Professional Documents
Culture Documents
Soft Robotics Paper 3
Soft Robotics Paper 3
Soft Robotics Paper 3
ORIGINAL ARTICLE
Abstract
The past decade has witnessed tremendous progress in soft robotics. Unlike most pneumatic-based methods, we
present a new approach to soft robot design based on precharged pneumatics (PCP). We propose a PCP soft
bending actuator, which is actuated by precharged air pressure and retracted by inextensible tendons. By pulling
or releasing the tendons, the air pressure in the soft actuator is modulated, and hence, its bending angle. The
tendons serve in a way similar to pressure-regulating valves that are used in typical pneumatic systems. The
linear motion of tendons is transduced into complex motion via the prepressurized bent soft actuator. Fur-
thermore, since a PCP actuator does not need any gas supply, complicated pneumatic control systems used in
traditional soft robotics are eliminated. This facilitates the development of compact untethered autonomous soft
robots for various applications. Both theoretical modeling and experimental validation have been conducted on
a sample PCP soft actuator design. A fully untethered autonomous quadrupedal soft robot and a soft gripper
have been developed to demonstrate the superiority of the proposed approach over traditional pneumatic-driven
soft robots.
Keywords: precharged pneumatics, soft actuator, controllable actuation, tendon control, soft robotic gripper,
untethered soft robot
1
2 LI ET AL.
To eliminate tethering to any hard robotic control systems the soft actuator design (Fig. 1A(1)), it does not need any
and power sources, a soft robot with microfluidic control and connection tubes for air supply and release. Instead, the air
powered by gas from chemical reaction is reported in Ref.22 chamber is completely enclosed, like a basketball or a tire,
The gas generated from fuel combustion inflates the fluidic leaving only a small orifice (check-valve) for air charge
networks downstream of the reaction sites, resulting in actua- (Fig. 1A(2)). As such, once the actuator is charged, it can be
tion. However, microfluidic logic can only provide very simple used for months before the next charge is required. We call
control strategies. Given the large strain of the soft robot, this precharged pneumatics (PCP) soft actuator, inspired by
clogging of the microfluidic channels poses a great challenge. air gun design. In operating an air gun, a cocking drive is
Furthermore, the force that the soft robot can exert on the pulled back to compress air in a pump. By pulling the trigger,
environment is very small, limiting its practical applications. air pressure is released and shot out the pellet by pneumatic
To increase the load capability of untethered soft robots, power. When the PCP soft actuator is charged, the tendons
Trolley et al. developed a large (0.65 m long) soft robot, (cables) pull back the already bent PCP actuator to a desired
actuated by pneumatic power.18 The large size is made pos- position. By pulling or releasing the tendons, the actuator’s
sible by fabricating the robot using composite materials air pressure is modulated, and hence, its deformation, as
comprising elastomer, polyaramid fabrics, and hollow glass shown in Figure 1B, where the actuator exhibits different
microspheres. Even at this large size, the robot still has dif- degrees of bending due to different tendon lengths. At each of
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
ficulties in encapsulating all the necessary power supplies the actuator position, its air pressure is different. The PCP
(batteries and compressed gas cylinders) and other compo- soft actuator has three controllable variables that can be
nents (valves, electric motor-driven air pump, circuit board) controlled one at a time: the tendon pulling force F, the
needed for autonomous operations, as shown in Supple- tendon pulling/releasing speed V, and the tendon pulling/
mentary Figure S1 (Supplementary Data are available online releasing displacement D. Such controllability allows the
at www.liebertpub.com/soro). Furthermore, the robot’s lo- actuator to be used for various applications. The control
comotion speed is very slow, as slow as 2.0 m/h in an un- system is also much easier and simpler, as shown in
dulating gait limited by the small flow rate of compressed air. Figure 1C, where the complicated pneumatic system (com-
To increase the flow rate, larger compressors, valves, and pressor, valves, and tubes) is completely eliminated. This
tubing are needed, which may cripple the robot. greatly facilitates the development of untethered autonomous
In this research, we report a novel approach to pump-less soft robots, as shown in the later parts of the article.
and untethered autonomous soft robotics. Although the pro- Tendon control is much easier and more accurate than
posed approach adopts the commonly practiced methods for pneumatic control. This is because gas is compressible, and
thus, a pneumatic system is regarded a nonlinear system.23 that of the pads) are installed inside the holes to reduce the
Since the gas pressure can hardly be controlled quickly, it is friction between the tendons and the silicone rubber pads, as
difficult to control the deformation of a soft actuator accu- well as to protect the pads from being cut by the cord. Through
rately.24,25 Moreover, when the tendons are long and their the tendon guide inside the actuator base, the two tendons
winding requires more sliding contact, the tendon driving converge to a single point for better control. After assembly,
mechanism may have nonlinearity and insensitivity problems the total length of the actuator is about 140 mm (Fig. 2C).
when the load is large.26,27 In contrast, the proposed PCP soft We use a pedal air pump connected with a gas needle to
actuator is driven by precharged air, not by the tendons. inflate the actuator (Fig. 2D). After inflating the actuator, the
air pump is removed and the actuator is at the bending state.
We design a valve pattern that is quite similar to the bas-
Actuator design and analytical modeling
ketball check-way valve structure, and mold it with silicone
A sample PCP soft actuator has been designed for analysis rubber (Fig. 2E). Figure 2F shows the check-valve cross-
of its characteristics, which is validated by a set of experiments sectional view and the working principle. The valve channel
using a prototype fabricated from the design. Figure 2 shows has different diameters, which gradually decrease from the
the 3D schematic drawing and details of the sample design. bottom (3 mm) to the top (0.5 mm). A mechanical clamper is
The actuator is composed of two parts, the soft actuator and the designed for clipping the smaller channel completely air-tight
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
actuator base. The soft actuator is a classic fiber-reinforced soft (Fig. 2F). When a gas needle is inserted into the valve, the
actuator design.1 Its cross section is semicircle with a radius small channel will stretch a bit for pumping air into the
16 mm (Fig. 2A). Three silicone rubber pads are molded onto chamber. When the actuator bends to a desirable angle, the gas
the back of the soft actuator for the tendon guide. We use the needle is pulled out and the clamper squeezes the valve air-
Kevlar cord as the pulling tendons because of its extremely tight, trapping the air in the chamber. Due to the small hardness
high tensile strength-to-weight ratio. (SHORE A 10D) of silicone rubber used for making the pro-
Each of the silicone rubber pads has two through holes totype, the air pressure that the valve can sustain is about 0.10
(Fig. 2B). Silicone rubber tubes (with higher hardness than Mpa, and the actuator bends about 100 under this pressure.
Given a PCP bending actuator design and a precharged air controlled by step motors are fixed vertically on the table. A
pressure, we can model the bending angle, the tendon pulling force gauge is installed on the sliding block with a hook head.
force, and the corresponding bending force.28 Supplementary The actuator tendons are anchored on the hook head (Fig. 3A).
Figure S2 shows the soft actuator model and its parameter The force gauge can be controlled to move forward and
definitions. A detailed analytical modeling is given in the backward at different speeds. We use a high-resolution
Supplementary Data. camera to record the actuator trajectory. The tendon force
can be read from the force gauge. To analyze the actuator
trajectory, we use the software Kinovea to measure the
Experimental validation
bending angle and the displacement of the tendon from
We build a test platform shown in Figure 3A to measure the video record.
the actuator’s characteristics. First, we use the platform to The experiment starts from the fully actuated state of the
measure the tendon pulling force and the actuator bending actuator (Fig. 3A). The force gauge is pulled at 10 mm/step
angle against the tendon displacement. Two linear guides for nine steps at a speed of 6.0 mm/s. For each step, there is a
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
FIG. 3. Experimental validation. (A) Experimental setup, bending angle at fully actuated state: h 98 . (B) Actuator
bending trajectory. (C) Actuator end pad tilt angle measurement. (D) Pulling force changes tendon displacement. (E)
Relationship between the tendon displacement D and the bending angle h. (F) Relationship between the tendon dis-
placement D and tendon pulling force F. (G) Bending force measuring experimental schematic and setup. (H) Relationship
between pulling force F and bending force FB. Color images available online at www.liebertpub.com/soro
PCP SOFT ACTUATORS AND UNTETHERED SOFT ROBOTS 5
pause of 2 s for the actuator to stabilize. The pulling forces are and the tendon displacement is linear. As a result, the bending
measured and the moving trajectories are video recorded. angle of the PCP actuator can be linearly controlled by the
When the actuator reaches the initial state (straight), the force tendon displacement.
gauge releases the tendon to the fully actuated state. The The relationship between the tendon pulling force F and
entire process is repeated six times. the displacement D is shown in Figure 3F. The pulling force
is measured when the PCP actuator is pulled from the fully
Bending trajectory and actuator end pad tilt. The bending actuated state to the initial state, and the releasing force is
trajectory of the actuator is shown in Figure 3B. The red dash measured when the PCP actuator is actuated from the initial
line shows the trajectory of the endpoint of the actuator state to the fully actuated state. The pulling force increases
during the tendon pulling and releasing processes. The tra- linearly against the tendon displacement during the pulling
jectory shows good consistency. It should be noted that while process. When the tendon changes from pulling to releasing
the actuator is controlled by the tendons, the actuation is (D changes from 90 to 80 mm), the tendon force curve shows
powered by the precharged air. This is quite different from a rapid decline. The rest of the curve (D changes from 80 to
the direct tendon-driven actuators.29 The body of the pro- 0 mm) in the releasing process remains a steady gradient.
posed PCP soft actuator exhibits a continuous and smooth More discussions are provided in the Supplementary Data.
bending curve.
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
Another major advantage of our proposed gripper is that width W1 ¼ 40 mm, gripper open width W2 ¼ 80 mm, and
the PCP actuators are precharged and can be easily controlled finger length l ¼ 120 mm.
by pulling and releasing the tendons. Our gripper eliminates When the shaft rotates clockwise, the tendons will be
the need of air supply, and hence, the bulky and heavy air rolled onto the shaft until the PCP actuators become straight,
pump, for its applications. Compared with air source, which is the initial state (Fig. 4C). It should be noted that if
electrical power source is more compact, efficient, easily we keep on rotating the tendon shaft, the PCPs can bend
available, and easy to set up. Therefore, our gripper is more further outward (Fig. 4D) to increase the gripper open width
versatile for various soft robotic applications. (W2¢ > 80 mm). At the fully actuated state, the tendons are
The gripper prototype is shown in Figure 4A. It consists of completely released, returning the actuators to the natural
a gripper base and three PCP actuators. The three actuators bending state without any extra power input (Fig. 4E).
are installed with two on one side and one on the other side. In We test the gripper’s performance by grasping different
Figure 4B, the PCP’s tendons are wound to the same shaft at objects with different sizes, shapes, and mechanical proper-
different positions, while the shaft rotates to roll/release the ties. The gripper grasps a water bottle (500 g), a piece of tofu
tendons. The shaft is actuated by a DC motor through a pair of (260 g), and a roll of sticky tape (D ¼ 100 mm > W2 ). When
gears. Based on the pulling force required for controlling the grasping the water bottle (Fig. 4F), like other soft grippers,
PCPs, we can select a motor with the right power and torque. the gripper just needs to approach the bottle from a proper
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
The DC motor is controlled by a controller that can control position and fully release the tendons. The largest payload
the rotation speed and direction. Thus, the actuators’ bending that our soft gripper can hold is about 600 g. Grasping tofu
speed and angle can be controlled accordingly. As shown in and sticky tape requires special control processes. Here we
Figure 4B, the basic configuration parameters of the gripper focus on describing the grasping procedure of tofu and
are as follows: finger assembled angle a ¼ 80 , gripper base the tape.
FIG. 4. Prototype soft robotic gripper. (A) The prototype gripper. (B) Schematic diagram of tendon driving mechanism.
(C) Initial state of the gripper. (D) Gripper fingers bending outward. (E) Actuated state of the gripper. (F) Grasp a bottle of
water (weight 500 g). (G) Grasp fragile object Tofu or bean curd (weight 260 g). (H) Grasp a roll of tape (diameter 100 mm).
Color images available online at www.liebertpub.com/soro
PCP SOFT ACTUATORS AND UNTETHERED SOFT ROBOTS 7
Tofu, or bean curd, is very fragile. Grasping tofu can For simplicity, the prototype has only two motors. Actuator
therefore test the gripper’s controllability. When approaching 1 and Actuator 2 are controlled by one servomotor, while
the tofu, the PCP actuator is controlled to move slowly to Actuator 3 and Actuator 4 by another servo motor. The total
result in a very gentle touch. As force sensors are not used in length of the robot is 200 mm and its weight is about 1.1 kg.
the grasp tests, the grasp experiment is guided by human The robot body contains a mobile power (10000mAh Mi
observation. After several trials, we have identified appro- Power Bank), an Arduino Mega board, and two servomo-
priate values for two control variables in tofu grasping: ten- tors (MG995 servomotor). The servomotors rotate clock-
don releasing speed at below 10 mm/s and tofu indent depth wise and counter-clockwise in the range of 180. The initial
(5 mm) when the fingers have contact with it. When these two installed states of the four actuators are different (Fig. 5B),
values are observed, the sample tofu can be easily grasped that is, Actuator 1 and Actuator 4 are at fully bent state,
without damage, as in Figure 4G. while Actuator 2 and Actuator 3 are at the initial straight
We further test the actuator by grasping a tape roll with a state.
diameter D ¼ 100 mm, which is a little bit larger than the The walking gait of the robot is shown in Figure 5C(1–6).
gripper’s initial open width W2 ¼ 80 mm. The grasp may be The diagonal actuators are always at the same actuation
unstable as only the fingertips of the gripper are in contact state. When Actuator 1 and Actuator 4 are bent, Actuator 2
with the tape. To increase the contact area, the tendons are and Actuator 3 are pulled to the initial state, and vice versa
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
pulled to open the gripper further, as shown in Figure 4D. (Fig. 5C(1) and (4)). The friction coefficient between sili-
The gripper then approaches the tape roll and wraps around cone rubber and wooded desk is quite large. Therefore,
it for a stable grasp, as shown in Figure 4H. This demon- when two actuators are bending backward, the reaction
strates that the gripper’s initial state can be changed for forces push the robot to lean forward (Fig. 5C(2) and (5)). At
different applications. the same time, the other two actuators are pulled to the
The grasping processes of objects in Figure 4 are recorded straight state, sweeping through the ground and moving
and shown in the Supplementary Videos S1–S6. forward along with the entire robot body. Sellotape is stuck
to the bottom of the actuator to decrease the friction
A fully untethered autonomous quadrupedal soft robot. We (Fig. 5B). The robot can move forward at an average speed
design and fabricate an untethered autonomous quadrupedal of 5 mm/s through periodically alternating movements. This
robot with four PCP actuators as limbs to further demonstrate speed is much faster than that reported in Ref.18 In fact, the
the actuators’ applications. The robot prototype is shown in walking speed can be further increased by using faster servo
Figure 5A. Its design schematic drawing is shown in Figure 5B. motors.
changes are in the following two aspects: (1) replacement of Appl Bionics Biomech 2008;5:99–117.
the air inlet with a check-valve and (2) adding tendons to pull 3. Laschi C, Mazzolai B, Cianchetti M. Soft robotics: tech-
back the actuator to its original position before precharged nologies and systems pushing the boundaries of robot
with air. Once charged, the air pump is not required. Ex- abilities. Sci Robotics 2016;1:1–11.
periments and experiences have shown that the check-valve 4. Sun Y, Song YS, Paik J. Characterization of silicone
sealing is very effective (basket balls and tires can last for rubber based soft pneumatic actuators. In IROS: Interna-
months before next charge). The bending (or actuating) tional Conference on Intelligent Robots and Systems,
process of the PCP actuator is controlled by pulling or re- Tokyo, Japan, 2013, pp. 4446–4453.
leasing the tendons that are controlled by servo motors. 5. Ilievski F, Mazzeo AD, Shepherd RF, Chen X, Whitesides
Controlling a servo motor is much easier and more accu- GM. Soft robotics for chemistry. Angew Chem Int Ed Engl
rate than controlling air pressure and flow rate. We can 2011;50:1890–1895.
control the pulling speed, the pulling displacement, and the 6. Suzumori K, Endo S, Kanda T, Kato N, Suzuki H. A
bending pneumatic rubber actuator realizing soft-bodied
pulling force of the PCP actuator relatively easily. These
manta swimming robot. In ICRA: International Con-
variables are directly associated with the soft actuator’s
ference on Robotics and Automation, Roma, Italy, 2007,
bending speed, bending angle, and bending force. This ac- pp. 4975–4980.
tuator and its control system can be regarded as a continu- 7. Martinez RV, Fish CR, Chen X, Whitesides GM. Elasto-
ously deformation nonlinear system constrained by a tendon meric origami: programmable paper-elastomer composites
that can be linearly controlled. We made a three-fingered as pneumatic actuators. Adv Funct Mater 2012;22:1376–
gripper prototype and a quadrupedal autonomous robot with 1384.
the PCPs. Through experiments, we found that the actuator 8. Tondu B, Lopez P. Modeling and control of McKibben
exhibits the advantages of both soft robots and tendon-driven artificial muscle robot actuators. IEEE Control Syst Mag
robots. The actuator shows great compatibility and control- 2000;20:15–38.
lability. It can not only bend slowly and gently to grasp 9. Connolly Y, Polygerinos P, Walsh CJ, Bertoldi K. Me-
delicate and fragile objects such as tofu but can also bend fast chanical programming of soft actuators by varying fiber
and provide enough power to drive a walking robot. angle. Soft Robotics 2015;2:26–37.
Although our current PCP actuator design shows good 10. Moseley P, Florez JM, Sonar HA, Agarwal G, Curtin W,
performance, there is room for improvement. First, the ac- Paik CJ. Modeling, design, and development of soft
tuator size and appearance can be redesigned and optimized. pneumatic actuators with finite element method. Adv Eng
The current design of the tendon guide support is made rel- Mater 2016;18:978–988.
atively sturdy to ensure certain strength. Therefore, the entire 11. Yap HK, Ng HY, Yeow CH. High-force soft printable
actuator looks bulky and clumsy, impeding its mobility and pneumatics for soft robotic applications. Soft Robotics
flexibility. The tendon guide can be redesigned and made 2016;3:144–158.
with stronger materials to reduce the size and enhance the 12. Wei Y, Chen YH, Ren T, Chen W, Yan CX, Yang Y, et al.
A novel, variable stiffness robotic gripper based on inte-
practicality of the actuator. Second, the hardness of the sili-
grated soft actuating and particle jamming. Soft Robotics
cone rubber used for the check-valve is quite small (10D,
2016;3:134–143.
Shore A), limiting the precharged air pressure to below 13. Mosadegh B, Polygerinos P, Keplinger C, Wennstedt S,
0.1 MPa and hence a maximum bending angle of about 98. A Shepherd RF, Gupta U, et al. Pneumatic networks for soft
more reliable check-valve design and the use of better ma- robotics that actuate rapidly. Adv Funct Mater 2014;24:
terials can contribute to a larger precharged air pressure, and 2163–2170.
hence, a larger bending angle and longer operation life. Fi- 14. Deimel R, Brock O. A novel type of compliant and un-
nally, although the actuator has eliminated the need for air deractuated robotic hand for dexterous grasping. Int J Rob
pumps and valves, a servomotor and battery are still required. Res 2016;35:161–185.
These parts are heavy as well. We are currently investigating 15. Polygerinos P, Galloway KC, Savage E, Herman M, Donnell
other types of tendons, such as shape memory alloys or shape KO, Walsh CJ. Soft robotic glove for hand rehabilitation and
memory polymers. task specific training. In ICRA: International Conference on
PCP SOFT ACTUATORS AND UNTETHERED SOFT ROBOTS 9
Robotics and Automation, Seattle, USA, 2015, pp. 2913– control for a pneumatic artificial muscle. IEEE Trans Ind
2919. Electron 2014;61:904–916.
16. Polygerinos P, Lyne S, Wang Z, Nicolini LF, Mosadegh B, 26. In H, Lee D, Cho KJ. Investigation of friction character-
Whitesides GM, et al. Towards a soft pneumatic glove for istics of a tendon driven wearable robotic hand. In Inter-
hand rehabilitation. In: IROS: International Conference on national Conference on Control, Automation and Systems,
Intelligent Robots and Systems, Tokyo, Japan, 2013, Gyeonggi-do, Korea, 2010, pp. 568–573.
pp. 1512–1517. 27. Palli G, Borghesan G, Melchiorri C. Tendon-based
17. Nemiroski A, Shevchenko YY, Stokes AA, Unal B, Ainla transmission systems for robotic devices: models and
A, Albert S, et al. Arthrobots. Soft Robotics 2017;4:183– control algorithms. In ICRA: IEEE International Con-
190. ference on Robotics and Automation, Kobe, Japan, 2009,
18. Trolley MT, Shepherd RF, Mosadegh B, Galloway KC, pp. 4063–4068.
Wehner M, Karpelson M, et al. A resilient, untethered soft 28. Ogden RW, Mineola NY. Non-Linear Elastic Deforma-
robot. Soft Robotics 2014;1:213–223. tions. New York: Dover Publications, 1997.
19. Bartlett NW, Tolley MT, Overvelde JTB, Weaver JC, 29. Manti M, Hassan T, Passetti G, D’Elia N, Laschi C, Cian-
Mosadegh B, Bertoldi K, et al. A 3D-printed, functionally chetti M. A bioinspired soft robotic gripper for adaptable and
graded soft robot powered by combustion. Science 2015; effective grasping. Soft Robotics 2015;2:107–116.
349:161–165. 30. Love AEH. Treatise on the Mathematical Theory of Elas-
Downloaded by SUNY Stony Brook package(NERL) from www.liebertpub.com at 06/20/18. For personal use only.
20. Tolley MT, Shepherd RF, Karpelson M, Bartlett NW, ticity. Cambridge, UK: Cambridge University Press, 1927.
Galloway KC, Wehner M, et al. An untethered jumping 31. Galloway KC, Becker KP, Phillips B, Kirby J, Licht S,
soft robot. In: IROS: International Conference on In- Tchernov D, et al. Soft robotic grippers for biological
telligent Robots and Systems, Chicago, 2014, pp. 561–566. sampling on deep reefs. Soft Robotics 2016;3:23–33.
21. Loepfe M, Schumacher CM, Lustenberger UB, Stark WJ. 32. Zhou X, Majidi C, O’Reilly OM. Soft hands: an analysis of
An untethered, jumping roly-poly soft robot driven by some gripping mechanisms in soft robot design. Int J Solids
combustion. Soft Robotics 2015;2:33–41. Struct 2015;64:155–165.
22. Wehner M, Truby RL, Fitzerald DJ, Mosadegh B, White-
sides GM, Lewis JA, et al. An integrated design and fabri-
cation strategy for entirely soft autonomous robots. Nature Address correspondence to:
2016;536:451–455. Yonghua Chen
23. Bobrow JE, McDonell BW. Modeling, identification, and Department of Mechanical Engineering
control of a pneumatically actuated, force controllable ro- The University of Hong Kong
bot. IEEE Trans Robotics 1998;14:732–742. Pokfulam Road
24. Caldwell DG, Medrano-Cerda GA, Goodwin M. Control of Hong Kong 852
Pneumatic Muscle. IEEE Control Syst 1995;15:40–48. Hong Kong
25. Andrikopoulos G, Nikolakopoulos G, Arvanitakis I, Man-
esis S. Piecewise affine modeling and constrained optimal E-mail: yhchen@hku.hk