Using Explosions to Power a Soft Robot

This manuscript describes the use of explosions to power a soft robot—one composed solely of organic elastomers (e.g., silicones). The robot has three pneumatic actuators (pneu-nets) in a tripedal configuration. Explosion of a stoichiometric mixture of methane and oxygen within the microchannels making up the actuators produced hot gas that rapidly inflated the pneu-nets, and caused the robot to launch itself vertically from a flat surface (e.g., to jump). A soft flap embedded in the pneu-net acted as the valve of a passive exhaust system, and allowed multiple sequential actuations. The flame and temperature increase from the explosions are short-lived, and do not noticeably damage the robots over dozens of actuation cycles.

Soft robots have emerged as a new set of machines capable of manipulation [1][2][3][4] and locomotion. [5][6][7][8] Pneumatic expansion of a network of microchannels (pneu-nets) fabricated in organic elastomers, using low-pressure air (<10 psi; 0.7 atm; 71 kPa), provides a simple method of achieving complex movements: [1,5] grasping and walking. Despite their advantages (simplicity of fabrication, actuation, and control; low cost; light weight), pneu-nets have the disadvantage that actuation using them is slow, in part because the viscosity of air limits the rate at which the gas can be delivered through tubes to fill and expand the microchannels. Here we demonstrate the rapid actuation of pneu-nets using a chemical reaction (the combustion of methane) to generate explosive bursts of pressure.
Although the combustion of hydrocarbons is ubiquitous in the actuation of hard systems (e.g., in the metal cylinder of a diesel or spark-ignited engine [9]), it has not been used to power soft machines. Here, we demonstrate that explosive chemical reactions [10] producing pulses of high temperature gas for pneu-net actuation provides simple, rapid, co-located power generation, and enables motion, in soft robots. In particular, we used the explosive combustion of hydrocarbons triggered by an electrical spark to cause a soft robot to "jump" (a gait previously only demonstrated for hard systems [11][12][13][14][15][16]).
We fabricated a tripedal robot ( Fig. 1; Fig. S4) using soft lithography. [1] This robot incorporated a passive valving system (Fig. 1a, inset) that allowed us to (i) pressurize the pneunets easily, (ii) exhaust the product gases automatically (without external control), and (iii) actuate the same pneu-net repeatedly. By actuating all three legs simultaneously, we caused the robot to jump more than 30 times its height in less than 0.2 s, at a maximum vertical velocity of ~3.6 m/s. Our choice of explosive chemical reactions for actuation was based on several factors, one being their high volumetric energy density (in units of MJ/L). The energy density of a compressed gas, which we previously used to power soft robots, is ~0.1 MJ/L at 2,900 psi from the potential for mechanical work, w, done by the change in pressure (P), and volume (V) when decompressed to atmospheric pressure; combustible gases, like CH 4 , can also be burned to release heat, q, resulting in an energy density of w+q ~ 8.0 MJ/L (SI). We used a stoichiometric mixture of methane and oxygen (1 mole CH 4 : 2 moles O 2 ) to power the jumps. This stoichiometry minimized the formation of soot, and prevented the contamination of the channels and the clogging of the valves by carbon deposits. We used an electrical spark to ignite the mixture inside the chambers because spark gaps are (i) easily incorporated into soft robots, (ii) controlled in their timing with msec precision, and (iii) faster than other means of ignition (e.g., resistive heating).
We chose pure oxygen instead of air to maximize the energy density of the mixture; air contains only ~21 wt % O 2 . Methane was the fuel because it is (i) readily available, (ii) a gas over all temperatures useful for soft robots, and easily pumped through tubing and pneu-nets, (iii) easily controlled to ignite an explosion (i.e., rapid burning) rather than a detonation (i.e., a shock wave), [10,17] (iv) sufficiently exothermic in combustion that it releases enough energy (890 kJ/mol; Eq. 1[18]) for actuation, but not enough to damage the channel or passive valve, and (iv) converted by combustion into products (CO 2 (g) and H 2 O (g)) that allow rapid depressurization of the actuator at the end of each cycle through a soft microvalve.

Eq. 1
We fabricated the robot using soft lithography (SI). [1,5] Each leg of the tripedal robot was a hollow chamber with a stoichiometric mixture of CH 4 and O 2 entering from one side, and gases exiting through a valved opening at the other. At the gas-input side of each pneu-net, we placed computer controlled electrodes that triggered a spark (SI).
Explosion Inside an Elastomeric Balloon. The high temperatures of explosive reactions (T>2,500 K in air [17,19,20]) seem incompatible with the low service temperature of silicone elastomers (most degrade at T< 600 K). [21] To a first approximation, the temperature within a pneu-net during the explosion can be estimated using Eq. 2, [20] Eq. 2 where = 28.6 J/mol·K and = 74.5 J/mol·K, = 20 mol and = 40 mol, and ≈ 18 J based on our gas-flow rates and channel dimensions; a more detailed analysis (SI) based on Eq. 2 that includes second-order effects predicts a temperature of 3,000 K (2,800 o C) immediately after ignition. In our pneu-nets, a thin layer of silicone may decompose on exposure to high temperatures, [22] and form a surface layer of silica; this layer may insulate the surface from the radiant heat of the flame. [23] The duration of the explosion is short. The temperature of the gas is quickly reduced as it expands and the pneu-net inflates. To try and capture the kinetics of the rapid combustion, we used a combination of high-speed infrared (IR) imaging (155 fps; T620; FLIR, Inc.) and bimetallic temperature probes in the interior of a pneu-net. We do not know how the IR intensity detected by the camera partitions between emissions from the hot gas, the surface, and the bulk polymer of the robot, but empirically, three ms after ignition, we detected IR temperature in excess of 500 o C (Fig. 2a,b). After ten ms, the temperature measured by the IR camera fell below 300 o C (below the decomposition temperature of silicone; Fig. 2a-d). The IR imaging also established that we could actuate a single pneu-net independently, by adjusting the delays between sparks in them (Fig 2e-h).
To measure the temperature of the exhaust gases during a sequence of actuations (one actuation every two seconds; Fig. 2i), we used a thermocouple (SI; Omega Instruments) placed inside the pneu-net. Because the response of the thermocouple is relatively slow, the measurements are averages over tens of milliseconds, and report the temperature of the exhaust gases (which do not exceed peak temperatures of ~125 o C, and cool to <50 o C prior to subsequent explosions). In principle, the stability of the CH 4 /O 2 mixture in the absence of initiating events, and the small increase in temperature in the pneu-nets (~25 o C above room temperature; SI) after the flame is extinguished, makes these systems safe to handle (Video S1).
We emphasize, nonetheless, that a mixture of CH 4 /O 2 is highly dangerous, and should only be handled by experienced personnel.
Using isothermal nanocalorimetry, we measured the heat evolution during actuation with i) that results from the ~11,000 fold increase in power causes rapid actuation of the pneu-nets, and enables the soft robot to jump.
Passive Valve for Release of Exhaust. After the methane has burned in the pneu-net, it is necessary to remove the waste products to ensure an appropriate ratio of CH 4 and O 2 for the next actuation. To purge the waste CO 2 and H 2 O vapor, we embedded exit channels into the ends of the legs of the tripod. To actuate a leg, we allowed fresh methane and oxygen to flow into the pneu-net and expel the exhaust gases.
The heat from the combustion reaction increases the pressure of gas in the pneu-net. To limit the expanding gas from leaking before the explosion was complete, we embedded a passive valve-a soft flap molded directly into the pneu-net-immediately before the exit channel ( Approximately seven ms after the spark ignited the CH 4 /O 2 , the pressure generated by the exploding gas caused the leg to inflate (Fig. 4a), and then, ~50 ms after ignition, to extend ~5 mm; this extension, in turn, caused the actuator to bend downward ( Fig. 4b-e).
Despite the large pressures generated during the explosions, the pneu-nets (fabricated from a stiff silicone rubber, Young's modulus ~ 3.6 kPa; SI) withstood multiple (>30) explosive actuations before failure. These failures typically occurred from the charring of the gas input lines and, occasionally, from tearing of the elastomers at the interface between the actuation layer and the strain-limiting layer.
The toughness and resilience of these silicone elastomers was further evident when we actuated all three legs simultaneously. The tripedal robot contained the three simultaneous explosions and used the energy they generated to jump over 30 times its body height (that is, 30 cm) in under 0.2 s ( Fig. 5; Video S3). We used high-speed video to estimate the instantaneous velocity after actuation: the robot jumped 2.5 cm in 8.25 ms, with a resultant velocity of 3.6 m/s (13 km/h). The 30 cm height was, in reality, limited by the height of the safety chamber we used to enclose the jumping robot; we estimate that the robot would actually have reached a height of 60 cm in a taller chamber, and without the weight of the attached tubing (SI).
The use of explosions for actuation is compatible with soft machines. Explosive power allowed a soft robot to jump 30 times its height with an initial speed of 3.6 m/s; a mobile robot powered by compressed air moved much more slowly (walking at ~0.03 m/s). [5] The siliconebased robot, whose body design we made no attempt to optimize, withstood the tensile forces and temperatures generated by igniting a mixture of methane and oxygen within its pneu-nets.
The heat capacity of the robot (~44 J/K; SI) was enough to absorb the heat generated from the rapidly burning gas (~18 J).
Recently, the use of jumping in hard robotic systems (e.g., the "Sandflea" by Boston Dynamics) has been demonstrated as a way to navigate obstacles. We believe that soft robots powered by explosive actuation, with future improvements in design and control, [25] could be autonomous and able to use their ability to jump to navigate obstacles in search and rescue missions; additionally, the cost of these robots (~$100; see SI for estimate) would be sufficiently low that they could be considered disposable, with insignificant loss if they were destroyed during use.
The soft robot described in this work can be further developed to convert chemical potential into useful mechanical work (see SI for efficiency calculation). By tailoring the timing in the sparks, it will be possible to increase the jumping height, improve energy efficiency, and direct the jump of the robot. Liquid butane (LB) and other liquid fuels (e.g., gasoline) have even greater volumetric energy densities than gaseous methane, and will be usable as fuels with improved design.

Estimate of Pressure Immediately After Ignition.
Using the unpressurized channel volume of 1.0 mL, and our theoretically estimated temperature immediately after ignition of 2,800 o C, we calculated the maximum pressure, using the ideal gas law, to be ~1 MPa (~140 psi). Though this value is an overestimate (it does not take into account the channel volume expansion and the gas cooling, a complex calculation that is beyond the scope of this paper), we use it to illustrate the quick impulse of high pressure to use a passive valving system to control the flow of gas out of the channel after combustion and cooling. pneu-nets composed of DS-10 could withstand the large forces generated within the channels during the explosion better than Ecoflex; in addition, DS-10 has a high resilience, which allowed the pneu-nets to release stored elastic energy rapidly for propulsion (Fig. S1). To seal the pneunet, we bonded a compliant and relatively inextensible silicone rubber (Sylgard 184; Dow Corning) to the actuation layer using a thin layer of uncured silicone (Sylgard 184) and then allowed the silicone to cure at room temperature over 12 hours.
We mixed the methane and oxygen gases off-board the robot and injected the mixture, separately, into each leg, at a rate of 12 mL/min. In order to assure that we delivered a stoichiometric mixture of CH 4 and O 2 , we used mass-flow controllers (100SCCM; MKS Instruments). We used capacitive discharge modules (CDIs), available from the hobby radiocontrolled airplane industry (part# RCEXL; Paragon RC, Inc.), to generate the large potentials (~6.6 kV at 2 mm electrode separation, or ~33 kV/cm, 10 times the approximate breakdown voltage of these gases [1]) to produce the sparks to ignite the gas mixture (Fig. S3). We threaded a single ground wire through all three pneu-nets of the tripod and we threaded the positive electrodes-coaxially-through each of the (three) gas delivery tubes (Figure 1b). We used an Arduino control board to trigger the CDIs to generate the spark between the desired positive electrode(s) and the common ground wire. We used this value to estimate the efficiency of the system; efficiency = (mechanical energy out / chemical energy in)*100%, which we evaluated as ~0.7%.

Measurement of the Heat Evolution During Explosive Actuation Using Nanocalorimetry.
The nanocalorimeter we used (1 nanoWatt sensitivity) measures the heat flow in units of (J/s) as a function of time, and the integral of this curve is the heat (q) evolved or absorbed [2]. Due to the volume constraint of the cylindrical cell of the calorimeter (the cylinder was 1 cm in diameter by 5 cm long cylinder) we used a smaller actuator volume (125 L) than the one we used for the jumping robot (1.5 mL).
To measure the heat evolved by actuation with compressed air, we assume that the calorimeter is adiabatic, and thus the q we measure is equivalent to the mechanical work, -w, done by the pneu-net including frictional losses. We determined -w by injecting 750 L of air via a syringe pump (Harvard Apparatus), the pressure within the pneu-net increased by ~1 psi and we detected a q = -3.3 mJ (Fig. 3a). After actuation, the pneu-net then slowly leaked air into the larger volume calorimeter cell (via diffusion through the porous silicone [3]) and we measured the heat absorbed (from the expansion of gas and coiling of the polymer chains) during the de-actuation to be q = 2.5 mJ (Fig. 3a). There is thus an 18% loss in converting the potential energy of the compressed gas into mechanical work in the actuator.
In the second experiment, by threading electrical wire into the calorimeter cell, we were able to trigger the combustion of premixed methane/oxygen gas inside the small pneu-net. We filled the volume of the pneu-net with stoichiometric methane/oxygen and triggered an explosion. The heat evolved during the combustion of the gas was q = 350 mJ (Fig. 3b).
The actuation time for compressed air to drive the 125 L pneu-net is ~ 1 second [4]; the resulting power supplied to the pneu-net is 3.3mJ/1s = 3.3 mW. The time required for an explosion to actuate the small pneu-net is ~10 ms (Fig. 2a- We used a pressure of 1 atm, an initial temperature of 298.15 K, fuel species of CH 4 , an air molar ratio of 0, and an equivalence ratio of 1, and the computed adiabatic flame temperature was 3,052 K. This value represents a lower limit of temperature because our process is not isobaric.

Estimate for Average Change in Temperature of the Robot Itself.
The specific heat of PDMS is 1.46e3 J/(kg K) and the robot is composed of 30 g of silicone, and thus has a net, equilibrated heat capacity of ~45 J/K. The average change in temperature, for the actuation of all its legs is then the difference of the heat generated from the methane combustion (Q; Matlab script below) and the mechanical energy from the jump (U; Matlab script below) divided by the heat capacity.
The average change in temperature is thus ~0.41 K for simultaneous actuation of all three legs.    Video S1. This video demonstrates that the mini explosions are small enough for the robots to be handled by experienced personnel only.