3216 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 3, NO. 4, OCTOBER 2018

Soft Inflatable Sensing Modules for Safe andInteractive Robots

Taekyoung Kim , Student Member, IEEE, Sohee John Yoon , and Yong-Lae Park , Member, IEEE

Abstract—Recent advancements of technologies in human–robotinteraction (HRI) significantly expanded the applications of robotsfrom specialized areas to our daily lives. However, safety is not al-ways guaranteed in traditional robots since they are mostly madeof rigid materials and structures, which cause serious injuries tohumans upon physical collisions. In this letter, we propose a softinflatable module with self-contained tactile sensing for safe HRIapplications. The proposed sensing module is made of a highlystretchable elastomer skin containing embedded microfluidic tac-tile sensors and a rigid bone structure. The module not only ac-tively inflates and deflates, but also detects surface contacts usingmicrofluidic soft pressure sensors to prevent human injuries causedby unintended collisions with robots and allowing follow-up actionsto be taken, such as halting and evading. This letter describes thedesign and fabrication of the proposed inflatable sensing module,characterizes the performance of the soft sensors with different in-flation levels, and evaluates the safety aspect of the module throughcollision tests. Finally, we present an example of applications forthe proposed system by integrating multiple modules into a roboticarm with 4 DOF. The arm was tested for evasion at different levelsof contacts.

Index Terms—Soft material robotics, force and tactile sensing,physical human-robot interaction, robot safety.

I. INTRODUCTION

R ECENT advancements of technologies in human robot in-teractions (HRI) significantly expanded the applications

of robots from specialized areas, such as industrial and medical

Manuscript received February 24, 2018; accepted June 7, 2018. Date ofpublication June 27, 2018; date of current version July 13, 2018. This letter wasrecommended for publication by Associate Editor W. Wang and Editor Y. Sunupon evaluation of the reviewers’ comments. This work was supported in partby the Technology Innovation Program under Grant 2017-10069072 fundedby the Ministry of Trade, Industry & Energy, South Korea, and in part by theNational Research Foundation under Grant NRF-2016R1A5A1938472 fundedby the Korean Government (MSIP). (Corresponding author: Yong-Lae Park.)

T. Kim and S. J. Yoon are with the Department of Mechanical and AerospaceEngineering, Soft Robotics Research Center, and Institute of AdvancedMachines and Design, Seoul National University, Seoul 08826, South Korea(e-mail:,kimtkm@snu.ac.kr; jylight@gmail.com).

Y.-L. Park is with the Department of Mechanical and Aerospace Engineer-ing, Soft Robotics Research Center, and Institute of Advanced Machines andDesign, Seoul National University, Seoul 08826, South Korea, and also withthe Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213 USA(e-mail:,ylpark@snu.ac.kr).

This letter has supplementary downloadable material available at http://ieeexplore.ieee.org, provided by the authors. The material consists of a videothat demonstrates the inflation and deflation performance of soft inflatable sens-ing modules, and shows evasive responses depending on the magnitude of thecontact forces detected by the modules. The size of the video is 2.58 MB. Con-tact kimtkm@snu.ac.kr and jylight@gmail.com for further questions about thiswork.

Digital Object Identifier 10.1109/LRA.2018.2850971

fields, to more common areas, such as our daily lives. Examplesinclude assistive robots for elderly people independently livingat home [1]–[3], healthcare robots through social communica-tion [4], [5], and interactive educational robots [6], [7]. Whendesigning such robots, one of the most important requirementsis safe physical interaction with human users. However, safety isnot always guaranteed in traditional robots, since they are mostlymade of rigid materials and structures which cause serious in-juries to humans upon physical collisions. To avoid dangerouscontacts, efforts have been made in preventing collisions withhumans using information from the surroundings collected bydifferent types of sensors, such as vision systems, range finders,or force and tactile sensors [8], [9]. These methods, however,cannot mitigate damage from collisions that occur due to controlor system errors. Therefore, the use of soft robotics technologieshere becomes attractive because it can significantly reduce oreven eliminate such serious physical injuries.

In soft robotics, highly deformable elastomers are widelyused for developing stretchable sensors [10]–[12] and compli-ant actuators [13]–[15]. Robots may increase their safety levelswhen combined with these types of soft components [16]. It ispossible to make the structures of robots entirely of soft mate-rials, as shown with soft manipulators [17]–[19] and wearablerobots [20]–[22]. Another approach is to cover or build eitherdesignated parts or the entire robot body using inflatable struc-tures. Ohta et al. have proposed such methods by using inflat-able sleeves made of heat sealable polymer materials to coverthe rigid bone structure of a robotic arm [23]. Qi et al. have de-veloped lightweight inflatable arms for interactive robots [24],and Hawkes et al. have also used an inflatable structure for theextension and navigation of a robotic arm [25].

Although the above approaches provide a certain level ofstructural safety, most of them lack the capability to directlysense external contacts through their body structures, whichwould significantly increase robot responsiveness and autonomythrough control.

We therefore propose a soft inflatable module with self-contained sensing. The proposed inflatable sensing moduleis composed of a rigid bone structure and a soft sensingskin. The skin contains embedded microchannels filled with aroom-temperature liquid metal, which in this case is eutecticgallium-indium (EGaIn). Microchannel geometries, when de-formed, change its electrical resistance and allow the detectionof physical contacts. Additionally, the skin is made of a highlystretchable elastomer that enables it to not only easily expandwith inflation, but also shrink with deflation if a compact form

2377-3766 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

KIM et al.: SOFT INFLATABLE SENSING MODULES FOR SAFE AND INTERACTIVE ROBOTS 3217

Fig. 1. Concept of soft inflatable sensing modules integrated with a roboticarm for increased human safety in HRI applications.

factor is more desirable depending on applications. During thisinflation and deflation, the fluidity of the liquid conductor main-tains the electrical conductivity of the microchannels. Fig. 1shows the overall concept of the proposed inflatable sensingmodule.

The rest of the letter is organized as follows. We first describethe main design features and the manufacturing processes inSections II and III, respectively. We then experimentally charac-terize the proposed system in Section IV and demonstrate safetythrough collision tests in Section V. Finally, we discuss appli-cation scenarios of multiple modules integrated into a roboticarm in Section VI followed by future work and conclusion inSection VII and Section VIII.

II. DESIGN

A. Inflatable Module

The inflatable module is made of a rigid plastic structurecomposed of a center column and two caps at either ends. Thecenter column, or bone, is covered with a soft skin layer madeof a hyperelastic elastomer (Dragon Skin 10, Smooth-On), asshown in Fig. 2(a). One of the caps has a pneumatic port thatis connected to an air tube. Since only the top and the bottomareas of the skin are glued to the caps to make hermetic seals,the rest of the skin is free to stretch when air is injected betweenthe bone and the skin. The skin, when rolled into a tube, has aninner diameter the same as the diameter of the center column,which makes a tight fit so that the module is compact when not

Fig. 2. Design of the inflatable sensing module: (a) exploded view of aninflatable sensing module, (b) example of multi-module assembly, and (c) actualsoft skin with embedded microchannel sensors and complete module prototype.

Fig. 3. Circuit diagram of sensing skin showing the connection to ADC andmicro-controller (left). Multi-contact pressure sensing for both location andmagnitude (right).

inflated. Modularized designing makes it easy to reconfigureor increase the number of the modules, as shown in Fig. 2(b).Actual prototypes of the soft skin and the assembled module areshown in Fig. 2(c). The skin is first made in a flat configurationbefore being rolled up into a tube.

B. Soft Sensing Skin

The soft skin plays the roles of touch sensing and stretch-ing for inflation. For touch sensing, microfluidic channels areembedded in the skin. These microchannels are filled with aroom-temperature liquid metal (EGaIn) and change their elec-trical resistance upon deformation. By measuring the resistancechange, the skin can detect collisions or light touches [26],as demonstrated in Fig. 3. When flat, the skin has the dimen-sions of 100 mm × 85 mm × 2.5 mm, and the embedded

3218 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 3, NO. 4, OCTOBER 2018

Fig. 4. Fabrication process: (a) molding and casting of the microchannel layer,(b) sealing of exposed microchannels using spin-coating, (c) injection of liquidmetal (EGaIn) into microchannels, (d) rolling up and bonding of sensing skinfor a cylindrical shape, (e) 3D printing of plastic bone structure and end caps,(f) and (g) insertion of signal wires and skin-bone assembly, and (h) completeprototype of inflatable sensing module.

microchannels have rectangular cross-sections of 250 µm ×500 µm. The skin was equally divided into nine square sensingareas containing a serpentine channel pattern with alternatingorientations, as shown in Fig. 2. This square serpentine patternmaximizes the sensing area in the skin and makes the skin sen-sitive to external contacts [27]. All the microchannel patternswere connected in series, and signal wires were added betweeneach pair of all the patterns. When a constant current is ap-plied to the microchannel, the voltage drop across each patternis measured independently, which provides information on thelocations of the contacts as well as their force magnitudes, asshown in Fig 3.

III. FABRICATION

The sensing skin is fabricated using a layered molding andcasting method [11], as shown in Fig. 4. First, two 3D-printed

Fig. 5. Experimental setup for characterizing the inflatable sensing module.Digital pressure controllers regulate the levels of inflation of the module, anda robotic arm with an attached load cell and indenter applies an external force.During this force application, sensor output data is collected through a DAQdevice.

(Object 30, Stratasys) molds are prepared, one with microchan-nel patterns and the other without patterns. A skin layer withmicrochannels is made by pouring uncured silicone elastomer(Dragon Skin 10, Smooth-On) in the patterned mold (Fig. 4(a))and curing in a convection oven. To close the top of the mi-crochannels, the patterned layer is covered with a thin pattern-less layer that was previously spin-coated with the same elas-tomer. Curing the two layers together produces a planar siliconeskin embedded with air-filled microchannels (Fig. 4(b)). Theseenclosed microchannels are then filled with EGaIn using twohypodermic syringes-one that injects EGaIn and another thatremoves the captured air (Fig. 4(c)). The flat silicone matrix isthen rolled up into a cylindrical shape, after which the seam isglued together by applying the same uncured silicone materialand curing it in an oven (Fig. 4(d)). A rigid bone structure andtwo caps are printed using a 3D printer (Cubicon Single Plus,Hyvision system) (Fig. 4(e)). The skin is placed on the bone,and both annular openings are glued to the caps using a slightlystiffer elastomer (Dragon Skin 30, Smooth-On), as shown inFigs. 4(f) and 4(g). Fig. 4(h) shows the complete module.

IV. CHARACTERIZATION

A. Experimental Setup

To characterize the sensing performance of the module ininflation and deflation, an experimental setup was prepared(Fig. 5). The internal air pressure of the module was adjustedusing a digital regulator (ME-5000sp, MUSASHI). A simplevoltage divider circuit was constructed to measure the voltagechanges across each of the nine sensing areas of the module,collected through a data acquisition (DAQ) board (USB NI-6211, National Instrument) at a sampling rate of 25 Hz. Forthe characterization of the soft sensors, external forces were ap-plied to the sensing skin using a commercial multi-joint robot

KIM et al.: SOFT INFLATABLE SENSING MODULES FOR SAFE AND INTERACTIVE ROBOTS 3219

Fig. 6. Curvature change of the inflatable skin for different pressure levels(top) and the corresponding sensor outputs (bottom).

arm (UR3, Universal Robots) with a cylindrical flat-end inden-ter (diameter: 6 mm) attached at the end. A commercial loadcell (RFT60-HA01, ROBOTOUS) was also installed with theindenter to collect force data during contact, as shown in Fig. 5.

B. Inflation and Deflation

The inflatable module was experimentally characterized bysupplying air with different pressure levels (0, 10, 13, and16 kPa). Fig. 6 (top) shows the expansion of the inflatablemodule with different air pressure levels and corresponding cur-vatures of the skin. Inflation was not visually noticeable untilpressures reached 8 kPa.

During inflation and deflation, we also measured voltagechanges of the sensing area at the center of the module afterinflation and deflation to different air pressure levels (Fig. 6).As the module was inflated incrementally, the voltage output ofthe center sensor also increased and maintained correspondingsteady state levels for the different air pressure levels. When themodule was deflated back to previous pressure levels, sensorvoltage also decreased with the reduced pressures. However,these voltages did not match those previously measured dur-ing inflation. This discrepancy describes the hysteresis in thesensor signal throughout an inflation and deflation cycle, show-ing slightly increased voltage outputs during deflation. We alsotested the reliability of the sensing module. The module wasinflated and deflated in cycles from full deflation (0 kPa) to twodifferent pressure levels, 13 kPa and 16 kPa, respectively. Atotal of 50 cycles of inflation and deflation for each pressurelevel was conducted and the sensor output was measured duringthese cycles. The results in Fig. 7 show that the sensor signalsare both repeatable and reliable, which also shows the physicalrobustness of the inflatable skin.

Fig. 7. Results of cyclic tests of inflation and deflation of the skin module withtwo different pressure levels (13 kPa and 16 kPa) and corresponding sensoroutputs.

TABLE ITIME RESPONSE OF SKIN SENSOR

C. Contact Sensing

Soft sensors on the inflatable skin were evaluated for theirsensing performance. Of the nine sensors, the one located inthe middle of the module was first tested at different inflationlevels controlled by air pressure (0, 10, 13, and 16 kPa). Normalforces up to 5 N were applied and released through the inden-ter. Fig. 8(a) shows the mean value results with the standarddeviation bands of five trials at each pressure level. Contactsensitivity was highest when the skin was not inflated, sincethe bone structure provided a rigid backing to the skin. Wheninflated, the sensor increased in sensitivity with increasing infla-tion pressures. Although the skin was most sensitive to externalcontacts when not inflated, hysteresis was also the highest dueto the low stiffness of the skin. Considering these changes inthe sensor characteristics due to different inflation states, main-taining sensor accuracy requires a re-calibration process at eachinflation state. Such calibrations would amplify the gains forlowered sensor outputs at higher inflation pressures, as wellas eliminate voltage offsets due to inflation. In addition, theresponse times of the sensor were measured at each pressurelevel. High air pressures improved sensor performance whiledecreasing response times, as summarized in Table I.

Sensors at four different locations were also tested. In thiscase, the air pressure was fixed to 13 kPa, and forces up to 5 Nwere applied and removed for approximately 0.5 seconds, asshown in Fig. 8(b). Although all four sensors showed similarresponses, the sensitivities of two sensors were much higherthan the other two. This is rather because of the microchannelpattern orientation than because of the sensor location. Whilethe sensors with higher sensitivities have horizontal channelpatterns, the other sensors with low sensitivities have vertical

3220 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 3, NO. 4, OCTOBER 2018

Fig. 8. (a) Force responses of the skin sensor located in the middle of themodule at different pressure levels. (b) Sensor outputs of skin sensors in differentlocations with a fixed pressure level (13 kPa).

patterns. The skin inflates much more in a lateral direction thanin a longitudinal direction because the vertical length of the bonestructure is fixed. As a result, horizontal channels grow thinnerwith inflation and consequently become more sensitive, whilevertical channels grow expanded with inflation and thereforebecome less sensitive to contacts.

V. SAFETY TEST

To evaluate the safety performance of the inflatable module,we prepared a setup for collision testing. The module was at-tached to a single revolute joint linkage structure connected toa latex band. By pulling and releasing the link, a controlledand repeatable collision was applied to the module. The result-ing impact forces were measured using a fixed force sensor, asshown in Fig. 9.

We conducted six impact tests with different inflation pressurelevels, as shown in Fig. 10(a). As the air pressure increased,the peak impact force measured by the force sensor tended todecrease. However, the peak force increased again at pressureshigher than 13 kPa, since the skin became too stiff at high airpressure levels although the impulses were in the same level inthe entire pressure range, as shown in Fig. 10(b). Although the

Fig. 9. Collision test setup for evaluating safety performance of inflatablesensing module. Elastic energy stored in the latex band creates a high-speedimpact of the module to the load cell.

Fig. 10. Collision test results: (a) Impact force responses during collision withdifferent inflation conditions. (b) Comparison between peak force and impulsewith different pressure levels.

module was equipped with skin sensors, impact forces were onlymeasured by the external force sensor because we were mainlyinterested in how energy from the impact dissipated through theinflated skin.

VI. APPLICATION

A. Integration of Robotic Arm

To demonstrate an application example of the proposed in-flatable sensing module, we integrated a robotic arm with two

KIM et al.: SOFT INFLATABLE SENSING MODULES FOR SAFE AND INTERACTIVE ROBOTS 3221

Fig. 11. Four-DOF robotic arm integrated with two inflatable sensing mod-ules, each acting as a link between the motor-operated joints: the initial positionof the robot arm without inflation (left) and with inflation and actuation (right).Dotted red lines show the axes of the joints.

modules. The arm had two links, each having a sensing module,and a total of four servo motors to prescribe four DOFs. Twomotors (DRS-0602, Dongbu Robot) were used for the shoulder,the other two (DRS-0201, Dongbu Robot) were used for ac-tuating the elbow and the tendon-driven underactuated gripper,respectively. Because the module had rigid bone structures, theycould directly serve as links to the robot arm. While the boneand arm structure was made of rigid plastic, the gripper wasmade of a flexible material, such as thermoplastic polyurethane(TPU), for flexion and extension of finger joints. Fig. 11 showsthe integrated robotic arm with the proposed inflatable sensingmodules.

B. Sensor Data Processing

Each module has nine sensing areas that are connected inseries, providing a total of eighteen individual voltage signalsfor the two modules in the robot arm. Analog voltage signalswere converted to digital signals by a 16-bit ADC (Ads1115,Adafruit) before being processed by two micro-controllers(Arduino Mega, SparkFun Electronics). Initial voltage readingsare calibrated to zero whenever a new inflation state is initiated.Since sensor sensitivities differ depending on channel patternorientation and location, a different amplification gain was ap-plied to each sensor. Making the sensitivities of all the sensorsthe same will make the calibration process simpler in the future.

C. Air Supply System

To inflate the modules, compressed air was provided throughthe system shown in Fig. 12 (bottom) and two solenoid valves(X-Valve, Parker Hannifin) that were controlled by pulse-width-

Fig. 12. System block diagrams describing the control flow of the module-integrated robotic arm and a schematic diagram showing the air flow system.

Fig. 13. Six basic evasion sequences (indicated by arrows) of the robot armbased on contact (indicated by the red starbursts) sensed and localized by themodules.

modulation (PWM). Precision pressure sensors (33A-150G-2210, Smate) were used to control the constant air pressurein the module.

D. Planning and Control

We constructed a simple control logic based on the sensordata received from the modules, as shown in Fig. 12 (top). Themain controller collects sensor signals, motor encoder data, andair pressure information, whereby it determines collisions andplans an evasive motion sequence for the motors. When the

3222 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 3, NO. 4, OCTOBER 2018

Fig. 14. Motor encoder data in response to external contacts detected by the skin sensor for an evasion control sequence. (a) A large contact force causing a largeand fast evasive movement. (b) A small contact force causing a less drastic evasive movement. Images are taken from a video sequence at instances indicated bythe dotted lines.

upper link module detects a contact, the arm performs corre-sponding evasive motions by operating the two motors (motor 1and motor 2) at the shoulder joint. When the lower link moduledetects a contact, the top shoulder motor (motor 1) and the elbowmotor (motor 3) are operated to evade the contact. A sequenceof six basic evasive motions for different contact locations areshown in Fig. 13.

Evasion speeds and distances were set to change dependingon the magnitude of the contact forces detected by the modules.Responses of the arm are examined by comparing motor en-coder data with the skin sensor data. Large contact forces causethe robot arm to evade further away with an increased speed(Fig. 14(a)), while small contact forces make the arm moveonly a short distance with a relatively slow speed (Fig. 14(b)).The average time response between the onset of collision readby the sensor and that read by the encoder was 280 ms. Uponan external stimulus input to the skin sensor, joint motors beganmoving in an evasive manner until the contact was completelyremoved. Since the external contact (the human finger in thiscase) remained in position after contact, drop of the sensor sig-nal to the original value indicates a successful evasion. Althoughevasion is performed after a collision occurs, not preventing thecollision from happening, such motions prevent the arm frommoving further toward the contact. Furthermore, detection offorce magnitude from the sensors allows appropriate actions tobe taken depending on the severity of the impact.

VII. FUTURE WORK

One of our immediate areas of future work will be optimiza-tion of the orientation and the configuration of the embedded

microchannel patterns, since these design parameters signifi-cantly affect the performance of the soft sensors with differentinflation levels. We will also investigate sensor responses for dif-ferent contact sizes that include areas larger than an individualsensing area. Since we only used an indenter whose contact sizeis smaller than each sensing area, we expect sensor signals tobe read across multiple areas with a larger indenter. In addition,we will minimize the areas of dead zones of contact sensing,which are currently located between sensing areas, by improv-ing the design of the microchannels. Making the channels morecontinuous will be an example in this case. Employing machinelearning will also help more accurately localize the contacts, asdiscussed in [27].

Another area of future work will be full integration of theinflatable sensing modules for a more complex robotic systemwith improved adaptability to various surroundings, expandingthe areas for applications of robots in daily life. We also plan tocover the entire robot with inflatable sensing skin in the futurefor increased safety.

VIII. CONCLUSION

We propose a soft inflatable robotic module with self-contained sensing through embedded microfluidic channelsfilled with a liquid conductor. The module is composed of arigid plastic bone structure and a highly stretchable elastomerskin that can be actively inflated and deflated. In addition, theliquid filled microfluidic channels embedded in the skin candetect both the magnitudes and the locations of external forcesregardless of its shape and size due to the continuity of theliquid conductor. The modular design not only allows for the

KIM et al.: SOFT INFLATABLE SENSING MODULES FOR SAFE AND INTERACTIVE ROBOTS 3223

expansion of the number modules but also enables their easyreconfigurability. A robotic system composed of the proposedmodules will provide safety by significantly reducing the impactfrom any accidental collisions.

The robot will be also more responsive and autonomousby detecting various physical contacts from the environment.The sensing capability of the module was characterized throughloading and inflation/deflation tests. At different inflation levels,the skin sensors were able to measure and localize externallyapplied forces. During the cyclic tests, the module also showedsensing reliability as well as physical robustness. Collision testsof the inflated module demonstrated the safety performance ofthe module, reducing the peak impact force by distributing theimpact stress over a longer period of time. To demonstrate anexample of its application, we integrated two inflatable sens-ing modules with a 4-DOF robotic arm and implemented acontrol logic for evasion based on sensory information fromthe skin.

ACKNOWLEDGMENT

The authors would like to thank Ms. V. A. Cunningham forher feedback on technical writing.

REFERENCES

[1] Z. Mohamed and G. Capi, “Development of a new mobile humanoidrobot for assisting elderly people,” Procedia Eng., vol. 41, pp. 345–351,2012.

[2] M. Vincze et al., “Towards a robot for supporting older people to staylonger independent at home,” in Proc. 41st Int. Symp. Robot., 2014,pp. 1–7.

[3] D. Fischinger et al., “Hobbit, a care robot supporting independent livingat home: First prototype and lessons learned,” Robot. Auton. Syst., vol. 75,pp. 60–78, 2016.

[4] C. Jayawardena, I.-H. Kuo, E. Broadbent, and B. A. MacDonald, “Sociallyassistive robot healthbot: Design, implementation, and field trials,” IEEESyst. J., vol. 10, no. 3, pp. 1056–1067, Sep. 2016.

[5] D. Feil-Seifer and M. J. Mataric, “Socially assistive robotics,” IEEE Robot.Autom. Mag., vol. 18, no. 1, pp. 24–31, Mar. 2011.

[6] D. F. P. Granados, B. A. Yamamoto, H. Kamide, J. Kinugawa, and K.Kosuge, “Dance teaching by a robot: Combining cognitive and physicalhuman–robot interaction for supporting the skill learning process,” IEEERobot. Autom. Lett., vol. 2, no. 3, pp. 1452–1459, Jul. 2017.

[7] M. Fridin, “Storytelling by a kindergarten social assistive robot: A toolfor constructive learning in preschool education,” Comput. Edu., vol. 70,pp. 53–64, 2014.

[8] D. Kulic and E. Croft, “Pre-collision safety strategies for human-robotinteraction,” Auton. Robots, vol. 22, no. 2, pp. 149–164, 2007.

[9] R. Schiavi, A. Bicchi, and F. Flacco, “Integration of active and passivecompliance control for safe human-robot coexistence,” in Proc. IEEE Int.Conf. Robot. Autom., 2009, pp. 259–264.

[10] T. Kim and Y.-L. Park, “A soft 3-axis load cell using liquid-filled 3-Dmicrochannels in a highly deformable elastomer,” IEEE Robot. Autom.Lett., vol. 3, no. 2, pp. 881–887, Apr. 2018.

[11] Y.-L. Park, B.-R. Chen, and R. J. Wood, “Design and fabrication of soft ar-tificial skin using embedded microchannels and liquid conductors,” IEEESens. J., vol. 12, no. 8, pp. 2711–2718, Aug. 2012.

[12] S. Yao and Y. Zhu, “Wearable multifunctional sensors using printedstretchable conductors made of silver nanowires,” Nanoscale, vol. 6, no. 4,pp. 2345–2352, 2014.

[13] J. Wirekoh and Y.-L. Park, “Design of flat pneumatic artificial muscles,”Smart Mater. Struct., vol. 26, no. 3, 2017, Art. no. 035009.

[14] P. Polygerinos et al., “Modeling of soft fiber-reinforced bending actuators,”IEEE Trans. Robot., vol. 31, no. 3, pp. 778–789, Jun. 2015.

[15] L. Paez, G. Agarwal, and J. Paik, “Design and analysis of a soft pneumaticactuator with origami shell reinforcement,” Soft Robot., vol. 3, no. 3,pp. 109–119, 2016.

[16] D. Shin, I. Sardellitti, Y.-L. Park, O. Khatib, and M. Cutkosky, “Designand control of a bio-inspired human-friendly robot,” Int. J. Robot. Res.,vol. 29, no. 5, pp. 571–584, 2010.

[17] J. M. A. Palacio, A. Riwan, N. Mechbal, E. Monteiro, and S. Voisembert,“A novel inflatable actuator for inflatable robotic arms,” in Proc. IEEEInt. Conf. Adv. Intell. Mechatronics, 2017, pp. 88–93.

[18] T. Ranzani, G. Gerboni, M. Cianchetti, and A. Menciassi, “A bioin-spired soft manipulator for minimally invasive surgery,” BioinspirationBiomimetics, vol. 10, no. 3, 2015, Art. no. 035008.

[19] J. D. Greer, T. K. Morimoto, A. M. Okamura, and E. W. Hawkes, “Seriespneumatic artificial muscles (sPAMs) and application to a soft continuumrobot,” in Proc. IEEE Int. Conf. Robot. Autom., Singapore, May 2017,pp. 5503–5510.

[20] Y.-L. Park et al., “Design and control of a bio-inspired soft wearablerobotic device for ankle–foot rehabilitation,” Bioinspiration Biomimetics,vol. 9, no. 1, 2014, Art. no. 016007.

[21] Y.-L. Park, J. Santos, K. G. Galloway, E. C. Goldfield, and R. J. Wood,“A soft wearable robotic device for active knee motions using flat pneu-matic artificial muscles,” in Proc. IEEE Int. Conf. Robot. Autom., 2014,pp. 4805–4810.

[22] Y. Menguc et al., “Wearable soft sensing suit for human gait measure-ment,” Int. J. Robot. Res., vol. 33, no. 14, pp. 1748–1764, 2014.

[23] P. Ohta et al., “Design of a lightweight, soft robotic arm using pneu-matic artificial muscles and inflatable sleeves,” Soft Robot., vol. 5, no. 2,pp. 204–215, 2018.

[24] R. Qi, A. Khajepour, W. W. Melek, T. L. Lam, and Y. Xu, “Design,kinematics, and control of a multijoint soft inflatable arm for human-safeinteraction,” IEEE Trans. Robot., vol. 33, no. 3, pp. 594–609, Jun. 2017.

[25] E. W. Hawkes, L. H. Blumenschein, J. D. Greer, and A. M. Okamura,“A soft robot that navigates its environment through growth,” Sci. Robot.,vol. 2, no. 8, 2017, Art. no. eaan3028.

[26] Y.-L. Park, C. Majidi, R. Kramer, P. Berard, and R. J. Wood, “Hypere-lastic pressure sensing with a liquid-embedded elastomer,” J. Micromech.Microeng., vol. 20, no. 12, 2010, Art. no. 125029.

[27] S. Han, T. Kim, D. Kim, Y.-L. Park, and S. Jo, “Use of deep learning forcharacterization of microfluidic soft sensors,” IEEE Robot. Autom. Lett.,vol. 3, no. 2, pp. 873–880, Apr. 2018.