Pneumatic Neurons Make a Soft Robot Crawl Without a Central Controller

A rhythm generated inside the machine

Soft robots can bend around obstacles and tolerate contacts that would challenge rigid mechanisms, but they still need a way to coordinate motion. That normally requires valves, sensors, clocks or electronic controllers outside the deformable body. A preprint submitted on September 10 describes another approach: modules that use heat, vapor pressure and mechanical switches to pass activation around a robot in a repeating sequence.

The researchers call each module a Pneu-ron, short for pneumatic neuron. Three connected modules produced a traveling inflation wave without a central digital controller or feedback sensor. When the team attached simple legs, the same sequence became a slow crawling gait. The 180-millimeter robot moved 25 millimeters over six cycles during a 95-minute laboratory experiment.

That motion is proof of a control principle, not a practical transport system. Each cycle took approximately 15 minutes. The crawler depended on an external electrical supply and wires between modules, while temperature changes could stop oscillation or lock the system into a static state. Its value lies in showing that sensing, timing and actuation can be partly embodied in the physical response of a soft structure.

How an inflate-and-fire module works

Each Pneu-ron is a sealed thermoplastic polyurethane pouch divided into two chambers. It contains a low-boiling-point fluid, a 30-centimeter silicone-coated electrical heater and a two-part buckling-beam switch. Supplying current warms the fluid until some of it vaporizes. Rising internal pressure inflates the pouch, producing lateral expansion and linear contraction while mechanically changing the switch contacts.

The switch provides two pressure thresholds. At the lower inhibitory threshold, an inflating module can disconnect the heater in the preceding module. At the higher excitatory threshold, it closes a connection that activates its successor. A self-latching path keeps the active module heating long enough for the transition to complete. Excitation moves the wave forward, inhibition turns the previous module off, and self-latching prevents an incomplete handoff.

This arrangement resembles the functional pattern of biological neurons, but it does not reproduce their chemistry or information-processing complexity. Pressure replaces membrane voltage, inflation replaces firing and wires carry electrical power between heaters. The demonstrated network still needs an external direct-current source. It is therefore more accurate to describe the design as distributed physical control than as electronics-free robotics.

A single module driven at 9 volts generated a reported maximum contraction force of about 0.5 newtons across a four-millimeter stroke. Its lateral expansion produced about four newtons. Thermal response was slow: the authors measured settling times of roughly 300 seconds at 9 volts and 600 seconds at 4.5 volts. Those figures explain both the sustained rhythm and the crawler's limited speed.

A ring passes activation from segment to segment

The researchers wired three modules into a ring and observed a traveling pattern in which two remained hot while the next transition occurred. A simplified mathematical model treated heating and cooling like charging and discharging in an electrical resistance-capacitance circuit. The model mapped a region between two failures: extinction, where a module cannot excite its neighbor, and saturation, where modules remain continuously active.

The physical response is more complicated than that first-order model. The pouches and buckling switches exhibit hysteresis, meaning their state depends partly on whether pressure is rising or falling. The model instead used fixed thresholds. It is useful for explaining the operating window, but it is not a complete representation of heat transfer, phase change, switch mechanics or material deformation.

A six-module ring demonstrated limited physical reconfiguration. It took about 14 minutes to complete one cycle at 9 volts. The researchers then separated it into two three-module rings, preserving the needed connections. One new ring required a brief manual compression of a module to restart, after which both oscillated for another 13 minutes. This is a controlled proof that the same modules can form smaller working loops, not evidence that arbitrary damage will automatically heal.

The network also continued oscillating when two 500-gram weights were placed on selected modules. Each load was about 20.8 times the 24-gram module body weight. The added force changed rather than erased the rhythm: reported rise time fell from about 25 seconds to 15 seconds, while the period increased approximately 34 percent, from 632 to 848 seconds. Physical resilience here means continued operation with altered timing, not unchanged performance.

Turning oscillation into crawling

For locomotion, the team fitted each of three modules with a 3D-printed L-shaped leg. Gripping tape formed a directional claw, while a flexible polyester hinge allowed the foot to move differently across phases of inflation and contraction. The resulting sequence converted the traveling pressure wave into forward displacement on a flat acrylic surface.

Powered at 9 volts, the robot completed six full cycles over 95 minutes and advanced 25 millimeters after a subsequent cooling period used before measurement. That corresponds to 0.0231 body lengths per cycle. The gait needed no separate digital trajectory generator because the event that deformed one segment also triggered the next.

This coupling is the experiment's most useful design idea. In many robots, a controller calculates timing and then sends commands to actuators. Here, the actuator's changing pressure participates directly in the timing logic. Such integration could eventually reduce the number of separate components needed in machines designed for constrained or difficult-to-access environments.

The crawler does not yet demonstrate those applications. Its external power and wiring would constrain mobility, and a 15-minute cycle is far too slow for most inspection or transport work. The experiment used one short robot on a controlled surface, without obstacles, steering, payload transport or an untethered energy source. Suggestions involving underground, underwater or space exploration remain future possibilities.

Temperature and deadlock expose the boundaries

The same thermal mechanism that enables activation creates pronounced environmental limits. At minus 20 degrees Celsius, the traveling wave extinguished within seconds because the modules could not reach the excitatory pressure threshold. Reducing the supply from 9 to 7.5 volts allowed approximately one additional cycle before oscillation decayed and stopped.

At the other extreme, placing the system on a 60-degree Celsius hotplate produced a saturated static arch instead of a traveling wave. That temperature exceeded the working fluid's reported 34-degree boiling point, so the authors treated the test as a qualitative failure case outside the model's quantitative range. Ambient conditions above the boiling point can activate modules without the intended sequence.

A separate experiment showed how self-latching can turn a disturbance into deadlock. Clamping one module in a four-segment ring generated two waves that collided, leaving two modules on and two off. Removing the clamp did not restart motion. The researchers used forced-air cooling to push one active module below its inhibitory threshold, break the latch and restore a single traveling wave.

That recovery required outside intervention. It demonstrates that the network has a physically understandable reset path, but not that it autonomously detects and repairs every fault. A deployable machine would need to manage temperature, voltage, fluid containment and stalled states without relying on a researcher holding an air source nearby.

An intriguing architecture with a materials problem

Materials set another boundary. Thermoplastic polyurethane was chosen because it could be sealed around the working fluid, but its heat sensitivity limits heater temperature and therefore actuation speed. The researchers report that earlier silicone versions suffered fluid diffusion, swelling and leakage that required replenishment. Faster operation will require a better balance among heating power, displacement, sealing and long-term reliability.

The paper is a first-version preprint based on author-run experiments, with no independent replication, endurance study or field deployment. It does not establish how many cycles the pouches, switches and seals can survive. Nor does it show that larger networks will preserve stable timing as manufacturing differences and heat exchange accumulate.

Even within those limits, the work advances a practical line of inquiry. It demonstrates that a soft actuator can also serve as a threshold sensor and a timing element, allowing a network of simple modules to organize rhythmic motion. The crawler is slow, tethered and thermally fragile, but its motion makes the broader proposition testable: some robotic control can be designed into matter instead of assigned entirely to a central computer.