A Genetic Deep Learning Model for Electrophysiological Soft Robotics

Abstract

Deep learning methods are modeled by means of multiple layers of predefined set of operations. These days, deep learning techniques utilizing unsupervised learning for training neural networks layers have shown effective results in various fields. Genetic algorithms, by contrast, are search and optimization algorithm that mimic evolutionary process. Previous scientific literatures reveal that genetic algorithms have been successfully implemented for training three-layer neural networks. In this paper, we propose a novel genetic approach to evolving deep learning networks. The performance of the proposed method is evaluated in the context of an electrophysiological soft robot like system, the results of which demonstrate that our proposed hybrid system is capable of effectively training a deep learning network.

A structural optimization analysis of cable-driven soft manipulator

“…[28][29][30] Specifically, genetic algorithms (GAs) have been leveraged to optimize inverse kinematics of concentric tube manipulators, 30 tune control parameters of flexible joint robots, 31 and boost the accuracy of motion prediction from electromyography data. 32 Swarm intelligence methods have also been explored to optimize the parameters of central pattern generators used in controlling bio-robotic fish. 33 customized GAs with specialized crossover and mutation schemes have demonstrated enhanced convergence for flexible manipulator optimization.…”