Biomimetic Robotic Hand Controlled via Deep Reinforcement Learning with Digital Twin

Lowell Welburn, Mechanical Engineering, Engineering, North Carolina Agricultural and Technical State University

Description

Robotic arms are frequently deployed in many manufacturing architectures, such as automotive, aerospace, semiconductor, and pharmaceutical industries. Most of these arms are limited to 6 or less degrees of freedom, restricting the fine control needed to manipulate delicate parts adequately. The purpose of this work is to present a physical biomimetic hand with a digital twin that are both trained and controlled via deep reinforcement learning. The expectation of this research is a hand that can perform a delicate operation such as picking up a bolt and screwing it into a threaded hole without the use of another tool. This hand could be implemented in any manufacturing industry where high dexterity is needed.