Abstract Biologists and engineers often attempt to develop biologically accurate neuromechanical models, but improving these models is challenging due to the high model dimensionality. To overcome this challenge, we present the Reinforcement-Learning-enabled Neuromechanical Model Analysis (RL-NMA) pipeline for targeted, demand-driven-complexity-based model improvements in neurorobotics and neuromechanics. This pipeline is agnostic to the model and system analyzed, allowing it to be broadly applied. We present two case studies of RL-NMA pipeline application. First, we assess a digital twin of a soft robot inspired by the feeding mechanism of the marine mollusk, Aplysia californica . Second, we perform iterative improvement of a computational neuromechanical model of Aplysia feeding to capture in vivo behavior. Third, we assess a different digital twin of a bioinspired soft grasper. Based on the pipeline’s recommendation, targeted model improvements led to improved correlations. These case studies demonstrate iterative application of the RL-NMA pipeline in both neurorobotics and neuromechanics, allowing researchers to achieve demand-driven model improvements in high-dimensional models.
Fernandez et al. (Tue,) studied this question.