SoRoForge: End-to-end soft actuator design

Abstract

Soft robots offer advantages over traditional rigid robots in terms of safe interaction and resilience to unexpected challenges. A significant limitation of soft robots is that they are difficult to design and fabricate, and most designs are created manually through trial and error. This article presents a novel approach to the automated design of pneumatic soft actuators. Our approach leverages genetic algorithms, novel generative encodings, simulation-based analysis via the finite-element method (FEM), and automated fabrication via multi-material additive manufacturing. We demonstrate the capabilities of this approach on a series of design tasks, ranging from matching the output of manually designed actuators to the much more challenging task of creating novel actuators with user-specified performance characteristics. We present actuators designed by our approach that demonstrate both high-force and high-displacement output, showing both the ability to match hand-designed actuators and to create novel designs that exceed the original performance specifications. Further, we demonstrate the ability to design actuators that produce complex 3-D trajectories, a design task that would be extremely challenging to solve using manual techniques. We also validate the fabricability of our automatically generated designs via 3-D printing and testing, with close correlation between simulated and measured performance.

BibTeX

@article{smith2023soroforge,
  title = {SoRoForge: End-to-end soft actuator design},
  author = {Smith, Lawrence and MacCurdy, Robert},
  journal = {IEEE Transactions on Automation Science and Engineering},
  volume = {20},
  number = {3},
  pages = {1475--1486},
  year = {2023},
  publisher = {IEEE},
  doi = {10.1109/TASE.2022.3217016},
}