Research · Building functional objects

Soft pneumatic actuators

Print-in-place soft actuators with embedded skeletons, and a differentiable simulator for designing actuators by gradient-based optimization.

The key idea. A good soft actuator has to be compliant enough to bend and strong enough to push. Multi-material printing lets one print combine soft and rigid parts, and simulation with gradients lets the computer work out which design best balances the two for a given task.

A translucent printed soft actuator with a row of ten air chambers on a white base, connected to blue tubing, on a black background
A print-in-place pneumatic actuator, inkjet-printed in soft and rigid materials. Pressurizing its chambers makes it bend.

A print-in-place actuator with a skeleton

With inkjet printing, a pneumatic actuator can be printed in one piece with an embedded watch-band endoskeleton. The skeleton resists bending off to the side while leaving the actuator compliant along the axis it should bend. The print combines soft, rigid, and breakable support materials so the mechanism prints in place, and it is compiled from a single parametric OpenVCAD script into roughly 40 million voxels of material assignments. The design is by Julian Hansell, an undergraduate student I mentor.

The underside of the printed actuator, showing a white segmented endoskeleton inside a translucent body, with blue tubing
The underside, showing the white watch-band endoskeleton printed inside the soft body.

A differentiable simulator for actuator design

Finding a compliant yet forceful actuator for each set of requirements means simulating many candidate designs. Standard simulators such as Abaqus report how well a design performs, but not its gradient: which way, and by how much, to turn each design parameter to improve it. Without that, the search has to guess. I am developing a differentiable shell-element simulator in JAX that returns these gradients for about the cost of one more simulation, so an optimizer can follow them directly. On the lab’s actuator designs it agrees with established solvers (FEBio and Abaqus).

  1. Design parametersThe knobs that shape the actuator
  2. Shell simulationInflate the actuator in a differentiable JAX solver
  3. PerformanceHow far it bends and how hard it pushes

The gradient tells the optimizer how to adjust every parameter at once.

Six renders of a finger-shaped actuator with five chambers, straight at first and curling further at each pressure, shaded from dark purple at the base to yellow at the tip
A simulated actuator inflating. A PneuNet-style actuator at 0, 20, 70, 100, 150, and 200 kPa, colored by how far each point moves. The gray outline is the actuator at rest.
Three renders of a ribbed tube-shaped actuator, gray at rest, curved when bending freely, and pressed flat against a pink plate
Bending and pushing. One simulated design at rest, bending freely, and pressing against a plate, where the contact force measures how hard it can push.