Safe Grasping with a Force Controlled Soft Robotic Hand
SSafe Grasping with a Force Controlled Soft Robotic Hand
Tran Nguyen Le, Jens Lundell and Ville Kyrki
Abstract — Safe yet stable grasping requires a robotic handto apply sufficient force on the object to immobilize it whilekeeping it from getting damaged. Soft robotic hands have beenproposed for safe grasping due to their passive compliance, buteven such a hand can crush objects if the applied force is toohigh. Thus for safe grasping, regulating the grasping force isof uttermost importance even with soft hands. In this work,we present a force controlled soft hand and use it to achievesafe grasping. To this end, resistive force and bend sensors areintegrated in a soft hand, and a data-driven calibration methodis proposed to estimate contact interaction forces. Given theforce readings, the pneumatic pressures are regulated using aproportional-integral controller to achieve desired force. Thecontroller is experimentally evaluated and benchmarked bygrasping easily deformable objects such as plastic and papercups without neither dropping nor deforming them. Together,the results demonstrate that our force controlled soft hand cangrasp deformable objects in a safe yet stable manner.
I. I
NTRODUCTION
A stable yet safe grasp requires a robotic hand to applysufficient force on the target object to not harm it but stillimmobilize it. Recently, soft hands have been proposed forsafe grasping due to their passive compliance. However, asshown in Fig. 1 even such hands can deform objects if theapplied force is too high. There is, therefore, a need to controlthe grasping force also in soft hands.Research on soft hands has mainly focused on hand design[1]–[10]. Only recently have works started to address theintegration of sensing capabilities such as position [11]–[14] and force sensing [15]. However, the authors foundthe existing solution for force sensing to be insufficient forcontact detection, especially for small objects [15]. In thiswork, we aim to show that force estimation is possible toallow both contact detection as well as control of contactforces for safe grasping.To this end, we propose to integrate off-the-shelf resistivebend and force sensors in each finger of a hand to estimatethe contact force and feed it into a force controller for eachfinger. In order to extract reliable force measurements, wepropose a data-driven sensor characterization and calibrationmethod for such sensors. Given the calibrated sensors, thereal contact force is estimated by subtracting the residualforce the sensor senses when bending in free space fromthe actual force reading. The residual force is also learned.The estimated force is then used in a Proportional-Integral(PI) controller to control the set-point force for each finger.
This work was supported by the Strategic Research Council at Academyof Finland, decision 314180.T. Nguyen Le, J. Lundell and V. Kyrki are withSchool of Electrical Engineering, Aalto University, Finland. { firstname.lastname } @aalto.fi Fig. 1: On the left is shown that grasping a fragile objectwith a high force will deform it completely while, as shownon the right, using force-feedback to control the graspingforce this is no longer the case.The complete force controller is experimentally validatedand compared to no force controller on a physical graspingexperiment where the objective is to grasp deformable andfragile objects in a stable manner without neither destroyingnor deforming them.The main contributions of this work are: (i) integration ofresistive bend and force sensors in a soft hand (Section III),(ii) calibration of the sensors to estimate contact force (Sec-tion IV), (iii) a force controller for the soft hand (Section V),and (iv) an empirical evaluation of the controller demonstrat-ing that using a force controller deformable objects can begrasped in a stable yet safe manner (Section VI).II. R
ELATED W ORK
Most of the research in soft robotic hands has focused onthe hand design itself where the main actuators have been theSoft Pneumatic Actuator (SPA) [16]. SPAs are constructed ofsoft and deformable matter and are pneumatically actuated.Such actuators were used in the pioneering soft hand designby Ilievski et al. in 2011 [3] where they presented a three-layer starfish inspired soft gripper using embedded pneumaticnetworks or PneuNets. The multilayer structures, where twoactive layers were separated by a passive layer, allowed thegripper to perform a wide range of motions, but due toits construction it could only withstand a small amount ofpressure resulting in a weak grasping force.Recently, alternative soft hand designs that can attainhigher grasping forces have been proposed, including thewell known anthropomorphic RBO Hand 2 [5] and DRL a r X i v : . [ c s . R O ] A ug oft Hand [17]. The RBO Hand 2 consists of seven fiber-reinforced pneumatic continuum actuators acting as fingersand palm. Although the anthropomorphic design enableddexterous grasping, one downside is the negative curvatureof the thumb, meaning that the backside of the thumb isthe primary contact surface rather than its front side, whichmakes it problematic to attach sensors to it. Compared tothe RBO Hand 2, the kinematics of the DRL Soft Hand,where two fingers are parallel and opposes the third, alloweasy sensor integration and was, therefore, the source ofinspiration for the hand proposed in this work.Despite the wide range of research in soft hand designs,few works have studied how to incorporate sensing capa-bilities in such hands. One possible reason for the lackof research in sensorized soft hands is the few existingcheap off-the-shelf sensors that can be integrated into aSPA. However, recently Elgeneidy et al. [11] proposed thefollowing sensors for measuring and controlling position ofSPAs: (i) conductive elastomer sensors, (ii) sensors madefrom liquid metal or (iii) resistive flex sensors.Conductive elastomer sensors were used in [18] to controlthe displacement of the soft gripper with an adaptive neuro-fuzzy inference strategy. Unfortunately, the accuracy of suchsensors quickly deteriorates as the distribution of carbonparticles inside the elastomer material is disturbed by theactuator’s repeated deformation. The more durable liquidmetal sensors were integrated into a soft gripper in [19] todetect the presence of grasped objects, and in [20] to acquirepressure, force, and position data which was subsequentlyfed to a Proportional-Integral-Derivative (PID) controllerto control the position of the finger and grasping force.However, to acquire the data, they used the eGain liquidmetal sensor [21] that was fabricated from scratch, which isa time-consuming and error-prone process. Additionally, theforce sensor was placed at the tip of the actuator, restrictingthe force controller to only work with pincer grasps or onlarge objects. In this work, we target objects of diverse sizes,including small objects and as such the eGain sensor is notapplicable.The third sensor type mentioned in [11], the resistive bendflex sensors, are cheap, widely available, and easy to integratein soft hands. For instance, resistive bend sensors have beenused for haptic identification of grasped objects [17] and forpredicting and controlling the position, i.e., , bending angleof a pneumatic-driven actuator [11]. The work in [17] waslater improved in [15] by adding a Force Sensitive Resistive(FSR) sensor to strengthen the grasp by detecting the contactbetween the hand and objects. They reported, however,that the force sensor provided unreliable data, resulting inextremely poor performance in contact detection, especiallyon small objects. This problem stems from placing the forcesensor only at the tip of the finger, which will not makecontact with small objects such as a tennis ball, resulting inno readings. In this work, we also use FSRs due to theirsimplicity and availability but opt for the kinds that measureforces along a wider area as they enable measuring forceseven on small objects. Fig. 2: A cross-sectional view of the actuator embedded withselected sensors.Fig. 3: The left figure shows the design of a resistive flexsensor and its behaviour in straight state. The right figureillustrates the behaviour of the sensor when it is bent in theright direction.III. R ESISTIVE B ASED S ENSING
To do complex manipulation tasks, robotic hands needposition and force feedback [22]. To achieve this for softhands, we propose integrating deformable resistive force andbend sensors onto the finger of the RBO Hand 2 proposedin [5]. Fig. 2 shows the sensors integrated into the fabricatedactuator. The body of the fabricated actuator is divided intotwo parts: an extensible and an inextensible layer.To keep the bend sensor in place, it was encapsulated inthe inextensible layer shown as the orange line in the figure.In contrast to the bend sensor, the force sensor illustrated bythe green line in the figure needs to be in contact with theenvironment to get measurements; thus it was glued directlyto the outer surface of the inextensible layer using freshlymixed Dragon Skin 10 silicone. Next, we describe the designand working principle of these sensors.
A. Position sensing
To measure the internal state (position) of SPAs we usedresistive flex sensors (bend sensors) . One side of the sensorconsists of a layer of polymer ink which is embedded withconductive particles, as shown in Fig. 3. The particles providethe ink with a certain amount of resistance when the sensoris straight. When the sensor bends away from the ink, theconductive particles move further apart as shown in Fig. 3which, in turn, increases the resistance. Once the sensorreturn to the initial pose, the resistance also returns to theoriginal value. Hence, the change in the resistance can beused to determine the curvature of the sensor. Flex Sensor 4.5”: ig. 4: The left figure shows the design of a strip FSR andits behavior when no pressure is applied. When pressure isapplied on the sensor, the conductive particles move andmake contact with the conductive film, resulting in moreconduction paths (lowering resistance).
B. Force sensing
To measure the grasping force applied by SPAs, we useFSRs. FSRs are designed to measure the presence andrelative magnitude of localized physical pressure. The FSRused to measure the force in [15] was placed only at thetip of the finger but, given the fact that soft hands typicallyare developed to mimic human’s grasping behaviour, i.e., wrapping an entire finger around the surface of an object,placing the force sensor at such a location is not optimalfor detecting contact between the hand and the object.Based on this, we opted for force sensors that can providemeasurements along the body of the whole finger rather thanonly at the tip. To meet this requirement, we chose a stripFSR with rectangle shape . The chosen sensor is a singlelarge sensing taxel with the force sensitivity ranges from . to N. In addition, the force resolution of the sensor isbetter than . of full use force.An FSR consists of a resistive polymer layer and aconductive film as illustrated in Fig. 4 and the workingprinciple is similar to that of the bend sensor. Specifically,the spacer placed in between the two layers to avoid contactbetween them results in a very high resistance when nopressure is applied. When pressure is applied, the resistivepolymer starts to make contact with the conductive film,which, in turn, reduces the resistance of the sensor. Thestronger the applied force is on the sensor’s active area, themore the resistance between the two terminals drops. Theactual applied force is mapped from the measured resistanceusing the resistance-force relationship graph provided in thedatasheet of the sensor .The design of the FSR enables easy contact detection instationary situations when the sensor is fixed to a surface.However, when the sensor is bent in free space the conduc-tive and resistive polymer layer can come in contact witheach other producing internal force readings which leadsto incorrect contact forces. In the next section, we proposea data-driven calibration method to correctly estimate thecontact force by compensating for the internal force. Force Sensitive Resistor - Long : https://cdn.sparkfun.com/datasheets/Sensors/Pressure/FSR408-Layout2.pdf IV. D
ATA -D RIVEN C ONTACT F ORCE E STIMATION
A. Internal Force Characterization
To characterize the internal force of the force sensor, afinger of our soft hand with integrated sensors was actuatedin free space with a 60 kPa ramp input. Once the internalpressure in the finger reached 60 kPa, the input was zeroed,and the finger returned to its initial position. The top left plotin Fig. 5 shows the force measurement of the force sensoragainst the bending angle after 35 repetitions. Theoretically,the force measurements should remain zero as there is nocontact between the finger and the environment. However,as seen in the figure, the force reading increases the morethe finger is bent. Due to such sensor characteristics, theFSR force measurements of a bent finger are incorrect. Thus,removing the effect of the internal force from the sensorreadings is crucial to enable force control of the fingers.
B. Contact Force Estimation
We model the actual contact force F c for the finger by thedifference between the measured force F m and the internalforce F i F c = F m − F i . (1)Thus, to estimate the actual contact force F c , we 1) learnthe internal force F i caused by bending in free space and2) subtract the estimated internal force from the measuredforce.We cast the learning of the internal force model F i foreach finger as a polynomial regression problem since therelationship between internal force and bending angle ofthe finger is non-linear (Fig. 5). Thus the internal force isexpressed as F i ( x ) = D (cid:88) d =0 w d x d = w T x , (2)where w = [ w , . . . , w D ] are the parameters to learn, x thebending angle of the finger, and D the maximum degree ofthe polynomial function. We estimate the model parametersby first mapping the bending angle x to a higher dimensionalfeature space using the feature map φ ( x ) = [ x d , . . . , T ,resulting in a transformed feature vector x = [ x d , . . . , T ,and then applying ordinary least squares so that the optimalparameters w ∗ can be found by w ∗ = ( X T X ) − X T F i .To choose the model complexity, that is the maximumdegree D of the polynomial function, we used the BayesianInformation Criterion (BIC) [23]BIC = ln( n ) k − L ) , (3)where ˆ L is the maximized value of the likelihood of themodel, n is the number of observations, and k is thenumber of parameters. The model with smallest BIC valueis considered the best.ig. 5: The upper row (from left to right) shows, as blue dots, the internal force for each finger (from finger 1 to finger 3)at different bending angles along with the fitted polynomial model in red. The bottom row shows the BIC values for eachfinger. V. FORCE CONTROLLERTo control the grasping force of the soft hand, we proposeusing a discrete PI controller u n = K p e n + K i T n (cid:88) k =1 e k , (4)where K p and K i are the proportional and integral termsrespectively, T is the sampling period, and e n is the forceerror between the target value and measured value at then-th moment of sampling. The output u n is a Pulse WidthModulation (PWM) signal. The reason why we chose to use aPI controller instead of a PID controller is that the derivativeaction is sensitive to noise. As the force measurement ofthe force sensor is usually noisy, the derivative action isneglected.VI. EXPERIMENTS AND RESULTSThe main questions addressed experimentally were:1) What is the accuracy of the contact force estimation?2) What is the accuracy of the force controller?3) Does the force controlled soft hand enable safe grasp-ing?In order to provide justified answers to these questions, weconducted three experiments. The first experiment examinesthe contact force estimation accuracy, the second one theproposed force controller accuracy while the third evaluatessafe grasping in terms of grasping deformable objects with-out neither damaging nor dropping them. A. Experimental setup
All the experiments were evaluated on a soft robotic handembedded with the bend and force sensors mentioned in Section III. For the hand design, we used the same fingerstructure as RBO Hand 2 [5] but opted for a kinematicstructure similar to the DRL Soft Hand [17] as it allowspower grasping. As a result, our soft hand consisted of threefingers, in which two fingers are on one side (we call them finger 1 and finger 2 ), and one finger is on the opposite side( finger 3 ). The final soft hand is shown in Fig. 1.To pneumatically control the hand, we used the softrobotics toolkit controller platform . The input pressure wasregulated with PWM, and the control rate was 60 Hz. How-ever, at such a high control rate, the high-speed switchingof PWM caused the hand to vibrate, which consequentlyinduced noise into the force readings. To decrease the noise,we implemented the pneumatic low pass filter proposed in[24]. B. Estimating contact force
Our hypothesis is that the internal force is correlated tothe curvature of the finger. To test this, we gathered forcedata at different bending angles for each finger. The datain Fig. 5 shows the force measurements for each finger atdifferent bending angles. Although the data for each finger,such as bending angle range differ from each other due to themanual fabrication process, we can clearly see the correlationbetween the bending angle and the force. Next, we fitted aseparate polynomial model (2) to the data for each finger.According to the BIC values presented in Fig. 5, a fourth-degree polynomial fitted the data for each finger best, whichis also indicated by the high R values. Fluidic Control Board, https://softroboticstoolkit.com/book/control-board ig. 6: Steps of the experiment conducted to evaluate theaccuracy of the estimated contact force using the proposedapproach.Fig. 7: Contact force estimation result of the force sensor.Next, using the models to predict the internal force, weevaluated the accuracy of the estimated contact force in(1). The steps of the experiment are illustrated in Fig. 6.The accuracy was measured by first actuating one fingerto press against a scale until the reading on the scalereached a predefined target set to an arbitrary value suchas 200 gram (approximately 2 N) in this case. Once thetarget was reached, the actual contact force was estimatedby subtracting the predicted internal force from the forcemeasurement obtained from the sensor. In addition, to eval-uate the accuracy of the proposed method at different fingerconfigurations, the distance d between the finger and thescale was increased after each experimental cycle.The estimated contact force against the reference contactforce at different bending angles is plotted in Fig. 7. Al-though the graph shows that the estimate contact force variesslightly around the actual contact force, the error only rangesfrom 0.01 to 0.15 N at most, which is within the tolerancefor the objects we will later grasp. All in all, estimating thecontact force by simply subtracting the internal force fromthe real measurement is sufficiently accurate. C. Force Control
Before conducting the experiment, we experimentally fine-tuned the proportional K p and integral K i gains of the forcecontroller to 10 and 1.5, respectively. To experimentallyevaluate the accuracy and stability of the force controller,each finger had to reach and track a variable reference signal.First, a finger was actuated with 65 kPa pressure to makecontact with a solid object. Then, the step reference signalwas increased from 0 to 3 N, and after 60 seconds the Fig. 8: Contact force response of the finger to step referencesignals.reference signal was set to 2 N. The sensory feedback fromthe embedded sensors was continuously fed to the internalforce predictive model to estimate the actual contact force.The difference between the target and current contact forcewas then fed to the PI controller in (4) whose output wasthe corresponding amount to be added to (or subtracted from)the current duty cycle signal. The experiment was repeated5 times. The contact force response of one finger is shownin Fig. 8. The force response of the finger closely followedthe step reference signals with a Root Mean Square (RMS)error of 0.1 N. Moreover, the settling time of the finger wasroughly 400 ms.To enhance the robustness of the grasping system, theforce controller should only be activated when contact isdetected. One option to achieve this is to utilize a switchingmechanism between position and force control. We exper-imentally tested such a switching mechanism by having afinger reach a target contact force of 2.5 N only after contactwas made. Initially, the finger was slowly actuated froma duty cycle of 0% until contact with a solid object wasdetected based on the estimated contact force. After contactwas established, the force controller was activated to regulatethe pressure to achieve the target contact force.The duty cycle output from the controller and the contactforce response are shown in Fig. 9. From the top figure,one can conclude that to achieve the target contact force thecontroller used a duty cycle value in the range of 45% - 55%.The main reason for the fluctuating duty cycle originatesfrom fluctuations in the estimated contact force responsecaused by residual oscillations in the sensory reading. Fromthe bottom figure, it is observed that the contact forceresponse settled to a value of approximately 2.5 N in roughly800 milliseconds. The RMS error between the measured andthe target contact force was 0.21 N. In conclusion, regardlessof the small fluctuations in the measured contact force andthe output duty cycle, the proposed control strategy wassuccessful in controlling the finger to achieve a target contactforce in reasonable settling time. Furthermore, the proposedcontrol strategy did not overshoot at the time of contact.Together, all these results confirm that it is possible tocontrol the interaction force between the soft hand andABLE I: The grasping results for the deformable objects. Target contact forceof the opposable finger (N) Empty plastic cup Empty paper cupDropped percentage Deformed percentage Dropped percentage Deformed percentage4 0% 100% 0% 100%3 0% 100% 0% 80%2 0% 90%
0% 20%
0% 10%
80% 0%0.5 60% 0% 100% 0%
Fig. 9: The top figure shows the change in duty cycle toachieve the desired contact force while the bottom figureshows the contact force response of the force controller. Theyellow region illustrates the duration the force controller wasactivated, the red dashed line represents the target contactforce of 2.5 N and the green arrow indicates the settlingtime.objects in a safe, accurate, and fast manner. Such featuresare of uttermost importance in safe grasping.
D. Safe Grasping
Based on the results from the previous experiments, ourproposed force controller exhibits characteristics for safegrasping. In safe grasping, the goal is to grasp deformableobjects that can easily deform, such as the plastic cup seenin Fig. 1, in a stable manner. We experimentally evaluated ifthe proposed force controlled soft hand can do safe graspingby grasping three sensitive objects: an empty plastic cup, anempty paper cup, and an empty eggshell all shown in Fig. 10.Of these objects, the empty plastic and paper cup representsdeformable objects while the empty eggshell represents afragile object.For this experiment, the soft hand was fixed to a handle,as shown in Fig. 1. As the soft hand had two fingers onone side and one on the opposite, to stabilize the objects Fig. 10: The target objects used in the safe grasping experi-ment.the targets contact forces of the three fingers were set insuch a way that the sum of the contact force of the twofingers was equal to that of the opposable finger. We testedsix different force set-points for the opposable finger: , , , . , , and . N. To evaluate a grasp, the hand first graspedthe object until the target contact force was achieved andthen we manually moved the handle upward 30 cm to liftthe object and finally shook it randomly. During the process,a grasp was successful if the object neither deformed in thecase of deformable objects nor broke in the case of fragileobjects nor slipped away from the hand. We evaluated tengrasps for each target force amounting to 60 grasps in total.The grasping result on the deformable objects are pre-sented in Table I. As suspected, a higher contact forcelowers the dropping rate but increases the deformation rate.The results show that even soft hands, which have beenproposed for safe grasping due to their passive compliance,can completely deform objects if the force is too high. Forinstance, the plastic cup, which is less durable than the papercup, can be deformed with as low a contact force as 1 N.The paper cup, on the other hand, can withstand up to 2 N ofcontact force. The minimum grasping force to successfullygrasp deformable object can also be deduced from the results.In addition, the experiment was also conducted on anempty eggshell to evaluate the proposed control strategyin the case of fragile objects. In this case, the eggshellnever broke not even with maximum target contact forces.However, with a soft hand that can generate much largergrasping forces, crushing fragile objects is indeed feasible.All in all, the results emphasize the need for integratingsensors into soft hands to manipulate deformable and fragileobjects.ig. 11: The top figure shows the two target objects in thisexperiment: a woolly hat and a spray can. Bottom figuresshow the experimental setup for both objects.
E. Object properties estimation
In addition to the previous experiments, we conductedone more extra experiment to show the potential of usingthe proposed soft hand to estimate object properties such ashardness. According to Yuan [25], the most important factorto estimate the hardness of an object is the relationship be-tween the geometry of the deformed object and the pressingforce. When pressing on harder objects, they deform lesscompared to soft objects, thus retaining larger slopes on thecontact surface [25]. In this work, the estimated contact forceis seen as the pressing force, and the bending angle of thefinger is considered as the deformation of the object. Thisexperiment investigates whether the relationship between thetwo can be used to realize the hardness of an object.For this experiment, the finger was fixed to a handle andactuated to make contact with two objects with differenthardness i.e., a solid spray can, and a woolly hat as shown inFig. 11. The target objects were placed in such a way thatthey will be in contact with the finger at 40% duty cycle.Both force and position sensory readings were then recordedfor analysis.Fig. 12 plots the relationship between the bending angleand the estimated contact force in both cases. The resultsshow that in the case of the spray can the bend angle remainsalmost constant while the contact force continues to increase.This means that the finger has been stopped by somethingstiff. And since the finger is kept actuating, it keeps pressingstronger against that stiff object resulting in the increaseof the contact force. However, in the case of the woollyhat, both the bending angle and the contact force increasesimultaneously after the contact. This indicates that the targetobject is not stiff enough to constrain the bending of thefinger after contact. Based on these results, it seems that the Fig. 12: The blue line represents the estimated contactforce when the finger bends in free space. The orange andgreen line represent the estimated contact force when thefinger makes contact with the spray can and the woolly hat,respectively.soft finger embedded with selected sensors can successfullydistinguish between a solid object and a soft object usingonly the sensory feedback.VII. CONCLUSIONSWe presented a force controlled soft robotic hand and usedit for safe grasping. The key component was to integrate bothresistive bend and force sensors onto the hand’s fingers andapply a data-driven method to estimate the actual contactforce between the fingers and the objects. The estimatedcontact force was then fed into a PI force controller to keepa constant grasping force. We experimentally validated ourforce controller by comparing it to no force controller ona grasping experiment where the objective was to graspdeformable objects in a stable manner without damagingthem. The results show that the force controlled soft handcould grasp the tested objects without neither dropping themnor causing significant deformation.All in all, the work presented here demonstrates that ap-plying a data-driven calibration method can make otherwiseunreliable force sensor readings reliable enough to be usedin a force feedback controller. This, in turn, poses newinteresting research questions. For instance, is it possibleto learn an accurate dynamics model from the force andbend data? If so, can such a model be used to do dexter-ous manipulation by a soft robotic hand with, e.g., modelpredictive control or model-based reinforcement learning?Dexterous manipulation, together with the safety inheritedfrom the softness properties of soft hands offer great potentialfor robots to interact with human in complex environments.R
EFERENCES[1] S. Schulz, C. Pylatiuk, and G. Bretthauer, “A new ultralight anthro-pomorphic hand,” in
Proceedings 2001 ICRA. IEEE InternationalConference on Robotics and Automation , vol. 3, May 2001, pp. 2437–2441.[2] A. M. Dollar and R. D. Howe, “The highly adaptive sdm hand: Designand performance evaluation,”
The International Journal of RoboticsResearch , vol. 29, no. 5, pp. 585–597, 2010.[3] F. Ilievski, A. D. Mazzeo, R. F. Shepherd, X. Chen, and G. M. White-sides, “Soft robotics for chemists,”
Angewandte Chemie InternationalEdition , vol. 50, no. 8, pp. 1890–1895, 2011.4] R. Deimel and O. Brock, “A compliant hand based on a novel pneu-matic actuator,” in , May 2013, pp. 2047–2053.[5] R. Deimel and O. Brock, “A novel type of compliant and underactuatedrobotic hand for dexterous grasping,”
Int. J. Rob. Res. , vol. 35, no. 1-3,pp. 161–185, Jan. 2016.[6] K. C. Galloway, K. P. Becker, B. Phillips, J. Kirby, S. Licht, D. Tcher-nov, 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.[7] J. Zhou, S. Chen, and Z. Wang, “A soft-robotic gripper with enhancedobject adaptation and grasping reliability,”
IEEE Robotics and Au-tomation Letters , vol. 2, no. 4, pp. 2287–2293, Oct 2017.[8] P. Glick, S. A. Suresh, D. Ruffatto, M. Cutkosky, M. T. Tolley, andA. Parness, “A soft robotic gripper with gecko-inspired adhesive,”
IEEE Robotics and Automation Letters , vol. 3, no. 2, pp. 903–910,April 2018.[9] P. Bryan, S. Kumar, and F. Sahin, “Design of a soft robotic gripperfor improved grasping with suction cups,” in , 2019, pp. 2405–2410.[10] T. Nishimura, Y. Suzuki, T. Tsuji, and T. Watanabe, “Fluid pressuremonitoring-based strategy for delicate grasping of fragile objects by arobotic hand with fluid fingertips,”
Sensors , vol. 19, p. 782, 02 2019.[11] K. Elgeneidy, N. Lohse, and M. Jackson, “Bending angle predictionand control of soft pneumatic actuators with embedded flex sensors -a data-driven approach,”
Mechatronics , vol. 50, pp. 234 – 247, 2018.[12] L. Chin, M. C. Yuen, J. Lipton, L. H. Trueba, R. Kramer-Bottiglio, andD. Rus, “A simple electric soft robotic gripper with high-deformationhaptic feedback,” in , May 2019, pp. 2765–2771.[13] G. Gerboni, A. Diodato, G. Ciuti, M. Cianchetti, and A. Menciassi,“Feedback control of soft robot actuators via commercial flex bendsensors,”
IEEE/ASME Transactions on Mechatronics , vol. 22, no. 4,pp. 1881–1888, 2017.[14] G. Soter, A. Conn, H. Hauser, and J. Rossiter, “Bodily aware softrobots: Integration of proprioceptive and exteroceptive sensors,” in , 2018, pp. 2448–2453.[15] B. S. Homberg, R. K. Katzschmann, M. R. Dogar, and D. Rus, “Robustproprioceptive grasping with a soft robot hand,”
Auton. Robots , vol. 43,no. 3, pp. 681–696, Mar. 2019.[16] J. Shintake, V. Cacucciolo, D. Floreano, and H. Shea, “Soft roboticgrippers,”
Advanced Materials , vol. 30, no. 29, p. 1707035, 2018.[17] B. S. Homberg, R. K. Katzschmann, M. R. Dogar, and D. Rus, “Hapticidentification of objects using a modular soft robotic gripper,” in , Sep. 2015, pp. 1698–1705.[18] D. Petkovi´c, M. Issa, N. D. Pavlovi´c, L. Zentner, and A. OjbaˇsI´c,“Adaptive neuro fuzzy controller for adaptive compliant robotic grip-per,”
Expert Syst. Appl. , vol. 39, no. 18, pp. 13 295–13 304, Dec. 2012.[19] R. Adam Bilodeau, E. L. White, and R. K. Kramer, “Monolithicfabrication of sensors and actuators in a soft robotic gripper,” in , Sep. 2015, pp. 2324–2329.[20] J. Morrow, H. Shin, C. Phillips-Grafflin, S. Jang, J. Torrey, R. Larkins,S. Dang, Y. Park, and D. Berenson, “Improving soft pneumatic actuatorfingers through integration of soft sensors, position and force control,and rigid fingernails,” in , May 2016, pp. 5024–5031.[21] Y. Park, B. Chen, and R. J. Wood, “Design and fabrication of softartificial skin using embedded microchannels and liquid conductors,”
IEEE Sensors Journal , vol. 12, no. 8, pp. 2711–2718, Aug 2012.[22] K. Hertkorn, M. A. Roa, and C. Borst, “Planning in-hand objectmanipulation with multifingered hands considering task constraints,”in ,May 2013, pp. 617–624.[23] G. Schwarz, “Estimating the dimension of a model,”
The Annals ofStatistics , vol. 6, no. 2, pp. 461–464, 1978.[24] M. Memarian, R. Gorbet, and D. Kulic, “Control of soft pneu-matic finger-like actuators for affective motion generation,” in , Sep. 2015, pp. 1691–1697.[25] W. Yuan, M. A. Srinivasan, and E. H. Adelson, “Estimating objecthardness with a gelsight touch sensor,” in2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)