A generative design pipeline for soft pneumatic actuators
Designing a soft actuator is usually a slow cycle of drawing, making a mold, casting silicone and testing. This project proposes replacing that cycle with a computational one: describe the actuator with parameters, automatically generate its geometry and mesh, simulate it and compare variants. The thesis develops the pipeline up to the simulation-ready volumetric mesh, and illustrates it with the finger of a pneumatic gripper.
Context
The advantage of soft robots, their ability to deform, is also what makes them hard to design. There are no closed-form formulas relating chamber geometry to the resulting curvature, and each change requires a finite element simulation with hyperelastic materials. If that simulation can be launched without manual intervention, many geometries can be explored and selected among.
The create, parameterize, simulate cycle
The process is organized in three steps repeated for each candidate: create the geometry, parameterize it, and simulate it in a defined environment.
Generative design process: from the parameterized robot to simulation data
From a robot idea with several parameters, a set of robots is generated, one for each combination of values.
Generation of a set of robots from parameter combinations
Case study: a gripper finger
The finger has two parts, the body and the internal cavity that is pressurized. Three parameters were chosen to vary: wall thickness, number of cavities and spacing between them. The remaining dimensions are tied to those three through constraints, so that any combination produces a valid geometry.
Finger cross-section with its parameters
Chained tools
- OpenSCAD generates the body and the cavity from code and exports them as surfaces.
Finger body generated from code
- Blender, driven by Python, remeshes the surface, which OpenSCAD delivers with very few points, and converts it to triangles. Two parameters control how fine it is.
Triangular surface mesh of the body and the cavity
- Gmsh builds the tetrahedral volume mesh between the two surfaces.
Generating the tetrahedral volume from the cavity and the body
- SOFA simulates the deformation with hyperelastic models as the cavity is pressurized.
The thesis points out the central trade-off of meshing. A finer mesh gives a more faithful simulation, but computing cost grows quickly. A mesh that is too coarse can lose geometric details, such as a cavity.
What is missing
The available version of the document goes as far as generating the volume mesh. It does not include simulation results, performance metrics or the comparison between variants, which is what would turn the pipeline into a complete generative design process. There is no validation with a fabricated actuator either. This post therefore describes infrastructure, not a design result.
How it fits in Robiolab
The project responds to a need that shows up in all of the lab's soft-actuator work, from the first silicone actuators to the soft tourniquet: each design iteration costs a mold and several days. The same idea, with a different toolset, was carried through to experimental validation in the parametric design of coral grippers.