Soft pneumatic actuators: fabrication, characterization and a robot that crawls

This project covers the full path of a soft actuator, from liquid silicone to a robot that moves. A fabrication process was documented, the deformation-pressure curve of ten actuators was measured, and a three-actuator robot was built whose gait was found by automatically trying combinations with a camera. The most useful finding is a limitation: the actuators stop working after 40 or 50 cycles.

Context

The actuator is a silicone chamber with one inextensible face. When pressurized, the free face stretches and the other does not, so the part arches. Here the inextensible layer is a sheet of bond paper impregnated with the same silicone.

Fabrication

The material is Ecoflex 00-30. The actuator base is cast in a 3D-printed mold, vacuum degassed and oven cured.

The paper sheet is then dipped in silicone, the chamber is closed with it, the hose is inserted and the part is cured again. The thesis describes the defects found and their cause: poorly impregnated paper, leaks at the corners, and a bump when the hose does not seat properly.

Characterization

Each actuator was inflated while a sensor recorded pressure and a camera tracked a marker at its highest point.

Maximum deformation of the ten actuators was between 541% and 571%. The silicone admits about 900% according to the manufacturer, but the actuator fails earlier, above 340 mbar, at the corners where the sheet joins the base. The limit is set by the joint, not the material.

Because manufacturing is manual, each actuator behaves differently and has to be characterized individually.

A three-actuator robot

Three actuators were joined into a T-shaped body.

The pneumatic system has pumps, a tank, three solenoid valves and an Arduino with relays.

To find how to activate the actuators, the six possible permutations were tried with several delays between valves. In each trial a camera measured the advance after 30 cycles using markers on the robot.

Results

• The best sequence was 1-2-3, with 0.5 s between intermediate valves and 0.6 s between the end and the start of the cycle. In that sequence at least two actuators are always partially inflated.
• Advance per trial was between 3 and 7 cm, and the same sequence did not always give the same value.
• In one minute, the robot advanced 4.3 cm on glass, 7.2 cm on wood and 7.6 cm on paper.
• Manual control does not produce continuous advance, because the timing required is too fine.

What is missing

Service life is the main problem. After 40 to 50 cycles the paper sheet deforms permanently, stops buckling, and the actuator loses its motion. Three of the five robots built were damaged during testing, and that deterioration explains part of the variation in advance. The gait search was done offline and there is no feedback during locomotion.

How it fits in Robiolab

Together with the peristaltic robot design, this is one of the two projects with which the lab started building soft robots, and it was presented at a mechanical engineering conference in 2015. It established the use of Ecoflex, printed molds and individual characterization of each actuator, practices that recur in later projects. The idea of letting the robot find its own gait by trial, measured with a camera, returns in the learning-based control of the peristaltic robot.