Multiscale Data-driven Models for Soft Robotic Actuators Based on Local Strain Fields

It is often easier to manufacture and experimentally characterize a soft robotic actuator than it is to construct a theoretical or finite element model to understand its mechanics. In many cases, soft actuators with reinforcement materials need to be homogenized to ascertain models for modeling and control, or a high fidelity model is needed, which is often computationally intractable to solve in real time. The modeling of soft robotic actuators should consider the micro-and mesoscale mechanics that impact the macroscale actuator response. Building on recent developments in both data-driven control and full-field measurement techniques of local strain deformations, we propose using Digital Image Correlation (DIC) to ascertain observables to relate the length scales in soft robotic actuators for rapid dynamic modeling. Using flexible structures with rotating squares as a demonstrating example, we fabricate tubular actuators and characterize their motions using DIC. Nonlinear models relating the length scales are constructed from observables ascertained via DIC using Sparse Identification of Nonlinear Dynamical Systems (SINDy). We demonstrate the goodness-of-fit of several models and discuss limitations in extending these models for control. This work provides a general recipe for the data-driven identification of dynamic models for soft robotic actuators from observables at multiple length scales.