11 short bellows (called as 11 segments in thispaper) with a wall thickness t = 2 mm. The two sections of thefinger are shared with the same length, which means l = l .The little gap between segments a is equal to 1 mm, which islimited by the thin-wall strength of the 3D printed molds. Thedistance between each short bellows b is the key parameter,which will affect the bending and deflection performance ofthe HBSF. A HBSF is divided into four air cavities.
3) Simulation:
Figure 3(c)-(i) present the simulation resultsof the soft fingers. Because the SOFA framework is developedfor simulation in real-time, the fibre-reinforcement structure isso thin to mesh it into element based on the FEM. Thus, it isomitted in the simulation. As a result, the input air pressurein SOFA simulation cannot fit the capability of the HBSF inreal-world, which limits the bending angle of the fingers inthe simulation results.The fingers’ deflection enables several useful collaborativemotions between the index and middle fingers. Although themotions in Figure 3(f) and (g) cannot work as a gripper dueto the limitation of the low stiffness of the soft fingers in theside-to-side axis, they play a key role in achieving the Feix[15] grasp postures 14, 17, 20, 21 and 23 in Figure 10.In addition, the deflection of the thumb acts similar to thefunctions shown in Figure 3(h) and (i). The thumb deflectstoward the palm for grasping small objects, while deflectsaway from the palm for grasping bigger objects.
C. Soft palm1) Function:
For human hand grasping, the palm alwayscollaboratively works with the fingers to implement all kindsof manipulation. In this paper, the functions of the soft palmis inspired from human palm motion, which include threemain functions: thumb abduction, four planar fingers splayingand palm bending. A key feature of the human hand is thethumb abduction and adduction. This feature allow the thumbto rotate and has a proper position and orientation with respectto the manipulated object, which is essential of grasping.The splaying of the four fingers enlarges the capabilityof grasping different size objects. By increasing the anglebetween the adjacent fingers, the contacting force from the
Fig. 3: The structure of the HBSF. (a) The schematic illustration of the components of HBSF. The fibre-reinforcement structure is presented by the thin raisedfeatures on the main body’s surface. (b) The sectional drawing (bottom view). The red chambers in (c)-(i) present the pressurized actuators in SOFA [16]. (c)The original morphology. (d) Pressurize the 2nd section of the index and middle fingers. (e) Pressurize the two sections together. (f) Deflection of two fingersto enlarge the grasping region. (g) Deflection of two fingers to decrease the gap between them. (h)-(i) Deflection of two fingers to achieve their wiggling tothe left and to the right. four planar fingers will evenly distributed on the objects andit also make it possible to grip larger objects using twoadjacent fingers. The bending of the whole plane of the palm isbeneficial to facilitate the grasping of smaller objects. Withoutthe help of palm bending, the thumb of soft hand cannot touchthe other fingers even all the fingers are pressurized under thehighest pressure and bending with 180 ◦ .
2) Design:
The method of modular design are applied inthe design of soft palm. The three functional goals are dividedand embedded into two parts: palm part A and B, as shownin Figure. 4. The structural design of both two parts areinspired by the mechanisms of Pneu-nets actuators, consistingof networks-like chambers inside the silicone body and thestrain-limiting layer on the bottom.Palm part A achieves the splaying of the four planar fingers,while part B implements the abduction of the thumb andoverall bending of the palm. Instead of bending as Pneu-netsactuators, the palm part A splays the four planar fingers witha small hump after pressurized. As shown in Figure 4(a), themain body has four air chambers to be actuated. The four largegroove in front are used to fix the four fingers. The bottomcavity is designed to arrange the pneumatic tubes and sensorwires inside the hands. As for part B, the bending motion ofthe palm is achieved by enlarging the horizontal width of thePneu-nets actuator. The width of the air chambers along thepalm were designed as wide as possible to generate enoughforce to bend the palm with four planar fingers. The tubingtunnel connecting with the bottom cavity in part A is used torun the tubes and lines through the palm.The soft palm uses the same key geometric parameters with the soft fingers, such as the wall thickness t and thelittle gap between segments a . The stiffness of the palm forenabling a reliable support for the other fingers is consideredand simulated in SOFA framework. In addition, the humanoidappearance design were implemented at last, after ensuringthe soft actuators work as intended and it will not affect theperformance of the pneumatic actuators.
3) Simulation:
Figure 4(b) and (d) present the simulationresults of the two parts of the soft palm. The three functionsof the palm can be clearly detected through the comparisonof each actuators before and after being pressurized.III. F
ABRICATION
A. Actuator body molding
The soft body was fabricated using the silicone rubber(Dragon Skin 10 Medium) by Smooth-On (Sil-Poxy) with a 10Shore-A hardness. Parts (A and B) were mixed (1:1 ratio) in aplastic cup and then put into a vacuum chamber. Using vacuumpump, the air trapped in mixed silicone materials will expandto bubble and finally collapse under about 0.9 bar vacuumpressure. Afterward, the silicone material was poured into themolds as described in Step 1 and 2 of finger and palm parts inFigure 5. The molds are generally left in an upright positionto cure for five hours at room temperature.
B. Strain-limiting layer sealing
The strain-limiting layer is used to ensure the actuatorsbending in the desired direction during pneumatic pressuriza-tion [4]. Its material is silkscreen fabric. It is placed in the
Fig. 4: The structure and simulation of the soft palm. (b) and (d) Show acomparison of the palm actuators under inflation and deflation states. The redchambers in (b) and (d) presents the pressurized actuators in SOFA. bottom mold of the Step 2 in Figure 5 and glued togetherwith the bottom sealing portions of all three components.Subsequently, the fabrication of the main body of the actuatorsis finished and the chambers inside the silicone body are allsealed. It is noted that the Step 3 of Palm part B does notneed molds. As shown in in Figure 5(h1)-(h2), the silkscreenfabric is added manually in the reserved gap. Then, afterpouring silicone into the gap and curing, the silkscreen fabricis combined with the main body.
C. Assembly
Figure 5(d1)-(d4) shows the main procedure of the handassembly. There are five soft fingers and two soft palm partsin this hand. We used the same silicone materials to bonddifferent components together by manually pouring liquidsilicone on their connecting surface. As shown in Figure 5(d1),the end of four planar fingers is fixed on the large groovementioned in Section II-C2. The tubes of each finger arearranged into the bottom cavity inside the palm part A.Figure 5(d2) presents the assembly result of the palm partB and a soft-rigid hybrid support. The support consists ofsilicone skin fitting with the shape of the palm part B andthe rigid 3D-printed skeleton inside. All the pneumatic tubeswill go through the support and run out the hand together.The yellow 3D-printed skeleton of the support is used as thebase to fix the hand with the robot arm. As a result, the threeparts in Figure 5(d1)-(d3) are assembled together into a softhumanoid hand, as shown in Figure 5(d4).IV. C
ONTROL
To control the proposed soft hand at different operatingconditions, a controller platform is constructed. As the de-signed soft actuators are pneumatic, controlling the pressureand the duration of the input pneumatic supply is needed.The controller platform is implemented based on the proposeddesign of the soft robotics toolkit . The architecture of ourcontroller board is shown in Figure 6.The controller board consists of a pneumatic regulator(which regulates the pressurized air to the system), a set ofsolenoid valves (which can open and close to direct the flowof air into the system). The valves are powered and directedby power FET switches. As we want the hand to have as muchdegree of freedom as possible, we use 20 solenoid valves intotal. Thus, each chamber of the finger can be pressurizedindependently, which in turn, allow the hand to have morepostures. Two Arduinos Mega 2560 REV3 controller are usedto enable users to interface with the hardware via a serialport connection. The board can be controlled manually (byadjusting switches and potentiometers) or automated via aprogrammed software.The system pressure is regulated with pulse-width mod-ulation (PWM), which basically controls the opening andclosing times of the valves, at a rate of 60 Hz through theArduino boards. PWM can be expressed as a technique forgetting analog results with digital means. One of the mostimportant terms in PWM is the duty cycle. The duty cycle isthe proportion of ’on’ time to the regular interval or ’period’of time. Duty cycle is expressed in percent, 100% being fullyon, and 0% being fully off. By modulating the value of theduty cycle, analog values can be achieved. For example, thevalve fully closes at 0% duty cycle, fully opens at 100% dutycycle and opens halfway at 50% duty cycle. Thus, the fixedregulated input pressure is set to the desired value based onthe duty cycle of the PWM signal. With this technique, thefinger can easily be controlled to a certain bending angle. Fluidic Control Board, https://softroboticstoolkit.com/book/control-board Step 1: Silicone bodyStep 1: Silicone bodyStep 1: Silicone body Step 2: Strain-limiting layer sealingStep 2: Strain-limiting layer sealingStep 2:Strain-limitinglayer sealing Step 3: Sensor & fibre-reinforcementStep 3: Silkscreen addition Step 4: Thumb base F i ng e r P a l m p a r t A P a l m p a r t B H a nd a ss e m b l y (a1) (a2)(a3) (a4) (b1) (b2)(b3) (b4) (c1) (c2)(c3) (c4)(d1) (d2)(d3) (d4) (e1) (e2)(e3) (e4) (j1) (j2)(j3) (j4)(f1) (f2)(f3) (f4) (g1)(g2) (g3)(g4) (g5) (h1)(h2) (i1) (i2)(i3) (i4) Fig. 5: The fabrication process of the soft hand.
Finger : (a1)-(a4) show that the main body of the HBSFs. It is made by first pouring silicone material in itsbottom mold and then press the top mold. In Step 2, the red part in (b3) is for clamping and positioning the main body. (c1)-(c4) is to fix the fibre and flexsensor to prevent relative displacement when the finger bends.
Palm part A : (d1) The molds. (d2) Pure silicone material. (d3)-(d4) Unmold. In Step 2, the3D printer support is used to hold and ensure the thickness of the bottom surface.
Palm part B : The process of making the main and bottom body is similarwith the other parts. The 2 red sticks in (g1)-(g5) are to reserve the space for air tubes and electric wire. The yellow support in (g2) works as a similar roleof the red in (b3).
Hand assembly : (j1) The assembly of palm and 4 planar fingers. (j4) The final result.Fig. 6: The architecture of the implemented controller board. For clarity, weonly show the connection scheme of the Arduino 1. The connection schemeof the Arduino 2 is identical to that of the Arduino 1.
V. E
XPERIMENTS
A. Analysis of the HBSF
The HBSFs in this paper have novel motion. They deflectin the side-to-side axis, while the bending motion in thegrasping axis is still the main function of the soft fingers. Afterpressurized, the adjacent inflated segments will generate amutual force, which achieve the deflection motion. In addition,the number of segments depends on the distance b , as shownin Figure 3(b). When the finger length is defined, larger b leads to a less segment number. When the finger only has 1segment, it changes back to the original paradigm of PneuFlexactuator with two chambers. In order to analyze the influenceof the segments number on the deflection, six soft fingerswith 1, 3, 5, 7, 9, 11 segments are fabricated and tested, asshown in Figure 7. The bending angle increases non-linearlywith the applied air pressure, with a relatively slow increasein low pressure region and a rapid increase afterward whenthe pressure reaches beyond 20 kPa. It is obvious that thevariation between the force and bending angle has a closepositive correlation. However, it is also noted that under thesame air pressure, the fingers with more segments generate alarger bending angle and force when pressurizing both left Fig. 7: The analysis of the effect of the segments number on the bending anddeflection motions of the six fingers. (a) The comparison of the deflectionextent from front to back, respectively S11, S9, · · · , S1; (b) The comparisonof the bending angle from front to back, respectively S1, S3, · · · , S11. (c) Thesectional drawing of the six fingers. (f) The experimental setup for measuringthe force of the fingers while the top layer of the fingers was constrained.Variation of the (d) bending angle and (g) generated force of fingers whileapplying air pressure equally to both left and right chambers. Variation of the(e) bending angle, (h) generated force of fingers and (i) displacement of thefingertip while applying air pressure to single chamber. and right chambers but generate a lower angle and forcewhen pressurizing single chamber. This phenomenon actuallyreflects the different mechanisms of the bending and deflectionmotion. The more segments means more grooves that mutuallyswell during inflation working as PneuNets actuators andless little sections that extend during inflation working asfibre-reinforcement actuators. As shown in Figure 7(h), thegenerated grasping force of the S1 finger becomes larger thanthose with multi-segments, when the applied air pressure goesfrom 20 to 50 kPa.Figure 7(i) shows the deflection displacement in the side-to-side axis. The S11 finger has significantly better performancefor deflecting up to 20 mm. Combining the experimentalresults of bending angle and force, the S11 finger ranks thesecond that is easily driven under double-chamber actuationand present the minimum bending angle under single-chamberactuation. Both characteristics are conducive to enhance thedeflection region of the HBSFs. It is also notable that the fingerwill first wiggle outwards in the initial stage of the actuationwith lower air pressures and turn back when the bending angleof the HBSFs becomes large at higher air pressure, resultingin losing efficacy of the deflection motion and even get justthe opposite displacement. As a conclusion, we selected theHBSF with 11 segments in our new hand.
B. Analysis of the palm
Figure 8 presents the experimental results of the soft palm.Different from the normal fixed palm, almost all the partsof our soft palm is deformable, which greatly increases thediversity of the hand posture.Figure 8(e) shows the relationship between the deformationperformance and the air pressure of the palm Part A. It
Fig. 8: The validation of palm performance on bending angle and output force.(a,e) palm splaying; (b,c,f) palm bending with palm down and palm up; (d,g)thumb abduction. indicates that the splaying angle and force increase slowlywhen the applied air pressure increases from 10 to 60 kPaand then achieve a rapid increase afterward. Figure 8(a) showsthe posture of the hand under 90 kPa air pressure with a50 ◦ splaying angle. The air pressure should not be more than100 kPa in order to avoid collapse.The influence of the gravity on palm bending can beobserved in Figure 8(b) and (c). The bending angles of thepalm with down pose are obviously larger than those withthe palm up pose. The tested maximum angle of the palmbending motion is up to 68 ◦ and of the thumb abduction isabout 90 ◦ . The two bending ranges can meet most of thecommon grasping. Without the fibre-reinforcement, the twoactuators of the palm part B should be driven under the saferange of air pressure less than 40 kPa. However, when theobject is grasped and in contact with the hand, the pressurecan be manually increased to enhance the reliability of thegrasping. C. Grasping in real-world thumb abduction.
Fig. 9: (a)-(e) present the capability of our soft hand prototype of performinglifting a watering can (143 g) by using 2 adjacent HBSFs (f) Grasping andlifting a chair (541 g) (g,h) Show the safely compliance in human-robotinteraction.
The deflection of the HBSFs can collaboratively work usedas a two-finger gripper to grasp and lift objects that allowthe insertion of fingers into the neck portion, as shown inFigure 9(a)-(e). From Figure 9(b) to (c), the index and middle
Fig. 10: Enacted grasps of the Feix taxonomy. Grasps are numbered according to the Feix taxonomy [15]. Please check the attached video for the 32 grasps,https://irobotics.aalto.fi/wp-content/uploads/2020/09/NewSoftHandvFast576p30.mp4
HBSFs illustrate the deflection motion as illustrated in Figure3(f). The gap between the index and middle fingers enlargesobviously. The bending motion can be used to provide thesupport force against gravity in case of the watering can.Besides, our hand is able to firmly grasp and lift the heavyand large-size chair in Figure 9(f). Figure 9(g) and (h) showthe safety and compliance of the soft hand in human-robotinteraction scenarios.
D. Grasp dexterity in Feix taxonomy
To test the grasping performance of our hand, we im-plemented the grasping experiments according to the Feixtaxonomy [15], which includes 33 comprehensive grasp types.For every cases, the pressures are adjusted to reach the desiredposture and the actuation sequences are generally follow the principle of actuating the palm first and then driving thefingers. The actuation of the soft palm includes the splayingof the planar four fingers, palm bending and thumb abduction.It help to firstly fit the shape and size of the target objects andmove the fingers in the proper position and orientation. Thenthe fingers are pressurized to implement the grasping tasks.The grasp quality is judged by moving up, down, and rotatingthe hand by Franka Emika Panda robotic arm under the speedof 40 mm/s. Furthermore, this procedure is repeated severaltimes to evaluate the quality of the grasp.Fig. 10 shows snapshots of 33 grasping posture types inFeix taxonomy [15]. Our hand failed to perform one graspingposture out of the 33. The ventral posture numbered as 32failed because the object (marker pen used in 32) is thin andlong, so the hand could not grasp it firmly. As the bending profile of the soft fingers has a circular shape, the innerdiameter of the bending fingers is larger than the diameter ofthe marker pen, even by reaching a maximum bending angle.VI. C
ONCLUSION
In this paper, we have successfully developed a new softhumanoid hand capable of grasping different kinds of objectsrobustly. The hand exhibits the advantages of large graspingforce, low cost, lightweight and potential applications to spe-cial cases. The pneumatic actuation enables the quick responseof grasping in high compliance without damaging the objects.Meanwhile, we proposed a new design of soft finger, HBSF,and a novel soft palm with 2 parts. The functions of eachsoft actuators were simulated based on FEM in SOFA. Theexperimental results on the 6 fingers with different accountof segments of the HBSF show that the soft finger with 11segments is the best choice for achieving both the bendingand deflection motions.The main advantage of this study is that the postures ofthe hand can be adjusted with the help of the soft palm, andthen we can use different configurations to realise stable anddexterous grasping. The hand allows us to achieve 32 out of33 grasp postures in Feix taxonomy, which demonstrates thepostural dexterity of our soft hand.A design limitation is that the thumb abduction angleand fingers splaying angle are limited by the silicone tensilestrength and wall thickness of the air chamber structure. Andthe Flex sensors embedded in soft finger are not well usedin the controller. Future work would address the optimisationof the soft palm to achieve that the thumb can touch to littlefinger and improve the control with safety and sensing.R
EFERENCES[1] R. Deimel and O. Brock, “A novel type of compliant and underactuatedrobotic hand for dexterous grasping,”
The International Journal ofRobotics Research , vol. 35, no. 1-3, pp. 161–185, 2016.[2] X. Liu, Y. Zhao, D. Geng, S. Chen, X. Tan, and C. Cao, “Soft humanoidhands with large grasping force enabled by flexible hybrid pneumaticactuators,”
Soft Robotics , ahead of print.[3] J. Shintake, V. Cacucciolo, D. Floreano, and H. Shea, “Softrobotic grippers,”
Advanced Materials , vol. 30, no. 29, p. 1707035,2018. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/adma.201707035[4] B. Mosadegh, P. Polygerinos, C. Keplinger, S. Wennstedt, R. F. Shep-herd, U. Gupta, J. Shim, K. Bertoldi, C. J. Walsh, and G. M. Whitesides,“Pneumatic networks for soft robotics that actuate rapidly,”
AdvancedFunctional Materials , vol. 24, no. 15, pp. 2163–2170, 2014.[5] K. C. Galloway, P. Polygerinos, C. J. Walsh, and R. J. Wood, “Me-chanically programmable bend radius for fiber-reinforced soft actuators,”in ,2013, pp. 1–6.[6] A. Lotfiani, H. Zhao, Z. Shao, and X. Yi, “Torsional Stiffness Improve-ment of a Soft Pneumatic Finger Using Embedded Skeleton,”
Journalof Mechanisms and Robotics , vol. 12, no. 1, 10 2019.[7] S. Abondance, C. B. Teeple, and R. J. Wood, “A dexterous soft robotichand for delicate in-hand manipulation,”
IEEE Robotics and AutomationLetters , vol. 5, no. 4, pp. 5502–5509, 2020.[8] Y. Sun, Q. Zhang, and X. Chen, “Design and analysis of a flexiblerobotic hand with soft fingers and a changeable palm,”
AdvancedRobotics , ahead of print.[9] C. B. Teeple, T. N. Koutros, M. A. Graule, and R. J. Wood, “Multi-segment soft robotic fingers enable robust precision grasping,”
TheInternational Journal of Robotics Research , ahead of print. [10] P. Polygerinos, Z. Wang, J. T. B. Overvelde, K. C. Galloway, R. J.Wood, K. Bertoldi, and C. J. Walsh, “Modeling of soft fiber-reinforcedbending actuators,”
IEEE Transactions on Robotics , vol. 31, no. 3, pp.778–789, 2015.[11] J. Fras and K. Althoefer, “Soft biomimetic prosthetic hand: Design,manufacturing and preliminary examination,” in , 2018,pp. 6998–7003.[12] M. H. Mohamed, S. H. Wagdy, M. A. Atalla, A. R. Youssef, and S. A.Maged, “A proposed soft pneumatic actuator control based on angleestimation from data-driven model,”
Proceedings of the Institution ofMechanical Engineers, Part H: Journal of Engineering in Medicine ,vol. 234, no. 6, pp. 612–625, 2020.[13] K. C. Galloway, K. P. Becker, B. Phillips, J. Kirby, S. Licht, D. Tchernov,R. J. Wood, and D. F. Gruber, “Soft robotic grippers for biologicalsampling on deep reefs,”
Soft Robotics , vol. 3, no. 1, pp. 23–33, 2016.[14] W. Hu and G. Alici, “Bioinspired three-dimensional-printed helical softpneumatic actuators and their characterization,”
Soft Robotics , vol. 7,no. 3, pp. 267–282, 2020.[15] T. Feix, J. Romero, H. Schmiedmayer, A. M. Dollar, and D. Kragic,“The grasp taxonomy of human grasp types,”
IEEE Transactions onHuman-Machine Systems , vol. 46, no. 1, pp. 66–77, 2016.[16] J. Allard, S. Cotin, F. Faure, P.-J. Bensoussan, F. Poyer, C. Duriez,H. Delingette, and L. Grisoni, “Sofa-an open source framework formedical simulation,” in
Medicine Meets Virtual Reality (MMVR15) ,2007.[17] C. Duriez, “Control of elastic soft robots based on real-time finiteelement method,” in , 2013, pp. 3982–3987.[18] E. Coevoet, T. Morales-Bieze, F. Largilliere, Z. Zhang, M. Thieffry,M. Sanz-Lopez, B. Carrez, D. Marchal, O. Goury, J. Dequidt, andC. Duriez, “Software toolkit for modeling, simulation, and control ofsoft robots,”