Bioinspired design of a tissue-engineered ray with machine learning
John F. Zimmerman, Daniel J. Drennan, James Ikeda, Qianru Jin, Herdeline Ann M. Ardoña, Sean L. Kim, Ryoma Ishii, Kevin Kit Parker
Abstract
In biomimetic design, researchers recreate existing biological structures to form functional devices. For biohybrid robotic swimmers assembled with tissue engineering, this is problematic because most devices operate at different length scales than their naturally occurring counterparts, resulting in reduced performance. To overcome these challenges, here, we demonstrate how machine learning-directed optimization (ML-DO) can be used to inform the design of a biohybrid robot, outperforming other nonlinear optimization techniques, such as Bayesian optimization, in the selection of high-performance geometries. We show how this approach can be used to maximize the thrust generated by a tissue-engineered mobuliform miniray. This results in devices that can swim at the millimeter scale while more closely preserving natural locomotive scaling laws. Overall, this work provides a quantitatively rigorous approach for the engineering design of muscular structure-function relationships in an automated fashion.