Active Learning Design: Modeling Force Output for Axisymmetric Soft Pneumatic Actuators
Gregory Campbell, Gentian Muhaxheri, Leonardo Ferreira Guilhoto, Christian Santangelo, Paris Perdikaris, James Pikul, Mark Yim
AI summary
Problem
Characterizing the force response of soft pneumatic actuators under external loads is laborious and computationally expensive, while existing theoretical models struggle with numerical stiffness and data-driven methods underperform in force prediction.
Approach
The authors combine energy minimization theory with an active learning framework to efficiently collect experimental data, then train a constrained neural operator model to predict force output across a parameterized design space and optimize membrane designs for specific lift trajectories.
Key results
- Developed a constrained neural operator model that accurately predicts force-pressure-height relationships for axisymmetric soft actuators.
- Demonstrated an active learning pipeline that efficiently explores the design space with minimal experimental trials.
- Achieved superior prediction accuracy compared to both theoretical energy-based models and naive regression baselines.
- Successfully optimized membrane designs to maximize lift height and execute targeted mass-lifting trajectories.
Why it matters
Enables rapid, data-efficient design and control of soft pneumatic actuators for precise physical interaction and object manipulation tasks.
Abstract
Soft pneumatic actuators (SPA) made from elas- tomeric materials can provide large strain and large force. The behavior of locally strain-restricted hyperelastic materials under inflation has been investigated thoroughly for shape re- configuration, but requires further investigation for trajectories involving external force. In this work we model force-pressure- height relationships for a concentrically strain-limited class of soft pneumatic actuators and demonstrate the use of this model to design SPA response for object lifting. We predict relation- ships under different loadings by solving energy minimization equations and verify this theory by using an automated test rig to collect rich data for n=22 Ecoflex 00-30 membranes. We collect data using an active learning pipeline to efficiently model the design space. We show that this learned model outperforms the theory-based model and a naive regression. We use our model to optimize membrane design for different lift tasks and compare this performance to other designs. These contributions represent a step towards understanding the natural response for this class of actuator and embodying intelligent lifts in a single- pressure input actuator system.